EPA/600/R-20/042
ERASC-020F
Summary Report
June 2020
SUMMARY REPORT
TERRESTRIAL METALS BIOAVAILABILITY: A LITERATURE-DERIVED
CLASSIFICATION PROCEDURE FOR ECOLOGICAL RISK ASSESSMENT
Ecological Risk Assessment Support Center
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
NOTICE
Mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
Preferred Citation:
U.S. EPA (U.S. Environmental Protection Agency). 2020. Summary Report. Terrestrial Metals
Bioavailability: A Literature-Derived Classification Procedure for Ecological Risk Assessment. Center
for Environmental Solutions and Emergency Response, Ecological Risk Assessment Support Center,
Cincinnati, OH. EPA/600/R-20/042.
11
-------
TABLE OF CONTENTS
Page
LIST OF TABLES iv
LIST OF FIGURES v
LIST 01 ABBREVIATIONS vi
AUTHORS, CONTRIBUTORS, AND REVIEWERS viii
ACKNOWLEDGMENTS x
PREFACE xi
EXECUTIVE SUMMARY xi
INTRODUCTION 1
METHODS 4
DATA SETS 4
DATA MANAGEMENT AND ASSUMPTIONS 6
APPORTIONMENT OF NUISANCE VARIATION 6
CLASSIFICATION PROCEDURE DEVELOPMENT AND VALIDATION 7
RESULTS AND DISCUSSION 8
APPORTIONMENT OF NUISANCE VARIATION 8
Bioaccumulation 8
Toxicity 9
CLASSIFICATION PROCEDURE FOR BIOAVAILABILITY 11
APPLICATION TO ECOLOGICAL RISK ASSESSMENT 12
REFERENCES 15
in
-------
LIST OF TABLES
No. Title Page
1. Summary of Data Used in Meta-analyses 25
2. Variable Selection Results from Apportionment Analysis for Both
Bioaccumulation and Toxicity Data Sets 28
3. Proposed Adjustments to Total Soil Concentrations That Account for Metals
Bioavailability Based on the Central Tendencies of the Terminal Nodes (i.e.,
Bioavailability Categories) from Figure 9 29
iv
-------
LIST OF FIGURES
No. Title Page
1. Mean (± 1 SE) Predicted Natural Log (nl) BAF Values by Dose, as Stratified by
Metal, from the Most Parsimonious Bioaccumulation Model Apportioning
Nuisance Variation 30
2. Mean (± 1 SE) Predicted Natural Log (nl) BAF Values by Endpoint (confounded
by receptor) from the Most Parsimonious Bioaccumulation Model Apportioning
Nuisance Variation 31
3. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Metal from the
Most Parsimonious Toxicity Model Apportioning Nuisance Variation 32
4. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Endpoint from the
Most Parsimonious Toxicity Model Apportioning Nuisance Variation 33
5. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Parameter from the
Most Parsimonious Toxicity Model Apportioning Nuisance Variation 34
6. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Receptor from the
Most Parsimonious Toxicity Model Apportioning Nuisance Variation 35
7. Relationships Between Model Residuals and Soil Organic Matter (SOM), Clay
Content, and pH for the Most Parsimonious Toxicity and Bioaccumulation Linear
Models Apportioning Nuisance Variation 36
8. Internal (Bioaccumulation Data Set) and External (Toxicity Data Set) Validation
Results of Cross-Validation Analyses Used to Determine the Optimal-Sized
Regression Tree 37
9. Regression Tree Results Constrained to Four Terminal Nodes 38
v
-------
AIC
AICc
BAF
BCF
C
CaCh
CART
Cd
CEC
CLAY
Co
DOSE
ECx
Eco-SSL
ENDPOINT
EPC
ERA
ERAF
KC1
LCx
LOEL
LOO
MAXDEPTH
METAL
Ni
nl
NOEL
PARAMETER
Pb
RECEPTOR
LIST OF ABBREVIATIONS
Akaike Information Criterion
Akaike Information Criterion with small-sample bias correction
bioaccumulation factor
bioconcentration factor
carbon
calcium chloride
classification and regression trees
cadmium
cation exchange capacity
variable that refers to soil particle sizes < 2 [j,m
cobalt
nuisance variable pertaining to the concentration the metal was either dosed or
measured in field-contaminated soils
effective concentration for X% of the population
Ecological Soil Screening Level
nuisance variable that specifies the measurement attribute of the organism
Exposure Point Concentration
ecological risk assessment
Ecological Risk Assessment Forum
potassium chloride
lethal concentration for X% of the population
lowest-observed-effect level
leave-one-out
variable that refers to the complexity of the regression tree. See text
nuisance variable that specifies the particular metal that was evaluated
nickel
natural log
no-ob served-effect level
nuisance variable that specifies the reported toxicity benchmark
lead
nuisance variable that specifies among plant, invertebrates, or microbes
vi
-------
LIST OF ABBREVIATIONS (continued)
RT
regression tree
SE
standard error
SOM
soil organic matter
SPECIES
nuisance variable that specifies the particular species that was evaluated
SSD
species sensitivity distribution
STUDY
nuisance variable that specifies study to be treated as a random effect
TYPE
nuisance spiked versus field-contaminated soils
Zn
zinc
vii
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS
(Reviewers of this Summary Report are marked with an asterisk)
AUTHORS
Richard H. Anderson1
U.S. Environmental Protection Agency
Office of Research and Development
National Center for Environmental Assessment
Cincinnati, OH 45268
David B. Farrar
U.S. Environmental Protection Agency
Office of Research and Development
Center for Public Health and Environmental Assessment
Cincinnati, OH 45268
Michael Kravitz
U.S. Environmental Protection Agency
Office of Research and Development
Center for Environmental Solutions and Emergency Response
Cincinnati, OH 45268
CONTRIBUTORS
Jeanmarie M. Zodrow2
U.S. Environmental Protection Agency
Office of Environmental Assessment
Region 10
Seattle, WA98101
REVIEWERS
Kirk Scheckel
U.S. Environmental Protection Agency
Office of Research and Development
National Risk Management Research Laboratory
Cincinnati, OH 45268
1 Present Address: Air Force Civil Engineer Center (AFCEC), Technical Support Branch, Lackland AFB, TX.
2 Present Address: Arcadia U.S., Inc., Lakewood, CO
viii
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
Matthew Etterson
U.S. Environmental Protection Agency
Office of Research and Development
National Health and Environmental Effects Research Laboratory
Duluth, MN 55804
Michael Griffith*
U.S. Environmental Protection Agency
Office of Research and Development
National Center for Environmental Assessment
Cincinnati, OH 45268
Tim Frederick*
U.S. Environmental Protection Agency
Region 4 Superfund Division
Atlanta, GA 30303
IX
-------
ACKNOWLEDGMENTS
This research was supported in part by an appointment to the Research Participation
Program at the U.S. Environmental Protection Agency (EPA) National Center for Environmental
Assessment (NCEA) administered by the Oak Ridge Institute for Science and Education
(ORISE) through an interagency agreement between the U.S. Department of Energy and EPA.
x
-------
PREFACE
A request was submitted by the Ecological Risk Assessment Forum (ERAF) to the Office
of Research and Development's Ecological Risk Assessment Support Center (ERASC) relating
to the issue of terrestrial metals bioavailability. The ERAF specifically requested a product that
characterizes typical aerobic soils in terms of their potential to mitigate metals bioavailability to
soil-dwelling biota in response to increased pressure from the academic community to explicitly
incorporate bioavailability concepts into the ecological risk assessment (ERA) process. An
exhaustive literature search and corresponding meta-analysis of the empirical data was
recommended and performed. The result is a quantitative tool that broadly accounts for metals
bioavailability and is proposed to augment the ERA process and risk-based remediation of
metals-contaminated soils. The tool, or classification procedure, presented here is suggested to
be used with other analyses such as direct toxicity testing of contaminated soils. The present
document summarizes the derivation and potential use of this tool as described in the ERASC
draft response and a peer-reviewed article (Anderson et al., 2013).
EXECUTIVE SUMMARY
Interstudy variation among bioavailability studies is a primary deterrent to a universal
methodology to assess metals bioavailability to soil-dwelling organisms and is largely the result
of specific experimental conditions unique to independent studies. The primary objective of this
review is to synthesize information in the open literature on the effects of soil chemical/physical
properties on metals bioavailability independent of extraneous variation due to the specific
attributes of individual studies. Accordingly, two data sets were established from relevant
literature; one includes data from studies related to bioaccumulation (total obs = 520), while the
other contains data from studies related to toxicity (total obs = 1,264). Experimental factors that
affect bioavailability independent of the effect of soil chemical/physical properties were
considered nuisance variables, i.e. variables not of direct interest in the context of this study, but
that need to be considered in analyzing the data. Variation associated with significant nuisance
variables was statistically apportioned from the variation attributed to soil chemical/physical
properties for both data sets using a linear mixed model. Residual bioaccumulation data were
XI
-------
then used to develop a nonparametric regression tree whereby bootstrap and cross-validation
techniques were used to internally validate the resulting classification procedure. A similar
approach was employed with the toxicity data set as an independent external validation. These
analyses obviously emphasize bioaccumulation as the primary metric for assessing
bioavailability but demonstrate concurrence with studies on toxicity. The validated classification
procedure is proposed as a quantitative tool that broadly characterizes typical aerobic soils in
terms of their potential to sequester common divalent cationic metal contaminants and mitigate
their bioavailability to soil-dwelling biota. This classification procedure is proposed to augment
other ERA approaches.
