Water Research 81 (2015) 137e148
Contents lists available at ScienceDirect
Water Research journal homepage: www.elsevier.com/locate/watres
Effect-based trigger values for in vitro bioassays: Reading across from existing water quality guideline values Beate I. Escher a, b, c, *, Peta A. Neale c, d, Frederic D.L. Leusch d a
UFZ e Helmholtz Centre for Environmental Research, Cell Toxicology, Leipzig, Germany Eberhard Karls University Tübingen, Environmental Toxicology, Center for Applied Geosciences, Tübingen, Germany c The University of Queensland, National Research Centre for Environmental Toxicology, Entox, Brisbane, QLD 4108, Australia d Smart Water Research Centre, School of Environment, Griffith University, Southport, QLD 4222, Australia b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 26 February 2015 Received in revised form 15 May 2015 Accepted 25 May 2015 Available online 27 May 2015
Cell-based bioassays are becoming increasingly popular in water quality assessment. The new generations of reporter-gene assays are very sensitive and effects are often detected in very clean water types such as drinking water and recycled water. For monitoring applications it is therefore imperative to derive trigger values that differentiate between acceptable and unacceptable effect levels. In this proofof-concept paper, we propose a statistical method to read directly across from chemical guideline values to trigger values without the need to perform in vitro to in vivo extrapolations. The derivation is based on matching effect concentrations with existing chemical guideline values and filtering out appropriate chemicals that are responsive in the given bioassays at concentrations in the range of the guideline values. To account for the mixture effects of many chemicals acting together in a complex water sample, we propose bioanalytical equivalents that integrate the effects of groups of chemicals with the same mode of action that act in a concentration-additive manner. Statistical distribution methods are proposed to derive a specific effect-based trigger bioanalytical equivalent concentration (EBT-BEQ) for each bioassay of environmental interest that targets receptor-mediated toxicity. Even bioassays that are indicative of the same mode of action have slightly different numeric trigger values due to differences in their inherent sensitivity. The algorithm was applied to 18 cell-based bioassays and 11 provisional effectbased trigger bioanalytical equivalents were derived as an illustrative example using the 349 chemical guideline values protective for human health of the Australian Guidelines for Water Recycling. We illustrate the applicability using the example of a diverse set of water samples including recycled water. Most recycled water samples were compliant with the proposed triggers while wastewater effluent would not have been compliant with a few. The approach is readily adaptable to any water type and guideline or regulatory framework and can be expanded from the protection goal of human health to environmental protection targets. While this work constitutes a proof of principle, the applicability remains limited at present due to insufficient experimental bioassay data on individual regulated chemicals and the derived effect-based trigger values are of course only provisional. Once the experimental database is expanded and made more robust, the proposed effect-based trigger values may provide guidance in a regulatory context. © 2015 Elsevier Ltd. All rights reserved.
Keywords: In vitro Bioassay Bioanalytical equivalent concentration Water quality Human health
1. Introduction 1.1. Chemical water quality guidelines to protect human health Health-based water quality guideline values (GV) are defined as concentrations of chemicals in drinking water that do not pose an * Corresponding author. Helmholtz Centre for Environmental Research e UFZ, Cell Toxicology, Permoserstraße 15, 04318 Leipzig, Germany. E-mail address: [email protected] (B.I. Escher). http://dx.doi.org/10.1016/j.watres.2015.05.049 0043-1354/© 2015 Elsevier Ltd. All rights reserved.
appreciable risk to health over a lifetime of exposure. GV are based on acceptable daily intake (ADI, also called reference dose RfD, or tolerable daily intake TDI). The ADI is typically extrapolated from ‘No Observed Adverse Effect Levels’ derived from a set of animal toxicity tests with extrapolation factors, accounting for uncertainties related to extrapolation from a model system to a human population (Ritter et al., 2007). Guideline values for individual chemicals were defined, for example, by the WHO for drinking water (WHO, 2011). Healthbased guidelines are also available for recycled water for indirect
138
B.I. Escher et al. / Water Research 81 (2015) 137e148
Abbreviations AO ADI ADWG AGWR BAC BEQ CarbEQ DEQ DexEQ DW EBT EEQ Eff EPA EC GV HC5 IR
advanced oxidation acceptable daily intake Australian Drinking Water Guidelines Australian Guidelines for Water Recycling biologically activated carbon bioanalytical equivalent concentration carbaryl equivalent concentration diuron equivalent concentration dexamethasone equivalent concentration drinking water effect-based trigger value estradiol equivalent concentration effluent Environmental Protection Agency effect concentration guideline value hazardous concentration to 5% of species induction ratio
potable reuse and for various other water types and environmental protection goals (Escher and Leusch, 2012). Snyder et al. (2008) proposed to derive drinking water equivalent levels for pharmaceuticals using the minimum therapeutic dose. This matches the approach proposed by Schriks et al. (2010) who developed a pragmatic approach to derive provisional drinking water guideline values from ADIs for unregulated chemicals immediately after their detection in drinking water. 1.2. Bioanalytical tools for water quality assessment Given the overwhelming number of chemicals present in water (Schwarzenbach et al., 2006; Villanueva et al., 2014), chemical analysis has been complemented for many years by in vitro bioassays which not only give information on the level of effect but also on the type of effect, i.e., the mode of action. Further, bioanalytical tools also integrate the mixture effects of chemicals that act according to the same mode of action in a concentrationadditive manner (Escher and Leusch, 2012). In vitro bioanalytical tools have also been recommended for the assessment of drinking water (van Wezel et al., 2010). Since 2008, the high-throughput in vitro toxicity program Tox21 of the National Institute of Health (NIH) and the US Environmental Protection Agency (EPA) has been and continues to provide quantitative information on the toxicity pathways of almost 10,000 organic compounds (Attene-Ramos et al., 2013b; Shukla et al., 2010). In vitro assays included in Tox21 are phenotypic assays (e.g., targeting mitochondrial toxicity (Attene-Ramos et al., 2013a)), pathway assays (e.g., targeting cell viability in specific cell lines) and nuclear receptor assays (e.g., targeting hormone receptors and metabolic pathways (Huang et al., 2011)) and various other endpoints (Tice et al., 2013). The chemicals and pathways screened for Tox21 are also relevant for water quality and some of the assays have been applied successfully to water quality testing together with more established in vitro assays (Escher et al., 2014). 1.3. Need for effect-based trigger values (EBT) Modern reporter gene assays and other cellular bioassays are so sensitive that even very clean waters often induce detectable
