Including Mixtures in the
Determination of Water Quality
Criteria for Herbicides in Surface
N A T H A L I E C H E Å V R E , *
C H R I S T I A N L O E P F E , H E I N Z S I N G E R ,
C H R I S T I A N S T A M M , K A T H R I N F E N N E R ,
A N D B E A T E I . E S C H E R
Swiss Federal Institute for Aquatic Science and Technology
(Eawag), 8600 Duebendorf, Switzerland
Monitoring programs throughout America and Europe
have demonstrated the common occurrence of herbicides
in surface water. Nevertheless, mixtures are rarely
taken into account in water quality regulation. Taking
mixtures into account is only feasible if the water quality
criteria (WQC) of the single compounds are derived by a
common and consistent methodology, which overcomes
differences in data quality without settling on the lowest
common denominator but making best use of all available
data. In this paper, we present a method of defining a
risk quotient for mixtures of herbicides with a similar mode
of action (RQm). Consistent and comparable WQC are
defined for single herbicides as a basis for the calculation
of the RQm. Derived from the concentration addition
model, the RQmcan be expressed as the sum of the ratios
of the measured environmental concentration and the
WQC for each herbicide. The RQmshould be less than one
to ensure an acceptable risk to aquatic life. This approach
has the advantage of being easy to calculate and
communicate, and is proposed as a replacement for the
current limit of 0.1 µg/L for herbicides in Switzerland. We
illustrate the proposed approach on the example of five
commonly applied herbicides (atrazine, simazine, terbuthyl-
azine, isoproturon, and diuron). Their risk profile, i.e., the
RQmas a function of time for one exemplary river, clearly
shows that the single compounds rarely exceeded their
seasonally applied herbicides, whose application periods
partially overlap, together with the continuously emitted
of the RQmthreshold value of one upon several occasions.
Pesticides, including herbicides, differ from most industrial
organic compounds in being introduced into the environ-
ment with the explicit intention of exerting effects on one or
toxic action only where they are applied, but can, through
persistence and transport, reach other compartments of the
and Europe have demonstrated the widespread presence of
pesticides in various freshwater bodies (1-6). Over the past
criteria (WQC) for each pesticide in surface waters. Within
the EU, these WQC are often equivalent to the predicted
no-effect concentration (PNEC), which aims to ensure the
overall protection of aquatic life (9, 12, 14). This parameter
is usually estimated by finding the lowest reliable aquatic
effect concentration and applying a safety factor to account
for various uncertainties, such as interspecies differences in
sensitivity, acute-to-chronic ratios, and laboratory-to-field
extrapolations (for review see refs 15-17). The drawback of
this approach is that the PNEC is derived from the lowest
value in the data set and therefore depends strongly on the
latter’s quality and quantity. Currently, large differences in
PNEC values for a given pesticide can be found in the
literature (see for example refs 9, 12, 14). More recently,
as WQC instead of PNEC (for review see ref 18). They are
derived from the species sensitivity distribution (SSD), a
statistical function describing the variation in toxicity
(generally the no-observed-effect concentration; NOEC) of
a certain compound among a set of species. The species set
should represent natural diversity and may be composed of
several species from a specific taxon, a selected species
the calculation of an HC that is assumed to protect a given
percentage of the species in a given environmental com-
protects 95% of the species.
in the ecosystem and organisms are typically exposed to
mixtures of pesticides. Even if each single compound of a
they may exhibit a significant effect as has been shown by
several laboratory (20-23) and field (24) studies with
for herbicides with similar modes of action (22, 25-27). For
this reason, several authors have discussed the necessity of
including mixtures of pesticides in the water quality regula-
tion (18, 22, 28-32). However, so far only Canada (13) has
proposed WQC for pesticide mixtures, and then only for the
short-term occurrence of herbicides (acute WQC).
One obstacle in considering mixtures of herbicides in a
available on herbicides varies greatly in terms of quality and
for single herbicides causes these values to be mutually
inconsistent. For example, in the case of two herbicides A
and B (where A is more toxic than B, A > B), B may have a
lower WQC than A (B > A) if a more extensive data set is
available for A, resulting in a smaller extrapolation factor or
are too incoherent for use in calculating WQC for pesticide
In this study, we propose a methodology to determine
consistent WQC for single substances, which can be com-
bined to calculate a risk quotient for mixtures of herbicides
with similar modes of action (RQm). This RQmmust be less
than one to ensure an acceptable risk to aquatic life and can
thus be considered as a risk indicator for the mixture in a
* Corresponding author e-mail: email@example.com; tel:
+41-44-823.55.75; fax: +41-44-823.55.47.
4269ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 2, 200610.1021/es050239l CCC: $33.50
2006 American Chemical Society
Published on Web 12/13/2005
of action. We did not consider the metabolites, which can
also have a similar mode of action as the parent compounds
(33) but for which very limited toxicity data are available.
