Is mixture toxicity measured on a biomarker indicative of what happens on a population level? A...
Transcript of Is mixture toxicity measured on a biomarker indicative of what happens on a population level? A...
ARTICLE IN PRESS
0147-6513/$ - se
doi:10.1016/j.ec
�CorrespondE-mail addr
Ecotoxicology and Environmental Safety 67 (2007) 323–332
www.elsevier.com/locate/ecoenv
Highlighted Article
Is mixture toxicity measured on a biomarker indicative of what happenson a population level? A study with Lemna minor
Nina Cedergreena,�, Majid Abbaspoora, Helle Sørensenb, Jens C. Streibiga
aDepartment of Agricultural Sciences, The Royal Veterinary and Agricultural University, Hoejbakkegaard Alle 13, 2630 Taastrup, DenmarkbDepartment of Natural Sciences, The Royal Veterinary and Agricultural University, Torvaldsensvej 40, 1870 Frederiksberg C, Denmark
Received 29 September 2006; received in revised form 30 November 2006; accepted 14 December 2006
Available online 29 January 2007
Abstract
For plants, pigment content has shown to be a remarkably consistent biomarker across chemicals with different modes of action. In
this study, we evaluated the use of pigment content as endpoint in binary mixture toxicity studies compared to three growth endpoints on
the floating plant Lemna minor. Six binary combinations of six herbicides with different mode of action were used. Data were tested
against both the concentration addition (CA) and independent action (IA) reference models. For CA, two statistical approaches were
used. The study showed that for some herbicide combinations the mixture toxicity measured on pigment content did not reflect the
results measured on plant population growth, emphasizing the importance of measuring growth in parallel with biomarkers. CA
explained the data just as well as IA, and the two different statistical models used to test the data in relation to CA showed very similar
results.
r 2007 Elsevier Inc. All rights reserved.
Keywords: Mixture toxicity; Endpoints; Concentration addition; Independent action; Statistical models; Herbicides; Lemna minor
1. Introduction
Studies of the biological effect of mixtures of chemicalsare an area of interest of companies and end users who aimat increasing the biological efficiency of chemicals, as wellas of risk assessors trying to quantify the effect of chemicalmixtures on non-target organisms. Previous studies haveshown that the biological effect of a mixture can depend onwhich trait is measured, even when only growth measuresare used (Cedergreen and Streibig, 2005). From a visualassessment of affected plants, it is clear that plant pigmentschange in different ways depending on the mode of actionof the toxicant. Biomarkers such as pigment content,primary chlorophyll and carotenoids, are commonly usedas endpoints for toxicity tests. Chlorophyll content is oneof the recommended endpoints in the standard OECD-guideline for Lemna-tests (Sims et al., 1999) and isfrequently used as an endpoint for other macrophytespecies (Brain et al., 2004; Turgut and Fomin, 2002; Viet
e front matter r 2007 Elsevier Inc. All rights reserved.
oenv.2006.12.006
ing author. Fax: +4535 28 21 75.
ess: [email protected] (N. Cedergreen).
and Moser, 2003). In a study with the macrophyteMyriophyllum aquaticum, including 17 pesticides represent-ing 12 different modes of action, eight endpoints weremeasured. The three pigment endpoints: chlorophyll a, b
and carotenoids, were the only endpoints sensitive enoughto determine an EC50 for all pesticides within the 2-weekduration of the experiments (Turgut and Fomin, 2002).Hence, changes in pigment concentration seem to occurconsistently across chemicals with different modes ofaction. Therefore, pigment changes could be an excellentendpoint when evaluating the joint toxicity of mixtures ofchemicals with different modes of action. In addition, theuse of chlorophyll fluorescence as a toxicity endpoint bothfor aquatic as for terrestrial plants is starting to emerge,and shows promising results in terms of correlations withbiomass growth both for individual compounds and forchemical mixtures (Christensen et al., 2003; Juneau et al.,2003; Marwood et al., 2001).Although many of these biomarkers are directly related
to the health of the organism in relation to collection ofenergy or protection of the cell, the question still remains asto whether the joint effect of chemicals of different mode of
ARTICLE IN PRESSN. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332324
action, measured using biomarkers such as pigmentcontents, corresponds to the joint effect of chemicals whenmeasured on individual plants or plant population growthrates.
To evaluate the joint effects of chemicals, comparison ofobserved results must be compared to a reference model.Two fundamental concepts are generally evaluated andeach be summarized as a model (Greco et al., 1995). Thefirst model, Concentration Addition (CA), assumes thatthe chemicals have the same molecular site of action andtherefore can be regarded as dilutions of one another. Thejoint effect of two chemicals is therefore expected to beequal to the effect of the sum of the chemicals, when thepotency of the chemicals (the degree of ‘‘dilution’’) hasbeen accounted for (Greco et al., 1995). If the effect of achemical mixture can be described with CA the chemicalsare said to be additive. Though the theory assumes similarmolecular target sites, it has been suggested that CA couldalso apply to mixtures of chemicals that are able to cause acommon toxicological response (Berenbaum, 1989). Whenit comes to integral endpoints such as death or growth, thismay apply to almost all chemicals (Faust et al., 2003). Themodel of independent action (IA) is based on test-systemswith binominal endpoints such as dead/alive and assumesthat the chemicals act independently. Under these assump-tions the fractional effects of individual chemicals (e.g.50% response) are expected to be independent of eachother in a probabilistic sense. This means that if twochemicals are added at a concentration that can kill 50% ofthe test-organisms when applied individually, chemical Awill kill 50% of the organisms and chemical B will kill 50%of what is left, leaving 25% of the test organisms alive(Greco et al., 1995). While IA is considered of theoreticalimportance, there is debate concerning its practicalrelevance when it comes to assessing joint effects oncontinuous endpoints measured on whole organisms(Berenbaum, 1989; Greco et al., 1995). Although bothmodels have proven to be valid in predicting joint effects ofchemicals on algae and bacteria (Altenburger et al., 2000;Backhaus et al., 2000; Faust et al., 2001, 2003), CA is oftenthe preferred model for risk assessment purposes. If achemical mixture gives a higher effect than predicted byone of the two models they are said to be synergistic inrelation to either CA or IA (or both). If the joint effect oftwo chemicals is lower than predicted, it is calledantagonism. Evaluating different endpoints, one couldexpect that biomarkers that are directly affected by achemical, such as chlorophyll often is by herbicidesaffecting photosynthesis, would follow the model of IA.Integrated endpoints, such as growth, could be more likelyto follow the model of CA.