Xll
-------
INTRODUCTION
Bioavailability of metal contaminants has been given much attention over the last decade.
In particular, a plethora of research has demonstrated that the uptake and subsequent effects of
toxic trace elements to soil-dwelling organisms are largely regulated by the specific
chemical/physical composition of contaminated soils. Despite these recent advances in assessing
terrestrial metals bioavailability (Scheckel et al., 2009), a disparity exists between the state of the
science and regulatory practice in terms of explicitly incorporating methods to assess
bioavailability into the ecological risk assessment (ERA) framework for metals. Although the
European Union has developed risk-based ecological soil standards using bioavailability models
derived from empirical relationships (Smolders et al., 2009; Semennzin et al., 2007), the United
States has yet to adopt a parallel approach (see U.S. EPA, 2007). In general, a universally
applicable methodology that accounts for metals bioavailability would augment the ERA process
and risk-based remediation of metals-contaminated soils.
Several methods are commonly applied to assess metals bioavailability. Sequential-,
parallel-, and single-chemical extractions characterize metal concentrations among various soil
geochemical phases. Relative advantages and disadvantages of chemical extraction methods are
reviewed elsewhere (Rao et al., 2008; Gleyzes et al., 2002). In general, however, current
extraction methods are considered insufficient to accurately assess metals bioavailability
simultaneously to multiple ecological receptors among heterogeneous soils (Peijnenburg et al.,
2007). Moreover, chemical extractants can modify chemical speciation and soil solution
chemistry, resulting in operationally defined phases with measured metals concentrations that
may not correlate with biological responses (Scheckel et al., 2009; 2003).
Alternatively, mechanistic models are predicated on theoretical thermodynamic and
kinetic principles, and most predict biological effects of metals exposure. Primary mechanistic
models applicable to terrestrial metals bioavailability include the free ion activity model (Hough
et al., 2005; Lofts et al., 2004; Parker and Pedler, 1997), the terrestrial biotic ligand model
(Thakali et al., 2006a; Thakali et al., 2006b; Steenbergen et al., 2005), and soil-water metal
equilibrium partitioning (Degryse et al., 2009). Mechanistic models are essential to a
comprehensive understanding of biological responses to soil contamination and have been
correlated with measurements from field-contaminated soils (Koster et al., 2006; Parker et al.,
1
-------
2001). However, application of mechanistic models can involve complicated input parameters
(often modeled from theoretical functions that assume equilibrium conditions), which could limit
their utility in regulatory practice.
Bioavailability models predicated on empirical data, on the other hand, normalize
experimental bioaccumulation or toxicity estimates from bioassays among contrasting soils (e.g.,
Criel et al., 2008; Rooney et al., 2007; Oorts et al., 2006). Bioassays with soil-dwelling
organisms are implicit indicators of bioavailability because they reflect cumulative mechanistic
soil biogeochemical processes (Basta et al., 2005; Lanno et al., 2004). Empirical models
quantify these processes, as well as random variability, unlike mechanistic models. Individual
empirical models presented throughout the open literature, although applicable to the specific
experimental conditions, may not provide tenable prediction across the entire range in
contaminated soils types. Moreover, preferentially utilizing a particular model in lieu of others
available, all else equal, may be programmatically difficult to justify.
Despite the preponderance of studies demonstrating systematic differences in uptake and
toxicity among contaminated soils, universal regulatory acceptance of a methodology to assess
metals bioavailability to soil-dwelling organisms is impeded in large part by irreproducibility
among studies, leading to the overarching issue of uncertainty (SERDP and ESTCP, 2008). The
extent to which inconsistencies in the literature can be resolved by meta-analysis of the empirical
data merits further investigation, given numerous experimental artifacts among independent
studies (e.g., Lowe and Butt, 2007; Sochova et al., 2006; Clark et al., 2004; Crouau and Cazes,
2003) and variations in standard test methods. Many standard test methods exist for terrestrial
toxicity testing (e.g., Environment Canada, 2005; ISO, 2005; ASTM, 2004; U.S. EPA, 1994).
Selection of test species, range of contaminant doses, measurement endpoint, and numerous
other experimental conditions are highly variable among studies and produce artifacts at the
discretion of the researcher. Moreover, comparison of exposure studies, as related to
bioavailability, is operationally challenged by the same issues. Application of current
bioavailability models, regardless of the metric used to assess bioavailability, is limited to a
specific suite of biotic and abiotic experimental conditions.
One method employed to evaluate results from multiple studies on metals bioavailability
is a modification of the species sensitivity distribution (SSD) approach (Smolders et al., 2009;
Semennzin et al., 2007). In a typical SSD, usually the 5th or 1st percentile of toxicity parameters,
2
-------
which reflect media-specific contaminant concentrations, from the cumulative probability
distribution of a sample of species is obtained (usually from the literature) and is considered
protective of 95 or 99% of species, respectively (Posthuma et al., 2002). In the context of
bioavailability, toxicity parameters used in the SSD can be normalized based on regression
models that include soil chemical/physical properties as predictors (Smolders et al., 2009;
Semennzin et al., 2007). Additional study-specific variation uncontrolled among bioavailability
studies (e.g., metal, endpoint, dose, etc.), however, should be accounted for, as well as
interspecies variation. Moreover, the effects-based SSD approach ignores results of studies that
report bioaccumulation as the metric of bioavailability. A meta-analysis that utilizes data with
both bioaccumulation and toxicity metrics that effectively apportions "nuisance variation" from
the effect of soil chemical/physical properties is paramount to accurately synthesize the literature
on terrestrial metals bioavailability universally applicable to ERAs.
The primary objective of this review is to quantify soil chemical/physical property effects
on terrestrial metals bioavailability by meta-analysis, controlling the confounding influence of
the specific attributes of individual studies independent of soil metal sequestration. In this
context, bioavailability is defined, simplistically, as the fraction of a total metal load (in soil) that
has the potential to traverse a biological membrane. In fact, membrane transport is inherent in
most, if not all, interpretations of bioavailability (Drexler et al., 2003). It follows that body
burden (i.e., bioaccumulation) is the appropriate metric for assessing bioavailability because
trace metals usually achieve steady-state concentrations in most soil-dwelling organisms,
especially essential trace elements (Mleczek et al., 2009; Nahmani et al., 2009; Vijver et al.,
2001; Peijnenburg et al., 2000; Spurgeon and Hopkin, 1999). However, the majority of literature
on the subject utilizes endpoints that reflect toxicity. Consequently, two data sets were
established from relevant literature; one includes information from studies related to
bioaccumulation (total obs = 520), while the other contains studies related to toxicity
(total obs = 1,264). Experimental factors that affect bioavailability independent of the effect of
soil chemical/physical properties were considered nuisance variables. Variation associated with
significant nuisance variables was statistically apportioned from the variation attributed to soil
chemical/physical properties for both data sets using a linear mixed model. Residual
bioaccumulation data were then used to develop a nonparametric regression tree whereby
bootstrap and cross-validation techniques were used to internally validate the resulting
3
-------
classification procedure. A similar approach was employed with the toxicity data set as an
independent external validation. These analyses obviously emphasize bioaccumulation as the
primary metric for assessing bioavailability but simultaneously tests concurrence with studies on
toxicity.
METHODS
DATA SETS
Two independent data sets were compiled from published studies on the effects of abiotic
soil factors on the bioavailability of the common divalent cationic metals cadmium (Cd), cobalt
(Co), copper (Cu), nickel (Ni), lead (Pb), and zinc (Zn) to soil-dwelling organisms available in
the open literature. Only studies that included multiple soil types were considered in order to
limit the scope of the review exclusively to studies that were designed to investigate metals
bioavailability in some capacity. The only exception was that all the data used in the
development of plant and invertebrate Ecological Soil Screening Levels (Eco-SSLs) for Cd, Co,
Cu, Ni, Pb, and Zn (U.S. EPA, 2005) were included since those metals were already being
considered. Applicable studies for additional metals were lacking. Some studies (e.g., Zhao et
al., 2006) were excluded because results were presented in a form such that did not allow
extraction. Most studies that reported internal tissue concentrations also reported
bioaccumulation factor (BAF) values computed as ratios of tissue metal concentration to total
soil concentration determined by vigorous acid digestion. In this context, BAF is used
interchangeably with bioconcentration factor (BCF) values. If BAF values were not reported,
they were computed and used for all analyses, which is comparable to analyses based on tissue
concentrations, given the objectives. For studies employing a dose-response design, reported
BAFs reflect a linear regression slope.
The data sets comprehensively summarize bioaccumulation (i.e., BAF values) and acute
toxicity (i.e., no-observed-effect level [NOEL], lowest-observed-effect level [LOEL], effective
concentration for 10% of the population [ECio], EC20, EC50, and lethal concentration for 50% of
the population [LC50] values) across 122 and 131 contaminated natural soils, respectively.
