Hormesis in mixtures — Can it be predicted?

SC IE N CE OF TH E TOTA L E N V I RO N ME N T 4 04 ( 20 0 8 ) 7 7– 8 7

a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m

w w w. e l s e v i e r. c o m / l o c a t e / s c i t o t e n v

Hormesis in mixtures — Can it be predicted? Regina G. Belz a,⁎, Nina Cedergreen b , Helle Sørensen c a

University of Hohenheim, Institute of Phytomedicine, Department of Weed Science, Otto-Sander-Straße 5, 70593 Stuttgart, Germany Department of Agricultural Sciences, Faculty of Life Sciences, University of Copenhagen, Højbakkegård Allé 13, 2630 Taastrup, Denmark c Department of Natural Sciences, Faculty of Life Sciences, University of Copenhagen, Thorvaldsensvej 40, 1871 Frederiksberg C., Denmark b

AR TIC LE I N FO

ABS TR ACT

Article history:

Binary mixture studies are well established for mixtures of pollutants, pesticides, or

Received 3 March 2008

allelochemicals and sound statistical methods are available to evaluate the results in

Received in revised form 5 June 2008

relation to reference models. The majority of mixture studies are conducted to investigate

Accepted 9 June 2008

the effect of one compound on the inhibitory action of another. However, since stimulatory

Available online 21 July 2008

responses to low concentrations of chemicals are gaining increased attention and improved statistical models are available to describe this phenomenon of hormesis, scientists are

Keywords:

challenged by the question of what will happen in the low concentration range when all or

Biphasic concentration–response

some of the chemicals in a mixture induce hormesis? Can the mixture effects still be

curve

predicted and can the size and concentration range of hormesis be predicted? The present

Concentration addition

study focused on binary mixtures with one or both compounds inducing hormesis and

Hormesis

evaluated six data sets of root length of Lactuca sativa L. and areal growth of Lemna minor L.,

Mixture toxicity

where substantial and reproducible hormetic responses to allelochemicals and herbicides have been found. Results showed that the concentration giving maximal growth stimulatory effects (M) and the concentration where the hormetic effect had vanished (LDS) could be predicted by the most-used reference model of concentration addition (CA), if the growth inhibitory concentrations (EC50) followed CA. In cases of deviations from CA at EC50, the maximum concentration M and the LDS concentration followed the same deviation patterns, which were described by curved isobole models. Thus, low concentration mixture effects as well as the concentration range of hormesis can be predicted applying available statistical models, if both mixture partners induce hormesis. Using monotonic concentration–response models instead of biphasic concentration–response models for the prediction of joint effects, thus ignoring hormesis, slightly overestimated the deviation from CA at EC20 and EC50, but did not alter the general conclusion of the mixture study in terms of deviation from the reference model. Mixture effects on the maximum stimulatory response were tested against the hypothesis of a linear change with mixture ratio by constructing 95% prediction intervals based on the single concentration–response curves. Four out of the six data sets evaluated followed the model of linear interpolation reasonably well, which suggested that the size of the hormetic growth stimulation can be roughly predicted in mixtures from knowledge of the concentration–response relationships of the individual chemicals. © 2008 Elsevier B.V. All rights reserved.

⁎ Corresponding author. Tel.: +49 711 459 23444. E-mail addresses: [email protected] (R. Belz), [email protected] (N. Cedergreen), [email protected] (H. Sørensen). 0048-9697/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2008.06.008

78

1.

SC IE N CE OF TH E TOTA L E N V IR O N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

Introduction

Some chemicals, which are toxic at high concentrations, can have a stimulatory effect on one or several traits in an organism when applied at low concentrations (Calabrese and Baldwin, 2001). This phenomenon is called hormesis and has been described for a range of substances and in a range of organisms for more than a century (Southam and Erlich, 1943; Calabrese, 2005; Cedergreen et al., 2007b). Great controversies have existed (and do exist) concerning the interpretation of these observations, spanning from believing hormesis is a general stress response occurring for all chemicals (and stresses) (Calabrese and Baldwin, 2001; Stebbing, 1998), to hormesis being nonexisting, but simply a result of poorly thriving control organisms or of resource allocations taking place within the organism (Forbes, 2000; Parsons, 2003). Evidence is, however, accumulating, showing hormesis to be true and reproducible for specific chemicals and organism traits (Calabrese and Baldwin, 2001; Duke et al., 2006; Cedergreen et al., 2007b; Cedergreen, in press). The prospect of hormesis being a true phenomenon has highly controversial implications within the areas of environmental and medical toxicology (Calabrese, 2005; Thayer et al., 2005; Kefford et al., 2008), as it questions the ways we set limit values for pollutants and toxins. And one thing is to set limit values for single substances; another is to take complex mixtures of chemicals into account especially within the framework of the hormetic dose response. As hormesis is a central concept in toxicology and pharmacology Calabrese (2008b) claimed a regular inclusion in dose–response assessments, including that of mixtures. However, what happens in the low concentration range when all, or some, of the substances in a mixture induce biphasic concentration–response relationships? Can the mixture effects then still be predicted, especially in the low concentration range? Can the size and concentration range of hormesis in the mixture be predicted from knowledge of the concentration–response relationship of the individual compounds? And how much does it affect joint effect predictions if the hormetic response is ignored? Answers to these questions are absent, although these issues are often addressed by human toxicologists and pharmacists who traditionally work with compounds (pharmaceuticals) that have beneficial effects at low concentrations while they have toxic effects at higher concentrations (Boekelheide, 2007). However, the issue of hormesis in mixtures is equally relevant to ecotoxicologists and agronomists working with the effect of mixtures of pollutants, pesticides or allelochemicals which all might induce biphasic concentration–response curves. The joint effect of chemicals in mixtures is traditionally evaluated and predicted by one of the two reference models, concentration addition (CA) or independent action (IA) (Greco et al., 1995). For binary mixtures, experiments can be designed to describe entire concentration–response surfaces and the joint effect of the two chemicals can be evaluated in relation to the reference models at different effects levels via isobolograms (Gessner, 1995; Greco et al., 1995). The traditional effect level evaluated is the 50% effect concentration or dose (EC/ED50), because the EC/ED50 is the EC/ED-value statistically determined with the highest precision. When it comes to