The aim of the present study was to evaluate the jointeffect of nine binary mixtures of herbicides with differentmode of action on three growth endpoints and two pigmentendpoints. The dose–response surface data was statisticallyanalyzed for synergy and antagonism both in relation toCA and IA, to evaluate which model best describes the
data for the different endpoints. In addition, data wasevaluated at an EC50 level in relation to CA usingisobolograms to visually quantify the degree of a potentialsynergistic or antagonistic interaction. Thereby the con-gruency of two recently published statistical methods forevaluating binary dose–response surfaces was evaluated(Jonker et al., 2005; Sørensen et al., 2007).
2. Materials and methods
2.1. Lemna test
The L. minor test closely follows the guidelines given by the
International Organization for Standardization (International Organiza-
tion for standardization, 2004), with a few modifications. The L. minor
plants were collected in Copenhagen in 1999, surface sterilized according
to Landolt and Kandeler (1987) and the resulting sterile clone was kept in
Ehrlenmeyer flasks containing ‘‘K’’-medium (Maeng and Khudairi, 1973).
Test conditions were held at pH 5, 24 1C and a continuous photon flux
density of 85–120mmolm�2 s�1 (PAR). The flasks and medium were
sealed with cotton wool and autoclaved before the weekly transfer of
plants to new media. Regular tests on K2Cr2O4, and on 3,5-dichlorophe-
nol (CAS-Nr.: 591-35-5) as part of the ISO-ring test, have proven the
Copenhagen L. minor clone equally sensitive to plants from standard
clones. The K-medium was used, as it has proven to give the highest
growth rates in the used experimental setup.
For the Lemna tests, three mixture ratios were used. Each concen-
tration–response curve consisted of 6 dilutions, each in three replicates.
There were 6–12 controls that were shared by all five concentration–re-
sponse curves. Nine binary combinations of the following six herbicides
were made: acifluorfen, diquat, glyphosate, mecoprop, mesotrione and
terbuthylazine. Details on the herbicides are given in Table 1. The
herbicide combinations were: acifluorfen/diquat, acifluorfen/mesotrion,
diquat/mesotrione, glyphosate/mecoprop, glyphosate/mesotrione, glypho-
sate/terbuthylazine, mecoprop/mesotrion, mecoprop/terbuthylazine and
mesotrion/terbuthylazien. Experiments using glyphosate/mecoprop, me-
coprop/mesotrion and mecoprop/terbuthylazine combinations were re-
peated to explore the reproducibility of the results. The dose ranges of the
herbicides were chosen based on previous experiments where area-specific
growth rates were measured (Cedergreen and Streibig, 2005). The
experiments were performed in 6-well tissue culture (TC) test plates
(CM. Lab. Aps, Vordingborg, Denmark) and were initiated by
transferring one L. minor frond to 10ml K-medium (Maeng and Khudairi,
1973) containing a selected concentration of herbicide. The plants were
photographed with a digital camera alongside a 1� 1 cm white plastic
square, and total frond surface area was determined by pixel counts using
the computer program Photoshop 5.0 (Adobe). Flasks were then placed in
a growth cabinet at 24 1C and a continuous photon flux density of
85–120mmolm�2 s�1 (PAR). Six samples of plants, similar to the
incubation plants were photographed, dried at 80 1C for 24 h and weighed,
to determine the start surface area/dry weight ratio. After 7 days, the
L. minor plants were photographed and frond surface area and frond
number were determined. The plants were then dried and weighed.
Relative growth rates were calculated according to: (lnAT–lnA0)/T, where
AT is the surface area, frond number or dry weight at time T and A0 is the
surface area, frond number or dry weight at the start of the experiment.
The three replicates per dose were pooled for dry weight determination in
order to obtain sufficient plant material for determining a reliable dry
weight on a five decimal balance (AX26 Coparator, Mettler Toledo,
Glostrup, Denmark).
2.2. Pigment analysis
The analysis of chlorophyll a and b and carotenoids were done on
ethanol extractions according to Lichtenthaler (1987). One to two
ARTICLE IN PRESS
Table 1
The molecular site of action and overall mode of action of the herbicides included in the study
Name Chemical group Site of action Mode of action Purity Source
Acifluorfen Diphenylether Protoporphyrinogen oxidase inhibitor Photosynthesis 40% BASF
Diquat Bipyridylium Photosystem I inhibitor Photosynthesis a Syngenta crop protection
Glyphosate Glycine derivative EPSPSb inhibitor Amino acid synthesis 95% Monsanto
Mecoprop Aryloxyalkanoic acid Synthetic auxin Plant hormone system 89% BASF
Mesotrion Triketone Inhibits carotenoid synthesis Photosynthesis 79% Syngenta crop protection
Terbuthyla-zine 1,3,5-triazine Photosystem II inhibitor Photosynthesis 496% Syngenta crop protection
aThe formulated product Reglone (200 gL�1 diquat) was used.b5-Enolpyruvylshikimate-3-phosphate synthase. Inhibition blocks the synthesis of aromatic amino acids.
N. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332 325
milligrams of freeze-dried L. minor fronds was weighed on a five decimal
balance and 1ml of 96% ethanol was added to the fronds. The vial was
sealed and placed in the dark for 24 h to allow for pigment extraction. The
contents of chlorophyll a, b and carotenoids were measured spectro-
photometrically on a Unicam-UV1 spectrophotometer (Thermo Spectro-
nic, Rochester, NY, USA), using wavelengths of 665, 654 and 470 nm,
respectively. The content was calculated according to Lichtenthaler (1987).
Two or three replicates from each treatment, depending on the amount of
plant material present, were extracted and measured.
2.3. Statistics
Growth rates or pigment contents as a function of herbicide
concentration were described by a four parameter logistic concentra-
tion–response relationship (Streibig et al., 1993). As all dose–response
curves of one mixture toxicity experiment shared the same controls, the
curves had a common upper limit. The model describes the expected
response, y, for the ith mixture ratio at the chemical concentration xi as
y ¼ ci þd � ci
1þ expðbiðlnðxÞ � lnðeiÞÞÞ, (1)
where d is the common upper limit of the response for chemical
concentrations (x) equal to zero and ci is the lower limit of the response
at infinite concentrations. The parameter ei is the concentration giving a
response of 50% of d–c (EC50) for the mixture ratio i and bi determines the
relative slope around ei. All five concentration–response curves of one
binary mixture toxicity experiment were tested for having similar lower
limits. This was done by describing data with Eq. (1) and thereafter with a
model where ci was reduced to c, and then comparing the two fits with an
F-test (Streibig et al., 1993). The subsequent dose–response surface
modeling and isobole analyses were only made on curves having similar
upper and lower limits.