The methodology developed and proposed in this paper
must satisfy three main criteria: (i) it should be based on
WQC to be calculated for herbicides with similar modes of
action, and should therefore allow the risk of mixtures to be
consistently assessed, and (iii) since it is intended for
application in Swiss regulations, this methodology should
also be easy to calculate and communicate to facilitate its
In the following two sections we present the concept for
the derivation of the RQmand the individual WQC values. In
the Concept section a recipe-style summary of the concept
is given. In the Materials and Methods section, all equations
are derived and all assumptions are stated. The details on
the datasets, statistics, and modeling are given in the
Supporting Information. In the Results and Discussion
section, we illustrate the concept using the example of two
families of photosystem II inhibitors: the triazines and the
phenylureas. These herbicides target photosystem II, where
they compete with plastoquinone, the electron transporter
the risks in a river of a small Swiss catchment area with
have been regularly detected by chemical analysis. We
conclude with a detailed critical discussion of each of the
improve the applicability of the methodology. In addition,
a sensitivity analysis is performed to better understand to
what extent the assumptions of concentration addition of
mixtures of compounds with similar mode of action influ-
ences the results and conclusions.
Concept for the Derivation of RQm. For a single compound
i, the site-specific risk for aquatic organisms is normally
evaluated by comparing the measured environmental con-
taken as equivalent to HC5-95%, which represents the
lower 95% confidence limit of the HC5 determined on the
(SSDNOEC), (19, 35). The ratio MECito HC5-95%i(RQi, eq 1)
an acceptable risk to the environment.
For mixtures of compounds exhibiting a similar mode of
action, the mixture effect can be predicted by the concept
of concentration addition (CA; 25). The risk quotient for
mixtures of these herbicides (RQm) thus corresponds to the
sum of the individual compounds RQi(eq 2). The RQmmust
remain less than one to ensure that a mixture of chemicals
presents an acceptable risk to the environment:
Concept for the Derivation of a Consistent HC5-95%
Value. The use of HC5-95% instead of HC5 as a WQC has
into account. It therefore reduces the risk of protecting
substantially less than 95% of the species, as could occur
with an HC5 determined from insufficient data. Indeed, the
confidence interval is greater and the HC5-95% value
consequently lower when the quantity of available data
decreases and/or the variability of these data increases (19,
35, 36). However, in some cases, not enough NOEC data are
that a minimum of 10 data points be used, as they showed
that results of SSD analysis appear to stabilize with 10-15
data points. Newman et al. (36) wrote that optimal sample
sizes range from 15 to 55 data points depending on the
chemical and data quality. The Technical Guidance Docu-
ment on Risk Assessment (38) recommends the use of at
least 10 NOEC values (preferably more than 15) for different
species covering at least 8 taxonomic groups. This require-
ment cannot be satisfied for all herbicides commonly found
in surface water. Furthermore, even if sufficient data are
result in HC5-95% values that are not comparable for each
herbicide in a mixture.
To make the HC5-95%iconsistent for each herbicide i
of a mixture even if not enough NOEC data are available,
we propose a methodology using EC50 data and following
three steps (Figure 1). Step 1: The derivation of a relative
potency (RPi) based on the SSD curves of acute effect
concentrations eliciting 50% lethality or 50% growth inhibi-
tion (EC50) after short-term exposure to each herbicide i of
the mixture (SSD-EC50i). The EC50 are used instead of the
Step 2: The derivation of the HC5-95% of the reference
compound (HC5-95%ref) based on the SSD of the NOEC for
this reference compound (SSD-NOECref). The reference
should be the compound for which the largest dataset is
available, in terms of both EC50 and NOEC. Step 3: The
based on its relative potency (RPi) and the HC5-95%ref.
This methodology relies on three hypotheses: (1) The
SSD-EC50 curves are parallel for herbicides with similar
mode of action. (2) The SSD-NOEC curves are parallel to the
(3) The RP of each given herbicide is constant for SSD-EC50
Materials and Methods
Curves. In this first step, the SSD curves are fitted to the
EC50 data to calculate the RPi of each herbicide i. As we
such as algae and aquatic plants, i.e., the most sensitive
species, have been used for calculation (see Supporting
Information). A precondition for the derivation of the RP is
the assumption of parallel SSD-EC50 curves (Hypothesis 1).
The RP is independent of the level of “fraction of species
affected” only if the cumulative SSD curves are parallel. The
SSD-EC50i of each herbicide i is fitted with a log-logistic
regression according to the following:
where HC50EC50,i) hazardous concentration affecting 50%
of species s, for herbicide i.
to have a common value for all the SSD-EC50 of a set of
similarly acting herbicides in a mixture and can be fitted by
the least-squares method to the entire data set of EC50i,s
values for all the herbicides under consideration (see SI for
more information and choice of the model given in eq 3).
< 1 (1)
1 + 10(log HC50EC50,i-log EC50i,s)*slope
VOL. 40, NO. 2, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY9427
The assumption that the slope is the same for all the
herbicides was tested with the extra sum-of-squares F-test
(R ) 0.05), (ref 39; see SI).
The RPiis calculated on the basis of the HC50EC50,i, which
is in the middle of the data distribution and therefore more
robust than extreme values such as the HC5EC50,i or the
i (RPi) with
where HC50EC50,ref) HC50 of the reference, and HC50EC50,i)
HC50 of herbicide i.
The standard deviation of the RPi, σlogRPi, can be
calculated by means of error propagation:
and the 95% confidence interval with
Step 2. Derivation of HC5-95% for the Reference
Compound from SSD-NOECref. In this second step, an SSD
curve is fitted to the NOEC data of a reference compound
(algae and aquatic plants) to calculate the HC5-95%ref(eqs
7 and 8). The reference compound is usually the compound
for which the most extensive EC50 and NOEC data set is
available. Note that in the case of standardized algae test,
the EC50 and the NOEC data are derived from the same
NOEC data were generally independent as provided by
different literature references.