evaluating the behaviour of biphasic concentration–response curves in mixtures, it is, however, the lower concentration/ dose range that is of interest. None of the current response surface models available include biphasic concentration– response curves (Jonker et al., 2005; White et al., 2004; Sørensen et al., 2007). Hence, to evaluate the hormetic response in mixtures, selected effect levels must be analyzed separately. Another matter of interest within the hormetic framework is the size of the hormetic response and its variability in case of chemical interaction. However, as the biological plasticity underlying hormetic effects is limited, it has been proposed that it is rather the size of the hormetic response that changes in case of a chemical interaction than the dose giving maximum stimulation (Calabrese, 2008a,b). Thus, chemical interactions at the lower end of the dose–response curve seem to involve a diminished combined dose in case of synergy (Calabrese, 2008a,b) and vice versa an increased dose in case of antagonism while the maximum response is not markedly changed. Based on this, we propose that the size of the hormetic response should change linearly with the mixture ratio at any kind of chemical interaction (antagonism, synergism, or additivism). This concept of hormetic interaction differs from the traditional concept of chemical interaction that becomes evident on the toxic response (Calabrese, 2008a,b). To get the most thorough understanding of how compounds inducing biphasic curves interact in mixtures with other compounds, either inducing or not inducing biphasic curves, binary mixtures describing whole concentration–response surfaces give the most complete picture (Greco et al., 1995). This study therefore focuses on binary mixtures with one or both compounds inducing hormesis. The aim of the study is to answer the following questions: 1) What are the consequences of using monotonic concentration–response models instead of biphasic concentration–response models for the prediction of joint effects of mixtures containing compounds inducing hormesis? 2) Can the size and concentration range of the hormetic growth stimulation be predicted in binary mixtures of two hormetic compounds? 3) Can the size and concentration range of the hormetic growth stimulation be predicted in binary mixtures of two compounds where one is hormetic, and the other is not? We investigated these questions using data of root length of lettuce (Lactuca sativa L.) and areal growth of duckweed (Lemna minor L.), where substantial and reproducible hormetic responses to phytotoxins have been found.

2.

Materials and methods

2.1.

Phytotoxicity bioassays

2.1.1.

Lactuca assay

The Lactuca assays were conducted as germination tests in non-pyrogenic polystyrene 6-well plates (BD Falcon™ multiwell cell culture plates, Falcon) with L. sativa var. capitata cv. Maikönig as the test species. Each well was prepared with 10 seeds on one layer of filter paper (Schleicher & Schuell) before 1.5 ml of test solution was added. The test solutions were mixed in deionized water from stock solutions of individual chemicals prepared in acetone to give a final acetone concentration of 1% (v/v) at each treatment. Controls were performed

SC IE N CE OF TH E TOTA L E N V I RO N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

with a 1% acetone solution. Each mixture experiment consisted of seven concentration–response curves each with 11–12 concentrations of the tested chemicals or their mixtures (0–6 mM). The seven concentration–response curves included the two chemicals tested separately and tested in five different fixed-ratio mixtures with a proportion p [p (% effect at EC50) = 100, 80, 65, 50, 35, 20, 0] of compound A and 100 − p of compound B. The effect ratios were determined from preliminary experiments. Each treatment used one plate with six replications and three to four plates for control treatments. Plates were sealed with nescofilm and cultivated in a completely randomized design in a growth cabinet (12/12 h, 24/18 °C, 50/0 µmol m− 2 s− 1). After five days, root length (≥1 mm) was recorded. The bioassays were conducted consecutively for each mixture and were repeated twice. Data of individual experiments were analyzed independently.

2.1.2.

Lemna assay

The L. minor test closely followed the guidelines given by the International Organisation for Standardization (International Organization for Standardization, 2004), with few modifications. Instead of 100 ml glass beakers and starting densities of four fronds per beaker, 10 ml 6-well TC-test plates (CM. Lab. Aps, Vordingborg, Denmark) and starting densities of one frond per well were used. Control plants grew exponentially throughout the experimental time (Cedergreen et al., 2004), and since surface area, and not frond number, was used as an endpoint, the coefficient of variance of the control treatment was kept below 10% (Cedergreen et al., 2007a). As only herbicides with a log Kow below 1.2 were used (Tomlin, 2002), adhesion to the multi-well plate material was considered to be insubstantial. The L. minor plants were collected in Copenhagen in 1999, surface sterilised with hypochlorite and the resulting sterile clone was kept in Erlenmeyer flasks in “K”medium (Maeng and Khudairi, 1973), at pH 5, 24 °C and a continuous photon flux density of 85–120 µmol m− 2 s− 1 (PAR). The K-medium was used, as it has proven to give the highest growth rates in the used experimental setup. 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-dichlorophenol (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. For the Lemna tests five or seven mixture ratios were used, including the single compounds. The mixture ratios were: p (% effect at EC50) = 100, 75, 50, 25, 0 of compound A and (100 − p) of compound B for the five mixture ratio design and p (% effect at EC50) = 100, 83, 67, 50, 33, 17, 0 for the seven mixture ratio design. The effect ratios were determined from preliminary experiments. Each concentration–response curve consisted of 6–8 concentrations, each in three replicates. There were 12 common controls. The experiments were initiated by transferring one L. minor frond to 10 ml K-medium 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). The plants were then placed in a growth cabinet at 24 °C and a continuous photon flux density of 85–120 µmol m− 2 s− 1 (PAR). After seven days the plants were photographed

79

again and frond surface area determined. Relative growth rates were calculated according to: (ln AT − lnA0) / T, where AT is the surface area at time T and A0 is the surface area at the start of the experiment.

2.1.3.

Chemicals

The study comprised two mixtures of compounds with known hormetic effects under the present experimental conditions (parthenin/tetraneurin-A, acifluorfen/mesotrione), and two mixtures where one compound induces hormesis while the other has a strictly decreasing concentration–response relationships (parthenin/caffeic acid, glyphosate/MCPA). The natural phytotoxins parthenin and tetraneurin-A were isolated from organic extracts of the leaf surface of plants from a South African population of Parthenium hysterophorus L. and fractionated by preparative high performance liquid chromatography (Varian model chromatograph) with UV detection (Varian UV–VIS detector model 345) as described by Belz et al. (2007). Caffeic acid was purchased (Roth, Germany). The synthetic herbicides were technical compounds provided by BASF (acifluorfen, 40%) and Syngenta Crop Protection (mesotrione, 79%; terbuthylazine, N96%).

2.2.

Statistics

2.2.1.