For the isobole analysis, data was analyzed at the EC50 response level
using the statistical method described in Sørensen et al. (2007)
programmed in the free-software programR (R Development Core Team,
2004). In short, data were described with a dose–response surface model
where the EC50’s of the five dose–response curves were confined to a linear
CA isobole and to another model where the EC50’s were confined to a
non-linear isobole model (Sørensen et al., 2007). The two surface models
were compared with a lack-of-fit F-test (Seefeldt et al., 1995) to test
whether the curvature parameter(s) improved the fit of the data
significantly. Two non-linear isobole models were used. The default
model included one parameter, l, describing the curvature of the isobole.
For very antagonistic isoboles, where one herbicide decreased the effect
of the other herbicide below the effect of the herbicide acting alone, a
two parameter isobole model was used with the parameters Z1 and Z2(Sørensen et al., 2007). A l-value smaller than one signifies antago-
nism while a l-value larger than one signifies synergism. Very different
Z-values give asymmetric, and in this study, antagonistic isoboles (see
Cedergreen et al., 2007a, b for examples). For further explanation and use
of the method see the following references (Cedergreen et al., 2007a, b;
Sørensen et al., 2007).
The evaluation of the dose–response surface in relation to CA and IA
was done using the method of Jonker et al. (2005) and the software
provided by the authors at http://www.ceh.ac.uk/sections/er/csvendsen.
html. The program provides tests for synergistic/antagonistic deviations
from both of the reference models and of dose level-dependent and effect
level-dependent deviations from the models. We only tested for
synergistic/antagonistic deviations from the two reference models, as
earlier studies with L. minor showed low reproducibility of both dose level-
dependent and effect level-dependent deviations with the number of
mixture ratios, doses and replicates used in this study (Cedergreen et al.,
2007a). We are aware of the simplifications thereby made and that
including dose level-dependent and effect level-dependent deviations
might improve the fits. But we feel it will add little extra biological
information, due to the low reproducibility of these asymmetries in the
used Lemna setup. In short, data was described with one of the two
reference models and with the same reference model including a deviation
function. These two models were then compared with a w2-test, to test
whether the deviation function improved the fit significantly. For further
explanation see Jonker et al. (2005). The probability of the data deviating
from the reference model is given together with the interaction parameter
a of the deviation function (Jonker et al., 2005). A positive interaction
parameter signifies antagonism while a negative parameter signifies
synergism.
3. Results
3.1. Growth rates and pigments
Control growth rates of L. minor were 0.3070.03,0.2870.04 and 0.2870.04 d�1 for area, dry weight andfrond number-specific growth rates (n ¼ 36). All growthrates were above the limit for minimal growth of 0.275 d�1
set by ISO (International Organization for standardization,2004). The chlorophyll a/b ratio was 1.6870.22 (n ¼ 624)irrespective of herbicides and dose, also for the herbicidesaffecting photosynthesis (diquat and terbuthylazine)(ANOVA on the chlorophyll a/b ratio as a function ofdose: p40:45). The chlorophyll a+b content was thereforeused in all further calculations. The average chlorophyllconcentration of the controls was 5.0570.80mg g�1
dry weight (n ¼ 36), which is within the range expectedfor L. minor (Cedergreen and Madsen, 2003; Maenpaa andAro, 1986). The total content of carotenoids was0.8770.16mg g�1 dry weight (n ¼ 36), which is in thelow end of what is found in L. minor and other species(Garcia-Plazaola et al., 2002; Kummerova and Kmentova,2004).
ARTICLE IN PRESSN. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332326
3.2. Response of endpoints towards herbicides
Not all endpoints achieved a full response at the highestdose tested within the 7-day growth period (Fig. 1, OnlineAppendix A). The area-specific and frond number-specificgrowth rates, on which basis the dose-range for theexperiments were chosen, decreased to approximately zeroat the highest dose for all herbicides tested. The fewexceptions of non-zero lower limits (Online Appendix A)were due to high variance at the high doses and were notconsistent with other studies (unpublished data). Dryweight-specific growth rates, chlorophyll content, andcarotenoid content in plants exposed to diquat andmecoprop decreased to approximately zero, though notconsistently for diquat. For acifluorfen, glyphosate andmesotrione dry weight-specific growth rates had onlydecreased to approximately 50% after 7 days at theconcentrations that had fully stopped area and frond-specific growth rates (Fig. 1, Online Appendix A).Chlorophyll and carotenoid content did not respond tothe terbuthylazine treatment within the experimental time,as shown in Fig. 1 with chlorophyll as an example. Therewas no significant difference in the EC50 values betweengrowth and pigment endpoints for any of the herbicideswhen tested separately (ANOVA: p40.05, data fromOnline Appendix A). The only exception was diquat,where the mean growth EC50 was 31.277.4 (n ¼ 6), whilethe mean pigment EC50 was 16.574.0 (n ¼ 4) (ANOVA:p ¼ 0:01). The average slope parameter b was 1.8570.97(n ¼ 90) excluding the 10 largest slopes which weretypically a result of large variability around EC50 (OnlineAppendix A).
As dose–response surface analyses using the presentedmodels can only be made on herbicide combinations withdose–response curves with similar upper and lower limits,analyses of combination effects on dry weight-specificgrowth rates could not be made for acifluorfen, mesotrioneand glyphosate combined with diquat, mecoprop orterbuthylazine. Likewise, combination effects on pigmentcontent could not be made for combinations withterbuthylazine. This gives a possibility for analyzing
a b
Fig. 1. Representative concentration–response curves for area-specific growth
n ¼ 1) and chlorophyll content (gray symbols, n ¼ 2 or 3) for the herbicides
relative to the controls as mean7standard deviation.
dose–response surfaces for 46 of the 60 combinations ofherbicides and endpoints.
3.3. CA versus IA
The results of the model of Jonker et al. (2005) showedthat the two reference models CA and IA gave fairlysimilar results for the Lemna data (Table 2). As aconsequence, 23 of 43 dose–response surfaces differedsignificantly from CA while 19 differed significantly fromIA (for the experiment diquat/mesotrione measured onarea and the two experiments with mecoprop/terbuthyla-zine measured on dry weight, the models did not converge).For all dose–response surfaces deviating from the referencemodels, the size of the deviation from the reference modelswere similar for the two models when comparing theinteraction term a. The difference between the twointeraction terms given as aCA–aIA was �0.0571.23(mean7SD) (n ¼ 11) for area-specific growth rates,�1.0276.05 (n ¼ 4) for dry weight-specific growth rates,�0.0270.05 (n ¼ 12) for frond number-specific growthrates, �0.3671.27 (n ¼ 7) for chlorophyll content and�0.6271.48 (n ¼ 7) for carotenoid content (Table 2). Asnone of these differences were significantly different fromzero, data will only be presented in relation to CA in thefollowing.