MCEQ MF MTEQ NIH P4EQ PTEQ PXR REF REP RfD RO SW TCDDEQ
metolachlor equivalent concentration membrane filtration malathion equivalent concentration National Institute of Health progesterone equivalent concentration parathion equivalent concentration pregnane X receptor relative enrichment factor relative effect potency reference dose reverse osmosis stormwater 2,3,7,8-tetrachlorodibenzodioxin equivalent concentration TDI tolerable daily intake TTEQ testosterone equivalent concentration ToxCast US EPA's program “Toxicity Forecaster” Tox21 NIH's program “Toxicology in the 21st Century” WHO World Health Organization
responses in vitro. High sensitivity is an advantage from the point of view of achieving the highest possible sensitivity and in the assessment of treatment efficacy (i.e., water treatment plants). In terms of water quality, however, a cellular response does not translate directly into a toxicological effect. A ‘simple’ yes or no answer will be defined by the sensitivity of the test and is not sufficient to evaluate if a water is “good” or “bad”. Therefore, trigger values are required that differentiate an acceptable response from an unacceptable response. Recent work that compared the bioassay response of chemicals detected in recycled water with the bioassay response of the entire water sample has shed light on knowledge gaps. For bioassays indicative of integrative effects (i.e., non-specific effects, reactive effects and adaptive stress responses) typically only a very small fraction (often less than 1% of the effect) can be explained by known and identified chemicals (Escher et al., 2013; Reungoat et al., 2012; Tang et al., 2013). Thus unidentified and unknown chemicals cause the majority of the biological response indicating the need to account for mixture effects in the derivation of EBTs. Also for these types of responses there is no clear-cut reference chemical available because many different chemicals can induce the effect. Therefore, we have previously proposed EBT-effect concentrations (EBT-EC) that account for mixture effects of these types of chemicals in non-specific bioassays (Escher et al., 2013; Tang et al., 2013). For example, for the derivation of the EBT for the bioluminescence inhibition test we assumed 1000 chemicals were present at 5% of their guideline value and applied a concentration-additive mixture toxicity model. The EBT-EC50 was expressed in relative enrichment factor units (REF), which describe how much the water sample was enriched by, e.g., solid phase extraction prior to the bioassay. The provisional EBT-EC50 for drinking and recycled water then came to 2.8 REF (Tang et al., 2013). In other words, if a sample that is roughly three times enriched (by solid phase extraction for example) shows 50% or more effect in this bioassay, then the EBT is exceeded; if on the other hand it induces less than 50% effect, then the water quality is acceptable. A similar approach for the oxidative stress response quantified with the AREc32 bioassay resulted in a provisional EBT-ECIR1.5 of 3 REF (Escher et al., 2013). Both provisional EBT-EC were able to suitably differentiate between water quality before and after advanced water treatment in recycling
B.I. Escher et al. / Water Research 81 (2015) 137e148
plants. In contrast to integrative assays, the majority of receptormediated effects (e.g., inhibition of enzymes or activation of hormone receptors) can typically be explained by known chemicals (Escher et al., 2011; Leusch et al., 2005, 2014b; Rutishauser et al., 2004; Tang and Escher, 2014). Effects can be expressed as socalled bioanalytical equivalent concentrations (BEQ), which represent the concentrations of a reference chemical that would elicit the same effect as the water sample of unknown composition. It has been suggested to apply concentration addition as the realistic worst-case scenario for mixture effects in risk assessment (Backhaus and Faust, 2012) and the BEQ concept is based on this mixture toxicity concept, which is valid for chemicals that have a common mode of action. As straightforward as it may seem, however, it would be wrong to simply compare the BEQ to the GV of the reference compound. While the BEQ of the water sample includes information about the mixture effect, the derivation of a single chemical GV is exclusively based on the effects of that single compound. 1.4. Framework for the application of EBT in water quality monitoring EBT are not meant to replace other monitoring tools but can be applied to complement chemical analysis and to decrease uncertainty in water quality assessment. A possible tiered approach to water quality assessment is shown in Fig. 1. In a first screening step, a limited number of indicator chemicals would be monitored and in parallel a limited number of indicator bioassays would be applied. Only if the measured chemical concentrations exceed one or more GV or the effects in the indicator bioassays are above the EBT, then a much larger list of chemicals with defined GV would need to be monitored to determine overall water quality. This is just one of many possible ways to implement EBTs in a water quality monitoring strategy. 1.5. Aims The aim of this study was to derive an algorithm for the derivation of EBT for receptor-mediated effects and to apply it as an illustrative example to Australian recycled water that was previously screened with 103 different bioassays (Escher et al., 2014). Eighteen of the 103 bioassays were selected for this exercise. The omitted bioassays were either not relevant for water quality (i.e., they had not shown any effects with water samples in the mentioned benchmarking study (Escher et al., 2014)), had no available data on reference chemicals or other regulated chemicals,
Fig. 1. Framework for the application of bioanalytical tools in water quality monitoring (C ¼ concentration, GV ¼ guideline value, EBT ¼ effect based trigger value).