Hormetic concentration–response model

The relationship between the expected response (y) and toxin concentration (x) was modeled by a logistic regression model allowing for hormesis [Schabenberger et al. (1999) based on Brain and Cousens (1989)]: y¼Cþ

D  C þ fd x  1þ 1þ d ebd ln ðx=EC50 Þ


2d f d EC50 DC

ð1Þ

where C denotes the expected response at indefinitely large concentrations, D denotes the expected response of the untreated controls, f denotes the initial rate of increase in response at subinhibitory concentrations, EC50 denotes the concentration causing 50% inhibition of the control response, and b denotes the rate of change around EC50. The expected response is assumed to be zero for infinitely large concentrations, that is, we assume C = 0. The model was fitted to data by non-linear regression analysis using SPSS®. Variance of responses was stabilized at each concentration by using the inverse standard deviation of replicates as weight. Significance of hormesis was assessed by calculation of the 95% confidence intervals for the estimates of the initial rate of increase f and was given for f N 0. In case of hormesis, the concentration giving maximum response (M), the maximal response at concentration M (ymax), and the concentration where the hormetic effect has vanished (limited concentration for stimulation — LDS) [corresponds to EC/ED0 or no-observable-adverse-effect-level (NOAEL) (Cedergreen et al., 2007b)] were estimated as well. This was done by reparameterizations of Eq. (1) [see Schabenberger et al. (1999)]. Similarly, additional inhibitory response levels ECx reducing plant response by x % were estimated to enable a joint-action analysis at different response levels. Concentration–response curves were also fitted with no hormesis effect (f = 0) in order to compare the joint-action analyses with and without allowing for hormesis.

80 2.2.2.

SC IE N CE OF TH E TOTA L E N V IR O N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

Joint-action analysis — isoboles

Once the concentration–response curves were fitted, the estimated concentrations were used in a subsequent isobole analysis, estimating the deviation from concentration addition (CA). We use CA as reference model as a necessary condition for the applicability of independent action (also called multiplicative survival model) is that effects are scaled between zero and 100% (Streibig and Jensen, 2000), which

restricts its applicability to monotonic concentration– response curves. Also, studies comparing CA and IA predictions on large numbers of mixture data found relatively small differences in mixture predictions between the two models for combinations of chemicals with different primary modes of action (Altenburger et al., 1996; Cedergreen et al., 2008). The procedure is described for the EC50–values in the following, but the same procedure was used for M, LDS and EC20. CA assumes

Table 1 – Parameters from the logistic concentration–response model (Eq. (1)) tested on growth of two plant species exposed to different phytotoxic mixtures Mixture

Lactuca sativa Parthenin versus tetraneurin-A (exp. 1)

Parthenin versus tetraneurin-A (exp. 2)

Parthenin versus caffeic acid (exp. 1)

Parthenin versus caffeic acid (exp. 2)

Lemna minor Acifluorfen versus mesotrione

Acifluorfen versus terbuthylazin

Ratio (%)

D

100:0 80:20 65:35 50:50 35:65 20:80 0:100 100:0 80:20 65:35 50:50 35:65 20:80 0:100 100:0 80:20 65:35 50:50 35:65 20:80 0:100 100:0 80:20 65:35 50:50 35:65 20:80 0:100

1.22 ± 0.02 (cm)

100:0 83:17 67:33 50:50 33:67 17:83 0:100 100:0 83:17 67:33 50:50 33:67 17:83 0:100

1.25 ± 0.02 (cm2)

1.31 ± 0.05 (cm)

1.46 ± 0.03 (cm)

1.23 ± 0.02 (cm)

1.18 ± 0.03 (cm2)

b

f

M

ymax

LDS

EC50

(mM)

(%)

(mM)

(mM)

5.50 ± 0.60 3.24 ± 0.30 3.38 ± 0.23 3.63 ± 0.28 3.05 ± 0.16 3.53 ± 0.17 2.36 ± 0.09 3.61 ± 0.48 3.00 ± 0.27 4.02 ± 0.61 3.69 ± 0.65 6.53 ± 0.83 6.04 ± 1.65 3.31 ± 0.49 3.87 ± 0.53 3.88 ± 0.56 3.86 ± 0.82 2.53 ± 0.31 4.52 ± 0.99 5.60 ± 1.24 4.51 ± 1.04 4.78 ± 0.47 3.96 ± 0.57 2.50 ± 0.18 3.37 ± 0.57 3.73 ± 0.47 6.55 ± 1.36 7.18 ± 1.61

01.67 ± 0.15 03.32 ± 0.39 04.76 ± 0.62 05.56 ± 0.74 05.98 ± 0.77 10.95 ± 0.44 16.32 ± 0.50 02.13 ± 0.44 06.44 ± 0.76 04.46 ± 0.77 03.32 ± 0.82 03.10 ± 0.18 02.61 ± 0.25 07.09 ± 1.60 03.17 ± 0.32 01.60 ± 0.28 00.83 ± 0.29 01.38 ± 0.31 00.32 ± 0.07 n.s. n.s. 01.66 ± 0.19 00.56 ± 0.16 01.62 ± 0.33 00.54 ± 0.13 00.44 ± 0.08 00.11 ± 0.04 n.s.

0.58 ± 0.04 0.32 ± 0.03 0.32 ± 0.03 0.32 ± 0.03 0.34 ± 0.03 0.27 ± 0.01 0.19 ± 0.01 0.56 ± 0.06 0.30 ± 0.02 0.36 ± 0.04 0.39 ± 0.06 0.50 ± 0.02 0.51 ± 0.06 0.30 ± 0.04 0.48 ± 0.05 0.65 ± 0.07 0.83 ± 0.14 0.79 ± 0.13 1.84 ± 0.26 – – 0.61 ± 0.03 0.72 ± 0.08 0.52 ± 0.06 0.73 ± 0.10 1.61 ± 0.14 2.15 ± 0.25 –

166.00 ± 3.65 159.96 ± 4.89 187.86 ± 4.12 206.19 ± 4.02 211.58 ± 5.24 271.78 ± 4.26 250.79 ± 3.31 165.68 ± 2.88 196.97 ± 2.81 192.93 ± 3.36 172.10 ± 2.31 200.61 ± 2.31 184.22 ± 8.10 212.50 ± 3.70 176.74 ± 4.16 152.99 ± 2.10 135.34 ± 3.46 145.31 ± 3.91 131.46 ± 3.16 100 100 162.93 ± 3.42 123.26 ± 2.93 139.38 ± 2.79 121.00 ± 3.09 139.87 ± 3.41 114.09 ± 4.48 100

0.95 ± 0.04 0.67 ± 0.05 0.69 ± 0.03 0.69 ± 0.03 0.83 ± 0.05 0.65 ± 0.02 0.72 ± 0.03 1.11 ± 0.08 0.72 ± 0.04 0.72 ± 0.04 0.78 ± 0.06 0.80 ± 0.03 0.82 ± 0.05 0.70 ± 0.05 0.93 ± 0.06 1.20 ± 0.07 1.48 ± 0.14 1.86 ± 0.19 3.06 ± 0.24 – – 1.05 ± 0.03 1.24 ± 0.08 1.20 ± 0.08 1.33 ± 0.12 2.96 ± 0.14 3.11 ± 0.24 –