3.4. Two methods for detecting deviation from CA
For the CA reference model, two methods were used fordetecting statistically significant differences from CA: Theisobole method described by Sørensen et al. (2007) and theresponse surface method described by Jonker et al. (2005).The isobole method seemed to be a little more conservativecompared to the method of Jonker et al. (2005). Using theisobole model, significant difference from CA was detectedin 18 out of the 44 dose–response surfaces analyzed (for oneof the experiments on glyphosate/mecoprop measured onchlorophyll and carotenoids, the models did not converge),whereas for the model of Jonker et al. (2005), 23 of 43dose–response models were tested to differ significantly from
c
rate (filled symbol, n ¼ 3), dry weight-specific growth rates (open symbols,
(a) diquat, (b) glyphosate and (c) terbuthylazine. All responses are given
ARTICLE IN PRESS
Table 2
Summary of statistical analyses of the dose-response surfaces using the method of Jonker et al. (2005)
Mixture Model RGRArea RGRDW RGRFrond Chl a+b Car
Acifluorfen/diquat CA p ¼ 0.001 po0.001 p ¼ 0.02 p ¼ 0.17
a ¼ 1.26 a ¼ 2.4 a ¼ 1.64 a ¼ 1.08
IA po0.001 po0.001 p ¼ 0.001 p ¼ 0.004
a ¼ 1.36 a ¼ 2.22 a ¼ 1.87 a ¼ 1.65
Acifluorfen/mesotrion CA p ¼ 0.19 p ¼ 0.27 p ¼ 0.008 p ¼ 0.06 p ¼ 0.009
a ¼ 0.53 a ¼ 0.86 a ¼ 1.18 a ¼ �1.91 a ¼ �2.98
IA p ¼ 0.40 po0.001 p ¼ 0.74 p ¼ 0.98 p ¼ 0.51
a ¼ 0.37 a ¼ 1.76 a ¼ 0.20 a ¼ 0.02 a ¼ �0.52
Diquat/mesotrione CA a po0.001 p ¼ 0.94 p ¼ 0.55
a ¼ 1.9 a ¼ 0.09 a ¼ �0.72
IA a po0.001 p ¼ 0.32 p ¼ 0.41
a ¼ 2.1 a ¼ 0.90 a ¼ 0.68
Glyphosate/mesotrione CA po0.001 po0.001 p ¼ 0.44 p ¼ 0.19 p ¼ 0.11
a ¼ 2.18 a ¼ 5.6 a ¼ 0.33 a ¼ �2.29 a ¼ �1.97
IA p ¼ 0.48 p ¼ 0.12 p ¼ 0.39 p ¼ 0.73 p ¼ 0.98
a ¼ 0.21 a ¼ 0.76 a ¼ 0.38 a ¼ �0.38 a ¼ �0.02
Glyphosate/terbuthylazine CA po0.001 p ¼ 0.08 p ¼ 0.008
a ¼ 1.81 a ¼ �0.97 a ¼ 1.03
IA po0.001 p ¼ 0.69 po0.001
a ¼ 4.04 a ¼ �0.37 a ¼ 2.26
Mesotrion/terbuthylazine CA p ¼ 0.005 po0.001 p ¼ 0.005
a ¼ �1.62 a ¼ �14.5 a ¼ �1.81
IA p ¼ 0.004 po0.001 p ¼ 0.008
a ¼ �1.73 a ¼ �5.12 a ¼ �1.85
Glyphosate/mecoprop 1 CA p ¼ 0.018 p ¼ 0.81 p ¼ 0.99 p ¼ 0.93
a ¼ 0.57 a ¼ 0.10 a ¼ �0.007 a ¼ 0.15
IA p ¼ 0.18 p ¼ 0.90 p ¼ 0.34 p ¼ 0.38
a ¼ 0.33 a ¼ 0.05 a ¼ �1.28 a ¼ �1.45
Glyphosate/mecoprop 2 CA po0.001 p ¼ 0.05 p ¼ 0.03 p ¼ 0.14
a ¼ 1.90 a ¼ 0.73 a ¼ �1.31 a ¼ �0.39
IA po0.001 p ¼ 0.3 p ¼ 0.001 p ¼ 0.07
a ¼ 4.06 a ¼ 0.42 a ¼ �2.23 a ¼ �1.31
Mecoprop/mesotrion 1 CA po0.001 p ¼ 0.12 p ¼ 0.60 p ¼ 0.005
a ¼ 1.63 a ¼ 1.13 a ¼ 1.05 a ¼ 1.65
IA p ¼ 0.08 p ¼ 0.01 p ¼ 0.70 p ¼ 0.76
a ¼ 0.64 a ¼ 1.28 a ¼ �0.86 a ¼ 0.06
Mecoprop/mesotrion 2 CA p ¼ 0.05 p ¼ 0.90 p ¼ 0.15 p ¼ 0.85
a ¼ 0.93 a ¼ �0.13 a ¼ 1.11 a ¼ �0.19
IA p ¼ 0.48 p ¼ 0.70 p ¼ 0.23 p ¼ 0.71
a ¼ 0.38 a ¼ 0.30 a ¼ 0.97 a ¼ 0.30
Mecoprop/terbuthylazine 1 CA p ¼ 0.04 a po0.001
a ¼ 0.95 a ¼ 2.73
IA p ¼ 0.002 a po0.001
a ¼ 1.25 a ¼ 2.52
Mecoprop/terbuthylazine 2 CA p ¼ 0.002 a po0.001
a ¼ 2.98 a ¼ 2.57
IA po0.001 a po0.001
a ¼ 2.74 a ¼ 2.46
The p-value for the hypothesis of concentration addition (CA) or independent action (IA) is given together with the interaction term a. Positive interaction
term indicates antagonism while a negative one indicates synergism. Analyses are made on the five endpoints: area-specific relative growth rates
(RGRArea), dry weight-specific relative growth rates (RGRDW), frond number-specific relative growth rates (RGRfrond), chlorophyll a and b content (Chl
a+b) and carotenoid content (Car).aData could not be described with the model.
N. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332 327
ARTICLE IN PRESS
Table 3
Results of the statistical analysis of the dose-response surfaces using the method of Sørensen et al. (2007)
Mixture RGRArea RGRDW RGRFrond Chl a+b Car
Acifluorfen/diquat po0.001 po 0.001 p ¼ 0.02 p ¼ 0.17
l ¼ 0.5670.13 l ¼ 0.1870.20 l ¼ 0.4470.28 l ¼ 0.6070.31
Acifluorfen/mesotrion p ¼ 0.60 p ¼ 0.98 p ¼ 0.02 p ¼ 0.91 p ¼ 0.11
l ¼ 0.8570.29 l ¼ 0.9870.66 l ¼ 0.6270.17 l ¼ 0.9870.25 l ¼ 1.4570.30
Diquat/mesotrione po0.001 po0.001 p ¼ 0.71 p ¼ 0.96
Z1 ¼ 11.876.0 Z1 ¼ 23.3752.1 l ¼ 0.8870.36 l ¼ 1.0170.34
Z2 ¼ 0.4570.14 Z2 ¼ 0.4770.16
Glyphosate/mesotrione p ¼ 0.79 p ¼ 0.90 p ¼ 0.85 p ¼ 0.43 p ¼ 0.62
l ¼ 0.9670.17 l ¼ 0.9670.31 l ¼ 0.9770.20 l ¼ 0.6270.57 l ¼ 1.4270.83
Glyphosate/terbuthylazine po0.001 p ¼ 0.27 po 0.001
Z1 ¼ 1.0170.12 l ¼ 0.5170.01 l ¼ 0.2170.22
Z2 ¼ 11.072.19
Mesotrion/terbuthylazine p ¼ 0.08 p ¼ 0.62 p ¼ 0.15
l ¼ 1.1370.19 l ¼ 0.7970.51 l ¼ 1.3270.23
Glyphosate/mecoprop 1 p ¼ 0.02 po 0.001 a a
l ¼ 0.7370.12 l ¼ 0.2170.20
Glyphosate/mecoprop 2 po 0.001 p ¼ 0.006 p ¼ 0.09 p ¼ 0.58
l ¼ 0.4270.15 l ¼ 0.4670.21 l ¼ 1.3570.20 l ¼ 1.1370.24
Mecoprop/mesotrion 1 p ¼ 0.007 p ¼ 0.20 p ¼ 0.50 p ¼ 0.70
l ¼ 0.5070.22 l ¼ 0.6670.31 l ¼ 1.2970.41 l ¼ 0.8370.48
Mecoprop/mesotrion 2 p ¼ 0.03 p ¼ 0.33 p ¼ 0.17 p ¼ 0.71
l ¼ 0.6770.17 l ¼ 0.5770.56 l ¼ 0.6670.25 l ¼ 1.1170.30
Mecoprop/terbuthylazine 1 po0.001 po0.001 po0.001
Z1 ¼ 0.8670.11 Z1 ¼ 0.6370.10 Z1 ¼ 0.8170.19
Z2 ¼ 12.772.9 Z2 ¼ 81.2767.2 Z2 ¼ 10.974.25
Mecoprop/terbuthylazine 2 p ¼ 0.02 p ¼ 0.06 p ¼ 0.15
l ¼ 0.6470.16 l ¼ 2.0970.46 l ¼ 1.3270.23
Data is related to the concentration addition model (CA) and evaluated at the 50% effect level. This is done by comparing a dose-response surface with
curved 50% isobole model with a a straight CA isobole (Figs. 3, 4). The p-value for the CA-hyphothesis is given together with the parameter(s) describing
the curvature of the isobole: l or Z1 and Z2 for very antagonistic isoboles. A l-value smaller than one signifies antagonism while a l-value larger than one
signifies synergism. Very different Z-values give asymmetric, and in this study, antagonistic isoboles. Analyses are made on the five endpoints: area-specific
relative growth rates (RGRArea), dry weight-specific relative growth rates (RGRDW), frond number-specific relative growth rates (RGRfrond), chlorophyll a
and b content (Chl a+b) and carotenoid content (Car).aData could not be described with the model.
N. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332328
CA (Tables 2 and 3). Both models did not converge with twoof the data sets, which was probably due to large variance inthe data. There was no difference between the models inrespect to detecting synergism or antagonism in relation toCA. The only exception was the dry weight-specific growthrates of the mixtures mesotrione/terbuthylazine and meco-prop/terbuthylazine, where the latter could not be describedwith the Jonker et al. (2005) model (Tables 2 and 3). Theinconsistencies were most likely due to unevenly spacedmixture ratios combined with the fewer replicates of the dryweight data. In the following we use the results from theisobole model unless otherwise noted.
3.5. Mixture toxicity dependence on endpoint
For all herbicide combinations the joint effect isobolesfor area and frond number-specific growth rates were
generally similar, though the level of significant deviationfrom CA varied (Table 3). Isoboles calculated from area-specific growth rates were antagonistic in relation to CA in10 out of 12 experiments, whereas they only deviatedstatistically from CA in 6 out of the 12 experiments usingfrond number-specific growth rates. For the four combina-tions where data on dry weight-specific growth rates couldalso be measured (excluding the mecoprop/terbuthylazinemixture), it was notable that even though some of the EC50
isoboles looked as if they deviated from CA, none of thesedeviations were significant (Table 3). This probably stemsfrom the dry weight determinations only having one replicaper dose instead of three, giving the tests much less power.For this reason dry weight-specific endpoints will beexcluded in the following. Turning to the chlorophyll andcarotenoid isoboles, only one of the 14 isoboles differedsignificantly from CA (acifluorfen/diquat measured on
ARTICLE IN PRESS
0 10000 250000
5000
15000
Glyphosate (µg/L)
Mec
opro
p (µ
g/L
)
0 20000 400000
10000
25000
Glyphosate (µg/L)
Mec
opro
p (µ
g/L
)
0 10000 250000
10000
25000
Glyphosate (µg/L)
Mec
opro
p (µ
g/L
)
0 200000
10000
25000
Glyphosate (µg/L)
Mec
opro
p (µ
g/L
)
RGRArea
Chl a+b
RGRFrond
Car
Fig. 3. An example of dissimilar isobolograms across endpoints here given
for the herbicide mixture glyphosate/mecoprop (experiment 2) measured
on: area-specific relative growth rates (RGRArea), frond number-specific
relative growth rates (RGRfrond), chlorophyll a and b content (Chl a+b)
and carotenoid content (Car). The solid curve describes the fitted isobole
and the dotted line describes the fitted concentration addition isobole.
Data is given 7standard error of the individually fitted curves.
N. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332 329
chlorophyll) (Table 3). For half the isoboles there was evena slight tendency towards synergism (l41), though notsignificantly. Tendencies towards synergy were onlyobserved for three out of the 24 isoboles based on dryweight and frond number-specific growth. The comparisonof interaction terms of the Jonker et al. (2005) model alsoshowed this trend of decreasing antagonism-approachingCA, when moving from growth endpoints towards thepigment endpoints (Table 2). The trend was not consistentfor all mixtures. For the six herbicide mixtures wheredose–response surfaces could be made for pigment content,only little change in isobole shape was observed whenshifting from one endpoint to the other for three of themixtures, namely acifluorfen/diquat, diquat/mesotrioneand mecoprop/mesotrione (Tables 2 and 3). The otherthree mixtures, acifluorfen/mesotrione, glyphosate/meso-trione and glyphosate/mecoprop, were the ones that hadthe marked decrease in the degree of antagonism whenshifting from growth endpoint to pigment endpoint(Tables 2 and 3). The three mixtures that were repeatedshowed that the isobole shapes could be reproduced. Anexample of an herbicide mixture where the shape of theisobole was similar across endpoints is given in Fig. 2,whereas an example of a mixture where the isoboles differacross endpoints is given in Fig. 3.
4. Discussion
4.1. Mixture toxicity dependence at endpoint
The study showed that mixture toxicity results measuredon various biomarkers are not necessarily indicative of
0 5 15 250
200
400
600
Diquat (µg/L)
Aci
fluo
rfen
(µg
/L)
0 20 400
500
1500
Diquat (µg/L)
Aci
fluo
rfen
(µg
/L)
0 5 10 200
200
600
Diquat (µg/L)
Aci
fluo
rfen
(µg
/L)
0 5 15 250
400
800
Diquat (µg/L)
Aci
fluo
rfen
(µg
/L)
RGRArea
Chl a+b
RGRFrond
Car
Fig. 2. An example of similar isobolograms across endpoints here given
for the herbicide mixture diquat/acifluorfen measured on: area-specific
relative growth rates (RGRArea), frond number-specific relative growth
rates (RGRfrond), chlorophyll a and b content (Chl a+b) and carotenoid
content (Car). The solid curve describes the fitted isobole and the dotted
line describes the fitted concentration addition isobole. Data is given
7standard error of the individually fitted curves.
what happens on a plant population level. Three of the six-herbicide combinations had mixture toxicity results onpigments that differed markedly from those on growthrates (acifluorfen/mesotrione, glyphosate/mesotrione andglyphosate/mecoprop). No particular combination in termsof mode of action (Table 1) appeared to explain thesedifferences when compared to the three mixture combina-tions where toxicity results were similar across endpoints(acifluorfen/diquat, diquat/mesotrione and mecoprop/me-sotrione). The timing of when a specific herbicide affects acertain biochemical pathway could play a role for thecombinations effect measured at different endpoints. Someherbicides are likely to damage pigments before they affectcell growth (diquat, acifluorfen and mesotrione), whileothers directly or indirectly might affect cell growth beforepigments (mecoprop). If a low EC50 is indicative of aprocess being affected quickly, differences in EC50 betweenendpoints could indicate which processes were the first tobe affected or at least the most sensitive for the differentherbicides. The evaluation of the EC50 across endpoints,however, showed no significant differences between pig-ment and growth endpoints for any of the herbicides exceptfor diquat, where pigment endpoints, as expected (Cobb,1992), were more sensitive than growth endpoints. Forterbuthylazine the opposite was the case, as pigmentendpoints were not affected at concentrations stoppinggrowth. This was surprising since terbuthylazine, as diquat,is a photosynthetic inhibitor, though it affects photosynth-esis at another target (Cobb, 1992). Other studies withbiomarkers and mixtures also show discrepancies betweenmixture effects measured on biomarkers and populationgrowth (Rendon-von et al., 2005) or show that biomarkers
ARTICLE IN PRESSN. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332330
can be affected differently by different toxicants (Brodeuret al., 2005; Tardif and Charest-Tardif, 1999). Cautionshould therefore be taken when extrapolating from biomar-ker endpoints to growth of individuals or populations.
Even though the mixture toxicity results differedbetween endpoints for some herbicide combinations, themixture toxicity results were all either additive orantagonistic. So from a regulatory perspective, CA couldbe used as a conservative reference model for all endpoints.It must be held in mind though, that the results are snap-shots of simultaneously inhibitory and recovery processestaken at a specified time. It can therefore not be excludedthat the results would have looked differently had theexperiments been run for a shorter or longer time.
4.2. Pigment content as a biomarker in plants
Pigment content was chosen as a biomarker in this studyas it is a commonly measured physiological endpoint (Simset al., 1999) and since earlier studies on the aquatic plantM. aquaticum had shown pigment contents to be affectedvery consistently by toxicants of different mode of action(Turgut and Fomin, 2002). In those experiments growthendpoints for some herbicides did not respond. Amongthose were the triazine herbicide terbutryn and thesynthetic auxins 2,4-D, dichlorprop and dicamba (Turgutand Fomin, 2002). The present study did not show thesame consistency in pigment response across herbicides.Most surprising was the lack of response in pigmentcontent for the Photosystem II inhibitor terbuthylazine,which is known to induce the production of reactiveoxygen species in the chloroplast (Cobb, 1992). One causefor the lack of pigment response could be the shorterexperiment time (1 week versus 2 weeks for M. aquaticum),which might not have allowed the full effect of terbuthy-lazine of pigment content to take place. Alternatively itcould be hypothesized that terbuthylazine affects the plantsdifferently from terbutryn, the triazine herbicide tested inthe study of Turgut and Fomin (2002). The latter thoughseems very unlikely. Light intensities, which affect the rateof production of reactive oxygen species, were similar inthe two studies.
Another reason for studying the relation betweenpigment content and growth in plants affected by differenttoxicants is the increased use of photosynthesis relatedparameters as toxicity endpoints (Christensen et al., 2003;Juneau et al., 2003; Marwood et al., 2001; Petersen andKusk, 2000). Photosynthetic rates are closely related to thecontent of light harvesting pigments. Particularly the use ofchlorophyll fluorescence parameters provides a quick andnon-destructive tool for toxicity assessments in plants,which could make it possible to track the development ofmixture toxicity effects over time. Despite the failure ofterbuthylazine to induce pigment reductions at the highestdoses in this study, it is noteworthy that all the otherherbicides, representing very different modes of action, didaffect pigments within the concentration range affecting
growth. And even though terbuthylazine did not affectpigment content in this study, other studies on terbuthy-lazine and L. minor have showed a rapid response onchlorophyll fluorescence parameters (Cedergreen et al.,2004). We therefore believe that both pigment contentand possibly photosynthesis related parameters have afuture as robust and reliable endpoints both in the study ofindividual chemicals and chemical mixtures. Studies of thetime dependence of the specific endpoint and correlationswith growth parameters should however always be made.