139
or the BEQ concept could not be applied to these bioassays. We included bioassays without any direct connection to human health endpoints to test whether non-target bioassays could serve as purely (bio)analytical tools for the quantification of the presence of certain groups of chemicals. For example, the mixture effect of the herbicides included in a guideline can be evaluated using their effects on algae because this is a sensitive and quantitative indicator for this chemical group despite its lack of direct human health relevance. We applied the proposed EBT derivation method to GV from the Australian Guidelines for Water Recycling: Augmentation of Drinking Water Supplies (AGWR) (NRMMC & EPHC & NHMRC, 2008), which was adopted in the Public Health Regulation, Schedule 3B Standards for quality of recycled water supplied to augment a supply of drinking water in Queensland, Australia (Queensland Government, 2005). The water samples tested were all sampled in Queensland (Escher et al., 2014). The GV in the Australian Drinking Water Guidelines (ADWG) (NHMRC, 2011) are a subset of the GV in the AGWR thus the proposed EBTs should be applicable to both recycled and drinking water in Queensland and other States of the Commonwealth of Australia. Without much effort, the approach can be adapted to guidelines from other countries and regions, including the WHO Guidelines for DrinkingWater Quality (WHO, 2011).
1.6. Approach One previous study has derived bioassay trigger values for bioassays indicative of receptor-mediated effects in drinking water. Brand et al. (2013) proposed to derive effect-based trigger for the CALUX bioassays from the ADI of the reference compound using bioanalytical equivalents (BEQ) that account for the concentrationadditive effect of chemicals that act similarly but may have different toxicokinetics (Punt et al., 2013). The toxicokinetic differences between in vivo and in vitro were considered by accounting for oral bioavailability and binding to plasma proteins but not for metabolism (Brand et al., 2013). The derived toxicokinetic factors were used to estimate the acceptable target concentrations of the reference compound, which in turn were translated to acceptable target concentrations of another compound than the reference compound in the associated CALUX assay by the relative effect potency in vitro. Finally, the water consumption and an allocation factor were used to derive the trigger value in the various CALUX bioassays. An alternative approach would be to apply bioassays purely as an analytical tool and as a means to capture the mixture effects of chemicals in water. In this case one does not need to invoke any in vitro to in vivo extrapolation, which avoids the associated uncertainties. The algorithm we propose starts with existing sets of chemical GV and directly read across from the GV to the EBT. The only input data required are the GV and published effect concentrations (EC) of the regulated chemicals. Thus the proposed method is simple and can be easily adapted if water quality guidelines change or bioassays are applied in another region of the world because they can be easily transferred from one regulatory framework to another. Here we develop this read-across approach and illustrate it with examples for chemicals included in the AGWR. The approach is inspired by the work of Jarosova et al. (2014), who predicted safe concentrations of estrogenic activity in wastewater using similar in vitro bioassays as applied in the current study by reading directly across from the environmental predicted no effect concentrations that are derived from in vivo data.
140
Table 1 Bioassays, reference chemicals and summary of EBT-BEQs for all endpoints; m ¼ available number of effect data points; n ¼ number of data EC values included in the distributions. Bioassay
References bioassay method
Literature source of EC value collection
Reference chemical for EBT
Ab-brevi-ation
EC (reference chemical) (M)
1 2 3
PXR-cisFACTORIAL PXR-transFACTORIAL HG5LN PXR
(Romanov et al., 2008) (Romanov et al., 2008) (Lemaire et al., 2006)
Metolachlor Metolachlor Metolachlor
MC MC MC
ECIR1.5 ECIR1.5 EC50
4 5
PPARg-transFACTORIAL AhR-CAFLUX
(Romanov et al., 2008) (Nagy et al., 2002)
(Martin et al., 2010) (Martin et al., 2010) (Creusot et al., 2010; Lemaire et al., 2006) (Martin et al., 2010) (Kruger et al., 2008; US EPA, 2008)
MT TCDD
6
H4IIEluc
(Jarosova et al., 2012; Sanderson et al., 1996)
(Lee et al., 2013)
7 8
(Romanov et al., 2008) (Escher et al., 2008)
(Martin et al., 2010) (Tang and Escher, 2014)
(Ellman et al., 1961)
11 12
AR-GeneBLAzer ERa-GeneBLAzer
(Huang et al., 2011) (Huang et al., 2011)
13
ER-CALUX
(Sonneveld et al., 2005)
14
E-SCREEN
(Soto et al., 1995)
15
YES
(Routledge and Sumpter, 1996)
16 17
hERa-HeLa-9903 PR-CALUX
(OECD, 2009) (Sonneveld et al., 2005)
18
GR-CALUX
(Sonneveld et al., 2005)
(Mishra et al., 2002; Neale and Escher, 2013) (Houtman et al., 2009; Leusch et al., 2014b; Sonneveld et al., 2006) (Huang et al., 2011) (Huang et al., 2011), this study (Escher et al., 2014; Houtman et al., 2009, Houtman et al. 2006; Legler et al., 2002; Leusch et al., 2010, Leusch et al. 2014b; Schenk et al., 2010; Schreurs et al., 2005; Sonneveld et al., 2005, Sonneveld et al. 2006; van der Burg et al., 2010) (Behnisch et al., 2001; Escher et al., 2014; € rner et al., 2001; Ko Leusch et al., 2010; Soto et al., 1995) (Escher et al., 2014; Leusch et al., 2010; Rutishauser et al., 2004; Sanseverino et al., 2005; Vinggaard et al., 2000) (Takeyoshi, 2006) (Houtman et al., 2009; Leusch et al., 2014b) (Houtman et al., 2009)
Parathion
10
AhR-cisFACTORIAL Algae photosynthesis inhibition Acetylcholinesterase inhibition AR-CALUX
Malathion 2,3,7,8Tetrachlorodibenzo-dioxin 2,3,7,8Tetrachlorodibenzo-dioxin Carbaryl Diuron
9
(Sonneveld et al., 2005)
m
n
EBT-BEQ
2.4 106 2.3 105 1.2 106
65 32 18
18 4 1
MCEQ MCEQ MCEQ
ECIR1.5 EC50
1.2 106 1.5 1011
37 5
2 1
N.D. N.D.