However, due to co-use of select soils among multiple studies, a total of 189 independent soils
were collated across both data sets. The bioaccumulation data set contains BAF values for
4
-------
11 plants, 2 earthworms, and 1 springtail, while the toxicity data set contains toxicity values for
19 plants, 20 earthworms, 1 springtail, and microbes. In addition to soil characterization
variables, both data sets were developed with generalized variables ubiquitous to all studies;
METAL, SPECIES, ENDPOINT, RECEPTOR (i.e., plant, invertebrate, or microbe), and TYPE
(i.e., spiked vs. field-contaminated soils). The bioaccumulation data set also contains a variable
(DOSE) pertaining to the concentration the metal was either dosed or measured in
field-contaminated soils—by evaluating DOSE on BAF values as opposed to tissue
concentrations, nonlinear effects could be assessed more easily. If the study had a dose-response
design, the value of DOSE that was recorded is the highest dosed concentration. The toxicity
data set exclusively contains dose-response studies and a variable (PARAMETER) that reflects
the reported toxicity benchmark (e.g., NOEL, LOEL, ECio, etc.). All other data were considered
too inconsistent or sparse to consider applicable for coding and analyses.
Soil characterization data were highly inconsistent among studies, yielding only as many
as three matrix properties common to most studies; pH, clay content, and either organic carbon
(C) or soil organic matter (SOM). Consequently, these were the properties selected to evaluate
and were assumed to generally account for the cumulative and inherently interactive processes of
metal sequestration and attenuation (Hamon et al., 2007); although it is well known that metals
have variable affinities for soil geochemical phases (e.g., Saeki and Kunito, 2009), certain
generalizations regarding metals bioavailability can be made. In general, precipitation and solid
phase adsorption are primary mechanisms mitigating metals bioavailability, but the magnitude of
sequestration largely depends on both direct and indirect effects of soil reactivity; formation
constants vary as a function of soil pH among metal-bearing minerals, and ionized organic and
inorganic complexation sites within the soil matrix increase with soil pH due to deprotonation
(Sparks, 2003). Whereas, soil pH influences bioavailability directly by regulating the formation
of additional solid phase sorbents, such as carbonate minerals (Sipos et al., 2009, 2008), and the
chemical form of the metal species (Lofts et al., 2005; Nolan et al., 2003).
Admittedly, omission of other soil matrix properties that have been shown to interact
with metal solubility, thereby influencing bioavailability (e.g., amorphous oxides [e.g., Dayton et
al., 2006; Bradl, 2004; Peijnenburg et al., 1999a] and cation exchange capacity [CEC] [e.g.,
Anderson and Basta, 2009b; Criel et al., 2008; Rooney et al., 2006]) is an oversimplification of
soil metal sequestration. However, soil CEC, when measured at ambient soil pH, is usually
5
-------
correlated with total clay content and SOM because ionized organic functional groups and
aluminosilicate clay edges (both of which contribute to overall soil CEC) are pH dependent (Ge
and Hendershot, 2002; Basta et al., 1993). Also, amorphous oxide data (see McKeague and Day,
1993) were only reported in a small subset of studies. To preserve symmetry among the data
sets, promoting an unbiased quantitative assessment of the available data, only studies that
reported all three selected soil properties were utilized. The ranges in selected soil properties
within the published studies have been compiled for both data sets and are summarized in
Table 1.
DATA MANAGEMENT AND ASSUMPTIONS
Certain assumptions and conversions were necessary to accommodate variations in the
design and reporting among studies. Primary assumptions herein include (1) reported
measurements among all studies reflect steady-state toxicokinetics and (2) exposure occurs
through direct absorption (or adsorption if bioreactive upon contact) or dietary uptake (e.g., soil
invertebrates). All earthworms were depurated, but depuration and exposure times were also
assumed not to affect BAF values. Concentrations, including toxicity values, are on a dry
weight-basis and, if necessary, were converted to mg kg-1. Additionally, if organic C values
were reported, they were doubled to estimate SOM values according to an approximation of the
organic C content of SOM (Sleutel et al., 2007). Then, if necessary, SOM values were converted
to percentages as were clay content values. All pH values reflect either potassium chloride (KC1)
or calcium chloride (CaCh) extracts and were analyzed indiscriminately. No pH measurements
in deionized water were recorded.
APPORTIONMENT OF NUISANCE VARIATION
The term nuisance variable refers to uncontrolled experimental conditions that influence
BAF and/or toxicity measurements independent of the effect of soil metal sequestration. For
example, within a soil type, bioaccumulation is expected to be influenced by toxicity at higher
doses depending on the metal, species, and other factors. The term apportionment refers to the
statistical methodology whereby variation attributed to significant nuisance variables is
quantitatively partitioned from the effect associated with soil chemical/physical properties.
6
-------
Effects of nuisance variables were evaluated statistically using a linear mixed model.
This approach allows a choice in how to represent the effects of a given variable, as either a
random source of variation (e.g., among species) or as a set of fixed effects (e.g., of individual
species. Selection of nuisance variables to be included in the model was based on the Akaike
Information Criterion (AIC) with small-sample bias correction (AICc, Burnham and Anderson,
2002). The criterion can be applied in two ways (see Table 2). The single model with best
(lowest) AICc has been here termed the "most parsimonious model," because of penalization of
fit for the number of variables included in the model. Alternatively, the use of AICc model
weights may allow some consideration of model uncertainty in the selection of a single best
model. Additional details of the methodology are contained in Anderson et al. (2013).
CLASSIFICATION PROCEDURE DEVELOPMENT AND VALIDATION
Residual values (observed values of log BAF minus predicted values) from the best
supported linear model describing nuisance variation were recovered for both data sets.
Variation in residual values is assumed to be attributed solely to bioavailability differences
among soils and random or latent error. Residuals from the bioaccumulation data set were used
as the response variable in the development of a regression tree (RT) methodology. The RT
methodology is essentially a case of the general classification and regression tree (CART)
algorithm of Breiman et al. (1984), also see Ripley (1996), as implemented with the rpart()
package Version 3.1-42 (Therneau and Atkinson, 2010; Faraway, 2006), with
specially-programmed extensions (R Development Core Team, 2008). An extension of the
methodology was to compute bootstrap support for particular variables, equal to percentages of
bootstrap samples where a variable was selected in the CART algorithm. Bootstrap samples
were of the set of studies (all data or no data from a given study were selected into a given
bootstrap sample). While the CART algorithm is relatively well established, a brief summary of
the standard algorithm, and our specific implementation, is provided in Anderson et al. (2013).
7
-------
RESULTS AND DISCUSSION
APPORTIONMENT OF NUISANCE VARIATION
Mixed-model analysis of the nuisance variables resulted in four candidate models for the
bioaccumulation data set and three candidate models for the toxicity data set with cumulative
Akaike weights >99% (see Table 2). All models presented in Table 2 are considered significant,
given that the null models (i.e., intercept only) were also evaluated (Burnham and Anderson,
2002). The additive fixed effects of the variables METAL, RECEPTOR, ENDPOINT, and
PARAMETER resulted in the most parsimonious model for the toxicity data set and accounted
for 91% of the overall model weight. The most parsimonious model for the bioaccumulation
data set coincidentally also accounted for 91% of the overall model weight and contained the
variables METAL, ENDPOINT, DOSE, and the METAL x DOSE interaction. All four of the
candidate bioaccumulation models contained the two-way METAL x DOSE interaction,
reflecting a relatively strong influence on BAF values (see Table 2). Significant random
variation due to SPECIES was observed for both bioaccumulation (p = 0.0189) and toxicity
(p = 0.0247) data sets, whereas, significant interstudy heterogeneity was only observed in the
toxicity data set (p = 0.0129) as determined by variance component estimation (Lindsey, 1997).
The variable TYPE only occurred in the third best model for the bioaccumulation data
set, which only accounted for 3% of the overall model weight (see Table 2). Although there is
evidence to conclude that the contamination source may have influenced BAF values, the effect
was minor relative to other experimental variables evaluated, similar to the results of Peijnenburg
et al. (2000). Most researchers subject artificially contaminated soils to various wet-dry cycles to
simulate aging, attenuating the effect of the metal (Orrono and Lavado, 2009; Si et al., 2006).
Metal salt-amended soils that are sufficiently "aged" can produce similar ecotoxicological effects
of field-contaminated soils (Smolders et al., 2009). Subsequent discussion and analyses pertain
to the most parsimonious models (see Table 2).
Bioaccumulation
Variation in BAF values was predominantly apportioned by metal-stratified doses or the
METAL x DOSE model parameter. In general, mean predicted intrametal BAF values
decreased with increasing dose, indicating constant (or declining) internal concentrations with
8
-------
increasing total soil concentrations (see Figure 1). Metal accumulation is related to metal
solubility/speciation (McLaughlin, 2002), membrane transport (Welch and Norvel, 1999;
Kochian, 1993), and uptake kinetics (Nahmani et al., 2009; Peijnenburg et al., 2000) and declines
with saturation of detoxification mechanisms, resulting in inherent toxicity thresholds that can be
quantified (Anderson et al., 2008). Threshold metal accumulation is referred to as the critical
body residue (Ma, 2005; McCarty and Mackay, 1993) and correlates with the onset of a toxic
response (Conder et al., 2002; Lanno et al., 1998). Critical body residues can impede further
metal accumulation at toxic doses (e.g., Hasaan et al., 2009; Kumar et al., 2008). Obviously,
toxicity affects BAF values and is usually the reason for the "plateau effect" in metal
salt-amended soils (McLaughlin, 2002; Hamon et al., 1999). Although linear metal
accumulation has been reported in plants and soil invertebrates at steady state (e.g., Yanai et al.,
2006; Spurgeon and Hop kin, 1996), results usually reflect subtoxic soil concentrations.