4.3. CA versus IA
One of the aims of this study was also to test theapplicability of the two mixture toxicity reference modelsCA and IA. The comparison of data to the two referencemodels proved to be remarkably similar both in respect to thenumber of mixtures that could be described by the models andto the deviations from the reference models. This similaritymost probably stems from the relatively low slopes of thedose–response curves of the individual herbicides of 1.85which gives IA isoboles only slightly antagonistic compared tothe CA isobole (Cedergreen and Streibig, 2005; Dreschner andBoedeker, 1995). In other studies with chemicals withdissimilar mode of action, IA has proven to be the model tobest predict the joint effect of chemicals compared to CA(Backhaus et al., 2004, 2000; Faust et al., 2003). Also, severalmodels have been proposed that separate chemicals in groupsof similar mode of action. Within each group CA is used forestimating the joint effect, where after the joint effect of thegroups of dissimilar mode of action is estimated with IA(Altenburger et al., 2004; Ra et al., 2006; Rider and LeBlanc,2005). These combined models have also proven to betterpredict the effect of mixtures of chemicals with dissimilarmode of action. All these studies have measured mortality ineither algae or bacterial populations or on Daphnia magna;test systems which all consist of populations of independentorganisms. This is in contrast to the Lemna test system, whereonly a single or few plants are exposed to the toxicants at theexperimental start and continuous endpoints as growth andpigment content is measured on individuals (including theiroff-spring). As IA is developed for test systems with binominalendpoints in populations of independent individuals (Grecoet al., 1995), this might form part of the explanation for whyIA does predict the mixture toxicity of the differently actingchemicals well for the algae and bacteria test systems, but notfor L. minor.
4.4. Comparing statistical models
The last comparison of this paper was between the twostatistical dose–response surface models proposed bySørensen et al. (2007) and Jonker et al. (2005), respectively.The two models gave very similar results with the Sørensenet al. (2007) model being slightly more conservative whentesting for CA. This is not surprising as both models areone-parameter generalizations of the CA-model and both
ARTICLE IN PRESSN. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332 331
rely on the same dose–response relationship for theherbicides used separately. The models differ in their wayof describing interaction. Sørensen et al. (2007) takes theisobole description as a starting point and models theisobole directly. This has the advantage that the estimatedisobole is easily plotted as an extremely useful visualizationof the interaction in the case of binary mixtures. The modelof Jonker et al. (2005) describes the relationship betweenthe components of a certain mixture and the expectedresponse more directly, but it comes at the expense ofestimating the isoboles. It should be noted that Sørensenet al. (2007) specify the same log-logistic curve type for thedose–response relationship for all mixtures. In the Jonkeret al. (2005) model the dose–response curves are modelledindirectly for the two-component mixtures, which poten-tially can lead to non-logistic dose–response curves.Whether these non-logistic dose–response curves differunrealistically from the expected sigmoid dose–responserelationship, we cannot judge.
5. Conclusion
The study showed that for some herbicide combinationsthe mixture toxicity measured on pigment content did notreflect the results measured on population growth. Cautionshould therefore be taken when extrapolating frombiomarker endpoints to growth of individuals and popula-tions. The IA model did not explain data better than theCA model even though all the tested herbicides possessed adifferent biochemical target site. The two different statis-tical models used to test the data in relation to CA showedvery similar results, validating the use of both.
Acknowledgments
We are grateful to BASF, Monsanto and Syngenta forproviding technical herbicides. Also Anja Weibell Stubbeand Arne Raun should be acknowledged for their help inthe laboratory, and Jason Belden for constructive com-ments on the manuscript.
Funding source: The work was supported by the DanishEnvironmental Protection Agency (Grant No. M 7041-0468).
Appendix A. Supplementary materials
Supplementary data associated with this article can be foundin the online version at doi:10.1016/j.ecoenv.2006.12.006.
References
Altenburger, R., Backhaus, T., Boedeker, W., Faust, M., Scholze, M.,
Grimme, L.H., 2000. Predictability of the toxicity of multiple chemical
mixtures to Vibrio fischeri: mixtures composed of similarly acting
chemicals. Environ. Toxicol. Chem. 19, 2341–2347.
Altenburger, R., Walter, H., Grote, M., 2004. What contributes to the
combined effect of a complex mixture? Environ. Sci. Technol. 38,
6353–6362.
Backhaus, T., Altenburger, R., Boedeker, W., Faust, M., Scholze, M.,
Grimme, L.H., 2000. Predictability of the toxicity of multiple mixture
of dissimilarly acting chemicals to Vibrio Fischeri. Environ. Toxicol.
Chem. 19, 2348–2356.
Backhaus, T., Arrhenius, A., Blanck, H., 2004. Toxicity of a mixture of
dissimilarly acting substances to natural algal communities: predictive
power and limitations of independent action and concentration
addition. Environ. Sci. Technol. 38, 6363–6370.
Berenbaum, M.C., 1989. What is synergy? Pharmacol. Rev. 41, 93–141.
Brain, R.A., Johnson, D.J., Richards, S.M., Hanson, M.L., Sanderson,
H., Lam, M.W., Young, C., Mabury, S.A., Sibley, P.K., Solomon,
K.R., 2004. Microcosm evaluation of the effects of an eight
pharmaceutical mixture to the aquatic macrophytes Lemna gibba
and Myriophyllum sibiricum. Aquat. Toxicol. 70, 23–40.
Brodeur, J.C., Woodburn, K.B., Klecka, G.A., 2005. Potentiation of the
vitellogenic response to 17 alpha-ethinylestradiol by cortisol in the
fathead minnow Pimephales priomelas. Environ. Toxicol. Chem. 24,
1125–1132.
Cedergreen, N., Madsen, T.V., 2003. Light regulation of root and leaf
NO3� uptake and reduction in the floating macrophyte Lemna minor.
New Phytol. 161, 449–457.
Cedergreen, N., Streibig, J.C., 2005. Can the choice of endpoint lead to
contradicting results of mixture toxicity experiments? Environ.
Toxicol. Chem. 24, 1676–1683.
Cedergreen, N., Andersen, L., Olesen, C.F., Spliid, N.H., Streibig, J.C.,
2004. Does the effect of herbicide pulse exposure on aquatic
plants depend on Kow or mode of action? Aquat. Toxicol. 71,
261–271.
Cedergreen, N., Kudsk, P., Mathiassen, S.K., Sørensen, H., Streibig, J.C.,
2007a. The reproducability of binary mixture toxicity studies. Environ.
Toxicol. Chem. 26 (1), 149–156.
Cedergreen, N., Kudsk, P., Matthiasen, S., Streibig, J.C., 2007b.
Combination effects of herbicides: do species and test system matter?