The fraction affected is expressed as follows:
of species s for the reference compound.
By implementing a fraction affected of 5% in eq 7 and
The 95% confidence interval of the HC5refis equal to
where σlogHC5refis the standard deviation of the logHC5ref.
limit of the HC5ref
Step 3. Prediction of HC5-95%ifor each Herbicide i of
the Mixture from SSD-NOECrefand RPi. For this last step,
we assume that the SSD-EC50iand SSD-NOECiare parallel
on the basis of SSD-EC50 is equivalent to the RPicalculated
on the basis of SSD-NOEC (Hypothesis 3).
Following these considerations, the HC5-95% of a
compound i (HC5-95%i) is calculated as follows:
As the HC5-95%ref is already a lower confidence limit,
the standard deviation of logHC5-95%i is equal to the
standard deviation of logRPigiven by eq 5.
The 95% confidence interval of the predicted HC5-95%i
can therefore be calculated from the following:
This value (HC5-95%i) corresponds to the WQC for
herbicide i and will be used to estimate the RQi(eq 1) and
RQm(eq 2) in surface water.
to triazines and phenylureas, on the statistical analysis, and
FIGURE 1. Three-step method to derive the HC5-95% for each
compound in a mixture of herbicides with similar mode of action.
Step 1: Derivation of the relative potency RPifor each herbicide
i of the mixture based on the SSD-EC50 curves of the herbicide i
(SSD-EC50i) and the reference (SSD-EC50ref). Step 2: Derivation of
(SSD-NOECref). Step 3: Prediction of the SSD-NOECi curves and
HC5-95%ifor each herbicide i in the mixture based on its relative
potency i and the SSD-NOECrefcurve, respectively HC5-95%ref.
σlog RPi)?σlog HC50EC50,ref
+ σlog HC50EC50,i
log RPi( 1.96 × σlog RPi
1 + 10(log HC50NOEC,ref-log NOECref,s)*slope(7)
log HC5ref) log HC50NOEC,ref-(
log HC5ref( 1.96 × σlog HC5ref
log HC5-95%ref) log HC5ref- 1.65 × σlog HC5ref(10)
σlog HC5-95%i) σlog RPi
log HC5-95%i( 1.96 × σlog HC5-95%i
4289ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 2, 2006
on the herbicide measurement can be found in the Sup-
Results and Discussion
Concept Applied to Triazines and Phenylureas. The meth-
odology was applied to the triazine and the phenylurea
herbicides registered in Switzerland (Table 1). The mixture
of triazines has been shown to be concentration additive
to follow CA for periphyton as well as the epipsammon
as test organism, however, CA and independent action (IA)
have been found to be equally valid to predict the mixture
toxicity of phenylureas (27). Both groups of herbicides are
photosystem II inhibitors, but strictly speaking do not have
the same mode of toxic action, as the target site on the
TABLE 1. Features of the Triazine and Phenylurea Herbicides Registered in Switzerlanda
aFrom http://www.faw.ch/index.htm, state in 2003. They all act as photosystem II inhibitors. Chlorbromuron, dimefuron, and metobromuron
are also registered in Switzerland. However, they are not included in the list because of insufficient ecotoxicity data (both EC50 and NOEC) to apply
the approach described in this study. The herbicide usage data for the Greifensee Catchment area are shown, as well as whether it was measured
there. The herbicides prometryn and propazine were measured before 1999 and found in concentrations just above the detection limit. However,
they are no longer registered in Switzerland and are therefore not included in the list.
VOL. 40, NO. 2, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY9429
photosystem II is not exactly the same, but in very close
proximity (plastoquinone binding sites QA and QB, the
acceptor sides on PSII; (41)). However, it has been observed
that if both groups of herbicides are mixed together, the
assumption of grouping them together in a group of similar
mode of action is justified.
and the phenylureas are parallel, as discussed above, the
slope parameter was adjusted to 1.4 (Table 2), which
corresponds to the best common slope estimate for all the
slope is the same for all the herbicides, which assumption
was rejected (extra sum-of-squares F-test, R ) 0.05; see SI),
but the assumption that the best fit slope is equal to the
common slope for each dataset was accepted for two-thirds
of the herbicides. However, the extra sum-of-squares F-test
is an asymptotic test, which means that it provides only an
approximation of the “true” P value. This approximation
improves with increasing observations and can be biased if
only few data are available (43), which is the case here for
some herbicides. Examination of curves by eye, even if the
visual perception of a graph could be discussed, suggests
fairly parallel curves. Furthermore, as we will show below,
the SSD-NOEC curves predicted with this approach are in
good agreement with the available NOEC data of some
herbicides. Since the assumption that the SSD-EC50 curves
are parallel is essential to our approach and as insufficient
to apply the best fit common slope. Future research might
be necessary to test this assumption.