TCDD
EC50
1.2 1011
6
0
N.D.
Carb D
ECIR1.5 EC50
1.2 106 1.5 108
17 11
3 7
CarbEQ DEQ
18 0.6
mg/L mg/L
PT
EC50
1.9 107
PTEQ
26
mg/L
10
5
10
3
1
59 N.D. N.D.
mg/L
Testosterone
TT
EC50
9.1 10
Testosterone 17b-Estradiol
TT E2
EC50 EC50
1.6 109 6.5 1011
6 22
4 6
TTEQ EEQ
14 1.8
ng/L ng/L
17b-Estradiol
E2
EC50
6.4 1012
28
7
EEQ
0.2
ng/L
17b-Estradiol
E2
EC50
7.1 1012
16
6
EEQ
0.9
ng/L
17b-Estradiol
E2
EC50
3.2 1010
14
5
EEQ
12
ng/L
17b-Estradiol Progesterone
E2 P4
EC50 EC50
8.2 1012 7.0 1010
8 1
7 0
EEQ P4EQ
0.6 N.D.
ng/L
Progesterone Dexa-methasone
P4* Dex
EC50
7.2 107 3.9 1010
1
1
P4EQ* DexEQ
230 150
mg/L
N.D.
ng/L
N.D. not defined. *Because the conventional reference compound for GR reporter gene assays (dexamethasone) does not have a drinking water guideline value, progesterone (P4, which is biologically active in both PR and GR reporter gene assays) was used for the GR-CALUX.
B.I. Escher et al. / Water Research 81 (2015) 137e148
#
B.I. Escher et al. / Water Research 81 (2015) 137e148
141
2. Material and methods 2.1. Collection of literature data The effect concentrations of chemicals listed in the AGWR for the selected 18 bioassays were collected from the literature. All bioassays included are described in detail in a previous paper (Escher et al., 2014, Table 1). In a few cases, review papers were used but mostly original publications were selected. If more than one EC value was available, the median was used. All median EC values are listed in the Supplementary Data, Table S1 and S2. Additional experiments were performed according to the experimental methods described in Escher et al. (2014). With exception of the FACTORIAL assays, all EC values refer to EC50. The FACTORIAL assays are biosensors with multiplexed library of transcription factors that allow the simultaneous quantification of response of 25 nuclear receptors and 48 transcription factor response elements to chemicals (Martin et al., 2010) or water samples (Escher et al., 2014). Only a few endpoints of the FACTORIAL assays were active when exposed to water samples and we included these in the analysis (Table 1, bioassays # 1,2,4,7). To enable comparison with water samples for which the ECIR1.5 were available, the raw concentration-effect data for FACTORIAL assays given in Martin et al. (2010) were reevaluated using induction ratios (IR) according to Escher et al. (2014), and the effect concentrations causing an IR of 1.5 (i.e. 50% over the control), ECIR1.5, were calculated for better comparability with the water samples' ECIR1.5.
2.2. Bioanalytical equivalent concentrations of water samples Ten water samples had been profiled with 103 bioassays in a previous study (Escher et al., 2014). The water data for 18 of these bioassays were selected for the current study to illustrate the application of the EBT-BEQ approach. Samples were taken at two recycling water plants that use different technologies: Secondary treated effluent from a conventional wastewater treatment plant (Eff-1) was the input to the first plant; samples were taken after membrane filtration (MF), after reverse osmosis (RO) and after advanced oxidation (AO). The treatment train of the second plant was based on oxidation of secondary effluent (Eff-2), and the recycled water was obtained after ozonation and biologically activated carbon filtration (O3/BAC). For comparison samples from a drinking water plant inlet (river) and outlet (DW), as well as stormwater (SW) were tested. The sample “blank” refers to a laboratory ultrapure water sample that underwent the same extraction and enrichment process as the samples. The EC values were previously published (Escher et al., 2014) and are summarized for convenience in the Supplementary Data, Table S3 (reprinted with permission from Escher et al., Benchmarking organic micropollutants in wastewater, recycled water and drinking water with in vitro bioassays. Environ Sci Technol. 48:1940e1956; 2014. Copyright 2014 American Chemical Society). The EC10 and ECIR1.5 values of the water samples have units of relative enrichment factors (REF), which have the net units of Lwater sample/Lbioassay, (Eq. (1)).
REF ¼
water volume equivalent transfered to biosassay total volume of bioassay
(1)
The EC10 and ECIR1.5 values were converted to BEQ using the same reference compounds as used to derive the EBT-BEQ. The BEQ was calculated from the EC of the reference compound (in mol/ Lbioassay) divided by the EC of the sample, which resulted in BEQs with the units mol/Lwater sample (Eq. (2)).
Fig. 2. Approach to derive EBT-BEQs for cell-based bioassays (GV ¼ guideline value, EC ¼ effect concentration, BEQ ¼ bioanalytical equivalent concentration, REP ¼ relative effect potency, n ¼ number of chemicals i, m ¼ number of available EC values per bioassay j, ref ¼ reference compound).