Pest Manage. Sci., in press.
Christensen, M.G., Teicher, H.B., Streibig, J.C., 2003. Linking fluores-
cence induction curve and biomass in herbicide screening. Pest
Manage. Sci. 59, 1303–1310.
Cobb, A., 1992. Herbicides that inhibit photosynthesis. In: Herbicides and
Plant Physiology. Chapman & Hall, London, pp. 36–79.
Dreschner, K., Boedeker, W., 1995. Assessment of the combined effects of
substances: the relationship between concentration addition and
independent action. Biometrics 51, 716–730.
Faust, M., Altenburger, R., Backhaus, T., Blanck, H., Boedeker, W.,
Gramatica, P., Hamer, V., Scholze, M., Vighi, M., Grimme, L.H.,
2001. Predicting the joint algal toxicity of multi-components-triazine
mixtures at low-effect concentrations of individual toxicants. Aquat.
Toxicol. 56, 13–32.
Faust, M., Altenburger, R., Backhaus, T., Blanck, H., Boedeker, W.,
Gramatica, P., Hamer, V., Scholze, M., Vighi, M., Grimme, L.H.,
2003. Joint algal toxicity of 16 dissimilarly acting chemicals is
predictable by the concept of independent action. Aquat. Toxicol.
63, 43–63.
Garcia-Plazaola, J.I., Hernandez, A., Artetxe, U., Becerril, J.M., 2002.
Regulation of the xanthophyll cycle pool size in duckweed (Lemna
minor) plants. Physiol. Plant. 116, 121–126.
Greco, W.R., Bravo, G., Parsons, J.C., 1995. The search for synergy: a
critical review from a response surface perspective. Pharmacol. Rev.
47, 332–385.
International Organization for standardization, 2004. Water quality—
duckweed growth inhibition. ISO/WD 20079, 1–18. 2004. Geneva,
International Organization for Standardization.
Jonker, M.J., Svendsen, C., Bedaux, J.J.M., Bongers, M., Kammenga,
J.E., 2005. Significance testing of synergistic/antagonistic, dose level-
dependent, or dose ratio-dependent effects in mixture dose–response
analysis. Environ. Toxicol. Chem. 24, 2701–2713.
ARTICLE IN PRESSN. Cedergreen et al. / Ecotoxicology and Environmental Safety 67 (2007) 323–332332
Juneau, P., Sumitomo, H., Matsui, S., Itoh, S., Kim, S.G., Popovic, R.,
2003. Use of chlorophyll fluorescence of Closterium ehrenbergii and
Lemna gibba for toxic effect evaluation of sewage treatment plant
effluent and its hydrophobic components. Ecotoxicol. Environ. Saf.
55, 1–8.
Kummerova, M., Kmentova, E., 2004. Photoinduced toxicity of
fluoranthene on germination and early development of plant seedling.
Chemosphere 56, 387–393.
Landolt, E., Kandeler, R., 1987. Physiological characteristics. In: Landolt,
E., Kandeler, R. (Eds.), The Family of Lemnaceae—A Monographic
Study. Stiftung Rubel, Zurich, pp. 54–113.
Lichtenthaler, H.K., 1987. Chlorophylls and carotenoids: pigments of
photosynthetic membranes. In: Packer, L., Douce, R. (Eds.), Plant
Cell Membranes. Academic Press, Cambridge, pp. 350–383.
Maeng, J., Khudairi, A.K., 1973. Studies on the flowering mechanism of
Lemna, I: amino acid changes during flower induction. Physiol. Plant.
28, 264–270.
Maenpaa, P., Aro, E.M., 1986. Chlorophyll–protein complexes, chlor-
ophyll a/b ratio and chloroplast ultrastructure in Lemna minor L.
grown under different light conditions. J. Plant Physiol. 123, 161–168.
Marwood, C.A., Solomon, K.R., Greenberg, B.M., 2001. Chlorophyll
fluorescence as a bioindicator of effects on growth in aquatic
macrophytes from mixtures of polycyclic aromatic hydrocarbons.
Environ. Toxicol. Chem. 20, 890–898.
Petersen, S., Kusk, K.O., 2000. Photosynthesis tests as an alternative to
growth tests for hazard assessment of toxicant. Arch. Environ.
Contam. Toxicol. 38, 152–157.
R Development Core Team, 2004. R: A Language and Environment for
Statistical Computing. R Foundation for Statistical Computing,
Vienna, Austria.
Ra, J.S., Lee, B.C., Chang, N.I., Kim, S.D., 2006. Estimating the
combined toxicity by two-step prediction model on the complicated
chemical mixtures from wastewater treatment plant effluents. Environ.
Toxicol. Chem. 25, 2107–2113.
Rendon-von, O., Ortiz-Arana, A., Guilhermino, L., Soares, A.M.V.M.,
2005. In vivo evaluation of three biomarkers in the mosquitofish
(Gambusia yucatana) exposed to pesticides. Chemosphere 58, 627–636.
Rider, C.V., LeBlanc, G.A., 2005. An integrated addition and interaction
model for assessing toxicity of chemical mixtures. Toxicol. Sci. 87,
520–528.
Seefeldt, S.S., Jensen, J.E., Fuerst, E., 1995. Log-logistic analysis of
herbicide dose–response relationships. Weed Technol. 9, 218–227.
Sims, I., Whitehouse, P., Lacey, R., 1999. The OECD Lemna growth
inhibition test. Technical Report EMA 003, 1-71. Bristol, USA,
Environmental Agency R&D Dissemination Centre.
Sørensen, H., Cedergreen, N., Skovgaard, I., Streibig, J.C., 2007. An
isobole-based statistical model and test for synergism/antagonism in
binary mixture toxicity experiments. Environ. Ecol. Stat., 14, in press.
Streibig, J.C., Rudemo, M., Jensen, J.E., 1993. Dose–response curves and
statistical models. In: Streibig, J.C., Kudsk, P. (Eds.), Herbicide
Bioassay. CRC Press, Boca Raton, FL, pp. 29–55.
Tardif, R., Charest-Tardif, G., 1999. The importance of measured end-
points in demonstrating the occurrence of interactions: a case study
with methylchloroform and m-xylene. Toxicol. Sci. 49, 312–317.
Turgut, C., Fomin, A., 2002. Sensitivity of the rooted macrophyte
Myriophyllum aquaticum (Vell.) Verdcourt to seventeen pesticides
determined on the basis of EC50. Bull. Environ. Contam. Toxicol. 69,
601–608.
Viet, U., Moser, H., 2003. Ecotoxicological test asays with macrophytes.
Fresenius Environ. Bull. 12, 545–549.