Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach

Environmental Pollution xxx (2017) 1e12

Contents lists available at ScienceDirect

Environmental Pollution journal homepage: www.elsevier.com/locate/envpol

Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach*
a Beaumelle, Claire Della Vedova, Karine Beaugelin-Seiller, Jacqueline Garnier-Laplace, Le Rodolphe Gilbin* ^t. 159, BP 3, 13115 St Paul-lez-Durance, France Institute for Radioprotection and Nuclear Safety (IRSN), PRP-ENV/SERIS/LRTE, Cadarache, Ba

a r t i c l e i n f o

a b s t r a c t

Article history: Received 26 January 2017 Received in revised form 31 May 2017 Accepted 1 September 2017 Available online xxx

A main challenge in ecological risk assessment is to account for the impact of multiple stressors. Nuclear facilities can release both radiological and chemical stressors in the environment. This study is the first to apply species sensitivity distribution (SSD) combined with mixture models (concentration addition (CA) and independent action (IA)) to derive an integrated proxy of the ecological impact of combined radiological and chemical stressors: msPAF (multisubstance potentially affected fraction of species). The approach was tested on the routine liquid effluents from nuclear power plants that contain both radioactive and stable chemicals. The SSD of ionising radiation was significantly flatter than the SSD of 8 stable chemicals (namely Cr, Cu, Ni, Pb, Zn, B, chlorides and sulphates). This difference in shape had strong implications for the selection of the appropriate mixture model: contrarily to the general expectations the IA model gave more conservative (higher msPAF) results than the CA model. The msPAF approach was further used to rank the relative potential impact of radiological versus chemical stressors. © 2017 Elsevier Ltd. All rights reserved.

Keywords: Mixture effects Species sensitivity distribution Ionising radiation Multiple stressors

1. Introduction The impact of multiple stressors on ecosystems is a challenging research area. While nuclear facilities release both radiological and chemical stressors to the environment under normal operating conditions, ecological risk assessment procedures still focus on the separate risks engendered by single types of stressors. It is widely recognised that (i) the effect of a mixture of substances or of stressors is different from the simple summation of the individual effects of each of its components, and (ii) that a mixture can exert a significant toxicity to organisms even if all of its components are present at concentrations below the thresholds that individually do not provoke any toxicity (i.e. NOECs: No observed effect concentration) (Kortenkamp et al., 2009; Backhaus et al., 2010). In a recent extended inter-laboratory experiment on various species, Carvalho et al. (2014) demonstrated that regulatory safety concentrations may not provide sufficient protection when chemicals occur in mixture. Posthuma et al. (2016) further demonstrated that chemical mixtures significantly impacted aquatic communities in several

*

This paper has been recommended for acceptance by B. Nowack. * Corresponding author. E-mail address: [email protected] (R. Gilbin).

field-based studies. It is thus crucial to develop ecological risk assessment tools accounting for the impact of mixture of stressors. Two classical mixture models proved their worth predicting the € ewe and effect of chemical mixtures: concentration addition (CA (Lo Muischnek, 1926)) and independent action (IA (Bliss, 1939)). In a recent review, Vanhoudt et al. (2012) have concluded that no conceptual limitation prevents the use of these two general concepts to model the effects of mixture of radioactive and stable substances and address the challenging issue of assessing the environmental impact of mixtures that include radioactive substances. CA and IA both formulate the assumption of additivity or non-interaction (i.e. the components of the chemical mixture act without diminishing or enhancing each other's toxicity). Although this assumption has been challenged both theoretically and empirically (e.g. for mixture of metals (Vijver et al., 2011) and mixture of radionuclides and metals (Margerit et al., 2015)), recent reviews concluded that synergistic effects (i.e. positive interaction between substances leading to a higher mixture effect than predicted by the additivity models) were rare at environmentally relevant concentrations (Cedergreen, 2014; Kortenkamp et al., 2007). Additivity models are thus not expected to severely underestimate the effect of chemical mixtures. The CA model has been repeatedly recommended as a reference model for risk assessment purposes (Cedergreen, 2014; Backhaus

https://doi.org/10.1016/j.envpol.2017.09.003 0269-7491/© 2017 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

2

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

et al., 2010). The rationale is that CA generally provides more conservative predictions than IA (i.e. CA predicts higher mixture effects than IA). The CA model assumes that the substances in the mixture have a similar mode of toxic action. This assumption is difficult to verify because for a number of chemicals the toxic mode of action is unknown and varies according to the species, the life stage, or even from one organ to another within the same organism (Syberg et al., 2009; Kortenkamp et al., 2009; Margerit et al., 2016). Thus, in the case of complex mixtures such as the releases of nuclear facilities, it is not clear whether the CA model is the best option. The IA model assumes that all the compounds in the mixture have different toxic modes of action. Cedergreen et al. (2008) showed that both CA and IA models are equivalent in terms of prediction accuracy, thus the choice of the mixture model to apply is not straightforward. Both models can be used to provide a prediction window in which the ’true’ effect of a mixture can be expected to fall (Backhaus et al., 2010; Altenburger et al., 2013). Other approaches also rely on combination of CA and IA models in which CA is applied on groups of substances sharing a common mode of toxic action and IA is applied across groups of substances with distinct modes of toxic action (e.g. de Zwart and Posthuma, 2005; Smetanov
a et al., 2014). The CA and IA mixture models can be combined with species sensitivity distributions (SSD) models to provide a single indicator of ecological risk associated with mixtures: the multisubstance potentially affected fraction of species (msPAF) (de Zwart and Posthuma, 2005). SSDs are widely used to account for the variation in species’ sensitivity when deriving maximum acceptable concentrations of a substance (prospective risk assessment) and in retrospective risk assessment approaches (Jesenska et al., 2013). Based on ecotoxicological data (NOEC, EC10 or EC50 (Effective concentration affecting 10 or 50 % of the test population)) for different species and different taxonomic groups, a SSD model expresses the proportion of species potentially affected by the exposure concentration of a given chemical (PAF). In a recent review, Posthuma et al. (2016) showed the interest of multisubstance SSD as a useful lower-tier risk assessment model providing a good proxy (msPAF) of the potential ecological impact of environmental stressors. The disadvantages of SSD modelling reviewed by Forbes and Calow (2002) notably include a lack of ecological realism (lack of data for representative species, ecological interactions not taken into account). Posthuma and de Zwart (2012) proposed to view msPAF as a metric of the toxic pressure characterising a water sample, that does not predict directly the ecological impact, but whose value can be related to the ecological impact. Indeed, several studies have shown significant relationships between msPAF and observed taxon abundance (Posthuma et al., 2016; Posthuma and de Zwart, 2012) or functional diversity (Smetanov
a et al., 2014; Jesenska et al., 2013). Finally, the msPAF approach is useful to rank substances and/or stressors within mixture according to their relative contribution to the ecological impact (Posthuma et al., 2016). The msPAF is thus a promising proxy for screening-level ecological risk assessment (ERA) of chemical mixture, but it has never been applied to mixtures that include radionuclides. Until now, only a few attempts have considered radiological and chemical stressors altogether in ERA. Garnier-Laplace et al. (2009) provided a screening level ERA of the liquid effluents from nuclear power plants that integrated radioactive and stable compounds based on ecotoxicological data. In that study, SSD curves were simplified as linear relationships inspired from a life cycle analysis (LCA) approach (Pennington et al., 2004). That previous study was a first attempt to integrate radiological and chemical stressors in an ERA approach that needs to be refined, notably by including the recent advances in the field. Garnier-Laplace et al. (2009) applied