BEQ ¼
EC50 ðreference compoundÞ EC50 ðsampleÞ
(2)
The EC50 of the reference compounds were converted to EC10 values assuming a slope of the log-logistic concentration-effect curve of one, which is a reasonable assumption for most in vitro assay data in our experience. While EC50 values had been used for the EBT-BEQ derivation the experimental samples had low effects and often did not reach the 50% effect level. To avoid extrapolations, the 10% effect level was reported for the water samples in the previous study (Escher et al., 2014). However, in principle, the BEQ are independent on the effect levels they were derived from, provided that the concentration-effect curves of reference compound and samples have the same (or at least a similar) slope (Villeneuve et al., 2000). The minimum detectable BEQ was calculated from the highest tested REF of the water samples. 3. Theory/calculation 3.1. General approach to derive EBT-BEQs For each bioassay j, a separate EBT-BEQ must be derived (Fig. 2). As a first step, effect concentrations were collated from literature for each bioassay and for as many chemicals as possible for which
142
B.I. Escher et al. / Water Research 81 (2015) 137e148
GV are defined (Fig. 2, “Input data”). The GV are not associated with specific modes of action but generally protective of human health. In vitro bioassays quantify one step in the toxicity pathway but not necessarily the one that is most toxicologically relevant for a given chemical. The effects observed in the in vitro bioassays thus often do not match the GV. Therefore, as the second step, it is important to filter the available number of effect data points (m) to reduce to the number of chemicals i (n) for which the bioassays are indicative of the corresponding GVi (Fig. 2, “Filtering”). EC values were excluded if they were smaller or larger than the GV by an order of magnitude, because if they were included the BEQ distribution would be skewed (as discussed in detail in the next sections below). In the next step, the relative effect potencies REPi were calculated in each bioassay j for each chemical I (Fig. 2, “Bioanalytical equivalents”). Each GVi was then converted to the corresponding BEQi by multiplication with the corresponding REPi. Finally, a cumulative distribution of BEQi was plotted and the EBT-BEQ derived from the fifth percentile of this distribution (Fig. 2, “Cumulative distribution”). The 5th percentile was selected as this will be protective for the majority of chemicals and will account for the mixture effects that chemicals may have in the given bioassays.
3.2. Filtering: selection of appropriate bioassay data As the cell-based bioassays are used as analytical tools, they must be sufficiently responsive to the chemicals selected from the guideline for the derivation of EBT-BEQ. The ECi of the m chemicals, for which EC values were available in the literature, were plotted i against the GVi (Fig. 2). Those chemicals outside a band of 0.1 GV ECi 10 were excluded from the analysis, resulting in n chemicals going forward. This band was chosen to ensure that only chemicals that target the mode of action or receptor of relevance in the bioassay were included. The point of departure for the derivation of the GVi does not necessarily need to be consistent with the endpoint of the bioassay but with the filtering step, we ensure that only those chemicals that are responsive in the given bioassay are selected. If GVi < 0.1 ECi, the bioassay is not sensitive enough to capture the effects caused at the GV-concentrations in the bioassay. While there are EC available, these chemicals are not potent enough to trigger the target mode of action of the bioassays at the GVi concentrations and are thus not suitable for inclusion into the derivation of that specific EBT-BEQ. If GVi > 10 ECi then the bioassay would be sensitive enough to capture the effect at the GV-concentrations in the given bioassay but chemical i may dominate the overall effect. This could be due to the chemical being more potent in vitro than in vivo or the GVi being based on a different mode of action. Therefore these values were also excluded to assure a fairly symmetric distribution of GVi/ECi ratios around 1, which also resulted in a more robust BEQi distribution. The filtering step has the advantage that it recognizes multiple modes of action of a chemical and accounts for a range of intrinsic potencies. The disadvantage is that it is possible that highly potent chemicals, even well-known reference chemicals, can be excluded in the filtering step if they have no relationship to the point of departure for the GVi.
REPi ¼
ECref ECi
(3)
Typically used reference chemicals for bioassays were only used if these reference chemicals were also included in the selected guideline. Where these were not included, instead a potent (but not necessarily the most potent) chemical was used as the reference chemical. If several bioassays were available for the one endpoint, a consensus reference chemical was selected that had available EC values in all bioassays for this endpoint. The BEQi for each chemical i were then calculated by multiplying the GVi with the REPi (Eq. (4)).
BEQ i ¼ REPi GVi
(4)
The choice of the reference compound is crucial. In relation to bioassay performance a reference compound needs to fulfill a number of performance criteria such as selectivity, specificity and have a clear mode of action, while as a trigger value it should also have chemical and biological relevance. 3.4. Distribution of BEQ and derivation of EBT-BEQ Biological data are typically log-normally distributed (Limpert et al., 2001) and therefore cumulative distributions of the log BEQi were used to derive the EBT-BEQ. The distributions were calculated according to the US EPA SSD generator (http://www.epa. gov/caddis/da_software_ssdmacro.html). The calculated bioanalytical equivalents BEQi were ranked and then the rank converted to proportions with Eq. (5). The logBEQi were plotted against the proportioni (Eq. (6)). The 5th percentile (Eq. (7)) was defined as EBT-BEQ and was calculated from a linear regression through zero.
proportioni ¼
rank 0:5 n
propotioni ¼ slope logBEQ i logEBT BEQ ¼
0:05 slope
(5) (6) (7)
In an ideal world, if the given bioassay was fully representative of the mode of action for which the health-based GV was derived, the BEQi would be equal for all chemicals i. This is not the case because the bioassay endpoints do not directly match the ultimate health effects underlying the derivation of GV even if they are related after the filtering step in Section 3.2. In reality, as will be discussed in the results section, the BEQi varied by a factor 10 to 100 between different chemicals for a given bioassay. As a realistic measure of the mixture effect one could use the 50th percentile (or median) of the distribution of EBT-BEQ. However, in a precautionary approach and to account for the possibility that in a realistic mixture chemicals with a higher potency contribute more to the mixture than low-potency chemicals, we chose the lower 5th percentile as EBT-BEQ. This aligns with the HC5 concept of species sensitivity distributions where HC5 stands for hazardous concentration to 5% of the species (Posthuma et al., 2002). 4. Results and discussion
3.3. Calculation of bioanalytical equivalent concentrations
4.1. Development of EBT-BEQs
Each ECi was converted to relative effect potencies REPi in relation to a reference compound with ECref (Eq. (3)).