Toxicity-induced nonlinear metal accumulation has also been demonstrated in many exposure
studies (e.g., Anderson and Basta, 2009b; Smilde et al., 1992).
Differences among SPECIES and ENDPOINT further apportioned variation in BAF
values. Approximately half of the bioaccumulation data set contains BAF values for vascular
plants. Among these, only two studies reported endpoints other than metal accumulation in
aboveground biomass; one evaluated metal accumulation in the grain of several agronomic
species (Smilde et al., 1992), while the other evaluated cumulative metal levels in the shoots and
roots of Avena sp. (Bjerre and Schierup, 1985). Conversely, BAF values for soil invertebrates
reflect whole body residues. So, SPECIES essentially contrasts aboveground bioaccumulation
among plants and whole body residues among soil invertebrates, which confounded evaluation
of RECEPTOR and ENDPOINT differences. Mean predicted BAF values were in the following
order: aboveground plant biomass < soil invertebrate whole body residues (see Figure 2). Thus,
although plants and soil invertebrates may be surrogate receptors for estimating potential
terrestrial metals exposure (Scott-Fordsmand et al., 2004), results depend on the vegetative tissue
analyzed and/or the specific metal-sensitivity of the species evaluated.
Toxicity
As expected, variation in toxicity values was apportioned by METAL. Higher predicted
toxicity values reflect relatively less toxicity because higher total soil concentrations were
9
-------
required for equivalent toxicity. Thus, the rank order of mean predicted toxicity values
illustrates the observed relative potencies of the six metals evaluated in order of least potent to
most potent. Mean predicted toxicity values were in the order Cu = Pb = Zn > Ni > Co > Cd (see
Figure 3). Cadmium toxicity is expressed in numerous endpoints in plants (Hasaan et al., 2009)
and soil invertebrates (Roh et al., 2006) and is routinely reported as a relatively potent cationic
trace metal in ecological toxicity testing (e.g., Anderson and Basta, 2009b; Athar and Ahmad,
2001; Bowers et al., 1997). Predicted metal potencies are roughly consistent with trends in
Eco-SSLs for plants and soil invertebrates (U.S. EPA, 2005).
Moderate variation in toxicity values was apportioned by ENDPOINT. Although some
endpoints are specific to a receptor, evaluating them independently, in conjunction with
RECEPTOR, accounted for the most variation. As expected, reproduction-related endpoints
(cocoon and grain production) were the most sensitive, while mortality (LCso values) was the
least sensitive (see Figure 4). Studies with microbial endpoints (e.g., nitrification,
mineralization, and respiration) tended to be relatively sensitive, underscoring their relevance to
bioavailability studies, which have become commonplace in the literature (e.g., Magrisso et al.,
2009; Oorts et al., 2006; Smolders et al., 2004; Giller et al., 1999).
Obviously, the reported toxicity parameter from a dose-response curve reflects the
magnitude of an effect. Consequently, PARAMETER was crucial to apportioning nuisance
variation (see Figure 5). As expected, LCso values were the least sensitive, coinciding with the
mortality endpoint. However, an unexpected result was that, on average, mean predicted ECio
values were lower than mean predicted NOEL and LOEL values. ECio values are estimated
from fitted dose-response curves while NOEL and LOEL values are ordinarily determined by
statistical comparisons of treated to negative control groups (Eaton and Klaassen, 2001), with
possibly limited sensitivity. Thus, ECio estimates may be lower than NOEL values suggesting
that quantitative dose-response evaluation can be a more conservative methodology, especially
when low-dose values are scant.
The variable RECEPTOR was a relatively minor source of nuisance variation. Mean
predicted toxicity values among receptors were in the following order:
microbes < plants < springtails < earthworms (see Figure 6). Thus, in general, microbes tend to
produce more conservative toxicity estimates when endpoint, metal, and toxicity parameters
have been accounted for. However, the variance component estimate for species was significant
10
-------
(p = 0.0247), illustrating intrareceptor differences in toxicity among species as are usually
observed (Clark et al., 2004). Accounting for specific sensitivities among species is critical to
summarizing results from multiple toxicity studies (Anderson et al., 2008).
CLASSIFICATION PROCEDURE FOR BIOAVAILABILITY
Residuals from the most parsimonious models apportioning nuisance variation in the
toxicity and bioaccumulation data sets were recovered. Relationships between model residuals
and SOM, CLAY, and pH are illustrated in Figure 7. Significant (p < 0.0001) positive trends
were associated with all three variables for the toxicity data set. Similarly, significant
(p < 0.0001) negative trends were associated with the bioaccumulation data set. Collectively,
both data sets tenably illustrate the effect of soil metal sequestration on bioavailability. Though a
high degree of variation is evident in the data sets, the significant ^-values serve to support
further analysis.
A final regression tree (RT) was identified using selected soil properties exclusively. Use
of internal and, in particular, external validation suggested no increase in predictive capacity
beyond four terminal groups (see Figure 8). The default four-group solution, which split first on
pH, then CLAY, and again on pH, is shown in Figure 9. However, subsequent bootstrap analysis
suggested that SOM had substantial support for trees of MAXDEPTH 3 (50%). (Precise
definitions of MAXDEPTH and other CART parameters are found in documentation of R
package rpart.) Detailed examination of the results revealed that at the second split (based on
CLAY), approximately the same improvement was obtained by a split on SOM. Therefore,
although default results are presented, similar solutions to the final RT (see Figure 9) could use
the variable SOM in some way.
The ability to use potentially important variables is limited by whether data are adequate
to characterize their specific influence. In particular, certain soil properties are often observed
intercorrelated; e.g., stable clay-organic matter complexes are typical (Stevenson, 1994). In fact,
numerous bioavailability studies have demonstrated intercorrelation among the
chemical/physical properties of their respective experimental soils (e.g., Anderson and Basta,
2009a,b; Bradham et al., 2006; Dayton et al., 2006). However, no significant intercorrelation
was observed among the selected properties of the experimental soils collated in the current
study, presumably due to tremendous diversity. It just so happened in these particular analyses
11
-------
that CLAY was selected by the CART algorithm instead of SOM at the second split, whereas
bootstrap results suggest approximately the same improvement with both variables. Regardless,
pH was the primary variable modifying BAF values beyond nuisance variation and is generally
considered the master variable regulating metal equilibria in contaminated soil systems.
Therefore, the final RT presented in Figure 9, though not uniquely supported by the available
data (other solutions could be equally valid), is a simple, yet robust tool broadly applicable to the
assessment of terrestrial metals bioavailability composed of exposure studies independently
validated by studies on toxicity.
Central tendencies among the terminal nodes (i.e., bioavailability categories) were used
to determine relative differences in bioavailability according to the proposed classification
scheme. Analysis of variance of the residual values among bioavailability categories indicated
highly significant (p < 0.0083) differences among all ordinal pair-wise combinations. Table 3
illustrates mean differences in residual BAF values for each category of increasing
bioavailability. Our results suggest a maximum 3.53-fold effect of soil metal sequestration on
terrestrial metals bioavailability. When normalized to the highest category of bioavailability,
relative differences equate to 70%, 41%, and 28% bioavailability for the Medium-High,
Medium-Low, and Low categories, respectively (see Table 3). Overall, the validated
classification procedure is proposed as a quantitative tool that broadly characterizes typical
aerobic soils in terms of their potential to sequester common divalent cationic metals and
mitigate their bioavailability to soil-dwelling biota.
APPLICATION TO ECOLOGICAL RISK ASSESSMENT
Although Screening Level ERAs are conservative and, hence, assume
100%) bioavailability, the ecological relevance of Baseline ERAs depend entirely on the extent to
which bioavailability of contaminants is accurately quantified. Direct toxicity testing of
contaminated soils is the superior approach but can prohibitively add to assessment costs. If
direct toxicity tests are employed, interpreting results can be operationally difficult without
accounting for potential spatial patterns in bioavailability that may result from soil heterogeneity.
In either case, the quantitative tools presented in the current study (see Figure 9 in conjunction
with Table 3) are proposed to augment terrestrial ERAs of cationic metals contaminated soils,
e.g., in support of the interpretation of toxicity testing.
12
-------
Typical Baseline ERAs include site characterization, risk assessment, and risk
management phases (e.g., U.S. EPA, 1998). Results of the current study can be directly applied
to the site characterization and risk assessment phases. For example, in the site characterization
phase, soil composition data can be included to refine Conceptual Site Models in the context of
bioavailability according to the classification procedure presented in Figure 9. Based on the
results of this investigation, risk assessors should, thus, routinely include relatively standard test
methods for soil pH and total clay content (and SOM) in concert with total metals analyses.