The relative ranking of toxicity between the triazines and
by the SSD curves in Figures 2A and 3A is as follows: diuron
< terbutryn < isoproturon ) linuron < terbuthylazine )
metribuzine < cyanazine ) chlortoluron < atrazine <
metoxuron < simazine. Faust et al. (22) found about the
same ranking based on EC50 in the growth inhibition test
with the algae Scenedesmus vacuolatus for the same herbi-
cides, except for isoproturon and metoxuron. In this study,
less toxic than simazine. However, Michel et al. (44) found
that isoproturon was more toxic than atrazine, with an RP
factor of about 3 for Lemna paucicostata in a 7-day growth
test. These results show that the RP should not be based on
a single toxicity value. When more than one species is taken
into account, the SSD-EC50 of isoproturon and metoxuron
clearly show that they are more toxic than atrazine, and
consequently than simazine (Figure 3A).
In this study, atrazine has been chosen as the reference
set for EC50 (30 data) and NOEC (18 data; see SI). The HC5-
TABLE 2. Parameter Estimates for the Best Fit Model (Slope and HC50 as Adjustable Parameters) and for the Model Fitted with
Constraint (Slope Fixed, HC50 as Adjustable Parameter) for (A) the Triazines, and (B) the Phenylureasa
(reference) terbuthylazine simazinecyanazine metribuzine terbutryn
number of EC50
30310 316 3
(slope fixed to 1.4)
linuronchlortoluron metoxuron diuronisoproturon
number of EC50
aAlso given are the RPivalues with atrazine as reference (column 2) in (A) as well as the HC5 estimate for this reference. The last line shows
the HC5-95% for the various herbicides. All the values are given with their 95% confidence interval in parentheses.bNot determined.c18 NOEC
4309ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 2, 2006
95% for atrazine was calculated on the basis of the SSD-
NOEC constructed with ecotoxicity values collected by Van
den Brink and Brock (45). To ensure consistency of the
methodology, the SSD-NOEC curve for atrazine was fitted
with the same slope as the SSD-EC50 (1.4; hypothesis 2).
the best fit slope value of 1.42 (see Supporting Information).
For atrazine, the ratio between the HC50EC50 and the
HC50NOEC, which corresponds to the acute-to-chronic ratio
(ACR) discussed for hypothesis 3, is equal to 8. It is of the
same order of magnitude as the ACR of about 10 evaluated
by Posthuma et al. (19). However, this latter ACR value was
derived from SSD curves fitted to data representing all the
species and not only primary producers, which can explain
the small difference with the ACR obtained in this study.
The HC5-95% (SSD-NOEC) of each triazine and phen-
ylurea is calculated on the basis of the HC5-95% of the
reference atrazine and of the RP value for each herbicide
(Table 2). These predicted values are proposed to be used
as WQC for the single herbicides. The predicted SSD-NOEC
3B and 4B, respectively. To assess the adequacy of the
predictions, the predicted SSD-NOEC were compared with
as no statistical comparison is possible due to the few
available data, shows that the agreement is quite good.
Summing up all evidence, we can conclude that the choice
of the assumptions is justified.
Dynamic RQm of Triazines and Phenylureas in Aa
diuron, that were continuously measured in the Aa Mo ¨n-
chaltorf river in 1999. The risk for the aquatic ecosystem is
acceptable if RQm has a value less than one. As shown in
Figure 4A, the RQmexceeded the critical value of one several
individual RQiremain less than one (with two exceptions).
This finding is due to two principle reasons:
overlapping of the risk of seasonal herbicides, and (2) the
constant risk contribution of nonseasonal herbicides.
As shown in Figure 4A, isoproturon occurred in the river
FIGURE 2. (A) Species sensitivity distribution (SSD) of the EC50
data [µg L-1] for the triazines and (B) predicted SSD of the NOEC
data [µg L-1] based on the RPiand with atrazine as reference; (C)
comparison of the predicted SSD-NOEC curves for simazine and
cyanazine with NOEC data.
FIGURE 3. (A) Species sensitivity distribution (SSD) of the EC50
data [µg L-1] based on the RPiand with atrazine as reference; (C)
comparison of the predicted SSD-NOEC curves for isoproturon and
linuron with NOEC data.
VOL. 40, NO. 2, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY9431
period; the RQmtherefore shows a first increase during this
period, reaching a RQmvalue of 2.5. Atrazine occurs later,
RQmin the period from June to July. Nonpoint releases of
herbicides after application have been studied extensively
of herbicide concentrations in surface water may present a
risk to the aquatic life. In addition to the risk from single
seasonal herbicides, however, there is an overlapping phase
of atrazine and isoproturon in May, which contributes to
maintaining the risk above the threshold of one even in
periods of low concentrations for both seasonal substances.
These results highlight the fact that, even if herbicides are
and atrazine), and even if the risk of a single herbicide is not
ecosystem, we should control the simultaneous occurrence
of herbicides with similar mode of action in the water and
pay attention to herbicide mixtures applied directly to the
The constant presence of diuron (Figure 4B) contributes
to a further increase of the RQmduring the whole period, as
it is continuously present in surface water. This continuous
occurrence of diuron may result from its use in paints and
coatings: it is leached from facades by rainwater and enters
the surface water through the urban drainage system (51).