1.20 ± 0.04 0.98 ± 0.05 0.98 ± 0.04 0.94 ± 0.04 1.22 ± 0.06 0.87 ± 0.03 1.22 ± 0.06 1.55 ± 0.12 1.07 ± 0.08 0.95 ± 0.06 1.07 ± 0.08 0.94 ± 0.05 0.98 ± 0.07 0.98 ± 0.09 1.26 ± 0.08 1.66 ± 0.10 2.13 ± 0.18 3.23 ± 0.31 4.21 ± 0.24 3.70 ± 0.33 3.67 ± 0.28 1.34 ± 0.06 1.84 ± 0.10 2.13 ± 0.11 2.13 ± 0.16 4.24 ± 0.23 4.12 ± 0.17 4.46 ± 0.22

2.03 ± 0.12 2.13 ± 0.08 2.11 ± 0.09 2.28 ± 0.14 2.69 ± 0.22 2.18 ± 0.11 2.12 ± 0.18 2.03 ± 0.11 1.86 ± 0.12 2.06 ± 0.10 1.98 ± 0.09 1.93 ± 0.09 1.98 ± 0.09 2.18 ± 0.43

05.93 ± 1.98 03.09 ± 1.20 02.36 ± 0.63 02.04 ± 0.50 01.36 ± 0.41 02.37 ± 0.64 00.54 ± 0.41n.s 09.22 ± 3.83 03.85 ± 1.87 20.96 ± 4.86 12.54 ± 4.72 02.42 ± 1.19 02.44 ± 0.57 n.s.

0.13 ± 0.02 0.22 ± 0.02 0.27 ± 0.03 0.47 ± 0.05 0.51 ± 0.06 0.32 ± 0.04 0.24 ± 0.07 0.12 ± 0.02 0.14 ± 0.03 0.14 ± 0.02 0.14 ± 0.03 0.21 ± 0.04 0.34 ± 0.05 –

130.68 ± 5.43 128.36 ± 7.49 126.89 ± 4.28 142.63 ± 5.08 135.19 ± 7.55 132.96 ± 4.50 105.57 ± 4.05 146.57 ± 9.07 121.17 ± 6.51 228.08 ± 5.90 175.16 ± 8.61 120.56 ± 6.49 134.60 ± 4.50 100

0.33 ± 0.03 0.53 ± 0.04 0.66 ± 0.07 1.17 ± 0.07 1.10 ± 0.09 0.79 ± 0.07 0.50 ± 0.15 0.34 ± 0.04 0.37 ± 0.06 0.61 ± 0.05 0.51 ± 0.05 0.52 ± 0.08 0.92 ± 0.11 –

0.73 ± 0.05 1.13 ± 0.07 1.43 ± 0.10 2.22 ± 0.15 1.89 ± 0.11 1.62 ± 0.12 1.54 ± 0.15 0.72 ± 0.07 0.99 ± 0.10 1.19 ± 0.10 1.07 ± 0.08 1.31 ± 0.13 2.07 ± 0.16 1.65 ± 0.22

The parameters are given ± standard error. In the logistic model D denotes the upper limit (here C = 0), b is proportional to the slope around the concentration reducing the response by 50% (EC50), f initial rate of increase in response at subinhibitory concentrations (n.s. = f not significantly different from zero), M gives the concentration of maximal hormetic growth stimulation (ymax), and LDS is the concentration where the hormetic effect has vanished.

SC IE N CE OF TH E TOTA L E N V I RO N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

81

that two chemicals act as dilutions of each other, that is, combinations (d1, d2) with 50% effect satisfy

(100 − p) of compound B, then the linear model states that the expected maximum response μp at p is given by

d1 d2 þ ¼1 EC50;1 EC50;2

lp ¼ lA d

ð2Þ

where d1 and d2 are the concentrations of chemicals 1 and 2 in a mixture and EC50,1 and EC50,2 are the 50% concentrations of the chemicals when tested separately. Deviations from CA were described by one of two isobole models. The simpler model has one parameter describing the curvature of the isobole (λ) (Hewlett, 1969; Sørensen et al., 2007): 

1=k  1=k d1 d2 þ ¼ 1: EC50;1 EC50;2

ð3Þ

In this model λ = 1 corresponds to CA, whereas λ N 1 and λ b 1 correspond to synergism and antagonism, respectively. Using this model, the sum of toxic units (TU) for a 50:50% mixture is 21 − λ. The model is best visualized with isobolograms, showing the combinations of concentrations yielding the same effect. In an isobologram the CA-isobole is a straight line. Eq. (3) is not appropriate for data showing extreme antagonism. For these types of data the Vølund model with two parameters describing the isobole curvature (ŋ1 and ŋ2) is used (Vølund, 1992; Streibig et al., 1998): 

d1 EC50;1

g1 

d1 d2 þ EC50;1 EC50;2

1g1     d2 g2 d1 d2 1g2 þ þ ¼ 1: EC50;2 EC50;1 EC50;2

ð4Þ

Apart from allowing extreme antagonism, the Vølund model can also describe asymmetric isoboles. To test whether an isobole model was significantly different from the CA model, a non-linear regression model (the isobole model) for the EC-estimates was fitted and compared to the CA model. The variance of the EC-estimates as well as a random effect of mixture were incorporated in the model. To be specific, let zi be the estimated logarithmic EC50-value for the ith mixture (measured in standard units) and let wi be the corresponding standard error. The joint-action analysis was carried out in the model   zi e N li ;w2i þ r2 : ð5Þ Here μi =μi(EC50,1, EC50,2, λ) or μi =μi(EC50,1, EC50,2, η1, η2) incorporates the relevant isobole model, the known estimation error wi describes the within-curve variation, whereas the extra variance parameter σ describes the between-curves variation. The parameters of the model, (EC50,1, EC50,2, λ, σ) for the Hewlett model and (EC50,1, EC50,2, η1, η2, σ) for the Vølund model, were fitted simultaneously, and the approximate asymptotic standard errors were reported. Note, however, that the asymptotic normality of the estimates is somewhat dubious due to the small number of observations (seven or even five for our data sets). If the actual test for synergism/antagonism (hypothesis λ = 1 or η1 =η2 = 1) is important then a likelihood ratio test taking into account the effect of the between-curves variation should be performed [see Sørensen (2008) for details].

2.2.3.