However, additional variables, not included in this investigation, may provide additional
value-added information. Specifically, the composition of SOM and total clay content can vary
greatly, which can affect soil metal sequestration (Lock and Janseen, 2001). Mineralogical
analysis of clay-sized particles including amorphous oxide content can also improve on
bioavailability estimates (e.g., Dayton et al., 2006; Bradl, 2004; Peijnenburg et al., 1999a). The
results of this investigation are, therefore, couched as the extent to which the current literature
on soil metal bioavailability could be synthesized by meta-analysis. Although a more
exhaustive dataset including additional soil chemical/physical properties would likely improve
the overall applicability, the results of the current study are a starting point for which additional
research may improve upon pending the quantification of relevant variables in future studies.
In the risk characterization phase, Figure 9 can be used in conjunction with some
categorization of total soil concentrations to develop a matrix of risk categories. For example,
low concentration and low bioavailability equates to low risk. Conversely, high concentration
and high bioavailability equates to high risk. Mapping heterogeneous sites by relative risk
category could focus subsequent remedial efforts. Additionally, adjustment factors presented in
Table 3 can be used to normalize total soil concentrations for the development of site-specific
Exposure Point Concentrations (EPCs) for soil-dwelling organisms according to the
classification scheme presented in Figure 9. For example, if the soil concentration of metal X
has been sampled and determined to be 1,500 mg/kg, and the soil has a pH greater than 4.7 but
less than 6.5 and clay content less than 26%, the bioavailability category (see Figure 9) would be
Medium-High, leading to a factor difference of 1.42 (or 70% bioavailable) relative to the High
category (see Table 3), which would result in a bioavailability-adjusted EPC of 1,050 mg/kg
(1,500 x 0.70). Alternatively, if the soil concentration of metal X has been sampled and
determined to be 1,500 mg/kg, and the soil has a pH greater than 4.7 and clay content greater
13
-------
than 26%, the bioavailability category (see Figure 9) would be Low, leading to a factor
difference of 3.53 (or 28% bioavailable) relative to the High category (see Table 3), which would
result in a bioavailability-adjusted EPC of 420 mg/kg (1,500 x 0.28). The adjustment factors in
Table 3 are calculated relative to the highest-bioavailability category. The approach does not
assume that the highest bioavailability values in the data would be used in all situations.
Bioavailability-adjusted EPCs for soil-dwelling organisms can be used as input to trophic
transfer models to predict site-specific exposures for higher-order wildlife species explicitly
applying soil bioavailability concepts to risk estimates for the relevant receptor(s) evaluated at a
given site.
14
-------
REFERENCES
Anderson RH and Basta NT. 2009a. Application of ridge regression to quantify marginal effects
of collinear soil properties on phytoaccumulation of arsenic, cadmium, lead, and zinc. Environ
Toxicol Chem 28:619-28
Anderson RH and Basta NT. 2009b. Application of ridge regression to quantify marginal effects
of collinear soil properties on phytotoxicity of arsenic, cadmium, lead, and zinc. Environ Toxicol
Chem 28:1018-27
Anderson RH, Basta NT, and Lanno RP. 2008. Partitioning species variability from soil property
effects on phytotoxicity: ECx normalization using a plant contaminant sensitivity index. J
Environ Qual 37:1701-9
Anderson RH, Farrar DB, and Zodrow JM. 2013. Terrestrial metals bioavailability: A
comprehensive review and literature-derived decision rule for ecological risk assessment. Hum
Ecol Risk Assess 19:1488-1513
ASTM International (American Society for Testing and Materials). 2004. ASTM E1440-91
Standard guide for acute toxicity test with the rotifer brachionus. In: Book of Standards, vol.
11.06. ASTM International, Philadelphia, PA, USA
Athar R and Ahmad M. 2001. Heavy metal toxicity: effect on plant growth and metal uptake by
wheat, and on free living azotobacter . Water Air Soil Pollut 138:165-80
Basta NT, Pantone DJ, and Tabatabai MA. 1993. Path analysis of heavy metal adsorption by soil.
Agron J 85:1054-7
Basta NT, Ryan JA, and Chaney RL. 2005. Trace element chemistry in residual-treated soil: key
concepts and metal bioavailability. J Environ Qual 34:49-63
Bjerre GK and Schierup HH. 1985. Uptake of six heavy metals by oat as influenced by soil type
and additions of cadmium, lead, zinc and copper. Plant Soil 88:57-69
Bowers N, Pratt JR, Beeson D, et al. 1997. Comparative evaluation of soil toxicity using lettuce
seeds and soil ciliates. Environ Toxicol Chem 16:207-13
Bradham KD, Dayton EA, Basta NT, et al. 2006. Effect of soil properties on lead bioavailability
and toxicity to earthworms. Environ Toxicol Chem 25:769-75
Bradl HB. 2004. Adsorption of heavy metal ions on soils and soils constituents. J Colloid
Interface Sci 277:1-18
15
-------
Breiman L, Freidman JH, Olshen RA, etal. 1984. Classification and Regression Trees. Chapman
& Hall/CRC, Boca Raton, FL, USA
Burnham KP and Anderson DR. 2002. Information and likelihood theory: A basis for model
selection and inference. In: Model Selection and Multimodel Inference. A Practical Information-
Theoretic Approach, 2nd edit, pp 49-96. Springer, New York, NY, USA
Clark J, Ortego LS, and Fairbrother A. 2004. Sources of variability in plant toxicity testing.
Chemosphere 57:1599-612
Conder JM, Seals LD, and Lanno RP. 2002. Method for determining toxicologically relevant
cadmium residues in the earthworm Eiseniafetida. Chemosphere 49:1-7
Criel P, Lock K, Eeckhout HV, el al. 2008. Influence of soil properties on copper toxicity for
two soil invertebrates. Environ Toxicol Chem 27:1748-55
Crouau Y and Cazes L. 2003. What causes variability in the Folsomia Candida reproduction test?
Appl Soil Ecol 22:175-80
Dang YP, Chhabra R, and Verma KS. 1990. Cadmium-enriched sewage sludge effect of Cd, Ni,
Pb and Zn on growth and chemical composition of onion and fenugreek. Commun Soil Sci Plant
Anal 21:717-35
Dayton EA, Basta NT, Payton ME, et al. 2006. Evaluating the contribution of soil properties to
modifying lead phytoavailability and phytotoxicity. Environ Toxicol Chem 25:719-25
Degryse F, Smolders E, and Parker DR. 2009. Partitioning of metals (Cd, Co, Cu, Ni, Pb, Zn) in
soils: Concepts, methodologies, prediction and applications—A review. Eur J Soil Sci 60:590-
612
de Haan S, Rethfeld H, and van DrielW. 1985. Acceptable Levels of HeavyMetals (Cd, Cr,
Cu,Ni, Pb, Zn) in Soils, Depending on Their Clay and Humus Content and Cation-Exchange
Capacity, pp 1-42. Instituut voor Bodemvruchtbaarheid, Haren, The Netherlands
Donkin SG and Dusenbery DB. 1993. A soil toxicity test using the nematode Caenorhabditis
elegans and effective method of recovery. Arch Environ Contam Toxicol 25:145-51
Donkin SG and Dusenbery DB. 1994. Using the Caenorhabditis elegans soil toxicity test to
identify factors affecting toxicity of four metal ions in intact soil. Water Air Soil Pollut 78:359-
73
16
-------
Drexler J, Fisher N, Henningsen G, et al. 2003. Issue Paper on the Bioavailability and
Bioaccumulation of Metals. Submitted by ERG, Lexington, MA to U.S. Environmental
Protection Agency, Risk Assessment Forum, Washington, DC, USA
Eaton DL and Klaassen CD. 2001. Principles of toxicology. In: Klaassen CD (ed), Cassarett and
Doull's Toxicology: The Basic Science of Poisons, 6th edit, pp 11-34. McGraw-Hill, New York,
NY, USA
Elmosly WA and Abdel-Sabour MF. 1997. Transfer characteristics and uptake of nickel by red
clover grown on nickel amended alluvial soils of an arid zone. Agric Ecosyst Environ 65:49-57
Environment Canada. 2005. Biological Test Method: Test for Measuring Emergence and Growth
of Terrestrial Plants Exposed to Contaminants in Soil. Report EPA l/RM/45. Environment
Canada Press, Ottawa, ON, Canada
ESG International Inc. and Aquaterra Environmental Consulting. 2000. Assessment of the
Biological Test Methods for Terrestrial Plants and Soil Invertebrates: Metals. Environmental
Technology Centre, Environment Canada, Ottawa, ON, Canada
Faraway JJ. 2006. Extending the Linear Model with R: Generalized Linear, Mixed Effects and
Nonparametric Regression Models. Chapman and Hall/CRC Press, Boca Raton, FL, USA
Ge Y and Hendershot W. 2002. Evaluation of soil surface charge using the back-titration
technique. Soil Sci Soc Am J 68:82-8
Giller KE, Witter E, and McGrath SP. 1999. Assessing risks of heavy metal toxicity in
agricultural soil: Do microbes matter? Hum Ecol Risk Assess 5:683-9
Gleyzes C, Tellier S, and Astruc M. 2002. Fractionation studies of trace elements in
contaminated soils and sediments: a review of sequential extraction procedures. Trends Analyt
Chem 21:451-67