CA but not IA model and concluded that the impact of radionuclides was 5 orders of magnitude lower than the impact of stable chemical substances based on a linear approach. van de Meent and Huijbregts (2005) recommended using the non-linear msPAF approach to estimate effect factors in LCA. Indeed, the msPAF approach best reflects the distribution of ecotoxicological responses, and it further offers the possibility to aggregate other stressors (such as eutrophication or warming) meaningfully (van de Meent and Huijbregts, 2005; van Zelm et al., 2007). Integrating the non-linear patterns of the distribution of species sensitivity could thus (i) affect the ranking of the potential ecological impact of radiological and chemical stressors previously derived, and (ii) provide a ranking that takes advantage of all the information available within ecotoxicological data. The present study tests and discusses the msPAF approach on the routine releases from nuclear facilities, with the aim to contribute to the improvement of existing ERA tools and framework by explicitly addressing the effect of mixture of radiological and chemical stressors. The msPAF approach was applied for the first time to the case of the liquid effluents released under normal operating conditions in a large river by four nuclear power plants located in the watershed. The two mixture models CA and IA and their combination (i.e. CA applied on separate groups of substances and IA across groups) were compared for different exposure scenarios (dilutions of the liquid effluents in the river). The msPAF approach was further used to rank two categories of stressors within the liquid effluents according to their potential ecological impact: ionising radiation and stable chemicals. The ranking results were compared with the ranking based on the linear approach carried out by Garnier-Laplace et al. (2009). 2. Material and methods The general methodology of this study is illustrated in a flow diagram in Fig. 1. 2.1. Exposure scenarios 2.1.1. Characteristics of the liquid effluents This study focused on the liquid effluents from four French nuclear power plants (Bugey (P1), Saint Alban (P2), Cruas (P3) and Tricastin (P4), ordered from upstream to downstream) under ^ne river in 2013. The normal operating conditions into the Rho chemical and radiological composition of the effluents were retrieved from the EDF environment report (EDF, 2013). Table 1 shows the annual released quantities of the 22 stable chemicals and 13 radionuclides. The effluents of these four plants covered a range of different mixture compositions. For example the P3 plant released more stable chemicals than the three other plants (Table 1). The range of exposure scenarios was further extended using three different dilution scenarios of the effluents into the river: (i) ^ ne near the nuclear plant at the mean flow rate of the Rho (465 m3 s
1 for P1, 590 m3.s
1 for P2, 1480 m3.s
1 for P3 and re-Mondragon canal) (Banque 1167 m3.s
1 for P4 (in the Donze Hydro, 2016)), (ii) at the minimum flow rate above which the national nuclear safety authority (ASN) authorises to discharge effluents into the river (130, 255, 300 and 400 m3.s
1 respectively for the plants P1 (ASN, 2013), P2 (ASN, 2014b), P3 (ASN, 2014a) and P4 (ASN, 2008) and (iii) the pure effluents, considered here as a worstcase scenario to test the model predictions at maximal (unrealistic) exposure levels. The concentrations of the different chemicals and radionuclides were obtained considering a simple dilution model in which the releases were averaged by year (Garnier-Laplace et al., 2009):

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

3

4

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

QF Ci ¼ i rel Vtot Friv

(1)

where Ci is the concentration (mg.L
1 for stable chemicals) or activity (Bq.L
1 for radionuclides) of substance i, Qi is the cumulated quantity of substance i released throughout the year (in Bq or kg), Vtot is the cumulated volume of liquid effluents throughout the year (L), Frel is the flow rate of the releases (m3.s
1 considering a constant release, i.e. dividing the cumulated volume of liquid effluents by the number of seconds per year) and Friv is the flow rate of the ^ ne river (m3.s
1). The concentrations of boric acid were Rho expressed in terms of boron (B) equivalents using the conversion factor given in ECHA (2008b). We kept simple assumptions to calculate the exposure concentrations because the aim of the study was to test the msPAF approach and not to assess the ecological risks of the liquid effluents. The exposure scenarios (4 mixture scenarios (liquid effluents per nuclear plant) each with 3 dilution scenarios) were thus based on total concentrations in water. 2.1.2. Dose rate calculations The effect of radionuclides on organisms is mainly due to the ionising radiation they produce. For radionuclides, chronic effect data are thus related to dose rates (the energy deposited to the organism by unit time) and not directly to the radionuclide concentration in the exposure media. Dose rate calculation involves a species- and radionuclide-specific coefficient of conversion (Dose coefficient: DC) from concentration in medium (Bq.L
1) to dose rate absorbed by the organism (mGy.h
1) (Beaugelin-Seiller et al., 2006). The DC integrates the fact that organisms are exposed both externally (to water and sediment for freshwater organisms) and internally (through direct absorption and trophic pathway). We computed the total DC of each radionuclide for a set of 13 reference freshwater organisms (amphibians, benthic fish, bird, crustacean, insect larvae, mammal, bivalve mollusc, gastropod mollusc, pelagic fish, phytoplankton, reptile, vascular plant and zooplankton) according to the ERICA tool (an integrated approach to the assessment and management of environmental risks from ionising radiation: Beresford et al. (2007)). For each radionuclide and each reference organism, the total DC was computed as follows:

DCtotal ¼ OFw :DCext;w þ ð1
OFw Þ:Kd:DCext;s þ CR:DCint

(2)

where DCtotal is the total dose coefficient associated to external and internal exposures to ionising radiation (mGy.h
1/Bq.L
1); OFw is the occupancy factor of the organism in the water (equal to 1 for pelagic organisms, birds and to 0.5 for organisms that live upon the sediment); DCext,w is the DC associated with external exposure in water (mGy.h
1/Bq.L
1); Kd is the solid-liquid partitioning coefficient of the radionuclide considered (L.kg
1); DCext,s represents the dose coefficient associated with the external exposure to sediment (it was set to DCext,w according to Beresford et al. (2007)); CR is the concentration ratio related to the radionuclide transfer from water to the organism, including the trophic pathway (L.kg
1 fresh

Table 1 Cumulated quantities of radioactive and stable substances released in 2013 by four French nuclear power plants (P1, P2, P3, P4, see the text for the codification of the ^ne river and cumulated volums of the liquid effluents throughout plants) in the Rho the year. Substances shown in gray are the substances for which SSDs could not be derived.

110m

Ag (GBq) 14 C (GBq) 58 Co (GBq) 60 Co (GBq) 134 Cs (GBq) 137 Cs (GBq) 3 H (GBq) 131 I (GBq) 54 Mn (GBq) 63 Ni (GBq) 124 Sb (GBq) 125 Sb (GBq) 123m Te (GBq) Al (kg) Cr (kg) Cu (kg) Fe (kg) Mn (kg) Ni (kg) Pb (kg) Zn (kg) Hydrazine (kg) Boric acid (kg) Lithium hydroxide (kg) Morpholine (kg) NH4 (kg) Chlorides (kg) Sulphates (kg)

P1

P2

P3

P4

0.465 38.1 0.449 0.104 0.017 0.020 44300 0.016 0.017 0.130 0.046 0.081 0.013 16.77 0.000 16.56 53.90 3.960 0.020 0.000 7.22 2.81 9000 0.710 700 2758 44200 142000

0.011 30.9 0.039 0.065 0.011 0.013 55600 0.012 0.011 0.143 0.011 0.032 0.009 2.72 0.000 0.72 25.95 1.200 0.450 0.000 4.28 0.27 3730 0.072 177 421 33000 NA

0.130 46.7 0.129 0.181 0.042 0.039 58300 0.030 0.037 0.189 0.033 0.092 0.038 9.10 0.685 15100.00 49.58 8.428 1.317 1.427 6600.00 1.81 13700 NA 450 107 NA 3967000

0.054 44.6 0.055 0.163 0.032 0.038 51300 0.031 0.035 0.071 0.032 0.092 0.036 49.23 0.000 1.86 194.48 57.120 0.000 0.910 9.33 7.07 9050 0.200 792 NA 13900 98000

weight); and DCint is the dose coefficient associated with the internal exposure (mGy.h
1/Bq.kg
1 fresh weight). The parameter values were retrieved from the ERICA tool (Beresford et al., 2007; Brown et al., 2016). DC were weighted with regard to the difference of relative biological effectiveness between ionising radiation type (i.e. weight of 3 for low beta radiation and 10 for alpha radiation) according to the ERICA approach. Dose rates (mGy.h
1) were calculated by multiplying the activity of each radionuclide (Bq.L
1, equation (1)) by the corresponding DCtotal of the radionuclide (mGy.h
1 per Bq.L
1, equation (2)). A total dose rate associated with the exposure to the mixture of radionuclides was then calculated by summing the dose rates associated to each radionuclide according to the additivity assumption. Total dose rates are usually specific of the organism (through the use of specific DCtotal values). However, the calculation of the msPAF value requires a single total dose rate representative of multiple organisms. In order to calculate these global (multiple species) total (multiple radionuclides) dose rates, we used for each radionuclide the maximum DCtotal value across the DCtotal of all 13 reference organism instead of using the species-specific DCtotal values. The maximum value was chosen to adopt a conservative approach (i.e. using the maximum DCtotal maximizes the dose rate). Preliminary analyses showed little effect of the DCtotal parameter on the sum of dose rates and on the resulting msPAF value (specifically, using the

Fig. 1. Overview of the methodology used to compute msPAF for mixtures of radiological and chemical stressors. Hexagonal boxes: input data; rounded boxes: calculations; rectangular boxes: calculation outcomes. Reference is given to text (x), Tables, Equations (Eq.) and Figures. Gray areas represent the steps of the methodology: ① Dilution scenarios. Qi: cumulated quantity of substance i released throughout the year (in Bq for radioactive substances or kg for stable substances); Ci: concentration (mg.L
1 for stable substances) or activity (Bq.L
1 for radioactive substances) of substance i. ② Total dose rate calculation. DRi: dose rate (mGy.h
1) associated to external and internal exposures to ionising radiation resulting from radioactive substance i. The total dose rate represents the exposure to ionising radiation from all the radioactive substances. ③ SSD fit. NOECi/ECxi: chronic effect data of stable substance i for freshwater organisms; EDR10: chronic effect data of external gamma radiation for aquatic organisms; SSDi: individual SSD computed for each stable substance i or for external gamma radiation. ④ Independent Action. PAFi: Potentially affected fraction of species computed for each stable substance i or radioactive substances, used for msPAFIA calculations. ⑤ Concentration Addition. HU: Hazard units computed for each stable substance i or radioactive substances, used for msPAFCAC calculations (through the use of the CA-SSD).

