Fractions Affected and Probabilistic Risk Assessment of Cu, Zn, Cd

ACS2GO © 2018. ← → → ←. loading. To add this web app to the home screen open the browser option menu and tap on Add to homescreen...
0 downloads 0 Views 206KB Size
Download PDF
Environ. Sci. Technol. 2005, 39, 8533-8540

Fractions Affected and Probabilistic Risk Assessment of Cu, Zn, Cd, and Pb in Soils Using the Free Ion Approach S T E P H E N L O F T S , * ,† DAVID SPURGEON,‡ AND CLAUS SVENDSEN‡ NERC Centre for Ecology and Hydrology, Bailrigg, Lancaster LA1 4AP, United Kingdom, and NERC Centre for Ecology and Hydrology, Monks Wood, Huntingdon, Cambridgeshire PE28 2LS, United Kingdom

The free ion approach quantifies the toxic effects of cationic metals on soil biota as a function of chemistry. The approach is here extended to calculate the general relationship among toxic effects as the Fraction Affected (FA), soil solution pH, and soil organic matter content (SOM) for Cu, Zn, Cd, and Pb within toxicity data sets from literature. The dependence of FA on SOM is strong, with the FA decreasing as the SOM increases. The dependence of FA upon pH varies; Cd and Zn show strong dependences while for Cu and Pb dependences are weaker. The FA usually decreases as the soil pH increases. When the free ion approach is applied, risks across soils due to different metals can be compared using the FA. The free ion approach may also be applied to probabilistic risk assessment. Risk values, using the joint probability curve approach, were derived for Pb using two field soil data sets. One data set, with higher SOM than that of the Pb toxicity data set, gave a lower risk with the free ion approach than that when the soil chemistry was not considered. The other data set had lower SOM than that of the toxicity data set and gave a higher risk with the free ion approach. Since literature toxicity tests are biased toward low SOM soils of circumneutral pH, using such data to perform classical risk assessment for soils of differing chemical composition can lead to misestimation of risk due to neglecting soil chemistry, especially in soils with extreme pH and/or SOM.

Introduction It is well-known that the toxic impacts of cationic metals on soil organisms vary according to soil chemistry. The variability has been shown under laboratory conditions for organisms such as earthworms (1), springtails (2), and crop plants such as oats (3). The issue of accounting for this variability in metal risk assessment is a current topic of research in terrestrial ecotoxicology, drawing strongly upon parallel work on the chemistry and toxicology of metals in freshwaters, particularly on the Biotic Ligand Model (BLM) (4). Some authors (5, 6) have proposed empirical expressions for the variability in metal toxicity to single species, while others (2) * Corresponding author phone: +44-1524-595800; fax: +44-152461536; e-mail: [email protected]. † Centre for Ecology and Hydrology. ‡ Centre for Ecology and Hydrology. 10.1021/es048121c CCC: $30.25 Published on Web 09/24/2005

 2005 American Chemical Society

have proposed relationships between toxicity, metal body burden, and the free metal ion, suggesting that metal toxicity to terrestrial organisms may be amenable to description by the BLM. Recently, Lofts and co-workers (7) have proposed a method, based on the free metal ion and use of species sensitivity distributions (SSDs) to consider multispecies data, to derive functions for quality standards (critical limits) for copper, zinc, cadmium, and lead based on soil metal concentration, soil solution pH, and soil organic matter (SOM) content. When used to derive risk-based environmental quality standards, the SSD approach is used to derive a pollutant concentration not exceeding a given proportion of endpoint concentrations in toxicity tests (typically the proportion is set to 0.05). As a more general measure of risk, the SSD can be used to calculate the proportion of endpoint concentrations that are exceeded by a given field concentration of the pollutant. This use of SSDs for the calculation of the Fraction Affected (FA) (also termed the Potentially Affected Fraction, PAF) is useful for two reasons. First, it scales pollutant concentrations to a consistent index of risk (the proportion of endpoints exceeded), thus allowing a ranking of risks due to multiple pollutants in the same soil. Second, FAs for multiple pollutants in the same soil may be combined, assuming additivity of toxic effects, to give an overall FA for a contaminated site. The SSD also finds use in probabilistic ecological risk assessment (RA) at regional scales (8, 9, 10). Here the SSD is compared with the distribution of field concentrations of the pollutant to derive the risk of a randomly chosen field concentration exceeding a randomly chosen toxic endpoint. The resulting risk value may be compared with set levels defining risk categories. As the approach of Lofts et al. (7) uses the SSD concept for consideration of multispecies toxicity data, it may in principle be extended to allow the calculation of FAs for cationic metals and for use in probabilistic risk assessment. Here we show how this may be done, using both an empirical bootstrapping method to derive the SSD, as was done by Lofts et al. (7), and a parametric method. We will demonstrate the use of the free ion approach in probabilistic risk assessment by applying it to two example data sets, and we will show how the approach corrects for biases in the chemical properties of the toxicity and field data used, by comparing the approach with the “classical: approach of considering the soil metal concentration to be the relevant predictor of toxic effect.