Gunther P and Pestemer W. 1990. Risk assessment for selected xenobiotics by bioassay methods
with higher plants. Environ Manage 14:381-8
Hague A and Ebing W. 1983. Toxicity determination of pesticides to earthworms in the soil
substrate. Z Pflanzenkrankh (Pflanzenpathol) Pflanzenschutz 90:395-408 (German)
Hamon R, McLaughlin M, and Lombi E. 2007. Natural Attenuation of Trace Element
Availability in Soils. CRC Press, Pensacola, FL, USA
Hamon RE, Holm PE, Lorenz SE, et al. 1999. Metal uptake by plants from sludge-amended
soils: caution is required in the plateau interpretation. Plant Soil 216:53-64
17
-------
Hasaan SA, Fariduddin Q, Ali B, etal. 2009. Cadmium: Toxicity and tolerance in plants. J
Environ Biol 30:165-74
Hough RL, Tye AM, Crout NMJ, el al. 2005. Evaluating a "free ion activity model" applied to
metal uptake by Loliumperenne L. grown in contaminated soils. Plant Soil 270:1-12
Howcroft CF, Amorim MJ, Gravato C, etal. 2009. Effects of natural and chemical stressors on
Enchytraeus albidus: can oxidative stress parameters be used as fast screening tools for the
assessment of different stress impacts in soils? Environ Internat 35:318-24
ISO (International Organization for Standardization). 2005. ISO 11269-2:2005 Soil Quality—
Determination of the Effects of Pollution on Soil Flora—Part 2: Effects of Chemicals on the
Emergence and Growth of Higher Plants. ISO 11269-2. International Organization for
Standardization, Geneva, Switzerland
Janssen RPT, Posthuma L, Baerselman R, el al. 1997. Equilibrium partitioning of heavy metals
in Dutch field soils II. Prediction of heavy metal accumulation in earthworms. Environ Toxicol
Chem 16:2479-88
Kapustka LA, Eskew D, and Yocum JM. 2006. Plant toxicity testing to derive ecological soil
screening levels for cobalt and nickel. Environ Toxicol Chem 25:865-74
Kjaer C and Elmegaard N. 1996. Effects of copper sulfate on black bindweed (Polygonum
convolvulus L.). Ecotoxicol Environ Saf 33:110-7
Kochian LV. 1993. Zinc absorption from hydroponic solutions by plant roots. In: Robson AD
(ed), Zinc in Soils and Plants, pp 45-57. Kluwer, Dordrecht, The Netherlands
Korthals GW, van de Ende A, van MegenH, el al. 1996. Short-term effects of cadmium, copper,
nickel and zinc on soil nematodes from different feeding and life-history strategy groups. Appl
Soil Ecol 4:107-17
Koster M, de Groot A, Vijver M, et al. 2006. Copper in the terrestrial environment: verification
of a laboratory-derived biotic ligand model to predict earthworm mortality with toxicity observed
in field soils. Soil Biol Biochem 38:1788-96
Kumar S, Sharma V, Bhoyar RV, el al. 2008. Effect of heavy metals on earthworm activities
during vermicomposting of municipal solid waste. Water Environ Res 80:154-61
Lanno RP, LeBlanc S, Knight B, et al. 1998. Application of body residues as a tool in the
assessment of soil toxicity. In: Sheppard S, Bembridge J, HolmstrupM, el al. (eds), Advances in
Earthworm Ecotoxicology, pp 41-53. SETAC Press, Pensacola, FL, USA
18
-------
Lanno RP,Wells J, Conder J, etal. 2004. The bioavailability of chemicals in soil for earthworms.
Ecotoxicol Environ Saf 57:39-47
Li HF, Gray C, Mico C, el al. 2009. Phytotoxicity and bioavailability of cobalt to plants in a
range of soils. Chemosphere 75:979-86
Lindsey JK. 1997. Applying Generalized Linear Models. Springer-Verlag, New York, NY, USA
Lock K and Janseen CR. 2001. Effect of clay and organic matter type on the ecotoxicity of zinc
and cadmium to the potworm Enchytraeus albidus. Chemosphere 44:1669-72
Lofts S, Spurgeon DJ, Svendsen C, et al. 2004. Deriving soil critical limits for Cu, Zn, Cd, and
Pb: A method based on free ion concentrations. Environ Sci Technol 38:3623-31
Lofts S, Spurgeon D, and Svendsen C. 2005. Fractions affected and probabilistic risk assessment
of Cu, Zn, Cd, and Pb in soils using the free ion approach. Environ Sci Technol 39:8533-40
Lowe CN and Butt KR. 2007. Earthworm culture, maintenance and species selection in chronic
ecotoxicological studies: A critical review. Eur J Soil Biol 43:281-8
Ma WC. 1982. The influence of soil properties and worm-related factors on the concentration of
heavy metals in earthworms. Pedobiologia 24:109-19
Ma WC. 2005. Critical body residues (CBRs) for ecotoxicological soil quality assessment:
Copper in earthworms. Soil Biol Biochem 37:561-8
Magrisso S, Belkin S, and Erel Y. 2009. Lead bioavailability in soil and soil components.Water
Air Soil Pollut 202:315-23
McCarty LS and Mackay D. 1993. Enhancing ecotoxicological modeling and assessment: Body
residues and modes of toxic action. Environ Sci Technol 27:1719-28
McKeague JA and Day JH. 1993. Ammonium oxalate extraction of amorphous iron and
aluminum. In: Carter MR (ed), Soil Sampling and Methods of Analysis, pp 214-43. Lewis
Publishers, Boca Raton, FL, USA
McLaughlin MJ. 2002. Bioavailability of metals to terrestrial plants. In: Allen HE (ed),
Bioavailability of Metals in Terrestrial Ecosystems: Importance of Partitioning for
Bioavailability to Invertebrates, Microbes, and Plants, pp 39-68. SETAC Press, Pensacola, FL,
USA
Mleczek M, Rissmann I, Rutkowski P, el al. 2009. Accumulation of selected heavy metals by
different genotypes of Salix. Environ Exp Bot 66:289-96
19
-------
Nahmani J, Hodson ME, Devin S, et al. 2009. Uptake kinetics of metals by the earthworm
Eisenia fetida exposed to field-contaminated soils. Environ Pollut 157:2622-8
Nolan AL, Lombi E, and McLaughlin MJ. 2003. Metal bioaccumulation and toxicity in soils—
Why bother with speciation? Aust J Chem 56:77-91
Oorts K, Ghesquiere U, Swinnen K, et al. 2006. Soil properties affecting the toxicity of CuC12
and NiC12 for soil microbial processes in freshly spiked soils. Environ Toxicol Chem 25:836-44
Orrono DL and Lavado RS. 2009. Distribution of extractable heavy metals in different soil
fractions. Chem SpecBioavail 21:193-8
Parker DR and Pedler JF. 1997. Reevaluating the free-ion activity model of trace metal
availability to higher plants. Plant Soil 196:223-8
Parker DR, Pedler JF, Ahnstrom ZA, et al. 2001. Reevaluating the free-ion activity model of
trace metal toxicity toward higher plants: experimental evidence with copper and zinc. Environ
Toxicol Chem 20:899-906
Pedersen MB, Kjaer C, and Elmegaard N. 2000. Toxicity and bioaccumulation of copper to
black bindweed (Fallopia convolvulus) in relation to bioavailability and the age of soil
contamination. Arch Environ Contam Toxicol 39:431-9
Peijnenburg WJ, Baerselman R, de Groot AC, et al. 1999a. Relating environmental availability
to bioavailability: soil-type-dependent metal accumulation in the oligochaete Eisenia andrei.
Ecotoxicol Environ Saf 44:294-310
Peijnenburg WJ, Posthuma L, Zweers PG, et al. 1999b. Prediction of metal bioavailability in
Dutch field soils for the oligochaete Enchytraeus crypticus. Ecotoxicol Environ Saf 43:170-86
Peijnenburg WJ, Baerselman R, de Groot A, et al. 2000. Quantification of metal bioavailability
for lettuce (Lactuca sativa L.) in field soils. Arch Environ Contam Toxicol 39:420-30
Peijnenburg WJ, Zablotskaja M, and Vijver MG. 2007. Monitoring metals in terrestrial
environments within a bioavailability framework and a focus on soil extraction. Ecotoxicol
Environ Saf 67:163-79
Posthuma L, Traas TP, and Suter GW (eds). 2002. Species Sensitivity Distributions in
Ecotoxicology. CRC Press, Boca Raton, FL, USA
R Development Core Team. 2008. R: A Language and Environment for Statistical Computing. R
Foundation for Statistical Computing, Vienna, Austria
20
-------
Rao CR, Sahuquillo A, and Lopez-Sanchez JF. 2008. A review of the different methods applied
in environmental geochemistry for single and sequential extraction of trace elements in soils and
related materials. Water Air Soil Pollut 189:291-333
Reber HH. 1989. Threshold levels of cadmium for soil respiration and growth of spring wheat
(.Triticum aestivum L.), and difficulties with their determination. Biol Fertil Soils 7:152-7
Ripley BD. 1996. Pattern Recognition and Neural Networks. Cambridge University Press, New
York, NY, USA
Roh JY, Lee J, and Choi J. 2006. Assessment of stress-related gene expression in the heavy
metal-exposed nematode Caenorhabditis elegans: A potential biomarker for metal-induced
toxicity monitoring and environmental risk assessment. Environ Toxicol Chem 25:2946-56
Rooney CP, Zhao FJ, and McGrath SP. 2006. Soil factors controlling the expression of copper
toxicity to plants in a wide range of European soils. Environ Toxicol Chem 25:726-32
Rooney CP, Zhao FJ, and McGrath SP. 2007. Phytotoxicity of nickel in a range of European
soils: Influence of soil properties, Ni solubility and speciation. Environ Pollut 145:596-605
Saeki K and Kunito T. 2009. Estimating sorption affinities of heavy metals on humic acid and
silica using a constant capacitance model. Commun Soil Sci Plant Anal 40:3252-62
Scheckel KG, Impellitteri CA, Ryan JA, et al. 2003. Assessment of a sequential extraction
procedure for perturbed lead-contaminated samples with and without phosphorus amendments.