Joint-action analysis — size of hormetic response

For the mixture effect on the maximum response we investigated if ymax changes linearly with the mixture ratio. If two compounds A and B are mixed with a proportion p of A and

p 100  p þ lB d : 100 100

ð6Þ

Based on the estimates of μA and μB and their corresponding standard errors, we constructed pointwise 95% prediction intervals for every p between 0 and 100, and evaluated if the estimated ymax values were included in the prediction intervals. For a given p the prediction interval was computed as qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   ffi lˆ p F1:96 ðp=100Þ2 s2A þ ð1  p=100Þ2 s2B þ s2A þ s2B =2

ð7Þ

where sA and sB are the standard errors for the pure compounds. Assuming that the linearity model is true, a new value will be within the band with 95% certainty, so values outside the prediction intervals indicate that the linearity model is not correct.

3.

Results

3.1.

Concentration–response analysis

Analyzing the mixture-toxicity data for parthenin versus tetraneurin-A on L. sativa showed significant hormesis for all 14 curves (f N 0; Table 1). Single concentration–response evaluations revealed that tetraneurin-A was more active towards L. sativa than parthenin in stimulation of root growth. The magnitude of ymax was on average 232 ± 19% of control treatment growth for tetraneurin-A, while parthenin stimulated root growth just 166 ± 0.2% of control (Fig. 1A). The size of maximum growth stimulation ymax declined with decreasing mixture ratio of the more hormetic compound tetraneurin-A (Table 1). Fig. 1A shows this decline for one mixture experiment, where ymax decreased from 151 ± 3% stimulation compared to control at pure tetraneurin-A, over 106 ± 4% at the 50:0% mixture to 66 ± 4% at pure parthenin. The concentration interval of stimulation was fairly equal for both compounds with an average upper limit (LDS) of 70 ± 6% of the EC50 (mean ± standard deviation; ranging from 59 to 79%). The same was observed regarding all 14 mixture-toxicity curves which had an average LDS of 73 ± 7% of the EC50 (ranging from 56 to 85%). As the single concentration–response curves for caffeic acid showed no significant hormesis in both experiments, f-values of three out of the 14 mixture-toxicity curves for parthenin versus caffeic acid were not significantly different from zero (Table 1). As a consequence of this, effective concentrations M and LDS describing the concentration range of hormesis could not be calculated for these curves. The size of the hormetic growth stimulation could, however, still be determined as in case of f = 0, ymax equals the expected response of the untreated control (D). Fig. 1B illustrates for one experiment that ymax declined similarly to the parthenin/tetraneurin-A mixture from 63 ± 3% stimulation for parthenin over 21 ± 3% at the 50:50% mixture to zero at pure caffeic acid. The average LDS for the 11 hormetic curves was 69 ± 7% of the EC50 (ranging from 58 to 78%). Evaluating the two herbicide mixtures on L. minor revealed significant hormesis for 12 of the 14 curves (Table 1). The

82

SC IE N CE OF TH E TOTA L E N V IR O N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

excluding hormesis in the concentration–response models shifted the isoboles towards slightly smaller concentrations, particularly at the EC20 level, but the curvature of the isoboles remained the same (Table 2). Comparing the shape of the isoboles at different effect levels, i.e. EC50, the concentration of maximal hormesis M, and the concentration where the growth stimulation has declined to control levels LDS, showed a similar pattern across effect levels for the two combinations of hormetic compounds tested (Fig. 3). For the parthenin/tetraneurin-A mixture all three response levels followed concentration addition (Table 3). For the herbicide mixture, where large antagonism was seen at EC50, similarly large antagonism was found at the concentrations giving hormesis (Fig. 3, Table 3).

3.3.

Fig. 1 – Selected concentration–response relationships for the inhibition of root length of Lactuca sativa. (A) Effect of parthenin, tetraneurin-A, and their 50:50% mixture (joint R2 = 0.94) and (B) effect of parthenin, caffeic acid, and their 50:50% mixture (joint R2 = 0.92).

hormetic curves showed an average growth stimulation of 41 ± 31% (ranging from 6 to 128%) which was considerably lower than the average growth stimulation observed for the phytotoxin mixtures of 72 ± 40% (ranging from 14 to 171%; n = 25). The variation in determining the ymax value was greater for the herbicide mixture-toxicity data than for the phytotoxin mixtures with an average coefficient of variation (CV calcupffiffiffi lated as sed n=estimate) of 11 ± 3% versus 6 ± 2%. The herbicide mixture-toxicity curves further showed a tendency of lower slopes with an average 1.9-fold lower b value compared to the phytotoxin mixtures. Therefore, the concentration interval of hormesis was located at relatively lower concentrations compared to the EC50. The average LDS was at 46 ± 7% of the EC50. Analyzing the variability of the determination of effective concentrations revealed no substantial differences between Lactuca and Lemna test systems. For M the CV was 27 ± 9% versus 26 ± 8%, for LDS 21 ± 11% versus 15 ± 5%, and for the EC50 14 ± 3% versus 15 ± 4%, for L. sativa and L. minor, respectively.

The hypothesis of a linear model relation for the size of the hormetic growth stimulation was tested on six data sets. The first two data sets comprised a binary mixture of only one hormetic compound, parthenin, versus caffeic acid tested on L. sativa (Fig. 4A,B). The observed ymax values essentially followed a linear pattern and declined with decreasing mixture ratio of the hormetic compound parthenin. However, merely six out of the ten ymax estimates observed at different mixture ratios were clearly located within the prediction interval, while two estimates were close to the lower bond and two differed atypically. Nevertheless, the majority of ymax observations for the mixture of parthenin and caffeic acid followed the model of linear interpolation well. Another two data sets comprised a binary mixture of two hormetic compounds (parthenin versus tetraneurin-A for L. sativa) with different ymax values (Fig. 4C,D). Although less consistent, the observed ymax values also tended to describe a linear pattern and gradually increased with increasing mixture ratio of tetraneurin-A. Five ymax values were within the prediction interval, while the other five were close to the lower bond of the prediction interval or differed atypically. Thus, the ymax observations for the mixture of two hormetic compounds also tended to follow the model of linear interpolation as the majority of observations was within or close to the prediction interval. Remarkably, two mixture ratios (80:20% and 20:80%) differed antipodal in the two experiments indicating some inter-experiment variation. Apart from this, both mixtures tested on L. sativa showed high reproducibility. The last two data sets comprised mixtures of the herbicide acifluorfen tested on L. minor either with the slightly hormetic herbicide mesotrione (Fig. 4E) or with the herbicide terbuthylazin, a compound that did not show significant hormetic effects in this setup (Fig. 4F). With three atypical deviations at both data sets the linearity model did not fit the data well.