Theory Free Ion Approach to Metal Toxicity. The basis of the free ion approach is the concept that variability in the toxic effect of a cationic metal, among soils of differing chemical composition, can be ascribed to two effects: the variation in the concentration of the free metal ion in the soil solution and the variations in the concentrations of the free forms of other soil solution cations (e.g., H+, Na+, Mg2+, and Ca2+), which “protect” the organism against the toxic effect of the free metal ion. Mathematically, the theory is expressed as

log Mfree,toxic ) R pHss + Σ{βi log[Ci]} + γ

(1)

where Mfree (mol dm-3) is the concentration of the free metal ion exerting a given level of toxic effect and the subscript “toxic” refers to a concentration exerting a given level of toxic effect, pHss is the soil solution pH, [Ci] is the concentration (mol dm-3) of a “protecting” cation, and R, βi, and γ are empirical constants. By assuming that other free “proVOL. 39, NO. 21, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8533

tecting” cations co-vary with H+ across soils, the expression may be simplified to

log Mfree,toxic ) R pHss + γ

(2)

in which case the pHss term integrates all of the protective effects. (Note that for the data calibrating data set values of R in eqs 1 and 2 would differ.) In soil toxicity tests, metal concentrations are conventionally expressed either as the metal added to the soil or as a concentration of metal extracted from the soil, as the amount of metal per unit mass of soil. Application of the free ion approach requires that the relationship between the soil metal and the free ion be known. For literature toxicity tests, where the detailed information required for speciation modeling of the free ion is usually absent, empirical regressions relating the two metal forms may be used. Lofts et al. (7) used regressions of the following form

log Mfree ) a pHss + b log(SOM) + c log Msoil,reactive + d (3) where Mfree is the free metal ion concentration (mol/dm3), SOM is the soil organic matter content (%), Msoil,reactive is the reactive pool of soil metal (mol/g dry soil), and a through d are empirical coefficients, derived from fitting the above expression to independent data sets of field soil pHss, SOM, Msoil,reactive, and Mfree. The reactive metal pool is that considered to be able to exchange with the soil solution and is typically measured in field soils by extraction with a dilute mineral acid (e.g., ref 11) or a chelating agent (e.g., 12). When this expression for toxicity data is used, it is assumed that where a toxic endpoint is given as added metal this approximates to the reactive pool. Combining eqs 2 and 3 gives

best consistency the endpoint metal should be expressed as the additional metal above the control concentration. On application to field conditions, consideration of the natural background metal (in reactive form) should be made. For a conservative approach, the background can be considered to be zero, otherwise background concentrations should be subtracted from the measured reactive concentrations before analysis. All references to field concentrations in this paper refer to background-corrected concentrations. Toxicity Data. Criteria for collation of toxicity data for Cu, Zn, Cd, and Pb were presented by Lofts et al. (7). Briefly, chronic endpoints (no observed effect concentrations (NOEC) and 10% effect concentrations (EC10)) for growth and population effects on plants, soil-dwelling invertebrates, and microbial processes were used. For this work, the toxicity databases used were harmonized with databases used for European Union risk assessment procedures (13). Quoted soil pH values, obtained with soil extractions using water or dilute salts, were converted to pHss, if required, using the expressions given in ref 14; these expressions are provided in the Supporting Information. As was done by Lofts et al. (7), endpoints for common effects on the same plant or invertebrate were weighted according to the number of points, ni, in each data group of common endpoints

wi ) 1/ni

(6)

where φ and ψ are new coefficients. In application to multispecies toxicity data, a value of Q for each endpoint concentration is derived and these are regressed against pHss to obtain φ and ψ. The scatter of points about the regression line is then assumed to be due to the distribution of species sensitivity. Quantification of the cumulative probability distribution of residuals R in Q then gives the expression

Weighting was not done for microbial process data since different components of the microbial community are regarded as responsible for the same process under varying soil conditions (15). Soil Reactive Metal-Free Ion Relationships. Soil reactive metal-free ion relationships (eq 3) for Cu, Zn, Cd, and Pb have been described previously by Lofts et al. (7) and are given in the Supporting Information. Parametric Calculation of the FA. For parametric calculation of the FA, we first apply the toxicity regression equation (eq 4) by weighted least-squares regression to obtain φ and ψ. We then assume that the regression residuals R are normally distributed with weighted standard deviation σR (the residual mean being zero by default). We must then find δp such that the field metal concentration for which the FA is to be calculated equals Msoil,reactive(p) in eq 5. To do this, we calculate δobs, which is effectively the value of R associated with the field metal concentration, if it were a toxicity endpoint