Environ Sci Technol 37:1892-8
Scheckel KG, Chaney RL, BastaNT, et al. 2009. Advances in assessing bioavailability of
metal(loid)s in contaminated soils. Adv Agron 107:10-52
Scott-Fordsmand JJ, Stevens D, and McLaughlin M. 2004. Do earthworms mobilize fixed zinc
from ingested soil? Environ Sci Technol 38:3036-9
Semennzin E, Temminghoff EJM, and Marcomini A. 2007. Improving ecological risk
assessment by including bioavailability into species sensitivity distributions: An example for
plants exposed to nickel in soil. Environ Pollut 148:642-7
SERDP (Strategic Environmental Research and Development Program) and ESTCP
(Environmental Security Technology Certification Program). 2008. SERDP and ESTCP expert
panel workshop on research and development needs for understanding and assessing the
bioavailability of contaminants in soils and sediments. In: Strategic Environmental Research and
Development Program and Environmental Security Technology Certification Program. August
21
-------
20-21, 2008, Annapolis, MD, USA. Available at http://www.serdp.org/News-and-
Events/C onferences-W orkshop s/Past-ERW orkshop s
Sheppard SC, Evenden WG, Abboud SA, etal. 1993. A plant life-cycle bioassay for
contaminated soil, with comparison to other bioassays: Mercury and zinc. Arch Environ Contam
Toxicol 25:27-35
Si JT, Tiam BG, and Wang HT. 2006. Effect of incubation temperature and wet-dry cycle on the
availablities of Cd, Pb, and Zn in soil. J Environ Sci (Beijing, China) 18:1119-23
Sipos P, Nemeth T, Kis VK, et al. 2008. Sorption of copper, zinc, and lead on soil mineral
phases. Chemosphere 73:461-9
Sipos P, Nemeth T, Kis VK, et al. 2009. Association of individual soil mineral constituents and
heavy metals as studies by sorption experiments and analytical electron microscopy analyses. J
Hazard Mater 168:1512-20
Sleutel S, de Neve S, Singier B, et al. 2007. Quantification of organic carbon in soils: a
comparison of methodologies and assessment of the carbon content of organic matter. Commun
Soil Sci Plant Anal 38:2647-57
Smilde KW, van Luit B, and van Driel W. 1992. The extraction by soil and absorption by plants
of applied zinc and cadmium. Plant Soil 143:233-8
Smolders E, Buekers J, Oliver I, etal. 2004. Soil properties affecting toxicity of zinc to soil
microbial properties in laboratory-spiked and field-contaminated soils. Environ Toxicol Chem
23:2633-40
Smolders E, Oorts K, Van Sprang P, etal. 2009. Toxicity of trace metals in soil as affected by
soil type and aging after contamination: Using calibrated bioavailability models to set ecological
soil standards. Environ Toxicol Chem 28:1633-42
Sochova I, Hofman J, and Holoubek I. 2006. Using nematodes in soil ecotoxicology. Environ
Internat 32:374-83
Sparks DL. 2003. Environmental Soil Chemistry. Academic Press, London, UK
Spurgeon DJ and Hopkin SP. 1996. 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 40:80-
96
22
-------
Spurgeon DJ and Hopkin SP. 1999. Comparisons of metal accumulation and excretion kinetics
in earthworms (Eisenia fetida) exposed to contaminated field and laboratory soils. Appl Soil
Ecol 11:227-43
Spurgeon DJ, Svendsen C, Rimmer VR, et al. 2000. Relative sensitivity of life-cycle and
biomarker responses in four earthworm species exposed to zinc. Environ Toxicol Chem
19:1800-8
Steenbergen NT, Iaccino F, de Winkel M, et al. 2005. Development of a biotic ligand model and
a regression model predicting acute copper toxicity to the earthworm Aporrectodea caliginosa.
Environ Sci Technol 39:5694-702
Stevenson FJ. 1994. Humus Chemistry: Genesis, Compostion, Reactions. John Wiley & Sons,
New York, NY, USA
Thakali S, Allen HE, Di Toro DM, et al. 2006a. A terrestrial biotic ligand model. 1.
Development and application to Cu and Ni toxicities to barley root elongation in soils. Environ
Sci Technol 40:7085-93
Thakali S, Allen HE, Di Toro DM, et al. 2006b. Terrestrial biotic ligand model. 2. Application to
Ni and Cu toxicities to plants, invertebrates, and microbes in soil. Environ Sci Technol 40:7094-
100
Therneau TM and Atkinson B; port by Brian Ripley. 2010. rpart: Recursive Partitioning. R
package version 3.1-46. Breiman L, Friedman J, Olshen R and Stone C. (1984). Classification
and regression trees. Statistics/Probability Series. Wadsworth & Brooks/Cole Advanced Books
& Software.
TN & Associates Inc. 2000. Plant Toxicity Testing to Support Development of Ecological Soil
Screening Levels. US Environmental Protection Agency, National Center for Environmental
Assessment, Washington, DC, USA
U.S. EPA (Environmental Protection Agency). 1994. Catalogue of Standard Toxicity Tests for
Ecological Risk Assessment. ECO Update, Intermittent Bulletin Volume 2, Number 2. Office of
Solid Waste and Emergency Response, Hazardous Site Evaluation Division (5204G),
Washington, DC. Publication 9345.0-051.
U.S. EPA. 1998. Guidelines for Ecological Risk Assessment. Risk Assessment Forum,
Washington, DC. EPA/630/R-95/002F.
U.S. EPA. 2005. Guidance for Developing Ecological Soil Screening Levels. Office of Solid
Waste and Emergency Response, Washington, DC. OSWER Directive 9285.7-55.
23
-------
U.S. EPA. 2007. Framework for Metals Risk Assessment. Office of the Science Advisor, Risk
Assessment Forum, Washington, DC. EPA/120/R-07/001.
van Gestel CAM and van Dis WA. 1988. The influence of soil characteristics on the toxicity of
four chemicals to the earthworm Eisenia fetida andrei (Oligochaeta). Biol Fertil Soils 6:262-5
Vijver M, Jager T, Posthuma L, et al. 2001. Impact of metal pools and soil properties on metal
accumulation in Folsomia Candida (Collembola). Environ Toxicol Chem 20:712-20
Vonk JW, Matla YA, van Gestel CAM, et al. 1996. The Influence of Soil Characteristics On the
Toxicity of Cadmium for Folsomia Candida, Eisenia fetida and Glutamate Mineralisation. MEP-
R96-144. Netherlands Organization for Applied Scientific Research, Delft, The Netherlands
Welch RM and Norvel WA. 1999. Mechanisms of cadmium uptake, traslocation and deposition
in plants. In: McLaughlin MJ and Singh BR (eds), Cadmium in Soils and Plants, pp 125-55.