4. 3.2.

Joint-action analysis — size of hormetic response

Discussion

Joint-action analysis — isoboles

The comparison of isoboles, constructed either on the basis of a model including hormesis or on a model excluding hormesis, showed no significant differences at the EC20 and EC50 effect level (Fig. 2, Table 2). The general trend was that

Although most environmental exposures to pollutants (e.g. pesticides) or allelochemicals are mixed exposures and low concentration stimulatory effects by these toxins have been shown for various test systems (Boekelheide, 2007; Calabrese and Blain, 2005; Cedergreen et al., 2007b), our understanding of

83

SC IE N CE OF TH E TOTA L E N V I RO N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

Fig. 2 – Isobolograms of the mixture parthenin/tetraneurin-A tested in two independent experiments (A, B) on Lactuca sativa. The filled dots give EC values fitted with the biphasic concentration–response model and the solid line describes the corresponding fitted isobole. The unfilled dots give EC values fitted with the monotonic concentration–response model and the dotted line describes the corresponding fitted isobole. Data are given as the mean ± standard error of the individually fitted curves.

mixed exposures to hormetic compounds is yet marginal. This study clearly demonstrated that hormesis is substantial, reproducible, and relevant for mixture-toxicity studies. Mixtures induced predictable hormetic effects, which adds to the large body of evidence that hormesis is a real phenomenon and renews the claim for a quantitative discussion of this phenomenon in relation to chemical mixtures. The study also showed that not all chemicals induce hormesis for all endpoints, even though it has been claimed by some that

hormesis is a general phenomenon that can be induced by all chemicals that cause toxicity (Calabrese and Baldwin, 2001). The observed induction of hormesis by acifluorfen and mesotrione has been previously reported by Cedergreen et al. (2007b), and hormesis by parthenin is known from studies of Batish et al. (1997b) and Belz et al. (2007). In contrast to this, induction of hormesis by tetraneurin-A, the compound showing the highest growth stimulation in this study, has not been previously documented. No significant hormesis was found in

Table 2 – Parameters for isoboles fitted to EC20 and EC50 values of concentration–response models either including a term for hormesis (Eq. (1)) or ignoring the possible presence of hormesis (Eq. (1) with f = 0) (Fig. 2) EC-level

EC20 EC50

Hormesis included

Yes No Yes No

Experiment 1 Intersection x-axis

y-axis

1.00 ± 0.08 1.08 ± 0.16 1.20 ± 0.11 1.23 ± 0.24

0.87 ± 0.07 0.79 ± 0.09 1.20 ± 0.12 1.15 ± 0.21

Experiment 2 λ

1.25 ± 0.12 1.45 ± 0.17 1.30 ± 0.14 1.30 ± 0.27

Intersection x-axis

y-axis

1.17 ± 0.10 0.90 ± 0.19 1.51 ± 0.11 1.64 ± 0.22

0.87 ± 0.07 0.74 ± 0.07 1.04 ± 0.08 0.91 ± 0.04

λ

1.28 ± 0.12 1.14 ± 0.21 1.35 ± 0.10 1.39 ± 0.14

The isobole model consists of three parameters: the intersection with the x-axis [parthenin (mM)], the intersection with the y-axis [tetraneurin-A (mM)] and a parameter describing the curvature of the isobole (λ). The parameters are given ± standard error.

84

SC IE N CE OF TH E TOTA L E N V IR O N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

Fig. 3 – Isobolograms of the mixture parthenin/tetraneurin-A tested on Lactuca sativa in two independent experiments (A, B) and of the mixture acifluorfen/mesotrione tested on Lemna minor (C). The solid curve describes the fitted isobole, and the dashed line describes the predicted concentration addition isobole. Data are given as the mean± standard error of the individually fitted curves.

this study, neither in the Lactuca assay for caffeic acid nor in the Lemna assay for terbuthylazine, despite the use of several no effect concentrations (Fig. 1B). This is consistent with previous reports, although terbuthylazine and other herbicides with similar mode of action have shown biphasic dose– response curves in some studies (Freney, 1965; Wiedman and Appleby, 1972; Cedergreen et al., 2005, 2007b; Jia et al., 2006). One of the questions asked, in relation to quantification of mixture studies including hormetic compounds, concerned the consequences of using monotonic concentration– response models instead of biphasic models when assessing

mixture-toxicity effects. The examples presented in this study showed that the consequences were marginal, as the adverseeffect concentrations only changed slightly depending on the type of model used. This reflects the fact that above the toxic threshold, the shape of the dose response is similar for monotonic and biphasic dose–response models (Calabrese, 2008a) and suggests that hormesis may not need to be taken into account in assessments of mixtures focusing on substantial adverse effects. In addition, it seemed that the isoboles for the concentration of maximal growth stimulation (M) and the concentration where the hormetic effect has

85

SC IE N CE OF TH E TOTA L E N V I RO N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

Table 3 – Parameters for isoboles fitted to the 50% inhibition concentration (EC50), the concentration giving maximal hormesis (M) and the concentration where the hormetic effect has vanished (LDS) Mixture

Parthenin versus tetraneurin-A (exp. 1)

Parthenin versus tetraneurin-A (exp. 2)

Acifluorfen versus mesotrione

Estimate

EC50 M LDS EC50 M LDS EC50 M LDS

Intersection x-axis

y-axis

1.20 ± 0.11 0.54 ± 0.07 0.91 ± 0.08 1.51 ± 0.11 0.44 ± 0.12 1.01 ± 0.11 262 ± 15 48.4 ± 5.48 120 ± 10

1.20 ± 0.12 0.23 ± 0.02 0.74 ± 0.05 1.04 ± 0.08 0.39 ± 0.08 0.79 ± 0.08 9.87 ± 0.86 1.36 ± 0.32 3.13 ± 0.61

λ

η1

η2

P-interval

0.95 ± 0.21 2.63 ± 1.34 2.69 ± 1.10

0.06–0.17 0.56–0.68 0.07–0.20 0.004–0.04 0.90–0.93 0.12–0.27 b 0.02 b 0.02 b 0.01

1.30 ± 0.14 1.09 ± 0.16 1.24 ± 0.11 1.35 ± 0.10 1.04 ± 0.35 1.24 ± 0.14 4.80 ± 1.25 2.55 ± 1.01 2.49 ± 0.78