Qp ) log Msoil,reactive,toxic(p) + (b/c)log(SOM) ) φ pHss + ψ + δp (5)

δobs ) log Msoil,reactive,field + (b/c)log(SOMfield) (φ pHss,field + ψ ) (7)

Here the term p refers to the cumulative probability in the distribution of R and δp is the corresponding value of the distribution. Since δp is a function of p, this is effectively a general expression for p as a function of Msoil,reactive, pHss, and SOM. It is the basis for the critical limit functions derived by Lofts et al. (7), where δp for a fixed p of 0.05 was found for each metal studied. Since the analysis of the residuals is equivalent to the analysis of toxic endpoints by the SSD approach, p for a given Msoil,reactive is the Fraction Affected and may be found by finding δp such that the field metal concentration equals Msoil,reactive,toxic(p). Analysis of the residuals distribution may be done by assuming a priori a given distribution type or by considering the distribution empirically and taking percentiles of residuals.

where the subscript “field” refers to field values. The p value for δobs must then be found. This is done based on the theory of extrapolation factors for defining hazardous concentrations (HCs) from a normally distributed set of toxic endpoints, as described by Wagner and Løkke (16) and Aldenberg and Jaworska (17), and is described in detail in the Supporting Information. Bootstrapped Calculation of the FA. The basis of the bootstrapping method is that the variability in a statistical property of a data set, for example, its mean, will be mimicked by the variability in the same property of a large number of samples of that data set (18). A bootstrap distribution of the property in question is derived by repeated resampling of the data set. Bootstrapped calculation of FAs further develops the method used by Lofts et al. (7) to calculate critical limit functions. Bootstrapping has the advantage of not a priori assuming a statistical distribution of regression residuals, thus avoiding potential bias in the results, and consideration of uncertainty in the input parameters can readily be incorporated. In the calculation of an FA by bootstrapping,

log Msoil,reactive,toxic + (b/c)log(SOM) ) φ pHss + ψ ) Q (4)

Application Consideration of Natural Background Metal. Endpoints in toxicity tests may be expressed as a total or an added metal concentration. Since the effect on the organism is always considered relative to its performance in the control soil, for 8534

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 21, 2005

uncertainty in the parameters b and c (eq 5) and in the calculated value of pHss may be incorporated. Essentially, for a soil with a given pHss and SOM, bootstrap distributions of log Msoil,reactive,toxic(p), for a defined range of p, are calculated by repeated parametrization of eq 5 with samples of the original toxicity data set. A percentile of each distribution is taken, based on the desired confidence level of the FA, to give an empirical distribution of log Msoil,reactive,toxic(p) against p. The FA is then found by linear interpolation of p values. The method is detailed in the Supporting Information. Calculations were done using a purpose-written program in Visual Basic, version 6.0 (Microsoft, Redmond, WA). Application to Probabilistic Risk Assessment. Probabilistic risk assessment is intended for use where an SSD is available for the pollutant of concern and where the field data set to be assessed has sufficient temporal and/or spatial variability to be assessed statistically. The methodology has been presented and discussed in detail previously (e.g., ref 10). Essentially, the method derives as a measure of risk the probability that a value chosen randomly from the distribution of field concentrations CENV will exceed a value chosen randomly from the distribution of endpoint concentrations (the SSD). This probability is equal to the area under the curve (AUC) of a joint probability curve (JPC), a plot of 1 p for the distribution of field concentrations (the exceedance exposure distribution, EED) against p for the SSD (the FA). From ref 10, the probability may be formally calculated from the distribution parameters, assuming the distributions to be normal

P(CENV > SSD) ) Φ0,1[(µ{log CENV} µ{log SSD})/x(σ{log CENV}2 + σ{log SSD}2)] (8) where Φs,t[r] is the cumulative probability of a normal distribution with mean s and standard deviation t at a value of r and µ{X} and σ{X} are the mean and standard deviation of the distribution X. Incorporating the parametric free ion approach into probabilistic RA is done by substituting the distribution of R for the distribution of log SSD. The distribution of log CENV is replaced by the distribution of δobs, calculated with eq 7. The expression for P(CENV > SSD) then becomes