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

median, mean or maximum DCtotal had little impact on the final dose rates and msPAF values). To give the order of magnitude of the total dose rates calculated, the total dose rates used to compute msPAF for the mean-flow scenario were: 0.024, 0.010, 0.007 and 0.008 mGy.h
1 for the P1, P2, P3 and P4 nuclear plants respectively. These values are far below the background levels reported by Brown et al. (2004) for naturally occurring radionuclides, ranging from 0.12 to 57 mGy.h
1 for pelagic fish and macrophytes respectively. 2.2. SSD 2.2.1. Data collection For each released substance, ecotoxicological data were collected from published SSDs based on chronic effect data for freshwater organisms (Table 2). Chronic effect data (NOEC/ECx) were mostly EC10 (EDR10 (Effective dose rate affecting 10 % of the test population) for ionising radiation), and NOEC or other equivalent endpoints (Table 2). Geometric means of NOEC/ECx were used when several values of similar endpoint types were obtained for different species. The effects considered were reproduction, mortality, morbidity and growth. We did not question the data selection used in the published SSDs and the selected sets of NOEC/ECx that were reported were used as such in the present study. Overall, data were available for ionising radiation (radionuclides) but only for 8 stable chemicals out of 22. Thus, one limitation of the study is that the msPAF approach was tested for a truncated mixture of these 8

5

stable chemicals and radionuclides. For lead (Pb) and zinc (Zn), the recent studies of Van Sprang et al. (2009, 2016) updated the datasets used in EU VRAR (2008a) and EU RAR (2010) with new ecotoxicological data and re-evaluated the previously selected NOEC, excluding, for example, studies that did not measure the nominal Zn or Pb concentration in the media. We relied on those recent reports for the present study. For B, data were retrieved from the Annex XV transitional report (ECHA, 2008b). As the dataset presented in this report lacks data for insects, it was completed according to Schoderboeck et al. (2011), taking into account two additional chronic effect data for non-aquatic insects in order to have a sufficient number of taxonomic groups. For chlorides and sulphates, chronic effect data covered various endpoints (IC10: 10 % inhibition concentration, IC25, EC50, MATC: Maximum acceptable toxicant concentration) as shown in Table 2. Elphick et al. (2011a, b) followed the guidelines of Environment Canada and not the European technical guidance document (ECHA, 2008a) recommendations. They selected ecotoxicological data with the following order of preference MATC > NOEC > LOEC (lowest observed effect concentration > EC50. When IC10 was lower than the NOEC, they selected IC25. For radionuclides, we relied on the previously published SSDs of g radiation extended to the case of ionising radiation (GarnierLaplace et al., 2010; IAEA, 2014). Those previous SSDs used the minimum EDR10 values for a given species when several EDR10 were available. In addition, both terrestrial and aquatic organisms were integrated. In the present study, we modified the EDR10

Table 2 Summary of ecotoxicological data used to derive the SSDs of ionising radiation and stable chemical substances. HC50/HDR50 and HC5/HDR5 and corresponding bootstrap 95 % confidence interval (CI) are the values calculated in the present study (log-normal SSD). Substance

Taxonomic groups (no. species) no. no. Endpoint no. type NOEC/ species taxonomic groups ECx

IAEA (2014); Garnier-Laplace et al. (2010) (we used fish(4); cladocerans (2); annelids (2); cyanobacteria (1); decapoda (1); molluscs only the data for aquatic organisms and calculated geometric means of EDR10 instead of using the (2) minimum values) fish (9); amphibians (1); cladocerans (3); INERIS (2005) and Maycock et al. (2007) (we used data for Cr(VI)) insects (1); coelenterates (2); molluscs (1); algae (8); macrophytes (4)

Ionising EDR10 radiation

67

12

6

Cr

NOEC, EC10, EC50

NA

29

8

Cu

NOEC

99

27

8

fish (10); cladocerans (3); insects (3); rotifers (1); amphipods (2); molluscs (4); algae (3); macrophytes (1);

Ni

NOEC 209 and EC10

31

10

Pb

NOEC 66 and EC10

25

8

fish (3); amphibians (3); cladocerans (8); insects (2); hydrozoans (1); molluscs (2); rotifers (1); amphipods (1); algae (8); macrophytes (2) fish (9); cladocerans (4); insects (3); rotifers (2); molluscs (2); amphipods (1); algae (3); macrophytes (1)

Zn

NOEC

19

7

fish (8); cladocerans (3); insects (1); rotifers (2); molluscs (2); amphipods (1); algae (2)

B

NOEC 53 and EC10

15

7

fish (5); amphibians (2); cladocerans (2); insects (2, non aquatic); bacteria (1); algae (2); macrophytes (1)

Chlorides

IC10, IC25, EC50, MATC NOEC, EC10, EC25

16

15

8

fish (2); cladocerans (3); insects (2); annelids (2); amphipods (1); rotifers (1); algae (3); macrophytes (1)

10

8

6

fish (2); amphibians (1); cladocerans (2); amphipods (1); rotifers (1); algae (1)

Sulphates

159

references

HC5 or HC50 or HDR50 (95 HDR5 (95 % CI) % CI)

2663.5 (473e14 294) mGy.h
1 1.6 10
1 (9.3 10
2 2.9 10
1) mg.L
1 EU VRAR (2008b) 2.3 10
2 (1.7 10
2 3.1 10
2) mg.L
1 EU RAR (2008) (we used the non-normalised 4.3 10
2 NOEC/EC10) (2.9 10
2 6.7 10
2) mg.L
1 Van Sprang et al. (2016); EU VRAR (2008a) 4.5 10
2 (3.1 10
2 6.7 10
2) mg.L
1 Van Sprang et al. (2009); EU RAR (2010) (we used 9.5 10
2 the non normalized NOEC) (6.0 10
2 1.5 10
1) mg.L
1 ECHA (2008b); Schoderboeck et al. (2011) 1.2 101 (7.3e1.9 101) mg.L
1 Elphick et al. (2011a) 1.2 103 (7.9 102e1.8 103) mg.L
1 Elphick et al. (2011) (we used the largest dataset 8.2 102 reported in the paper: for a water hardness of (6.1 102e80 mg.L
1) 1.1 103) mg.L
1

19.5 (1.9 e334.6) mGy.h
1 1.3 10
3 (5.8 10
3 3.22 10
2) mg.L
1 6.3 10
3 (4.2 10
2 1.0 10
2) mg.L
1 5.9 10
3 (3.2 10
3 1.2 10
2) mg.L
1 8.5 10
3 (4.8 10
3 1.6 10
2) mg.L
1 1.6 10
2 (8.4 10
3 3.6 10
2) mg.L
1 2.4 (1.2 e5.5) mg.L
1 3.1 102 (1.7 102e- 6.2 102) mg.L
1 4.1 102 (2.8 102e- 6.8 102) mg.L
1

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

6

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

selection compared to the work of Garnier-Laplace et al. (2010) and IAEA (2014) with the objective to harmonise the ecotoxicological data selection across substances. The present SSD only included aquatic organisms. We further derived the geometric means of EDR10 when several EDR10 of similar endpoint types were available for a given species. When several values (EDR10 and/or geometric means) for different endpoint types were available for a given species, we selected the minimum value across the set of EDR10 and/or geometric means for that species. Both freshwater and marine species were considered, as previous study reported no significant differences between freshwater and marine species sensitivity to ionising radiation (Larsson, 2008).