The isobole model consists of three or four parameters: The intersection with the x-axis [compound A (mM)], the intersection with the y-axis [compound B (mM)] and one or two parameters describing the curvature of the isobole. For slightly synergistic or antagonistic isoboles, the curvature parameter is λ (Eq. (3)), for very antagonistic or asymmetric isoboles, the parameters are η1 and η2 (Eq. (4)). The P-interval for the significance of the deviation parameter(s) being significantly different from concentration addition is also given. All parameters are given ± standard error.

vanished (LDS), were closely linked to the EC50 isoboles. This is reflected in the similar ratios between the parameters resulting in similar isobole shapes regardless of effect concentration (Fig. 3). These results demonstrate that it is

the total load of toxicant that gives the effect, and that low doses of a chemical that, when applied singly, gives hormetic effects does not alleviate the effect of another chemical applied at concentrations giving adverse effects. If this had

Fig. 4 – Maximum stimulatory response ymax in mixture experiments. Mixture of parthenin versus caffeic acid (A, B) or tetraneurin-A (B, C) tested on Lactuca sativa in two independent experiments. Mixture of acifluorfen versus mesotrione (E) or terbuthylazin (F) tested on Lemna minor. The solid line describes the mixture ratio model relation predicted based on the pure compounds, and the dashed lines describe the prediction interval at a probability level of 95%. Data are given as the mean ± standard error of the individually fitted curves.

86

SC IE N CE OF TH E TOTA L E N V IR O N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

been the case, the isobole shapes would have shown antagonistic interactions at low doses of the individual chemicals, and more so for the EC20 isoboles compared to the EC50 isoboles. But this was not seen for any of the isobolograms. We do not know the physiological mechanism behind the growth stimulations observed, but it could be hypothesised that for chemicals, which induce hormesis through a common mechanism, the presented observations will hold true. Of particular interest here is that even a very antagonistic isobole (Fig. 3C), indicating chemical interactions, was reproduced, also at hormetic concentrations. Differences in isobole shapes, depending on effect levels in other studies, have been associated with different modes of action at different concentrations leading to different types of interactions (Syberg et al., 2008). This does not seem to be the situation for the effect of acifluorfen and mesotrione on L. minor. The amplitude of hormesis was well described with a linear model for the L. sativa test, but less consistently for the L. minor test (Fig. 4). These results suggest that the size of the hormetic growth stimulation changes linearly with mixture ratio if there are no interactions between the chemicals, as is suggested by the isoboles of the mixtures tested on L. sativa, which do not deviate significantly from CA (Fig. 3). Thus, results confirm the concept of hormetic interaction for concentration addition. If this linearity hypothesis can be verified in future mixture studies, it paves the way for both qualitative and quantitative predictions of maximum growth stimulation of mixtures of non-interacting compounds. The lack of linearity of ymax for the mixtures tested in the L. minor test-system, could either be due to the apparent interaction between the chemicals resulting in antagonism (isobole data not shown for the acifluorfen/terbuthylazine mixture), implying that the size of the hormetic growth stimulation can be increased depending on the mixture compound and ratio. This would however contradict the concept of hormetic interaction for an antagonistic response. Alternatively, the inconsistency of the L. minor results could be due to the lower amplitude of hormesis in this test system, making the determination of ymax more susceptible to experimental variance. It has been shown that even the reproducibility of the shape of isoboles, at the EC50 level in the L. minor test systems and other plant systems, is very susceptible both to within and between experiment variance (Cedergreen et al., 2007a). Possible reasons for the observed difference in hormesis amplitude between the L. sativa and L. minor test-system may include a different test species, differences in endpoint or duration of the experiment (Cedergreen et al., 2007b), or the chemical-specific mechanisms behind hormesis (Calabrese and Baldwin, 2002). Evaluations of growth rates for example, based on the relative frond area in the Lemna assay, tend to give lower hormetic effects compared to standing stock endpoints as the root length in the Lactuca assay (Cedergreen et al., 2007b). This also applies for parthenin tested on L. minor, which only induced a 10% growth rate increase corresponding to a 30% increase in leaf area after 7 days incubation (unpublished data). Amplitudes of hormesis as observed for L. sativa root growth have not been observed for L. minor and only rarely for other plant test systems where the vast majority of hormetic responses lie in the interval of 110–150% of control treatments (Calabrese and Blain, 2005; Cedergreen et al., 2007b). The large response seen in the L.

sativa test could imply that the mechanism behind the stimulation of root growth was associated to the plant hormone auxin, which is known to promote root elongation up to several-fold (Crozier et al., 2000). Based on literature reports and their own findings, Batish et al. (1997a) also assumed that parthenin may act as plant growth regulator comparable to plant auxins. Without doubt, more mixture studies are needed to conclude on the concept of hormetic interaction in case of extreme antagonism. However, as a number of studies seem to follow this interaction pattern whether or not synergy existed (Calabrese, 2008a), the deviations observed in the present study may be rather due to experimental variance.

5.

Conclusions

The present study shows that both the concentration range and amplitude of hormesis can be determined in the case of mixtures of chemicals showing large and reproducible hormesis, and which appear to follow the model of concentration addition. Furthermore, for interacting chemicals, the examples presented showed that the hormetic concentration range could be predicted from knowledge of the size of interaction at the EC50 effect level, while the prediction of the amplitude was more dubious. More studies are certainly needed to make clear statements concerning the generality of the patterns found in this study. But the study clearly showed that predicting the hormetic effect of mixtures is possible. Hence, if stimulatory effects are to be viewed as a true low concentration phenomenon of many toxicants, then this study represents a first step towards an incorporation of hormesis into mixture-toxicity evaluations.

Acknowledgements We thank BASF and Syngenta Crop Protection for providing the technical herbicides. Technical assistance is acknowledged from Christine Metzger, Helena Mastel, Jochen Schöne, and Dr. Andreas Büchse.

REFERENCES Altenburger R, Boedeker W, Faust M, Grimme LH. Regulations for combined effects of pollutants: consequences from risk assessment in aquatic toxicology. Food Chem Toxicol 1996;34:1155–7. Batish DR, Kohli KH, Saxena DB, Singh HP. Growth regulatory response of Parthenin and its derivatives. Plant Growth Regul 1997a;21:189–94. Batish DR, Kohli KH, Singh HP, Saxena DB. Studies on herbicidal activity of parthenin, a constituent of Parthenium hysterophorus, towards billgoat weed (Ageratum conyzoides). Curr Sci 1997b;73:369–71. Belz RG, Reinhardt CF, Foxcroft L, Hurle K. Residue allelopathy in Parthenium hysterophorus L. — does parthenin play a leading role? Crop Prot 2007;26:237–45. Boekelheide K. Mixed messages. Toxicol Sci 2007;99:1–2. Brain P, Cousens R. An equation to describe dose responses where there is stimulation of growth at low doses. Weed Res 1989;29:93–6.