P(CENV > SSD) ) Φ0,1[(µ{δobs})/x(σ{δobs}2 + σR2)] (9) P(CENV > SSD) may also be calculated by bootstrapping. Discrete pairs of values were sampled from the observations, and δobs and R were calculated; the risk value is the proportion of pairs where δobs > R. Initial values showed that a large number of pairs were required to give a consistent reproducible risk value; a sample size of 1 million was used here. When the toxicity data sets were sampled, it was necessary to account for the differential weightings of individual points. This was done by first sampling a data group of common endpoints, then sampling a single endpoint from the group. To demonstrate these approaches, they were applied to two data sets of field Pb concentrations (reactive soil metal). The first data set comprised 80 measurements in upland soils of the U. K. (11), and the second comprised 84 measurements in lead-contaminated soils of Canada and the U.S.A. (19). The original U. K. data set, as used in ref 11, contained both soils chosen as representative of remote sites and soils from contaminated areas. To reduce bias in the data, 17 points representing specifically chosen contaminated sites were removed from the original data set. It should be emphasized, however, that the data sets as used here are not necessarily free of bias in terms of their spatial extent or soil properties. Therefore they are not necessarily representative of suitable databases for probabilistic RA and are used here

TABLE 1. Parameters for Parametric and Bootstrapped Regressions of Q for Toxicity Data against pHss parametric

bootstrapping

metal

O

ψ

σR

O

ψ

copper zinc cadmium lead

0.07 0.26 0.37 0.06

-6.62 -8.13 -9.62 -6.37

0.59 0.60 0.66 0.59

-0.01 0.18 0.34 0.09

-6.16 -7.59 -9.48 -6.55

purely for illustrative purposes. For simplicity, we have also assumed the background Pb to be zero in all of the soils. To compare the risk predicted from the free ion approach with the classical approach taking the soil metal as the index of toxicity, risk values (AUCs) were calculated using eq 8 with distributions of CENV and SSD as the field soil reactive Pb and the added toxic Pb, respectively. The toxicity data set was identical to that used for the free ion approach, and the mean and standard deviations were calculated using the same weights used for the free ion approach. Therefore, the comparison of risk values is entirely one of the two methodologies.

Results Toxicity Data. A summary of the toxicity data used and the chemical properties of the test soils are given in the Supporting Information. Toxicity Regression Parameters. Values of φ, ψ, and σR are given in Table 1. Values of σR are similar for all four metals, implying that the range of sensitivities of soil biota are similar, although this may be due to selection of similar species and processes in the literature toxicity tests. The pH term φ varies among the metals, being relatively small for Cu and Pb. A positive value of φ indicates that for a given SOM and p, Msoil,reactive(p) increases as pHss increases, and vice versa. This is so in all cases except for the bootstrap-calculated φ for Cu; however, this value does not differ significantly from zero and is somewhat smaller than the corresponding parametrically derived value. Statistical Analysis of Toxicity Endpoints. To check for bias introduced by combining toxicity test results for invertebrates, plants, and microbial processes and the use of two endpoint types, the toxicity endpoints (expressed as residuals in Q after performing the toxicity regression above) were analyzed for significant differences between types of endpoint and broad species/process type. Analysis was done by an analysis of variation (ANOVA) and ad hoc Tukey testing using MINITAB, release 14 (Minitab Inc., State College, PA). For Zn, Cd, and Pb, no significant differences across broad group or endpoint types were found (p ) 0.086, 0.0851, and 0.0387, respectively). For Cu, invertebrate EC10 values were significantly lower than plant NOECs and microbial process EC10 values (p < 0.05), and invertebrate NOECs were significantly lower than microbial process EC10 values (p < 0.05). Thus, while there is no apparent bias in the toxicity data for Zn, Cd, and Pb due to aggregation of different endpoints and broad species groups, this does not appear to be case for Cu. A large proportion of the invertebrate endpoints for Cu (seven of eight EC10 values and four of nine NOEC values) were from tests on Oligochaete worms (earthworms and enchytraeids). A number of species from these taxa have been shown to be sensitive to Cu, and this has been given as the reason for the reduced abundance and diversity of earthworms in Cu-treated vineyards (20, 21). This sensitivity is likely to be the reason for the observed bias. Some caution should therefore be applied to the results for copper until such a time as more toxicity data tests on Cu toxicity to a wider range of invertebrates, particularly those with haemocyanic blood, becomes available. VOL. 39, NO. 21, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8535

FIGURE 1. Variation of fraction affected with soil reactive metal and pH (left-hand panes) or SOM (right-hand panes): symbols, bootstrap-calculated FAs; lines, parametrically calculated FAs. Left-hand panes: pH 4 and 1% OM (closed circles); pH 5 and 1% OM (open circles); pH 6 and 1% OM (closed squares); pH 7 and 1% OM (open squares). Right-hand panes: pH 5 and 1% OM (closed circles); pH 5 and 3.2% OM (open circles); pH 5 and 10% OM (closed squares); pH 5 and 32% OM (open squares); pH 5 and 100% OM (closed triangles). Fraction Affected. Figure 1 shows predicted median FAs (i.e., at a confidence level of 50%) for the four metals under varying pHss and SOM, for reactive soil metal from 1 to 10 000 mg metal per kg soil. Parametric and bootstrapped FAs are generally very similar, with the exception of Zn at pH 4 and pH 5 where the parametrically calculated FAs are typically higher. The predicted FAs are dependent upon pHss and SOM. The dependence upon pHss is a function of φ (Table 1). The parametrically calculated φ increases in the order Pb < Cu < Zn < Cd, while for the bootstrap-calculated values the order is Cu < Pb < Zn < Cd. These differences in φ can be seen in the trends in FA. For example, the reactive soil metal for which the parametric FA ) 0.5 at pHss ) 4 and SOM ) 8536