Diuron is not the only herbicide with an urban source, and
may be important, especially in combination with peak
concentrations of seasonally applied agricultural herbicides
Critical Evaluation of the Hypotheses Underlying the
Concept. Hypothesis 1: The SSD-EC50 curves (and SSD-
NOEC curves) are parallel for herbicides with similar mode
of action. For this hypothesis, we assume that the sensitivity
of action is the same, and that the difference between
herbicides is only highlighted by a shift of the distribution,
which can be explained in terms of a higher potency of the
respective substance. In other words, the RP are assumed to
be independent of the level of “fraction of species affected”.
systematically studied SSD curves. Vaal et al. (56, 57) have
shown that specifically acting compounds had wider SSD
(mean ( standard deviation). However, these results are
In toxicity testing of single species, common slopes of
concentration-response curves have often been assumed
for compounds with the same mode of action although no
evidence has been found to confirm this assumption (58).
As a pragmatic solution to this dilemma, three options were
to one for all herbicides. Second, the best-fit slope of the
reference compound was used as a fixed slope for deriving
the SSD for the other compounds. Third, a common best-fit
slope was calculated for all the compounds. We opted for
this third approach because it satisfies the conditions for
deriving RP with the fewest assumptions and leads to
consistent results (see the concept applied to triazines and
phenylureas). However, there is undoubtedly a need for
Hypothesis 2: The SSD-NOEC curves are parallel to the
for SSD-EC50 and SSD-NOEC. If we combine both hypoth-
eses, we assume that a constant factor exists between EC50
and NOEC for each species tested, i.e., the ACR is constant
for all species and compounds with similar mode of action.
the ACR can be calculated based on the available EC50 and
NOEC data (59). A generic value of 10 is often used (16, 60).
16) for review) as the ACR value can greatly vary depending
on mode of action (61), or if the mode of action is different
in acute and chronic endpoints (60). For algal toxicity
however, which is relevant in the context of this study, a
constant ACR is more likely because EC50 and NOEC data
are typically both derived from the same growth inhibition
compounds showed that the assumption of a constant ACR
is reasonable for atrazine and malathion, but not for
chlorpyriphos (19), if all the available data are included in
study, an ACR of 8 was found for atrazine between SSD-
EC50 and SSD-NOEC which is in agreement with the results
by Posthuma et al. (19).
Sensitivity Analysis of the Model: Concentration Ad-
dition (CA) versus Independent Action (IA) for Predicting
the RQm. In this paper, we applied the CA model to estimate
the RQm of triazines and phenylureas (eq 2), which have
been shown to be concentration additive (see above).
one may argue that the use of CA is too conservative for risk
assessment and that the alternative model of independent
action (IA) should be used for the calculation. To test how
conservative the assumption of CA is compared to IA, we
also evaluated the effects of the five herbicides using the IA
model. This model assumes that the components of the
in Aa Mo 1nchaltorf (continuous line) compared to the single risk
(B) diuron (dashed line).
E(cmix) ) E(c1+ ... + cn) ) 1 -∏
[1 - E(ci)] (14)
4329ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 2, 2006
where E(cmix) is the overall effect (scaled to the range 0-1)
of a mixture composed of n compounds at a total concen-
tration of cmix. The application of this model allows the
risk, such as the RQm (eq 14). However, an RQm of one
corresponds to the percentage of species affected by a
concentration equal to the HC5-95%NOEC,ref (eq 2). In this
study, an RQmof one corresponds to 4% of species affected.
The comparison of the effects of the five herbicides with
bands of the predictions were estimated by the re-sampling
of eq 15 are assumed to follow a log-normal distribution,
to be normally distributed.
The IA prediction is always lower than the CA prediction,
which is in accord with mixture theory (32, 62). Compared
to the upper confidence limit of IA, CA predictions differ by
a maximum factor of 1.3 during the whole period of
the argument that CA is over-conservative compared to IA.
This leads to the conclusion that CA is a pragmatic and
justifiable assumption for estimating the risk of herbicides
with similar mode of action in surface waters.
The approach we present here allows a reliable and robust
ecotoxicological limit value to be calculated for single
herbicides and hence limit values to be defined for mixtures
of herbicides having the same mode of action in surface
water. In our opinion, the combination of both individual
and mixture criteria is better than the use of the current
criterion of 0.1 µg/L for individual pesticides in Switzerland
(63), which was arbitrarily defined without ecotoxicological
justification. One could argue that the proposed individual
WQC are higher than the current criteria, thus being less
protective, but this is no justification for not taking ecotoxi-
cases. Furthermore, for assessing risk of mixtures, it is
important to account for the differences of the relative
potencies of the different herbicides.
We are also convinced that this approach is better than
the proposed risk quotient for mixtures can only be applied
be grouped yet, an individual threshold value might still be
the risk quotients of the single substances (RQi) and the
mixture (RQm) to be assessed. The time course of RQmin a
small river typical of an agricultural region in Switzerland
highlights the strength of the proposed approach. In par-
ticular, the RQmclearly shows the concurrent occurrence of
various herbicides in surface water, resulting in a nonneg-
if the single herbicides do not present a risk on their own.
of action, i.e., groups of insecticides, or fungicides, should
be developed in a next step. In addition, metabolites, which
can occasionally be as equally toxic as their parent com-
pounds (33, 64), have not been considered here. Such
considerations show that defining limit values based on no-
effect values for single substances is only a first step to
protecting aquatic ecosystems.