2.2.2. Derivation of individual substance SSD A single distribution model needs to be adopted to perform multisubstance SSD with the CA model (see further explanation below). The log-normal model was used preferentially because (i) all the sets of NOEC/ECx were log-normally distributed according to Shapiro-Wilke, Anderson-Darling and Cramer von Mises tests (with an alpha of 0.05) and (ii) this model is recommended by ECHA (2008a) and several authors (e.g. van Zelm et al., 2007; Aldenberg and Jaworska, 2000; Wheeler et al., 2002). The log-normal distribution has two parameters: the scale parameter (s) is the standard deviation of the set of log-NOEC/ECx, and the location parameter (m) is the mean value of the set of log-NOEC/ECx. Individual SSDs were computed using the fitdistrplus package (Delignette-Muller and Dutang, 2015) of the R software (R Core Team, 2015). Once the normality of each set of log-NOEC/ECx was verified based on Shapiro-Wilk, Anderson-Darling and Cramer-von Mises tests (Gross and Ligges, 2015), the distributions were fitted by maximum likelihood estimation using the fitdist function (Delignette-Muller and Dutang, 2015) whose outputs were the two parameters of each SSD.

2.3. msPAF calculation Three mixture models were used to calculate msPAF: CA, IA and a combination of CA and IA.

2.3.1. Concentration addition (CA) model The CA model is used when substances are supposed to share a similar mode of toxic action. This toxicological assumption is translated into the mathematical assumption that the doseresponse curves of the different substances have similar slopes. Extending the CA model to SSD, the mathematical assumption no longer applies on the dose-response curve of individual species, but on the SSD curves that include multiple species. Using the CA model combined with log-normal SSDs consists in creating a multisubstance SSD curve having a scale parameter (s) that equals the mean scale parameters of the individual substance SSDs. It implies that the SSDs of the different substances have a similar s parameter (i.e. same toxic mode of action). de Zwart and Posthuma (2005) suggested that the slope parameters must not differ more than 0.1 order of magnitude from each other to apply CA with log-logistic SSD. In the log-normal model, as the s parameter of the SSD is the standard deviation of the log-NOEC/ECx values, it is possible to test the differences between scale parameters using a Bartlett's test of homogeneity of variance of the logNOEC/ECx values (Sala et al., 2012). In the CA model, the concentrations of different substances were expressed in a unit that accounts for the relative potency of each substance: hazard unit (HU), calculated as follows:

X

HU ¼

n X i¼1

Ci HC50i

(3)

where Ci is the concentration of substance i in the environment (or dose rate for ionising radiation), and HC50i is the hazardous concentration of substance i potentially affecting 50 % of the species (or the HDR50 for ionising radiation), calculated from the individual substance SSD using the quantiles function of the R package fitdistrplus. Table 2 shows the HC50 and HDR50 and the HC5 and HDR5 and their bootstrap 95 % confidence intervals computed using the bootdist function of the fitdistrplus package. PAF and msPAF were calculated according to the distribution function of the log-normal distribution, using the plnorm function in R: x¼X Z

PAFðXÞ ¼ x¼0

" # 1 ðlogðxÞ
mÞ2 pffiffiffiffiffiffi exp
dx 2s2 s 2p

(4)

where m and s are the two parameters of the log-normal distribution, and X is the exposure concentration. Under the CA model, msPAFCA were computed using equation (4) with m ¼ 0, s ¼ the mean of the si of the SSDs of the substances in the mixture, and for X ¼ S HU (Smetanov
a et al., 2014; Jesenska et al., 2013; de Zwart and Posthuma, 2005). It is noteworthy that the exposure to radionuclides is calculated according to an approach that is close to the CA model (i.e. by summing the individual dose rates associated to each radionuclide into a total dose rate associated with the exposure to the mixture of radionuclides). Indeed, in the case of radionuclides, the assumption of a common toxic mode of action is usually adopted: the chronic effects of radionuclides are linked to the energy of ionising radiation deposited in the organism by unit time (dose rate). All radionuclides contribute to the total dose rate, with a magnitude that is specific to each radionuclide (accounted for through the DCtot parameter). The total dose rate was thus calculated by summing the individual dose rates corresponding to each radionuclide activity. The PAF associated to the ionising radiation stressor was then calculated based on the total dose rate and on the single SSD of g radiation. This is close to the CA model reasoning in which all the components of a mixture contribute to the overall hazard (sum of HU, used to compute the msPAFCA) with a magnitude that depends on their toxicity (accounted for through the HC50 parameter, eq. (3). 2.3.2. Independent action (IA) model The IA model is used when stressors are considered to have dissimilar modes of toxic action. The underlying assumption of this model is that different substances will act according to the statistical concept of independent random events. The PAF of each substance was first calculated based on the individual substance SSD according to equation (4), then the msPAFIA was computed based on the following equation for n substances:

msPAFIA ¼ 1


n Y

ð1
PAFi Þ

(5)

i¼1

2.3.3. Combination of CA and IA models It is also possible to combine CA and IA models according to the mode of toxic action of each substance (de Zwart and Posthuma, 2005). The CA model is applied to groups of substances sharing a common mode of toxic action and an msPAFCA is calculated for each

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

group according to equation (3). Then, the msPAF of the total mixture is computed using the IA model (msPAFIA) based on equation (5) replacing the PAFi by the msPAFCA for the different groups of substances. As the modes of toxic action of the studied substances are not fully established, we proposed to group substances according to the scale parameter of their respective SSD. The groups are described in the results section.

2.4. Ranking radiological and chemical stressors The msPAF approach was further used to rank the potential impact of chemical and radiological stressors within the studied mixtures. For chemical stressors, two msPAFs were calculated according to the CA and IA models. For radiological stressors, the IA model was not considered relevant as discussed in section 2.3.1, and a single msPAF was calculated based on the SSD of g radiation and using the total dose rate associated to the mixture of radionuclides. Individual PAFs of each stable chemical and each radionuclide were also computed to identify which were the potentially most impacting substances. The calculation of individual PAF values for radionuclides was based on the dose rate associated to each radionuclide. The results were compared with those of a ranking approach based on linear relationship between the potential impact and the exposure concentration that was described in Garnier-Laplace et al. (2009). Briefly, the effect factor (EF or DPAF) was calculated according to the following linear equation: n X i¼1

Ci HC50i

(6)

The approach is based on the CA concept with the use of HU (see equation (3)), but the EF is calculated based on a linear model whose slope is 0.5. The multisubstance dose-response relationship is a straight line from 0 to 1 HU, and Garnier-Laplace et al. (2009) assumed that such a relationship would be more conservative than the msPAF approach. We compared msPAF and EF values for the chemical and radiological stressors in the liquid effluents from the nuclear power plants for the mean-flow scenario.

a

3. Results and discussion 3.1. Individual SSDs of chemical and radiological stressors: choice of mixture model Fig. 2 shows the individual SSDs of the substances considered centred on 1 HU (1 HU of any substance potentially affecting 50 % of the species). The multisubstance SSD under the CA model, whose scale parameter is the mean scale parameter of the different SSDs, is also shown Fig. 2. The mean scale parameter was 1.3 ± 0.9 across n ¼ 21 substances (13 radionuclides and 8 stable chemicals). It is noteworthy that the lowest s value was obtained for sulfates, characterised by a low number of ecotoxicological data (Table 2). Fig. 2 shows the strong differences between the shape of the SSD of g radiation (g-SSD) and that of the other SSDs. The value of the s parameter for the g-SSD was higher than for the other substances (Fig. 2 b), resulting from a higher variability of the log-EDR10 compared to the variability of the log-NOEC/ECx of other substances. This generated a g-SSD curve with a lower steepness compared to SSDs of stable chemicals (high s values are associated to low steepness in log-normal distributions). The steepness of the g-SSD was thus lower than that of the other curves, and resembled more a linear than an s-shape curve on a logarithmic scale (Fig. 2 a). This difference between the shape of the SSDs of radiological and chemical stressors may result from a difference in their toxic modes of action (de Zwart and Posthuma, 2005). de Zwart and Posthuma (2005) suggested that the slope parameters of individual log-logistic SSD curves should not differ more than 10 % from each other in order to apply the CA model. Applying this criterion to the s parameters of our log-normal SSDs, all the substances considered in the present study should be treated separately using the IA model. The Bartlett's test of homogeneity of variances used to compare the variances of the sets of log-NOEC/ ECx of the nine stressors, showed that the variances were not equal (p < 0.05). The assumption of a common s for the SSDs of the 8 stable chemicals and of g radiation (tested with the Bartlett's test) was not consistent with the data. Thus, the CA model may not be the best option to calculate the potential impact of mixtures that include such stressors.

b

Fraction of species affected (PAF)

1.00

0.75

3

0.50 Gamma CA model Cr Ni Zn Pb Boric acid Chlorides Cu Sulphates

0.25

0.00 0.001

0.01

0.1

1

10

σ parameter

EF ¼ 0:5

7

2 1 0

a Cr Ni Zn Pb cid es Cu tes a rid a ric hlo lph Bo C Su

mm

Ga

100

Hazard Units Fig. 2. Comparison of the SSDs of the 8 stable chemicals, g radiation, and the multisubstance CA-SSD under the assumption of the concentration addition (CA) mixture model: (a) in log-10 of hazard units (HU); and (b) s parameter values of each SSD with their associated standard errors (error bars) and where similar colour indicate similar s values (i.e. similar variances of log-NOEC/ECx according to a Bartlett's test of homogeneity of variances). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

8

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

The other options are to use the IA model or a combination of CA and IA models based on grouping substances sharing a common toxic mode of action. For the stressors considered in the present study, it is difficult to assign each substance to a given toxic mode of action, especially considering that one substance do not necessarily have the same toxic mode of action for different species within the SSD framework (Syberg et al., 2009). However, several substances exhibited SSDs with very close s parameters (Fig. 2). The s parameters of Ni, Zn, Pb, Boric acid, chlorides and Cu ranged between 0.8 and 1.2 and were not significantly different (Bartlett's test, p ¼ 0.22). Based on these results, we constituted four groups of substances according to the s parameters of their respective SSD. The groups are shown in different colours in Fig. 2b. Three mixture models were thus compared for the msPAF calculation: CA (considering a single SSD), IA (using each individual SSD) and a combination of CA and IA (based on the 4 groups of substances defined above).