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 21, 2005

1 is 29 for Cu, 5.3 for Zn, 0.8 for Cd, and 154 for Pb (all in mg/kg soil). For these reactive soil metal concentrations, the FAs for the metals at pHss ) 7 and SOM ) 1 are 0.35, 0.11, 0.043, and 0.38, respectively. The dependence of SOM upon the FA is consistently negative with the FA decreasing as SOM increases. For example, considering the FAs listed above at pHss ) 7 and SOM ) 1, for the same reactive metal concentrations, the FAs at pHss ) 7 and SOM ) 100 are all less than 0.01. Figure 2 shows uncertainties in the calculated FAs, expressed as the FAs with 95% confidence (upper error limits) and with 5% confidence (lower error limits). The uncertainties are typically similar, although for Zn the uncertainties calculated by the bootstrap method are notably larger. The

FIGURE 2. Uncertainty of fraction affected for Cu, Zn, Cd, and Pb. Left-hand panes: pHss ) 4 and SOM ) 1. Right-hand panes: pHss ) 7 and SOM ) 1. Symbols are bootstrap-calculated FAs. The error bar limits are point predictions of the FA with 95% confidence (upper) and the FA with 5% confidence (lower). Solid lines, parametrically calculated predictions of FA with 50% confidence; dashed lines, parametrically calculated predictions of FA with 95% confidence (upper lines) and 5% confidence (lower lines). magnitude of the uncertainties varies with the magnitude of the FA, with smaller uncertainties seen for low and high FAs; an example is shown in Figure 3. The parametrically calculated uncertainties display regular relationships with the FA, while the bootstrap-calculated uncertainties show an irregular relationship to FA, and for this example are highly asymmetrical and typically larger than the uncertainties calculated parametrically. The somewhat irregular nature of the bootstrap uncertainties is because the actual distribution of residuals in Q close to the calculated FA greatly affects the magnitude of the uncertainty and because the bootstrap method incorporates uncertainties in the values of pHss and Q that are not considered in the parametric approach. However, since the parametric method assumes a regular statistical distri-

bution of residuals in Q, the uncertainties in the FA are correspondingly regular. Additional plots of uncertainties in the FA for all of the metals are given in the Supporting Information. Probabilistic Risk Assessment. Ranges and medians of pHss, SOM, and Pbsoil,reactive in the field data sets are given in Table 2. Although the ranges of pHss in both data sets are similar to the range seen in the Pb toxicity data set, the U. K. data have a lower median pHss than either of the other two data sets. The U. K. data also tend to higher SOM than either of the other data sets, while the North American data tend to a slightly lower SOM than the toxicity data. The North American data have a similar range of Pb as the toxicity data, albeit with a somewhat lower median, while the U. K. data have lower median and maximum Pb. VOL. 39, NO. 21, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8537

FIGURE 3. Uncertainties in the fraction affected, for zinc in a soil of pHss ) 7 and SOM ) 1. The x-axis is the calculated FA at 50% confidence (FA50). The y-axis (∆FA) represents the absolute difference between the calculated FA at 50% confidence and the calculated FA at either 95% or 5% confidence (FA95 or FA5). The filled circles and the solid line represent the difference between FA95 and FA50, by bootstrapping and parametric methods, respectively. The open circles and dashed line represent the difference between FA5 and FA50, by bootstrapping and parametric methods, respectively.