Of the 103 bioassays applied to the ten water samples, only 65 bioassays targeted receptor-mediated effects. The other 38
B.I. Escher et al. / Water Research 81 (2015) 137e148
143
Fig. 3. Steps in the derivation of EBT-BEQ on the example of the E-SCREEN assays. A. All available experimental EC50 values; B. EC50 values after the filtering step; C. Cumulative distribution of estradiol equivalent concentrations (EEQi).
bioassays targeted reactive MOAs, adaptive stress response pathways and cytotoxicity, for which no EBT-BEQs can be derived, although EBT-ECs have been derived (Escher et al., 2013; Tang et al., 2013). For 18 out of the 65 receptor-mediated bioassays, we found experimental data for chemicals included in the AGWR (Table 1 and individual EC-values in Supplementary Data, Tables S1 and S2). Between one and 65 EC values were found per bioassay (median ¼ 13). In particular the assays that were included in ToxCast screening of large datasets had many matching data, e.g., 65 EC values were available for the PXR-cisFACTORIAL assay (Martin et al., 2010). The process of derivation of the EBT-BEQ is illustrated on the example of the E-SCREEN assay in Fig. 3 and depicted for all 18 bioassays in the Supplementary Data, Figure S1. 4.1.1. Filtering: selection of appropriate bioassay data After the filtering step, only 0 to 18 (median ¼ 5) EC values remained (Table 1). Large numbers of chemicals were removed by filtering for the bioassays indicative of metabolism (Table 1, Supplementary Data, Figure S1): for the pregnane X receptor (PXR), 65, 32 and 18 EC values were available in the PXRcisFACTORIAL, PXRtransFACTORIAL and HG5LN PXR assays, respectively. As the majority of assays were not sensitive to these chemicals in relation to the GV, only 18, 4 and 1 chemicals, respectively, remained after filtering. This is consistent with the general observation that PXR is a highly promiscuous receptor (Ihunnah et al., 2011). A similar observation was made for the peroxisome proliferatoractivated receptor gamma (PPARg), where only 2 out of 39 chemicals were within one order of magnitude of the GV. Thus no EBTBEQ could be derived for PPARg. The aryl hydrocarbon receptor (AhR) is mainly activated by dioxin-like chemicals but also by many more water-soluble chemicals with typically lower affinity (Denison and Nagy, 2003), but most chemicals in the AGWR had too low GVi to be included in the BEQ derivation. A different picture was observed for the algal photosynthesis inhibition test. This bioassay was included despite the fact that it has no direct relevance to human health, because eleven herbicides are included in the guideline. As we suggest using bioassays as bioanalytical indicators for the presence of mixtures of chemicals it is legitimate to apply a sensitive indicator bioassay as (bio)analytical tool even if it is not clearly related to a human health endpoint. Since the GV was derived for human health but the alga test targets photosynthesis, the bioassay was often more sensitive than the GV. Non-herbicides are also active in this bioassay but at low potency, therefore they were omitted upfront (Tang and Escher, 2014).
Filtering removed only 4 chemicals. For acetylcholinesterase inhibition, a similar observation as for herbicidal activity was made. Evidently this endpoint was not the basis of the derivation of the GV, therefore 5 out of 10 EC50 values had to be removed either as a result of too high or too low potency. For the endocrine active compounds, the available EC50 dataset had to be reduced during the filtering step but not to an as great extent as for the xenobiotic metabolism endpoints. For example, in the E-SCREEN assay (Fig. 3), 16 available EC50 values were reduced to 6 after filtering. Even 17b-estradiol had to be excluded because it fell into the category of GVi > 10 ECi. Most likely, this is because the GV of 17b-estradiol in the AGWR is based on its carcinogenicity not on estrogenic effects. However, 17b-estradiol was still used as reference compound to derived 17b-estradiol equivalents (EEQ). After filtering, no EC50 value remained for H4IIEluc and PRCALUX and those bioassays were not further considered. This does not mean that the endpoints of AhR activation and progesterone receptor response were considered to be irrelevant, only that the current potency dataset is insufficient to derive an EBT-BEQ value. 4.1.2. Distribution of BEQ and derivation of EBT-BEQ The lack of available data meant that EBT-BEQ could not be calculated for 7 bioassays. For the remaining 11 bioassays, linear regressions of high quality were obtained after plotting the log BEQ versus the cumulative proportions (Fig. 3 and Supplementary Data, Figure S1 and Table S4). The high correlation coefficients indicate that the BEQs were indeed log-normally distributed as had been anticipated from theoretical considerations. The EBT-BEQs were derived from the lower fifth percentile of the distribution and are listed in Table 1. 4.2. Modes of action 4.2.1. Xenobiotic metabolism 4.2.1.1. Pregnane-X receptor (PXR). Metolachlor (MC) served as the reference chemical for induction of the PXR. It was chosen as the one compound for which EC50 values were available in all three assays. Due to limited availability of experimental data a valid EBTMCEQ of 59 mg/L could only be derived for PXRcisFACTORIAL (Supplementary Data, Table S4 and Table 1). 4.2.1.2. Peroxisome proliferator-activated receptor (PPARg). For PPARg, malathion (MT) was chosen as reference compound because it was the chemical closest to the 1:1 line in the filtering plot. This is certainly not the prototype PPARg activator and in the
144
B.I. Escher et al. / Water Research 81 (2015) 137e148