3.2. Comparing msPAFs from different mixture models Fig. 3 shows the msPAFs calculated according to the three mixture models (CA, IA and a combination of CA and IA (CA-IA)), for the 12 exposure scenarios (three dilutions and four nuclear plants). Overall, the results showed no difference between msPAF calculated according to IA and CA-IA, except at the P3 nuclear plant ^ne was at the minimum flow allowed for discharges when the Rho (Lowest-flow scenario), and at the P2 and P4 plants in the pure effluents. In the case of the routine liquid effluents from nuclear plants, there may be no added value of grouping the substances to combine CA and IA models based on the resulting level of protection. However, for exposure scenarios with higher (unrealistic) concentrations, the results for the pure effluents illustrate that grouping substances and combining the IA and CA models to compute the msPAF give more conservative results than using the single CA or IA models. Contrarily to the general acceptation, the IA model was generally more conservative than the CA model in our study: msPAFIA were higher than msPAFCA, except in the case of the nuclear plant P3 for the mean- and low-flow dilution scenarios (Fig. 3). Brosche and Backhaus (2010) indicated that IA predicts higher toxicities compared to CA when the steepness of the dose-response curves of individual substances is low. In that paper, a low steepness was

msPAF (%)

Mean−flow scenario

expressed when the ratio EC50/EC05 exceeded a threshold value of 13.5 (i.e. if the ratio is lower the steepness is higher). In the present study, HC50/HC05 ratio of the SSD curves of stable chemicals ranged from 2 for sulphates to 12.5 for Cr, while the HDR50/HDR05 ratio was 136.8 for g radiation, much higher than the 13.5 threshold. This high HDR50/HDR05 value could explain that the IA model generally predicted higher msPAFs than CA in the present study. Brosche and Backhaus (2010) suggested that CA could be considered more conservative than IA in general because most dose response curves exhibit EC50/EC05 values below the 13.5 threshold. However, the present findings suggest that this is not always the case and that at least ionising radiation are an exception to that general rule. Duboudin et al. (2004) reported log-normal SSDs of 22 stable substances and the associated s parameters in log base 10: the antilog (10x) s values varied from 1.6 to 15.5 while the antilog (exponential) s values in the present study varied from 1.5 for sulphates to 19.9 for g radiation. The range of s values obtained here is thus consistent with the range reported by Duboudin et al. (2004) suggesting that other stressors than g radiation may exhibit SSD with elevated s parameter. Thus, the IA model may also be more conservative than CA for other mixtures than mixtures that include ionising radiation. The only scenarios in which the msPAFCA were higher than the msPAFIA were those of the P3 plant (Fig. 3). Fig. 2 shows that for the radiological stressor, a given dose rate (below 1 HU, i.e. for a PAF below 50 %) is associated to a lower potential impact (PAF) when based on the CA-SSD than when based on the g-SSD. Compared to the IA model (based on individual PAF values derived from the substance-specific SSDs), the CA model thus generally underestimated the impact associated to ionising radiation in the total mixture effect. Furthermore, the P3 nuclear plant released higher quantities of stable chemicals (notably Cu and Zn) than the other plants in 2013 (Table 1). As it can be seen on Fig. 2, for a given exposure concentration below 1 HU of a given stable chemical (e.g. Cu), the multisubstance CA-SSD predicts a higher impact (PAF) than the Cu-SSD. Thus, in the case of P3, the multisubstance CA-SSD probably overestimated the risk associated to stable chemicals in the mixture compared to the IA model. These results highlight the issues that can arise when using the CA model with SSD exhibiting various scale parameters.

Lowest−flow scenario

Pure effluents 100

10

10

1

1

10−1

10−1

80

10−2

10−2

70

10−3

10−3

10−4

10−4

60

10−5

10−5 P1

P2

P3

P4

90

50 P1

P2

CA

IA

P3

P4

P1

P2

P3

P4

CAIA

Fig. 3. msPAFs associated to the liquid effluents from four nuclear plants (P1, P2, P3, P4 see the text for the codification of the plants) calculated based on three different mixture ^ne river (mean flow rate; lowest flow rate authorised for discharges, and undiluted effluents (hypothetical models (CA, IA or CAIA) and under three different dilutions by the Rho maximal exposure scenario)). The figure represents the msPAF on a log-10 scale.

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

This study shows that no conceptual nor practical limitations prevent the use of the msPAF approach for mixture that include ionising radiation. The main weak point of the approach is the need for sufficient ecotoxicological data to derive the SSD. Indeed, several of the stable chemicals released by nuclear facilities could not be integrated in the present study due to the lack of chronic effect data. But several advances could help to fill this lack of data, notably acute-to-chronic conversion approaches (Duboudin et al., 2004), and new fitting methods such as hierarchical Bayesian models (Kon Kam King et al., 2015; Oldenkamp et al., 2015) and non-parametric SSDs (Gottschalk and Nowack, 2013; Gottschalk et al., 2013). In addition, in the approach we used, the SSDs were derived using one NOEC/ECx per species (minimum value across different endpoint types). The approach thus only integrates the inter-species variability but not the intra-species variability of sensitivities. According to Aldenberg and Rorije (2013), taking into account the intra-species variability of sensitivities in the SSD has only a low impact on the estimation and leads to less conservative SSD fits. Nevertheless, hierarchical Bayesian models could be useful to take into account intra-species variability and take advantage of all the information available in the ecotoxicological databases (Kon Kam King et al., 2015). Here we used the ”classical” approach in deriving the SSDs of the studied substances and fitted the lognormal distribution. However, this choice lead to discard several substances that were present in the releases of the nuclear plants but for which we had not enough data to fit a log-normal distribution. Non-parametric approaches such as the one described by Gottschalk and Nowack (2013) and Gottschalk et al. (2013) are wellsuited to cope with low data availability but they can not yet be fully implemented to integrate mixture effects with CA model. It is important to reinforce the statement that the msPAF approach is very interesting proxy of the ecological impact of mixture of stressors, but is not yet a fully predictive tool (Posthuma et al., 2016). The approach has several limitations and shortcomings. To our opinion, the main limitation of the SSD is the fact that it does not yet integrate species interactions. This has important repercussions on the predictions of ecological impact and may lead to underestimate the impact by neglecting the indirect effects of contaminants (e.g. through the trophic pathways) (De Laender et al., 2008). In addition, the lack of ecological realism of the species for which NOEC/ECx are available is a main concern. Recent studies suggest that laboratory species may not be representative of field species (e.g. Larras et al., 2016; Garnier-Laplace et al., 2013). Furthermore, the SSDs are based on different effect levels (EC10, NOEC, …) which is not the best approach to derive a pertinent proxy of potential impact. However, as stated by Batley et al. (2014), mixing EC10 and NOEC in SSD is inevitable at the moment. Another limitation of the msPAF approach was introduced by Gregorio et al. (2013): to apply the CA model in an msPAF approach, one needs to assume that the tolerances of different species toward different contaminants are correlated (i.e. the species are ranked in the same order of sensitivity for the different components of the mixture). This is rarely achieved and Gregorio et al. (2013) showed that when species tolerance are not correlated, applying SSD first and then applying the CA model to compute an msPAF (i.e. the ”classical” approach as used in the present study) leads to large discrepancies compared to applying first the CA model on each species and then deriving the SSD (i.e. the approach proposed by Gregorio et al. (2013)). In this study, it was not possible to verify whether the species tolerance were correlated because the species were different in each SSD considered as well as the number of species considered (Table 2). However, the substances considered (mainly radionuclides and metals) do not have a specific, targeted mode of toxic action (such as pesticides for example that target certain groups of organisms). Thus, in the present study, large