Luluwer, Dordrecht, The Netherlands
Weng L, Lexmond TM, Wolthoorn A, et al. 2003. Phytotoxicity and bioavailability of nickel:
Chemical speciation and bioaccumulation. Environ Toxicol Chem 22:2180-7
Yanai J, Zhao FJ, McGrath SP, et al. 2006. Effect of soil characteristics on Cd uptake by the
hyperaccumulator Thlaspi caerulescens. Environ Pollut 139:167-75
Zhao FJ, Rooney CP, Zhang H, et al. 2006. Comparison of soil solution speciation and diffusive
gradients in thin films measurements as an indicator of copper bioavailability to plants. Environ
Toxicol Chem 25:733-42
24
-------
Table 1. Summary of Data Used in Meta-analyses
Data set
Reference
Number of
Range in soil properties
Soils
pHa
Clay (%)
SOM (%)b
Bioaccumulation
(Yanai et al., 2006)
25
4.30-7.77
1.8-44
2.40-29.2
(Dayton et al., 2006)
21
4.00-7.90
10-60
0.400-5.80
(Bradham et al., 2006)
21
4.00-7.90
10-60
0.400-5.80
(Peijnenburg et al., 1999a)
20
3.81-7.43
0.20-47
0.300-15.1
(Peijnenburg et al., 1999b)
20
3.81-7.43
0.20-47
0.300-15.1
(Janssen et al., 1997)
20
3.81-7.43
0.20-47
0.300-15.1
(Vijver et al., 2001)
16
3.09-7.30
0.20-47
0.300-35.0
(Peijnenburg et al., 2000)
15
4.36-7.22
1.3-47
0.900-23.4
(Nahmani et al., 2009)
7
4.58-6.54
T
o
0.284-2.36
(Anderson and Basta, 2009a)
5
3.67-7.34
6.8-42
0.810-4.78
(Weng et al., 2003)
4
4.70-6.80
4.0-4.0
4.00-4.00
(Smilde et al., 1992)
3
4.20-7.20
3.0-40
3.70-7.00
(Elmosly and Abdel-Sabour, 1997)
3
7.40-8.10
5.5-20
0.0500-1.00
(Bjerre and Schierup, 1985)
3
6.20-7.50
4.9-8.0
1.90-17.7
Toxicity
(Oorts et al., 2006)
35
3.00-7.70
1.0-55
0.500-66.1
(Criel et al., 2008)
19
3.00-7.50
5.0-51
0.800-46.6
(Rooney et al., 2006)
18
3.40-7.50
5.0-51
0.760-46.6
(Rooney et al., 2007)
16
3.60-7.70
0.40-55
0.500-66.1
(Smolders et al., 2004)
12
3.00-7.50
5.0-51
0.800-46.6
25
-------
Table 1. Summary of Data Used in Meta-analyses (continued)
Data set
Reference
Number of
Range in soil properties
Soils
pHa
Clay (%)
SOM (%)b
Toxicity cont.
(Li et al., 2009)
10
4.30-7.53
1.0-48
1.60-10.6
(Spurgeon and Hopkin, 1996)
9
4.00-6.00
20-20
5.00-15.0
(de Haan et al., 1985)
6
4.60-5.60
4.0-58
1.60-19.4
(Anderson and Basta, 2009b)
5
3.67-7.34
6.8-42
0.812-4.78
(Vonk et al., 1996)
5
3.50-6.80
1.9-20
2.40-10.0
(Donkin and Dusenbery, 1994)
4
5.10-6.20
16-39
1.70-3.40
(Weng et al., 2003)
4
4.70-6.80
4.0-4.0
4.00-4.00
(Donkin and Dusenbery, 1993)
4
5.10-6.20
16-39
1.70-3.40
(ESG International Inc. and
Aquaterra Environmental
Consulting, 2000)
3
6.05-8.10
11-30
2.90-12.8
(Reber, 1989)
3
5.60-7.00
3.2-21
1.67-2.62
(Sheppard et al., 1993)
2
7.30-7.90
43-46
2.70-8.90
(Gunther and Pestemer, 1990)
1
6.10
9.9
1.31
(Korthals et al., 1996)
1
4.10
4.0
1.90
(Ma, 1982)
1
7.30
17
8.00
(van Gestel and van Dis, 1988)
1
7.00
4.3
1.70
(Spurgeon et al., 2000)
1
6.35
9.7
2.35
(Kjaer and Elmegaard, 1996)
1
6.40
11
3.40
26
-------
Table 1. Summary of Data Used in Meta-analyses (continued)
Data set
Reference
Number of
Soils
Range in soil properties
pHa
Clay (%)
SOM (%)b
Toxicity cont.
(Hague and Ebing, 1983)
1
6.10
3.2
1.00
(Dang et al., 1990)
1
8.30
24
0.560
(TN & Associates Inc., 2000)
1
6.32
3.2
0.200
(Pedersen et al., 2000)
1
6.70
14
9.00
(Kapustka et al., 2006)
1
6.32
3.2
0.100
(Howcroft et al., 2009)
1
5.60
13
4.72
a pH values include KC1 and CaCL extracts.
b Studies that only reported organic C values were doubled to convert to soil organic matter (SOM) values (Sleutel et al., 2007).
27
-------
Table 2. Variable Selection Results" from Apportionment Analysis for Both Bioaccumulation and Toxicity Data Sets
Data Set
Variables
Kb
-2(/maxc)
AICcd
AAICce
Toxicity
METAL + RECEPTOR + ENDPOINT + PARAMETER
27
3,785.9
3,841.1
0.0
0.91
METAL + ENDPOINT (RECEPTOR) + PARAMETER
31
3,782.2
3,845.8
4.7
0.09
METAL + ENDPOINT + PARAMETER
26
3,797.6
3,850.7
9.6
0.01
Bioaccumulation
METAL + ENDPOINT + DOSE + METAL x DOSE
31
1,580.7
1,646.8
0.0
0.91
METAL + DOSE + METAL x DOSE
29
1,591.5
1,653.1
6.3
0.04
METAL + TYPE + DOSE + METAL x DOSE
30
1,589.5
1,653.4
6.5
0.03
METAL + RECEPTOR + DOSE + METAL x DOSE
30
1,591.4
1,655.3
8.4
0.01
a Only those models with a cumulative total of Akaike weights >99% are presented.
b Number of model parameters.
0 Maximum Log-likelihood.
d Small-sample adjusted Akaike Information Criterion (AICc).
e AICc difference.
f AICc weight.
28
-------
Table 3. Proposed Adjustments to Total Soil Concentrations That Account for Metals Bioavailability Based on the Central
Tendencies of the Terminal Nodes (i.e., Bioavailability Categories) from Figure 9. (See Example)
Bioavailability Category
High
Medium-High
Medium-Low
Low
Factor difference
1.00
1.42
2.44
3.53
Percent bioavailable
100
70
41
28
Example: The value of 1.42 for the medium-high category can be obtained in two steps from
results displayed in Figure 9: First apply the antilogarithm (exponential) to the mean values for
"H" and "M-H" groups, which were calculated in the natural logarithm scale. Thus we compute
values 1.878 (H) and 1.323 (M-H). Finally, the value displayed is 1.42 = 1.878/1.323. The
corresponding percent bioavailable is the inverse of this value x 100: 70 = 100 x (1/1.42).
29
-------
Dose
Figure 1. Mean (± 1 SE) Predicted Natural Log (nl) BAF Values by Dose, as Stratified by Metal, from the Most Parsimonious
Bioaccumulation Model Apportioning Nuisance Variation.
30
-------
0.4 -
0.8
c/) 0.6 -
LL
<
GO
_ 0.2 H
c
"O
0
-t—f
o
TD
CD
0.0
-0.2 -
-0.4
Plant shoots
Invertebrate whole body residue
Figure 2. Mean (± 1 SE) Predicted Natural Log (nl) BAF Values by Endpoint (confounded by receptor) from the Most
Parsimonious Bioaccumulation Model Apportioning Nuisance Variation.
31
-------
Cu Pb Zn Ni Co Cd
Figure 3. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Metal from the Most Parsimonious Toxicity Model
Apportioning Nuisance Variation.
32
-------
O)
O)
E
C/)
CD
CO
>
g
x
o
•*—>
c
"O
CD
•*—>
o
TD
CD
^ ,
-------
8
o
'x
o
"D
Q)
O
~o
^ 4 -
2 -
0
EC10 NOEL LOEL EC20 EC50 LC50
Figure 5. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Parameter from the Most Parsimonious Toxicity
Model Apportioning Nuisance Variation.
34
-------
6.0
O)
jx.
O)
cn
0
CD
>
C
"D
0
-t—>
O
~o
0
5.9 -
& 5.8 -
o
X
o
5.7 -
5.6
Worms
Springtail
Plants
Microbes
Figure 6. Mean (± 1 SE) Predicted Natural Log (nl) Toxicity Values by Receptor from the Most Parsimonious Toxicity Model
Apportioning Nuisance Variation.
35
-------
-4
0 10 20 30 40 50 60 70
SOM (%)
10 20 30 40 50 60
Clay (%)
5 6
PH
Figure 7. Relationships Between Model Residuals and Soil Organic Matter (SOM), Clay Content, and pH for the Most
Parsimonious Toxicity and Bioaccumulation Linear Models Apportioning Nuisance Variation (see Table 2).
36
-------
0)
c
o
CD
"to
>
if)
V)
o
:
o
<1)
>
4->
_ro
d)
Ct
CD
d
Internal Validation Performance
CO
o
o —
LO
a
External Validation Performance
n i i i i i i i r
23456789 10
Number of Groups
12
14
Figure 8. Internal (Bioaccumulation Data Set) and External (Toxicity Data Set) Validation
Results of Cross-Validation Analyses Used to Determine the Optimal-Sized
Regression Tree. The upper figure relates the sum of squared error (from £-fold
cross validation) for an RT model, relative to the error from use of an RT with no
splits (i.e., use of the simple grand mean), to the number of terminal groups. The
lower figure relates the Spearman correlation between predictions and toxicity values
to the number of groups.
37
-------
Figure 9. Regression Tree Results Constrained to Four Terminal Nodes (see Figure 8).
Mean (m) and sample size (//) values are characterized for each terminal node, which
are classified into bioavailability categories of High (H), Medium-High (M-H),
Medium-Low (M-L), and Low (L). Relative RSS gives the sum of squared error for
an RT with a given number of splits, relative to an RT with no splits (i.e., simple use
of the grand mean).
38
------- |