Furthermore, it is important to note that we only
Deneer (65), and recently Junghans et al. (62), have shown,
for binary mixtures and environmentally relevant mixtures
and concentrations, respectively, the suitability of the
concentration addition model (CA) to predict the effects of
mixtures of herbicides even if they do not show a similar
mode of action. This is especially true if large numbers of
compounds are evaluated together (32). Therefore, if all
herbicides present in the river at the same time are taken
The inclusion of mixture considerations in the WQC for
herbicides in surface water is therefore strongly recom-
mended. However, the implications for agriculture could be
drastic, possibly leading to recommendations of load reduc-
tion or the ban of some herbicides. From a pesticide
management perspective, our results show that the applica-
of herbicides that are already problematic as single sub-
This project was funded by the Swiss Agency for the
Environment, Forests and Landscape (SAEFL). We thank
Edwin Mu ¨ller and Stephan Mu ¨ller from SAEFL for helpful
discussions. We also thank Marion Junghans (Eawag) and
Munro Mortimer (Queensland EPA, Australia) for reviewing
the paper, Alessandra Brazzale (Institute of Biomedical
Engineering, Padova, Italy) for reviewing the statistics, and
Christoph Steiner (Eawag) for his help with the illustrations.
Supporting Information Available
Description of the data used in this paper, information on
the statistical analysis performed, description of the herbi-
cides measurements in surface water, and a glossary. This
material is available free of charge via the Internet at
FIGURE 5. Prediction of the fraction of species affected at the
measured environmental concentrations with the models of (A)
lines) with their 95% confidence bands (continuous gray lines).
VOL. 40, NO. 2, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY9433
(1) Garmouna, M.; Teil, M. J.; Blanchard, M.; Chevreuil, M. Spatial
concentrations in the catchment basin of the Marne river
(France). Sci. Total Environ. 1998, 224, 93-107.
(2) Kreuger, J. Pesticides in streamwater within agricultural catch-
(3) Capel, P. D.; Larson, S. J.; Winterstein, T. A. The behaviour of
39 pesticides in surface waters as a function of scale. Hydrol.
Process 2001, 15, 1251-1269.
(4) Mu ¨ller, K.; Bach, M.; Hartmann, H.; Spiteller, M.; Frede, H.-G.
Point- and nonpoint-source pesticide contamination in the
Zwester Ohm catchment, Germany. J. Environ. Qual. 2002, 31,
(5) Leu, C.; Singer, H.; Stamm, C.; Mu ¨ller, S. R.; Schwarzenbach, R.
P. Simultaneous assessment of sources, processes, and factors
influencing herbicide losses to surface waters in a small
agricultural catchment. Environ. Sci. Technol. 2004, 38, 3827-
(6) Leu, C.; Singer, H.; Stamm, C.; Mu ¨ller, S. R.; Schwarzenbach, R.
P. Variability of herbicide losses from 13 fields to surface water
within a small catchment after a controlled herbicide applica-
tion. Environ. Sci. Technol. 2004, 38, 3835-3841.
(7) Stephan, C.; Mount, D.; Hansen, D.; Gentile, J.; Chapman, G.;
Brungs, W. Guidelines for Deriving Numerical National Water
Quality Criteria for the Protection of Aquatic Organisms and
Their Uses; NTIS PB85-227049; U.S. Environmental Protection
Agency: Washington, DC, 1985.
(8) Nowell, L.; Resek, E. National standards and guidelines for
pesticides in water, sediment, and aquatic organisms: applica-
tion to water-quality assessment. In Reviews of Environmental
Contamination and Toxicology, Volume 140; Ware, G. W., Ed.;
Springer-Verlag: New York, 1994; pp 1-164.
(9) Roussel, P. Syste `me d’e ´valuation de la qualite ´ des cours d’eau.
quality);Agencedel’eauLoire-Bretagne: Orle ´ans,France,1999.
(10) Zabel, T.; Cole, S. M. The derivation of environmental quality
standards for the protection of aquatic life in the UK. J. CIWEM
1999, 13, 436-440.
(11) Crommentuijn, T.; Sijm, D.; De Bruijn, J. H.; Van Leeuwen, K.;
Van de Plassche, E. Maximum permissible and negligible
concentrations for some organic substances and pesticides. J.
Environ. Manage. 2000, 58, 297-312.
(12) Jahnel, J.; Zwiener, C.; Gremm, T. J.; Abbt-Braun, G.; Frimmel,
F. H.; Kussatz, C.; Schudoma, D.; Rocker, W. Zielvorgaben fu ¨r
Pflanzenschutzmillelwirkstoffe und andere Schadstoffe in Ober-
fla ¨chengewa ¨ssern(Waterqualitycriteriaforpesticidesandother
pollutants). Acta Hydrochim. Hydrobiol. 2001, 29, 246-253.
(13) Ministe `redel’environnementduQue ´bec.Crite `resdequalite ´ de
l’eau de surface au Que ´bec (Water quality criteria for Que ´bec);
Ministe `re de l’environnement du Que ´bec: Que ´bec, Canada,
(14) Babut, M.; Bonnet, C.; Bray, M.; Flammarion, P.; Garric, J.;
Golaszewski, G. Developing environmental quality standards
for various pesticides and priority pollutants for French
freshwaters. J. Environ. Manage. 2003, 69, 139-147.
and Hazard Assessment: Sixth Symposium; Publication 802;
ASTM Special Technical Publication: Philadelphia, PA, 1983.
of safety (uncertainty) factors for ecological risk assessment.
Environ. Toxicol. Chem. 1998, 17, 99-108.