9

discrepancies between the orders of sensitivity of species were not expected. In the case of mixture of radiological and chemical stressors, the present findings suggest that the CA model might not be fully appropriate because the SSDs exhibited different slopes. Nevertheless, the CA model is considered to be the more conservative and its use is recommended at screening steps of ERA (Backhaus et al., 2010). The different exposure scenarios considered in the present study illustrate the interest of considering both mixture models. Combining these two concepts provide a prediction window (Backhaus et al., 2010; Altenburger et al., 2013) whose utility is illustrated for plant P3. Indeed, for plant P3, one of the mixture model (CA) indicates a significant ecological risk if the flow rate of the river reaches the minimum allowed flow rate (msPAFCA was 12.4 % and msPAFIA was 4.4 10-2 %). A more thorough assessment of the ecological impact of the effluents might thus be needed for this scenario of minimum flow rate of the river because the prediction window overlaps the threshold of 5 %. In that case a relevant first step is to check the variation in scale parameters and to use a combination of CA and IA models separating stressors according to their scale parameters if the parameters differ strongly and if the toxic modes of action are not fully established (Fig. 3). 3.3. Ranking chemicals and radionuclides within mixture The msPAF approach was further used as a mean to rank stressors within the studied mixtures according to their relative potential impact (PAF or msPAF). The ranking is illustrated for the case of the mean-flow scenario, which is the most realistic in terms of chronic exposure as assumed by SSDs. Fig. 4 shows the msPAF associated to the chemical and radiological stressors in the liquid effluents of nuclear plants. In our exposure scenarios, the msPAF associated with ionising radiation were very low (msPAF-radiological ranging from 8.6 10-4 to 5.1 10-4 % for P3 and P1 plants respectively). msPAF of ionising radiation were higher than the msPAF associated with stable chemicals for three nuclear power plants (P1, P2 and P4, Fig. 4). For the P3 plant, the msPAF of stable chemical stressors was 1.1 10-3 %, slightly higher than the msPAF of radiological stressors (8.6 10-4 %). The individual PAF associated to each radionuclide dose rate identified 14C as the prime contributor to the global impact in the 4 nuclear plants considered. This result is partly related to the high 14C activities released compared to other radionuclides (Table 1). The msPAF approach thus identified the radiological stressor as the main contributor to the global potential impact of the liquid effluents from the four studied nuclear plants. Fig. 4 shows the msPAFCA for stable chemicals. The msPAFIA of stable chemicals were calculated as well but they were 0 % except for the P3 nuclear power plant (3.4 10-6 %). The CA model gave slightly higher (yet low) msPAF values for stable chemicals (between 10-20 and 10-30 % for plants P1, P2 and P4, and of 1.1 10-3 % for the P3 plant). With msPAFIA, the difference between the potential individual impacts of radionuclides and stable chemicals would thus be even higher than with msPAFCA. The highest individual PAFs of stable chemicals were for Zn and boric acid for P1 and P4 nuclear plants whereas they were highest for Ni and Zn for plant P2 and Cu and Zn for plant P3. Contrarily to the other plants, for P3, the individual PAFs of Cu and Zn were higher than the individual PAF of certain radionuclides (data not shown). Finally, we compared the ranking results with a previous approach based on a simplified linear dose-effect relationship (EF) (Garnier-Laplace et al., 2009). The EFs of stable chemicals within the releases were higher than those of radionuclides for all the nuclear plants considered (Fig. 4). Garnier-Laplace et al. (2009) also found higher EF values for stable chemicals than for radionuclides and the EF for stable chemicals were between 104 and 106 times

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

10

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

msPAF (%)

P1

P2

P3

P4

1 10−2 10−4 10−10

10−20

ca l gi lo io ra d

m ic al ch e

ca l gi ra d

io

ch e

msPAF

lo

m ic al

ca l gi lo io ra d

m ic al ch e

ca l gi lo io ra d

ch e

m ic al

10−30

EF

Fig. 4. Ranking chemical and radiological stressors according to their potential ecological impact within the liquid effluents from four french nuclear power plants (P1, P2, P3 and P4). Comparison of the msPAF approach (msPAF) and a linear approach (EF) to ranking stressors. msPAFs of the mixture of stable chemicals were calculated according to the CA model. msPAF of ionising radiation associated to the mixture of radionuclides was calculated according to the SSD of g radiation. EF values were calculated for stable chemicals and ionising radiation based on the CA concept. msPAF and EF are represented on a log-10 scale.

higher than for radionuclides. The results obtained with the linear approach (EF) were thus consistent with the previous results of Garnier-Laplace et al. (2009). The msPAF-based ranking results presented in this study contrasted with the ranking approach based on a linear (EF) approach described in Garnier-Laplace et al. (2009). While the msPAF approach identified ionising radiation as a higher impact stressor compared to stable chemicals in the effluents, the linear approach concluded the opposite (Fig. 4). Depending on the stressor considered the linear approach over or underestimated the impact of individual stressor compared to the SSD-based msPAF approach: the linear approach largely overestimated the impact associated to stable chemicals while it slightly underestimated the impact associated to radionuclides compared to the SSD-based msPAF approach (Fig. 4). A graphical representation (Fig. SI, Supplementary Information) showed that the linear model was generally above the SSD of stable chemicals between 0 and 1 HU (thus generally predicting higher msPAF than the SSD) but that it was below the SSD of g radiation (thus predicting lower msPAF values for that stressor). Overall the present study demonstrates the interest of integrating the non-linear patterns of species sensitivity distributions to assess the ecological risks associated with the releases from nuclear facilities, and to rank the different stressors.

4. Conclusions This study is the first to apply the msPAF approach on mixture of radiological and chemical stressors. msPAF provided an integrated proxy of the toxic pressure associated with the mixture of substances released by nuclear facilities. The results show that the SSD of ionising radiation has a very different shape than the SSDs of 8 stable chemicals released by the studied nuclear facilities. This difference in shapes of the SSDs had implications for the msPAF approach. Indeed, contrarily to what is generally accepted, the CA model was not the most conservative in our studied case. Although this result was obtained for only twelve scenarios (four mixtures corresponding to different nuclear plants releases in three dilution scenarios), the studied cases covered environmentally relevant (i.e.

substances at very low concentrations and activities) to higher exposure scenarios (i.e. pure effluents). The present results are thus valuable for the ERA of the routine releases from nuclear plants, but they are also useful for msPAF users that face contrasted SSD slopes. CA model is generally recommended as a first approach in the ERA of mixtures because the model is supposed to be more conservative. Our results however point that for mixture of radiological and chemical stressors, the mixture model IA might be more conservative. The present study highlights the issues arising when CA model is used on substances that exhibit contrasting variances in chronic ecotoxicological data. It further underpins the interest of considering both CA and IA mixture models to determine a prediction window at screening steps of ERA. The msPAF approach also provided a ranking of the potential effects of radiological versus chemical stressors within mixtures. The results indicated a higher contribution of ionising radiation to the overall impact compared to stable chemicals, always with a negligible potential ecological risk in the context of routine effluents from nuclear plants. Such a tool can thus be useful for management purposes (substance prioritization). This approach was shown to be fully consistent with the existing ERA tools and frameworks for single chemicals (Technical Guidance Document ECHA (2008a)) and ionising radiation (ERICA Integrated Approach Beresford et al. (2007)). It provides a useful basis for the integrated assessment and management of environmental risks from ionising radiation and chemicals. The main challenge of the approach will be to be exhaustive because many substances lack sufficient chronic effect data to compute a robust SSD. However, the main advantage is that the approach offers a framework to integrate other environmental stressors, such as warming, a major environmental change driver particularly relevant in the context of nuclear facilities that release thermal effluents.

Appendix A. Supplementary data Supplementary data related to this article can be found at https://doi.org/10.1016/j.envpol.2017.09.003.