TABLE 2. Ranges and Medians (in brackets) of Chemical Parameters for the Pb Toxicity Data and the Two Field Datasets determinand

toxicity data

U. K. data

North American data

pHss 3.69-7.88 (5.99) 3.35-8.28 (4.37) 3.49-8.14 (7.34) SOM (%) 1.0-80 (6.2) 8.9-98 (41) 0.45-11 (2.5) Pbsoil,reactive 10-16571 (765) 11-9660 (83) 5.5-14860 (323) (mg/kg)

Cumulative probability plots of δobs (field data) and R (toxicity data) are shown in Figure 4. Of note is the high mean δobs of -0.20 for the North American data, which is close to the mean R of zero. The mean δobs of -1.44 for the U. K. data is over an order of magnitude lower. The standard deviations of δobs are 0.64 and 0.46 for the North American and U. K. data, respectively, and are reasonably similar to the value of 0.59 for R. The joint probability curves show the effects of accounting for soil chemistry, via the free ion approach, on the predicted risk when calculated parametrically. The predicted pH dependence of toxicity for Pb is low so the effect of the relatively low pH of the U. K. data, which would be expected to increase risk, is more than balanced by the counter effect of the high SOM in these data, so the predicted risk is overall lower than when calculated with the classical approach. When the parametric calculation method is used, risk values are 0.027 using the free ion approach and 0.19 using the classical approach. When the bootstrapping method is used, the risk values are 0.036 and 0.13, respectively. The calculated risk for the North American data are slightly higher when calculated with the free ion approach, due to the SOM in these data being slightly lower than that in the toxicity data. Risk values calculated using the parametric method are 0.41 and 0.37, respectively, for the free ion and classical approaches. When the bootstrapping method is used, the values are 0.41 and 0.33.

Discussion The underlying principle of the free ion approach is to provide a more scientifically robust approach to ecological risks of cationic metals than the classical approach. Lofts et al. (7) have already demonstrated the application of the approach to the calculation of quality standards (critical limits) for soils. Here we have shown the extension of the method to (i) the calculation of soil-dependent Fractions Affected and (ii) probabilistic risk assessment. The approach centers on deriving a best estimate expression for the dependence of 8538

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 21, 2005

toxicity on soil pH and organic matter (i.e., the dependence of toxicity on chemistry across the multispecies endpoint dataset) and then applying the species sensitivity distribution concept to describe the spread of individual species toxicity data around the average. We therefore first consider the effects of soil chemistry on the multispecies toxicity data set, before considering the intrinsic variability in individual species sensitivities. The curves describing the variation in FA with metal concentration for a given soil composition (Figure 1) are effectively the SSD normalized to that composition. Calculated FAs show clear dependences upon pH and SOM, with the FA increasing as the pH or SOM increase. Dependences for SOM are dependent entirely upon the calculated relationship between the soil reactive metal and the free ion; no dependence of SOM on free ion toxicity is assumed. The dependence on pH, however, is a combination of the dependence of pH on the reactive metal-free ion relationship (eq 3) and the overall “protective” effects of cations (eq 2). These two effects counterbalance each other, with the “protective” effect acting to ameliorate the increased concentration of toxic free ions at low pH, although in most cases the overall effect of increasing pH is to increase the FA somewhat. Cu and Pb show low overall effects of pH on FA, while Zn and Cd show higher effects. We have presented two methods for calculating the FA using the free ion approach. The parametric approach presented here is suitable where the residuals in the toxicity regression can be shown to follow a normal distribution; corresponding methods could be derived for other statistical distributions. The bootstrapping method, however, has the advantage of not assuming a priori a statistical distribution, but where the actual distribution approaches one that may be described statistically, it gives results similar to those obtained with a parametric method. The disadvantage of the bootstrap method is that it is difficult to extrapolate beyond the minimum and maximum endpoints of the toxicity data set. This is not a major issue in this case, since the data sets have reasonably large numbers of points, but it must be borne in mind when considering a methodology for small data sets. The main value of the Fraction Affected is not currently as an absolute indicator of harm to an ecosystem, although research into the links between the FA and harm is taking place (e.g., ref 22). The main value of the FA is as a method of ranking multiple risks, as explained in the Introduction. However, under the free ion approach, where the FA is a function of a number of soil chemistry variables, it is also necessary to use the FA to properly compare risks due to a single pollutant across different soils. This also holds for any other approach to risk incorporating variability in soils, whether chemical or otherwise. We have also demonstrated the use of the free ion approach in probabilistic risk assessment. Using two field data sets, we have shown how differences in predicted risks calculated with the free ion and classical approaches are dependent upon the range of soil chemistries found in the field and toxicity data sets. Since literature toxicity tests are biased toward low SOM soils of circumneutral pH, using such data to perform classical risk assessment for soils of differing chemical composition can lead to misestimation of risk. Since the free ion approach predicts that cationic metals become less toxic as soil pH and/or SOM increase, misestimation of risk under the classical approach is likely to be greatest where soils exhibit extremes of pH or relatively high SOM. The free ion approach is able to deal with this variability in soil chemistry and thus provides more scientifically robust and realistic estimates of risks. It is important to appreciate that the use in practice of risk measures such as the Fraction Affected, which are derived