(17) Whitehouse, P. Measures for protecting water quality: current
2001, 50, 115-126.
(18) Reiley, M. C.; Stubblefield, W. A.; Adams, W. J.; Di Toro, D. M.;
Hodson, P. V.; Erickson, R. J.; Keating, F. J. J. Reevaluation of
SETAC: Pensacola, FL, 2003.
(19) Posthuma, L.; Suter, G. W. I.; Traas, T. P. Species Sensitivity
Distribution in Ecotoxicology; Lewis Publishers: Boca Raton,
(20) Faust, M.; Altenburger, R.; Boedecker, W.; Grimme, L. H. Algal
toxicity of binary combinations of pesticides. Bull. Environ.
Contam. Toxicol. 1994, 53, 134-141.
M.; Grimme, L. H. Predictability of the toxicity of multiple
mixture of dissimilarly acting chemicals to Vibrio fischeri.
Environ. Toxicol. Chem. 2000, 19, 2348-2356.
(22) Faust, M.; Altenburger, R.; Backhaus, T.; Blanck, H.; Boedeker,
W.; Gramatica, P.; Hamer, V.; Scholze, M.; Vighi, M.; Grimme,
L. H. Predicting the joint algal toxicity of multicomponent
s-triazine mixtures at low-effect concentrations of individual
toxicants. Aquat. Toxicol. 2001, 56, 13-32.
W.; Gramatica, P.; Hamer, V.; Scholze, M.; Vighi, M.; Grimme,
L. H. Joint algal toxicity of 16 dissimilarly acting chemicals is
predicatble by the concept of independent action. Aquat.
Toxicol. 2003, 63, 43-63.
(24) Arrhenius, A.; Gro ¨nvall, F.; Scholze, M.; Backhaus, T.; Blanck,
H. Predicatbility of the mixture toxicity of 12 similarly acting
congeneric inhibitors of photosystem II in marine periphyton
M.; Grimme, L. H. Predictability of the toxicity of multiple
chemical mixtures to Vibrio fischeri: mixtures composed of
similarly acting chemicals. Environ. Toxicol. Chem. 2000, 19,
(26) Junghans, M.; Backhaus, T.; Faust, M.; Scholze, M.; Grimme, L.
H. Predictability of combined effects of eight chloroacetanilide
herbicides on algal reproduction. Pestic. Manage. Sci. 2003, 59,
(27) Backhaus, T.; Faust, M.; Scholze, M.; Gramatica, P.; Vighi, M.;
Grimme, L. H. Joint algal toxicity of phenylurea herbicides is
action. Environ. Toxicol. Chem. 2004, 23, 258-264.
(28) Calamari, D.; Pacchetti, G. Population parameters as indices of
effects of mixtures of chemicals on aquatic populations: the
of Chemicals; Vouk, V. B., Butler, G. C., Upton, A. C., Parke, D.
(29) Thompson, H. M. Interactions between pesticides; a review of
reported effects and their implications for wildlife risk assess-
ment. Ecotoxicology 1996, 5, 59-81.
(30) Grimme,L.H.;Altenburger,R.;Backhaus,T.;Faust,M.;Bo ¨deker,
W.; Scholze, M. Kombinationswirkungen von Umweltchemi-
kalien in der O ¨ kotoxikologie (Combined effects from environ-
mental chemicals in ecotoxicology). Z. Umweltchem. O ¨kotox.
2000, 12, 226-234.
(31) U.S. EPA. Guidance on Cumulative Risk Assessment of Pesticide
Chemicals that have a Common Mechanism of Toxicity; U.S.
Environmental Protection Agency, Office of Pesticide Pro-
grams: Washington, DC, 2002.
(32) Backhaus, T.; Altenburger, R.; Arrhenius, A.; Blanck, H.; Faust,
M.; Finizio, A.; Gramatica, P.; Grote, M.; Junghans, M.; Meyer,
W.; Pavan, M.; Porsbring, T.; Scholze, M.; Todeschini, R.; Vighi,
M.; Walter, H.; Grimme, L. H. The BEAM-project: prediction
Cont. Shelf Res. 2003, 23, 1757-1769.
Prentice-Hall: Englewood Cliffs, NJ, 1993.
(35) Aldenberg, T.; Slob, W. Confidence limits of hazardous con-
Ecotoxicol. Environ. Safe. 1993, 25, 48-63.
(36) Newman, M. C.; Ownby, D. R.; Me ´zin, L. C.; Powell, D. C.;
Christensen, T. R.; Lerberg, S. B.; Anderson, B.-A. Applying
species-sensitivity distributions in ecological risk assessment:
assumptions of distribution type and sufficient number of
species. Environ. Toxicol. Chem. 2000, 19, 508-515.
M. Species sensitivity distributions: data and model choice.
Mar. Pollut. Bull. 2002, 45, 192-202.
(38) European Commission. Technical Guidance Document on Risk
Assessment. TGD Part II; Institute for Health and Consumer
(EC): Ispra, Italy, 2003.
using Linear and Nonlinear Regression; GraphPad Software
Inc.: San Diego, CA, 2003.
Chemicals 201; Organisation for Economic Co-operation and
Development (OECD): Paris, France, 1984.
(41) Peterson, D. E.; Regehr, D. L.; Thompson, C. R.; Al-Khatib, K.