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

References Aldenberg, T., Jaworska, J.S., 2000. Uncertainty of the hazardous concentration and fraction affected for normal species sensitivity distributions. Ecotoxicol. Environ. Saf. 46 (1), 1e18. Aldenberg, T., Rorije, E., 2013. Species sensitivity distribution estimation from uncertain (QSAR-based) effects data. Altern. Laboratory Animals 41 (1), 19e31. Altenburger, R., Arrhenius, A., Backhaus, T., Coors, A., Faust, M., Zitzkat, D., 2013. Ecotoxicological Combined Effects from Chemical Mixtures. Part 1: Relevance and Adequate Consideration in Environmental Risk Assessment of Plant Protection Products and Biocides. Umweltbundesamt (UBA) report No. UBA-FB 00, Dessau-Roßlau, Germany.
cision 2008-DC-0101 de l'Autorite
de sûrete
nucle
aire du 13 mai 2008 ASN, 2008. De
s de pre
le vements et de confixant les prescriptions relatives aux modalite sommation d’eau et de rejets dans l’environnement des effluents liquides et
aires de base n. 87 et n. 88 exploite
es par Elecgazeux des installations nucle
de France (EDF-SA) sur la commune de Saint-Paul-Trois-Ch^ tricite ateaux
partement de la Dro ^me). (de
cision 2013-DC-0333 de l'Autorite
de sûrete
nucle
aire du 14 fe
vrier ASN, 2013. De
s de pre
le vement et de 2013 fixant les prescriptions relatives aux modalite consommation d’eau et de rejets dans l’environnement des effluents liquides et
aires de base n. 111 et n. 112 exploite
es par gazeux des installations nucle
de France e Socie
te
Anonyme (EDF-SA) sur les communes de Cruas, Electricite che) et la Coucourde (Dro ^me). Meysse (Arde
cision 2014-DC-0442 de l'Autorite
de sûrete
nucle
aire du 15 juillet ASN, 2014a. De
s de pre
le vement et de 2014 fixant les prescriptions relatives aux modalite consommation d’eau et de rejets dans l’environnement des effluents liquides et
aires de base n. 45, n. 78, n. 89 et n. 173 exploite
es gazeux des installations nucle
de France e Socie
te
Anonyme (EDF-SA) dans la commune de par Electricite
partement de l'Ain). Saint-Vulbas (de
cision 2014-DC-0469 de l'Autorite
de sûrete
nucle
aire du 2 ASN, 2014b. De
cembre 2014 fixant les prescriptions relatives aux modalite
s de pre
le vement de et de consommation d’eau et de rejet dans l’environnement des effluents liq
aires de base n. 119 et n. 120 exploite
es uides et gazeux des installations nucle
de France e Socie
te
Anonyme (EDF-SA) dans la commune de par Electricite
partement de l’Ise re). Saint- Alban-Saint-Maurice (de Backhaus, T., Blanck, H., Faust, M., 2010. Hazard and Risk Assessment of Chemical Mixtures under REACH: State of the Art, Gaps and Options for Improvement. Swedish Chemicals Agency, Bromma, Sweden. Banque Hydro, 2016. Online French Hydrological Database. http://www.hydro. eaufrance.fr/. Batley, G., Van Dam, R., Warne, M.S.J., Chapman, J., Fox, D., Hickey, C., Stauber, J., 2014. Technical Rationale for Changes to the Method for Deriving Australian and New Zealand Water Quality Guideline Values for Toxicants. Australian Government Standing Council on Environment and Water, Canberra. Beaugelin-Seiller, K., Jasserand, F., Garnier-Laplace, J., Gariel, J.C., 2006. Modeling radiological dose in non-human species: principles, computerization, and application. Health Phys. 90 (5), 485e493. Beresford, N., Brown, J., Copplestone, D., Garnier, J., Howard, B., Larsson, C.-m., €hl, G., Zinger, I., 2007. D-ERICA: an Integrated Approach to the Oughton, D., Pro Assessment and Management of Environmental Risks from Ionising Radiation. A Deliverable Report for the Project “ERICA” (Contract No. FI6R-CT2004e508847) within the European Commission’s VIth Framework Programme, Stockholm, Sweden. Bliss, C., 1939. The toxicity of poisons applied jointly. Ann. Appl. Biol. 26 (3), 585e615. Brosche, S., Backhaus, T., 2010. Toxicity of five protein synthesis inhibiting antibiotics and their mixture to limnic bacterial communities. Aquat. Toxicol. 99 (4), 457e465. Brown, J., Alfonso, B., Avila, R., Beresford, N., Copplestone, D., Hosseini, A., 2016. A new version of the ERICA tool to facilitate impact assessments of radioactivity on wild plants and animals. J. Environ. Radioact. 153, 141e148.
n, R., Thørring, H., i Batlle, J.V., 2004. Radiation doses to Brown, J., Jones, S., Saxe aquatic organisms from natural radionuclides. J. Radiological Prot. 24 (4A), A63. Carvalho, R.N., Arukwe, A., Ait-Aissa, S., Bado-Nilles, A., Balzamo, S., Baun, A., Belkin, S., Blaha, L., Brion, F., Conti, D., Creusot, N., Essig, Y., Ferrero, V.E.V.,
s, A., Flander-Putrle, V., Fürhacker, M., Grillari-Voglauer, R., Hogstrand, C., Jona Kharlyngdoh, J.B., Loos, R., Lundebye, A.-K., Modig, C., Olsson, P.-E., Pillai, S., Polak, N., Potalivo, M., Sanchez, W., Schifferli, A., Schirmer, K., Sforzini, S., Stürzenbaum, S.R., Sø fteland, L., Turk, V., Viarengo, A., Werner, I., Yagur-Kroll, S., Zounkov
a, R., Lettieri, T., Sep. 2014. Mixtures of chemical pollutants at European legislation safety concentrations: how safe are they? Toxicol. Sci. 141 (1), 218e233. Cedergreen, N., 2014. Quantifying synergy: a systematic review of mixture toxicity studies within environmental toxicology. PLoS ONE 9 (5). Cedergreen, N., Christensen, A.M., Kamper, A., Kudsk, P., Mathiassen, S.K., Streibig, J.C., Sørensen, H., 2008. A review of independent action compared to concentration addition as reference models for mixtures of compounds with different molecular target sites. Environ. Toxicol. Chem. 27 (7), 1621e1632. De Laender, F., De Schamphelaere, K.A., Vanrolleghem, P.A., Janssen, C.R., 2008. Do we have to incorporate ecological interactions in the sensitivity assessment of ecosystems? an examination of a theoretical assumption underlying species sensitivity distribution models. Environ. Int. 34 (3), 390e396. de Zwart, D., Posthuma, L., 2005. Complex mixture toxicity for single and multiple