FIGURE 4. Probabilistic risk assessment of two data sets of Pb soil concentrations: U. K. upland soils (top panes) and North American soils (lower panes). Left-hand panes: probability distribution of δobs values (open symbols) and the distribution of residuals in Q for toxicity data. Right-hand panes: joint probability curves (JPCs) for the parametric approach. Solid line, JPC derived from the free ion approach; dashed line, JPC derived from the classical approach. from a consideration of laboratory toxicity data, ought to be constrained by the limitations of such data. We accept that there are valid concerns regarding the NOEC concept as a valid toxicity endpoint given its dependence upon study design, the number of replicates used, and possibly other factors such as the intrinsic variability of organism response across replicates (e.g., refs 22 and 23). However, in many cases NOECs are the only form of endpoint available from literature toxicity studies. While some literature studies present the primary data, thus in principle allowing an ECx endpoint to be calculated, these are few compared to the number of studies quoting NOECs. Risk assessors have to make the best possible use of the best available data, even considering known shortcomings of such data, rather than make no attempt at all to assess risks. The SSD approach rests upon the assumption that the toxicity data used are a representative random sample of species and microbial processes occurring in the system under study. Bias may be introduced by an emphasis upon certain test species, for example, those that are easily cultivatable or are perceived to be of commercial interest (e.g., crop plants). Here we have shown that in the cases of Zn, Cd, and Pb there is no statistically identifiable bias in sensitivity across endpoint types and a broad organism group. For Cu, we have identified a bias in the sensitivity of selected test endpoints for invertebrates, plants, and microbes. This can be attributed to the high representation of Oligochaete worms in the invertebrate data set and the relative sensitivity to Cu of species within this taxa. As a result, the results for this metal should be interpreted with caution. The most suitable way to eliminate such bias would be the generation of additional toxicity data covering a wider range of invertebrate taxa than is currently available. In any case, the free ion approach focuses upon accounting for the effects of soil chemistry on metal toxicity and in this respect is a step forward from soil risk assessment methods that do not account for chemistry, given similar toxicity data as a starting point. These potential difficulties notwithstanding, it is important to appreciate that field risk evaluation using the SSD approach is not intended to be a substitute for detailed site-specific ecotoxicological

evaluations, rather it is intended as means of identifying sites or regions of potential risk prior to more detailed studies or alternatively in large-scale risk evaluations such as critical loads (e.g., ref 25) where detailed study is not feasible due to the scale of the evaluation required.

Acknowledgments We thank Ed Tipping, Mike Ashmore, Laura Shotbolt, Jane Hall, Joseph Fawehinmi, Michiel Rutgers, Natalia Naumova, and Tatiana Pampura for helpful discussions. This work was funded by the U. K. Department of the Environment, Food and Rural Affairs, the Scottish Executive, the National Assembly of Wales, the Department of the Environment (in Northern Ireland) (Contract EPG/1/3/188), and INTAS (Project No. 01poll-2213).

Supporting Information Available Summary of the toxicity data, conversion of soil pH to pHss, soil reactive metal-free ion relationships, parametric calculation of FAs, calculation of FAs by bootstrapping, and calculated uncertainties in the Fraction Affected. This material is available free of charge via the Internet at http:// pubs.acs.org.

Literature Cited (1) Spurgeon, D. J.; Hopkin, S. P. Effects of variations of the organic matter content and pH of soils on the availability and toxicity of zinc to the earthworm Eisenia fetida. Pedobiologia 1996, 40, 80-96. (2) Van Gestel, C. A. M.; Koolhaas J. E. Water-extractability, free ion activity, and pH explain cadmium sorption and toxicity to Folsomia candida (Collembola) in seven soil-pH combinations. Environ. Toxicol. Chem. 2004, 23, 1822-1833. (3) Weng, L. P.; Wolthoorn, A.; Lexmond, T. M.; Temminghoff, E. J. M.; Van Riemsdijk, W. H. Understanding the effects of soil characteristics on phytotoxicity and bioavailability of nickel using speciation models. Environ. Sci. Technol. 2004, 38, 156162. (4) Di Toro, D. M.; Allen, H. E.; Bergman, H. L.; Meyer, J. S.; Paquin, P. R.; Santore, R. C. Biotic ligand model of the acute toxicity of metals. 1. Technical basis. Environ. Toxicol. Chem. 2001, 20, 2383-2396. VOL. 39, NO. 21, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8539