future, fingerprinting of larger lists of chemicals should come up with a potent reference compound that is also included in the guideline. We recognize the relevance of this endpoint but conclude that at present it is premature to define an EBT for the PPARg-transFACTORIAL assay. 4.2.1.3. Aryl hydrocarbon receptor (AhR). After filtering only one data point remained for the AhR-CAFLUX assay (4-nonylphenol, Table 1). This value could theoretically be converted to 2,3,7,8-tetrachlorodibenzodioxin equivalents TCDDEQ, and used as tentative EBT-TCCDEQ. Clearly, this is unsatisfactory and more data on watersoluble compounds would need to be generated to obtain a reliable EBT-TCCDEQ. Data on activation of the AhR by waterborne chemicals in the AhR-CAFLUX and H4IIEluc remained too scarce to be able to derive an EBT at this stage. Further research on ligands relevant in the aqueous environment would be required before an EBT can be derived. However, the AhR-cisFACTORIAL was tested for a large number of pesticides (Martin et al., 2010), 17 of which overlapped with our guideline. After the filtering step only three chemicals remained. The most potent of the three, carbaryl (Carb), was chosen as reference compound and the EBT-CarbEQ was 18 mg/L, which is a preliminary EBT due to the low input data quality. 4.2.2. Specific receptor-mediated modes of action 4.2.2.1. Herbicidal activity. As discussed above, it is difficult to compare drinking water GVs, which are derived for human health protection with specific assays such as these based on photosynthetic organisms, algae. While it would of course be preposterous to use inhibition of photosynthesis as a relevant endpoint purely based on human health considerations, our premise was to apply bioassays as analytical tools to assess the mixture effects of chemicals relevant for water pollution. This approach renders this assay relevant as herbicides constitute a major threat to drinking water sources and are included in drinking and recycled water guidelines. We have previously evaluated the activity of herbicides and non-herbicides on photosynthesis inhibition and concluded that the mixture effects in realistic water samples are dominated by the herbicides (Tang and Escher, 2014). Therefore we included only the 11 herbicides from the AGWR in our analysis. While in most bioassays filtering removed chemicals that had too low potency in the bioassays, here the number of chemicals included was reduced from 11 to 7 by removing four herbicides whose GV was too high in relation to their activity in the bioassay. The remaining seven herbicides yielded a high quality distribution (Supplementary Data, Table S4). With the reference compound diuron (D) (EC50 15 nM, Supplementary Data, Table S1) a
Fig. 4. Comparison of BEQ distributions for the estrogenicity bioassays.
provisional EBT-DEQ of 0.6 mg/L was obtained (Table 1). 4.2.2.2. Acetylcholinesterase (AChE) inhibition. Many insecticides are inhibitors of the AChE but humans are not necessarily sensitive due to extensive detoxification, and the human health GV is therefore not necessarily closely related to the mode of action of AChE inhibition. Consequently only a small set of five insecticides could be included in the BEQ distribution. The EC50 of the reference compound parathion (PT) was 0.2 mM (Table 1) and the provisional EBT-PTEQ resulting from a high-quality linear regression was 26 mg/ L (Supplementary Data, Table S4 and Table 1). 4.2.3. Endocrine disruption 4.2.3.1. Androgenic activity. While the experimental database of the AR-CALUX assays was too small to derive a BEQ distribution, four chemicals from the AGWR remained after filtering in the ARGeneBLAzer assay allowing the calculation of an EBT-BEQ for that assay. The EC50 of the reference compound testosterone (TT) was 1.6 nM (Table 1) and the EBT-TTEQ was 14 ng/L (Supplementary Data, Table S4 and Table 1). 4.2.3.2. Estrogenic activity. Five bioassays for estrogenicity were included. The EC50 of the reference compound 17b-estradiol ranged over two orders of magnitude from 3 to 7 pM for ER-CALUX, ESCREEN and hERa-HeLa-9903 over 65 pM for ERa-GeneBLAzer to 320 pM for the YES assay (Table 1). The EEQ-distributions were comparable between ER-CALUX, E-SCREEN, hERa-HeLa-9903 and ERa-GeneBLAzer (Fig. 4 and Supplementary Data Table S4). The BEQ distributions for the YES assay were at the highest concentrations, resulting in an EBT-EEQ of 12 ng/L, which was almost an order of magnitude higher than the EBT-EEQ of the four other estrogenic assays, which ranged from 0.2 to 1.8 ng/L. As different numbers and types of chemicals were available, the chemicals included in the distribution varied, so that a direct comparison is limited. To assess the true differences in the BEQ distributions one would have had to measure EC values of the same chemicals. The comparison we have done is only valid if the experimental EC represented random samples out of the true distribution and this is less likely the fewer the number of EC values are available. In the ER-CALUX, we obtained an EBT-EEQ of 0.2 ng/L. For comparison, the independent derivation based on an entirely different rationale in Brand et al. (2013) resulted in a trigger value of 3.8 ng/L EEQ. A difference of a factor of 20 appears large but one needs to account for the fact that entirely different assumptions are underlying the two methods for EBT derivation. Safe concentration of estrogenic equivalents in wastewater determined from environmental predicted no-effect concentration and extrapolated to cell-based bioassays ranged from 0.2 to 0.4 ng/L EEQ for ER-CALUX, 0.1e0.3 ng/L for E-SCREEN and 0.1e0.2 ng/L EEQ for the YES (Jarosova et al., 2014), i.e., they cover very similar concentration ranges as the EBT-EEQs derived here. The protection target for these “safe concentration of estrogenic equivalents” is fish reproduction but since indirect potable reuse schemes rely on residence in an environmental buffer such as a reservoir or lake prior to reuse, it is interesting to note that the proposed EBT-EEQ proposed here would also be protective of ecological endpoints. 4.2.3.3. Progesterone and glucocorticoid receptor activity. Typically dexamethasone (Dex) is used as reference compound for glucocorticoid activity (Schriks et al., 2013). Since this chemical was not included in the AGWR, we used progesterone (P4, EC50 0.7 mM, Table 1) as the reference compound, as it is active in both the PR and GR reporter gene assays (Houtman et al., 2009). No distribution could be made because this was also the only EC50 available that
B.I. Escher et al. / Water Research 81 (2015) 137e148
145
Table 2 Summary of BEQs of the bioassays in relation to EBT-BEQs. Definition of sample abbreviation in Section 2.3. The abbreviations of the BEQ for the different bioassays are given in Table 1. The recycled water samples are in bold font. #