SC IE N CE OF TH E TOTA L E N V I RO N ME N T 4 04 ( 20 0 8 ) 7 7–8 7

Calabrese EJ. Paradigm lost, paradigm found: the re-emergence of hormesis as a fundamental dose response model in the toxicological sciences. Environ Pollut 2005;138:378–411. Calabrese EJ. Hormesis: why it is important to toxicology and toxicologists. Environ Toxicol Chem 2008a;27:1451–74. Calabrese EJ. Hormesis and mixtures. Toxicol Appl Pharmacol 2008b;229:262–3. Calabrese EJ, Baldwin LA. Hormesis: a generalizable and unifying hypothesis. Crit Rev Toxicol 2001;31:353–424. Calabrese EJ, Baldwin LA. Defining hormesis. Hum Exp Toxicol 2002;21:91–7. Calabrese EJ, Blain R. The occurrence of hormetic dose responses in the toxicological literature, the hormesis database: an overview. Toxicol Appl Pharmacol 2005;202:289–301. Cedergreen N. Herbicides can stimulate plant growth. Weed Res in press. Cedergreen N, Andersen L, Olesen CF, Spliid NH, Streibig JC. Does the effect of herbicide pulse exposure on aquatic plants depend on Kow or mode of action? Aquat Toxicol 2004;71:261–71. Cedergreen N, Ritz C, Streibig JC. Improved empirical models describing hormesis. Environ Toxicol Chem 2005;24:3166–72. Cedergreen N, Kudsk P, Mathiassen SK, Sørensen H, Streibig JC. The reproducibility of binary mixture toxicity studies. Environ Toxicol Chem 2007a;26:149–56. Cedergreen N, Streibig JC, Kudsk P, Mathiassen SK, Duke SO. The occurrence of hormesis in plants and algae. Dose-response 2007b;5:150–62. Cedergreen N, Christensen AM, Kamper A, Kudsk P, Matthiasen S, Streibig JC, Sørensen H. A review of independent action as a reference model for binary mixtures of compounds with different molecular target sites. Environ Toxicol Chem 2008;27:1621–32. Crozier A, Kamiya Y, Bishop G, Yokota T. Biosynthesis of hormones and elicitor molecules. In: Buchanan BB, Cruissem W, Jones RL, editors. Biochemistry and Molecular Biology of Plants. Rockville: American Society of Plant Physiologists; 2000. p. 850–928. Duke SO, Cedergreen N, Velini ED, Belz RG. Hormesis: is it an important factor in herbicide use and allelopathy? Outlooks Pest Manag 2006;17:29–33. Forbes VE. Is hormesis an evolutionary expectation? Funct Ecol 2000;14:12–24. Freney JR. Increased growth and uptake of nutrients by corn plants treated with low levels of simazine. Aust J Agric Res 1965;16:257–63. Gessner PK. Isobolographic analysis of interactions: an update on applications and utility. Toxicology 1995;105:161–79. Greco WR, Bravo G, Parsons JC. The search for synergy: a critical review from a response surface perspective. Pharmacol Rev 1995;47:332–85. Hewlett PS. Measurement of the potency of drug mixtures. Biometrics 1969;25:478–87. International Organization for standardization. Water Quality — Duckweed Growth Inhibition. ISO/WD 20079, 1–18. Geneva: International Organization for Standardization; 2004.

87

Jia C, Kudsk J, Mathiassen K. Joint action of benzoxazinone derivatives and phenolic acids. J Agric Food Chem 2006;54:1049–57. Jonker MJ, Svendsen C, Bedaux JJM, Bongers M, Kammenga JE. Significance testing of synergistic/antagonistic, dose level-dependent, or dose ratio-dependent effects in mixture dose–response analysis. Environ Toxicol Chem 2005;24:2701–13. Kefford BJ, Zalizniak L, Warne MSJ, Nugegoda D. Is the integration of hormesis and essentiality into ecotoxicology now opening Pandora's Box? Environ Pollut 2008;151:516–23. Maeng J, Khudairi AK. Studies on the flowering mechanism of Lemna I. Amino acid changes during flower induction. Physiol Plant 1973;28:264–70. Parsons PA. Metabolic efficiency in response to environmental agents predicts hormesis and invalidates the linear no-threshold premise: ionizing radiation as a case study. Crit Rev Toxicol 2003;33:443–9. Schabenberger O, Tharp BE, Kells JJ, Penner D. Statistical tests for hormesis and effective dosages in herbicide dose response. Agron J 1999;91:713–21. Sørensen H. Small sample distribution of the likelihood ratio test in the random effects model. J Stat Plan Inference 2008;132:1605–14. Sørensen H, Cedergreen N, Skovgaard IM, Streibig JC. An isobole-based statistical model and test for synergism/ antagonism in binary mixture toxicity experiments. Environ Ecol Stat 2007;14:383–97. Southam CM, Erlich J. Effects of extract of western red-cedar heartwood on certain wood-decaying fungi in culture. Phytopathology 1943;33:517–24. Stebbing ARD. A theory for growth hormesis. Mutat Res 1998;403:249–58. Streibig JC, Jensen JE. Actions of herbicides in mixtures. In: Cobb AH, Kirkwood RC, editors. Herbicides and their mechanisms of action. Sheffield: Sheffield Academic Press; 2000. p. 153–80. Streibig JC, Kudsk P, Jensen JE. A general joint action model for herbicide mixtures. Pestic Sci 1998;53:21–8. Syberg K, Engell-Kofoed AE, Pedersen H, Cedergreen N, Forbes VE. Mixture toxicity of three toxicants with similar and dissimilar mode of action to Daphnia magna. Ecotoxicol Environ Saf 2008;69:428–36. Thayer KA, Melnick R, Burns K, Davis D, Huff J. Fundamental flaws of hormesis for public health decisions. Environ Health Perspect 2005;113:1271–6. Tomlin CDS, editor. The e-Pesticide Manual — version 2.2. Surrey: British Crop Protection Council; 2002. Vølund A. Dose response surface bioassay. Proceedings XVIth International Biometric Conference, 7–11 December 1992, Hamilton, New Zealand; 1992. p. 249. White DB, Faessel HM, Slocum HK, Khinkis L, Greco WR. Nonlinear response surface and mixture experiment methodologies applied to the study of synergism. Biometrical J 2004;46:56–71. Wiedman SJ, Appleby AP. Plant growth stimulation by sublethal concentrations of herbicides. Weed Res 1972;12:65–74.