(5) Lock, K.; Janssen, C. R.; De Coen, W. M. Multivariate test designs to assess the influence of zinc and cadmium bioavailability in soils on the toxicity to Enchytraeus albidus. Environ. Toxicol. Chem. 2000, 19, 2666-2671. (6) Lock, K.; Janssen, C. R. Cadmium toxicity for terrestrial invertebrates: Taking soil parameters affecting bioavailability into account. Ecotoxicology 2001, 10, 315-322. (7) Lofts, S.; Spurgeon, D. J.; Svendsen, C.; Tipping, E. Deriving soil critical limits for Cu, Zn, Cd, and Pb: A method based on free ion concentrations. Environ. Sci. Technol. 2004, 38, 3623-3631. (8) Klepper, O.; van de Meent, D. Mapping the Potentially Affected Fraction (PAF) of Species as an Indicator of Toxic Stress; Technical Report 67504001; RIVM: Bilthoven, The Netherlands, 1997. (9) Klepper, O.; Bakker, J.; Traas, T. P.; van de Meent, D. Mapping the Potentially Affected Fraction (PAF) of species as a basis for comparison of ecotoxicological risks between substances and regions. J. Hazard. Mater. 1998, 61, 337-344. (10) Solomon, K. R.; Takacs, P. Probabilistic risk assessment using species sensitivity distributions. In Species Sensitivity Distributions in Ecotoxicology; Posthuma, L., Suter, G. W., II, Traas, T. P., Eds., Lewis Publishers: Boca Raton, FL, 2000; pp 345-382. (11) Tipping, E.; Rieuwerts, J.; Pan, G.; Ashmore, M. R.; Lofts, S.; Hill, M. T. R.; Farago, M. E.; Thornton, I. The solid-solution partitioning of heavy metals (Cu, Zn, Cd, Pb) in upland soils of England and Wales. Environ. Pollut. 2003, 125, 213-225. (12) During, R. A.; Hoss, T.; Gath, S. Sorption and bioavailability of heavy metals in long-term differently tilled soils amended with organic wastes. Sci. Total Environ. 2003, 313, 227-234. (13) European Commission. Technical Guidance Document in Support of Commission Directive 93/67/EEC on Risk Assessment for New Notified Substances and Commission Regulation (EC) No. 1488/94 on Risk Assessment for Existing Substances, Part II; Office for Official Publications of the European Communities: Luxembourg, 1996. (14) Manual on Methodologies and Criteria for Modelling and Mapping Critical Loads and Levels and Air Pollution Effects, Risks and Trends; ICP Modelling and Mapping: Laxenburg, Austria, 2004. http://www.icpmapping.org/html/body_manual.html. (15) Crommentuijn, T.; Polder, M. D.; van de Plassche, E. J. Maximum Permissible Concentrations and Negligible Concentrations for Metals, Taking Background Concentrations into Account; Report

8540

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 21, 2005

(16) (17) (18) (19) (20) (21)

(22)

(23) (24) (25)

No. 601501001; National Institute of Public Health and the Environment (RIVM): Bilthoven, The Netherlands, 1997. http:// www.rivm.nl/bibliotheek/rapporten/601501001.html. Wagner, C.; Løkke, H. Estimation of ecotoxicological protection levels from NOEC toxicity data. Water Res. 1991, 25, 12371242. Aldenberg, T.; Jaworska, J. S. Uncertainty of the Hazardous Concentration and Fraction Affected for normal species sensitivity distributions. Ecotoxicol. Environ. Saf. 2000, 46, 1-18. Chernick, M. R. Bootstrap Methods: A Practitioner’s Guide; Wiley: New York, 1999. Sauve´, S.; McBride, M. B.; Hendershot, W. H. Speciation of Pb in contaminated soils. Environ. Pollut. 1997, 98, 149-155. Streit, B. Effects of high copper concentrations on soil invertebrates (earthworms and oribatid mites)sExperimental results and a model. Oecologia 1984, 64, 381-388. Paoletti, M. G.; Sommaggio, D.; Favretto, M. R.; Petruzzelli, G.; Pezzarossa, B.; Barbafieri, M. Earthworms as useful bioindicators of agroecosystem sustainability in orchards and vineyards with different inputs. Appl. Soil Ecol. 1998, 10, 137150. Van Beelen, P.; Fleuren-Kemila, A. K.; Aldenberg, T. The relation between extrapolated risk, expressed as Potentially Affected Fraction, and community effects, expressed as pollutioninduced community tolerance. Environ. Toxicol. Chem. 2001, 20, 1133-1140. Laskowski, R. Some good reasons to ban the use of NOEC, LOEC and related concepts in ecotoxicology. Oikos 1995, 73, 140144. Koojiman, S. A. L. M. An alternative to the NOEC exists, but the standard model has to be abandoned first. Oikos 1996, 75, 310316. Hall, J. R.; Ashmore, M.; Fawehinmi, J.; Jordan, C.; Lofts, S.; Shotbolt, L.; Spurgeon, D. J.; Svendsen, C.; Tipping, E. Developing a critical load approach for national risk assessments of atmospheric metal deposition. Environ. Toxicol. Chem., in press.

Received for review November 29, 2004. Revised manuscript received August 23, 2005. Accepted August 25, 2005. ES048121C