11

species: proposed methodologies. Environ. Toxicol. Chem. 24 (10), 2665e2676. Delignette-Muller, M.L., Dutang, C., 2015. Fitdistrplus : an R package for fitting distributions. J. Stat. Softw. 64 (4), 1e34. Duboudin, C., Ciffroy, P., Magaud, H., 2004. Acute-to-chronic species sensitivity distribution extrapolation. Environ. Toxicol. Chem. 23 (7), 1774e1785. ECHA, 2008a. Technical Guidance Document: Guidance on Information Requirements and Chemical Safety Assessment: Characterisation of Doseresponse for Environment. European Chemicals Agency, Helsinki, Finland. ECHA, 2008b. Annex XV Transitional Report - Boric Acid. European Chemicals Agency. https://echa.europa.eu/information-on-chemicals/transitionalmeasures/annex-xv-transitional-reports.
aire et Environnement” 2013. Division Production EDF, 2013. Rapport ”Nucle
aire Electricite
de France (EDF) SA. Nucle Elphick, J.R., Bergh, K.D., Bailey, H.C., Jan. 2011a. Chronic toxicity of chloride to freshwater species: effects of hardness and implications for water quality guidelines. Environ. Toxicol. Chem. 30 (1), 239e246. Elphick, J.R., Davies, M., Gilron, G., Canaria, E.C., Lo, B., Bailey, H.C., Jan. 2011b. An aquatic toxicological evaluation of sulfate: the case for considering hardness as a modifying factor in setting water quality guidelines. Environ. Toxicol. Chem. 30 (1), 247e253. EU RAR, 2008. European Union Risk Assessment Report - Nickel and Nickel Compounds. European Chemicals Agency. https://echa.europa.eu/information-onchemicals/information-from-existing-substances-regulation. EU RAR, 2010. European Union Risk Assessment Report - Zinc Metal. European Commission - Joint Research Centre e Institute for Health and Consumer Protection. http://publications.jrc.ec.europa.eu/repository/handle/JRC61245. EU VRAR, 2008a. European Union Voluntary Risk Assessment Report - Lead and Lead Compounds. submitted to the European Chemicals Agency by the Lead Development Association International. http://echa.europa.eu/web/guest/ voluntary-risk-assessment-reports-lead-and-lead-compounds. EU VRAR, 2008b. European Union Voluntary Risk Assessment Reports d Copper and Copper Compounds. submitted to the European Chemicals Agency by the European Copper Institute, Brussels, Belgium. https://echa.europa.eu/web/ guest/copper-voluntary-risk-assessment-reports. Forbes, V.E., Calow, P., 2002. Species Sensitivity Distributions revisited: a critical appraisal. Hum. Ecol. Risk Assess. 8 (3), 473e492. Garnier-Laplace, J., Beaugelin-Seiller, K., Gilbin, R., Della-Vedova, C., Jolliet, O., Payet, J., 2009. A Screening Level Ecological Risk Assessment and ranking method for liquid radioactive and chemical mixtures released by nuclear facilities under normal operating conditions. Radioprotection 44 (5), 903e908. Garnier-Laplace, J., Della-Vedova, C., Andersson, P., Copplestone, D., Cailes, C., Beresford, N.A., Howard, B.J., Howe, P., Whitehouse, P., 2010. A multi-criteria weight of evidence approach for deriving ecological benchmarks for radioactive substances. J. Radiological Prot. 30, 215e233. Garnier-Laplace, J., Geras’kin, S., Della-Vedova, C., Beaugelin-Seiller, K., Hinton, T., Real, A., Oudalova, A., 2013. Are radiosensitivity data derived from natural field conditions consistent with data from controlled exposures? a case study of chernobyl wildlife chronically exposed to low dose rates. J. Environ. Radioact. 121, 12e21. Gottschalk, F., Kost, E., Nowack, B., 2013. Engineered nanomaterials in water and soils: a risk quantification based on probabilistic exposure and effect modeling. Environ. Toxicol. Chem. 32 (6), 1278e1287. Gottschalk, F., Nowack, B., 2013. A probabilistic method for species sensitivity distributions taking into account the inherent uncertainty and variability of effects to estimate environmental risk. Integr. Environ. Assess. Manag. 9 (1), 79e86. vre, N., Junghans, M., 2013. Critical issues in using the common Gregorio, V., Che mixture toxicity models concentration addition or response addition on species sensitivity distributions: a theoretical approach. Environ. Toxicol. Chem. 32 (10), 2387e2395. Gross, J., Ligges, U., 2015. Nortest: Tests for Normality. R package version 1.0-4. http://CRAN.R-project.org/package¼nortest. IAEA, 2014. Modelling of Biota Dose Effects. International Atomic Energy Agency, TECDOC Series, Vienna, Austria.
es toxicologiques et environnementales des subINERIS, 2005. Fiche de donne
rive
s. Institut National de l'Environnement stances chimiques : Chrome et ses de Industriel et des Risques. http://www.ineris.fr/substances/fr/page/21. Jesenska, S., Nemethova, S., Blaha, L., 2013. Validation of the species sensitivity distribution in retrospective risk assessment of herbicides at the river basin scale-the Scheldt river basin case study. Environ. Sci. Pollut. Res. 20 (9), 6070e6084. Kon Kam King, G., Larras, F., Charles, S., Delignette-Muller, M.L., 2015. Hierarchical modelling of species sensitivity distribution: development and application to the case of diatoms exposed to several herbicides. Ecotoxicol. Environ. Saf. 114, 212e221. Kortenkamp, A., Backhaus, T., Faust, M., 2009. State of the Art Report on Mixture Toxicity. The School of Pharmacy. University of London, ULSOP. Final Report, Study Contract Number 070307/2007/485103/ETU/D.1. Kortenkamp, A., Faust, M., Scholze, M., Backhaus, T., 2007. Low-level exposure to multiple chemicals: reason for human health concerns? Environ. Health Perspect. Perspect. 115 (Suppl. l), 106e114. vre, N., 2016. Comparison of Larras, F., Gregorio, V., Bouchez, A., Montuelle, B., Che specific versus literature species sensitivity distributions for herbicides risk assessment. Environ. Sci. Pollut. Res. 23 (4), 3042e3052. Larsson, C.-M., Sep. 2008. An overview of the ERICA Integrated Approach to the assessment and management of environmental risks from ionising

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003

12

L. Beaumelle et al. / Environmental Pollution xxx (2017) 1e12

contaminants. J. Environ. Radioact. 99 (9), 1364e1370. €ewe, S., Muischnek, H., 1926. Über kombinationswirkungen 1. mitteilung: hilfsLo mittel der fragestellung. Schmiedeb. Arch. Exp. Pathol. Pharmakol 114, 313e326. Margerit, A., Gomez, E., Gilbin, R., 2016. Dynamic energy-based modeling of uranium and cadmium joint toxicity to Caenorhabditis elegans. Chemosphere 146, 405e412. Margerit, A., Lecomte-Pradines, C., Svendsen, C., Frelon, S., Gomez, E., Gilbin, R., 2015. Nested interactions in the combined toxicity of uranium and cadmium to the nematode Caenorhabditis elegans. Ecotoxicol. Environ. Saf. 118, 139e148. Maycock, D., Sorokin, N., Atkinson, C., Rule, K., Crane, M., 2007. Proposed EQS for Water Framework Directive Annex VIII Substances : Chromium(VI) and Chromium(III) (Dissolved). Environment Agency, Bristol. Oldenkamp, R., Hendriks, H.W.M., van de Meent, D., Ragas, A.M.J., 2015. Hierarchical bayesian approach to reduce uncertainty in the aquatic effect assessment of realistic chemical mixtures. Environ. Sci. Technol. 49 (17), 10457e10465. Pennington, D.W., Payet, J., Hauschild, M., 2004. Aquatic ecotoxicological indicators in life-cycle assessment. Environ. Toxicol. Chem. 23 (7), 1796e1807. Posthuma, L., de Zwart, D., Sep. 2012. Predicted mixture toxic pressure relates to observed fraction of benthic macrofauna species impacted by contaminant mixtures. Environ. Toxicol. Chem. 31 (9), 2175e2188. Posthuma, L., Dyer, S.D., de Zwart, D., Kapo, K., Holmes, C.M., G. Allen Burton, J., 15 December 2016. Eco-epidemiology of aquatic ecosystems: multiple stressors Separating chemicals from multiple stressors. Sci. Total Environ. 573, 1303e1319. http://dx.doi.org/10.1016/j.scitotenv.2016.06.242. R Core Team, 2015. R: a Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project. org/. Sala, S., Migliorati, S., Monti, G.S., Vighi, M., May 2012. SSD-based rating system for the classification of pesticide risk on biodiversity. Ecotoxicology 21 (4), 1050e1062. Schoderboeck, L., Mühlegger, S., Losert, A., Gausterer, C., Hornek, R., 2011. Effects

assessment: boron compounds in the aquatic environment. Chemosphere 82 (3), 483e487.
, S., Bl
€fer, R.B., Beketov, M.A., 2014. Do predictions Smetanova aha, L., Liess, M., Scha from Species Sensitivity Distributions match with field data? Environ. Pollut. 189, 126e133. Syberg, K., Jensen, T.S., Cedergreen, N., Rank, J., 2009. On the use of mixture toxicity assessment in REACH and the water framework directive: a review. Hum. Ecol. Risk Assess. 15 (6), 1257e1272. van de Meent, D., Huijbregts, M. a. J., 2005. Calculating life-cycle assessment effect factors from potentially affected fraction-based ecotoxicological response functions. Environ. Toxicol. Chem. 24 (6), 1573e1578. Van Sprang, P.A., Nys, C., Blust, R.J.P., Chowdhury, J., Gustafsson, J.P., Janssen, C.J., De Schamphelaere, K.A.C., 2016. The derivation of effects threshold concentrations of lead for European freshwater ecosystems. Environ. Toxicol. Chem. 35 (5), 1310e1320. Van Sprang, P.A., Verdonck, F. a. M., Van Assche, F., Regoli, L., De Schamphelaere, K. a. C., 2009. Environmental risk assessment of zinc in European freshwaters: a critical appraisal. Sci. Total Environ. 407 (20), 5373e5391. van Zelm, R., Huijbregts, M. a. J., Harbers, J.V., Wintersen, A., Struijs, J., Posthuma, L., van de Meent, D., 2007. Uncertainty in msPAF-based ecotoxicological effect factors for freshwater ecosystems in life cycle impact assessment. Integr. Environ. Assess. Manag. 3 (2), 203e210. Vanhoudt, N., Vandenhove, H., Real, A., Bradshaw, C., Stark, K., Sep. 2012. A review of multiple stressor studies that include ionising radiation. Environ. Pollut. 168, 177e192. Vijver, M.G., Elliott, E.G., Peijnenburg, W.J., de Snoo, G.R., Jun. 2011. Response predictions for organisms water-exposed to metal mixtures: a meta-analysis. Environ. Toxicol. Chem. 30 (6), 1482e1487. Wheeler, J.R., Grist, E.P.M., Leung, K.M.Y., Morritt, D., Crane, M., 2002. Species sensitivity distributions: data and model choice. Mar. Pollut. Bull. 45 (1e12), 192e202.

Please cite this article in press as: Beaumelle, L., et al., Ecological risk assessment of mixtures of radiological and chemical stressors: Methodology to implement an msPAF approach, Environmental Pollution (2017), https://doi.org/10.1016/j.envpol.2017.09.003