KS, 2001. Accessible at www.oznet.ksu.edu/library/crpsl2/
4349ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 2, 2006
of multiple mixtures of triazines and phenylureas; Poster
presentation at the 9th Annual Meeting of SETAC Europe, 25-
29 May, 1999, Leipzig, Germany, 1999.
(43) Seber, G. A. F.; Wild, C. J. Nonlinear Regression; John Wiley &
Sons: New York, 1989.
(44) Michel, A.; Johnson, R. D.; Duke, S.; Scheffler, B. E. Dose-
of action and growth of Lemna paucicostata: an improved
ecotoxicological method. Environ. Toxicol. Chem. 2004, 23,
(45) Van den Brink, P. J.; Brock, T. C. M. Derivation of short- and
long-term quality standards for atrazine and simazine; Alterra
Green World Research: Wageningen, The Netherlands, 2002.
(46) Klaine, S. J.; Hinman, M. L.; Winkelmann, D. A.; Sauser, K. R.;
Martin, J. R.; Moore, L. W. Characterization of agricultural
nonpoint pollution: pesticide migration in the West Tennesse
watershed. Environ. Toxicol. Chem. 1988, 7, 609-614.
(47) Thurman, E. M.; Goolsby, D. A.; Meyer, M. T.; Kolpin, D. W.
Herbicides in surface waters of the midwestern united states:
(48) Garmouma, M.; Blanchard, M.; Chesterikokk, A.; Ansart, P.;
Chevreuil, M. Seasonal transport of herbicides (triazines and
phenylureas) in small stream draining an agricultural basin:
Me ´larchez (France). Water Res. 1997, 31, 1489-1503.
(49) Donald, W. W.; Hjelmfelt, A. T. J.; Alberts, E. E. Herbicide
distribution and variability across Goodwater Creek watershed
in North Central Missouri. J. Environ. Qual. 1998, 27, 999-
(50) Leu, C. Sources, processes and factors determining the losses
of atrazine, dimethenamid and metolachlor to surface waters:
Thesis no 15156, Swiss Federal Institute of Technology (ETH),
Zu ¨rich, Switzerland, 2003. Accessible at: http://e-collection.
ethbib.ethz.ch/cgi-bin/show.pl?type ) diss&nr ) 15156.
(51) Gerecke, A.; Scha ¨rer, M.; Singer, H. P.; Mu ¨ller, S. R.;
Sources of pesticides in surface waters in Switzerland: pesti-
cide load through waste-water treatment plants-current situ-
ation and reduction potential. Chemosphere 2002, 48, 307-
concentrations of urban pesticides. In Current Practices in
Modelling the Management of Stormwater Impacts; James, W.,
Ed.; Lewis Publishers: Boca Raton, FL, 1994; pp 85-98.
in eight U.S. urban streams. Environ. Toxicol. Chem. 2000, 19,
(54) Revitt, D. M.; Ellis, J. B.; Llewellyn, N. R. Seasonal removal of
herbicides in urban runoff. Urban Water 2002, 4, 13-19.
(55) Schiff, K.; Bay, S.; Stransky, C. Characterization of stormwater
toxicants from an urban watershed to freshwater and marine
organisms. Urban Water 2002, 4, 215-227.
(56) Vaal, M. A.; Van der Wal, J. T.; Hoekstra, J. A.; Hermens, J.
Variation in the sensitivity of aquatic species in relation to the
classification of environmental pollutants. Chemosphere 1997,
(57) Vaal, M. A.; Van Leeuwen, C. J.; Hoekstra, J. A. Variation in
sensitivity of aquatic species to toxicants: practical conse-
Manage. 2000, 25, 415-423.
(58) Smit, M. G. D.; Hendriks, A. J.; Schobben, J. H. M.; Karman, C.
C.; Schobben, H. P. M. The variation in slope of concentration-
effect relationships. Ecotoxicol. Environ. Safe. 2001, 48, 43-50.
(59) Brix, K. V.; DeForest, D. K.; Adams, W. J. Assessing acute and
chronic copper risks to freshwater aquatic life using species
Toxicol. Chem. 2001, 20, 1846-1856.
(60) Roex, E.; Van Gestel, C.; Van Wezel, A.; Van Straalen, N. M.
growth rates in relation to toxicant mode of action. Environ.
Toxicol. Chem. 2000, 19, 685-693.
Assessment of Chemicals; Hill, I. R., Heimbach, F., Leeuwangh,
P., Matthiessen, P., Eds.; Lewis Publishers: Boca Raton, FL,
1994; pp 379-397.
relevant toxicants: Validation of prognostic concepts for
Ph.D. Thesis, University of Bremen, Bremen, Germany, 2004.
Accessible at http://elib.suub.uni-bremen.de.
(63) SwissConfederation.Gewa ¨sserschutzverordnungofOctober28,
(64) Boxall, A. B. A.; Sinclair, C.; Fenner, K.; Kolpin, D.; Maund, S.
J. When synthetic chemicals degrade in the environment.
Environ. Sci. Technol. 2004, 38, 368A-375A.
(65) Deneer, J. W. Toxicity of mixtures of pesticides in aquatic
systems. Pestic. Manage. Sci. 2000, 56, 516-520.
Received for review February 4, 2005. Revised manuscript
received October 17, 2005. Accepted October 21, 2005.
VOL. 40, NO. 2, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY9435