1 7 8 9 11 12 13 14 15 16 18
Abbreviation
PXR-cisFACTORIAL AhR-cisFACTORIAL Algae photosynthesis inhibition Acetylcholinesterase Inhibition AR-GeneBLAzer ER-GeneBLAzer ER-CALUX E-SCREEN YES hERa-HeLa-9903 GR-CALUX
BEQ
Units
EBT-BEQ
jddRecycling plant 1ddj
jddRecycling plant 2dj
jddReference samplesddj
Eff-1
MF
RO
AO
Eff-2
O3/BAC
River
DW
SW
Blank
2000 17 0.2
1900 19 0.2
Contents lists available at ScienceDirect
Water Research journal homepage: www.elsevier.com/locate/watres
Effect-based trigger values for in vitro bioassays: Reading across from existing water quality guideline values Beate I. Escher a, b, c, *, Peta A. Neale c, d, Frederic D.L. Leusch d a
UFZ e Helmholtz Centre for Environmental Research, Cell Toxicology, Leipzig, Germany Eberhard Karls University Tübingen, Environmental Toxicology, Center for Applied Geosciences, Tübingen, Germany c The University of Queensland, National Research Centre for Environmental Toxicology, Entox, Brisbane, QLD 4108, Australia d Smart Water Research Centre, School of Environment, Griffith University, Southport, QLD 4222, Australia b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 26 February 2015 Received in revised form 15 May 2015 Accepted 25 May 2015 Available online 27 May 2015
Cell-based bioassays are becoming increasingly popular in water quality assessment. The new generations of reporter-gene assays are very sensitive and effects are often detected in very clean water types such as drinking water and recycled water. For monitoring applications it is therefore imperative to derive trigger values that differentiate between acceptable and unacceptable effect levels. In this proofof-concept paper, we propose a statistical method to read directly across from chemical guideline values to trigger values without the need to perform in vitro to in vivo extrapolations. The derivation is based on matching effect concentrations with existing chemical guideline values and filtering out appropriate chemicals that are responsive in the given bioassays at concentrations in the range of the guideline values. To account for the mixture effects of many chemicals acting together in a complex water sample, we propose bioanalytical equivalents that integrate the effects of groups of chemicals with the same mode of action that act in a concentration-additive manner. Statistical distribution methods are proposed to derive a specific effect-based trigger bioanalytical equivalent concentration (EBT-BEQ) for each bioassay of environmental interest that targets receptor-mediated toxicity. Even bioassays that are indicative of the same mode of action have slightly different numeric trigger values due to differences in their inherent sensitivity. The algorithm was applied to 18 cell-based bioassays and 11 provisional effectbased trigger bioanalytical equivalents were derived as an illustrative example using the 349 chemical guideline values protective for human health of the Australian Guidelines for Water Recycling. We illustrate the applicability using the example of a diverse set of water samples including recycled water. Most recycled water samples were compliant with the proposed triggers while wastewater effluent would not have been compliant with a few. The approach is readily adaptable to any water type and guideline or regulatory framework and can be expanded from the protection goal of human health to environmental protection targets. While this work constitutes a proof of principle, the applicability remains limited at present due to insufficient experimental bioassay data on individual regulated chemicals and the derived effect-based trigger values are of course only provisional. Once the experimental database is expanded and made more robust, the proposed effect-based trigger values may provide guidance in a regulatory context. © 2015 Elsevier Ltd. All rights reserved.
Keywords: In vitro Bioassay Bioanalytical equivalent concentration Water quality Human health
1. Introduction 1.1. Chemical water quality guidelines to protect human health Health-based water quality guideline values (GV) are defined as concentrations of chemicals in drinking water that do not pose an * Corresponding author. Helmholtz Centre for Environmental Research e UFZ, Cell Toxicology, Permoserstraße 15, 04318 Leipzig, Germany. E-mail address: [email protected] (B.I. Escher). http://dx.doi.org/10.1016/j.watres.2015.05.049 0043-1354/© 2015 Elsevier Ltd. All rights reserved.
appreciable risk to health over a lifetime of exposure. GV are based on acceptable daily intake (ADI, also called reference dose RfD, or tolerable daily intake TDI). The ADI is typically extrapolated from ‘No Observed Adverse Effect Levels’ derived from a set of animal toxicity tests with extrapolation factors, accounting for uncertainties related to extrapolation from a model system to a human population (Ritter et al., 2007). Guideline values for individual chemicals were defined, for example, by the WHO for drinking water (WHO, 2011). Healthbased guidelines are also available for recycled water for indirect
138
B.I. Escher et al. / Water Research 81 (2015) 137e148
Abbreviations AO ADI ADWG AGWR BAC BEQ CarbEQ DEQ DexEQ DW EBT EEQ Eff EPA EC GV HC5 IR
advanced oxidation acceptable daily intake Australian Drinking Water Guidelines Australian Guidelines for Water Recycling biologically activated carbon bioanalytical equivalent concentration carbaryl equivalent concentration diuron equivalent concentration dexamethasone equivalent concentration drinking water effect-based trigger value estradiol equivalent concentration effluent Environmental Protection Agency effect concentration guideline value hazardous concentration to 5% of species induction ratio
potable reuse and for various other water types and environmental protection goals (Escher and Leusch, 2012). Snyder et al. (2008) proposed to derive drinking water equivalent levels for pharmaceuticals using the minimum therapeutic dose. This matches the approach proposed by Schriks et al. (2010) who developed a pragmatic approach to derive provisional drinking water guideline values from ADIs for unregulated chemicals immediately after their detection in drinking water. 1.2. Bioanalytical tools for water quality assessment Given the overwhelming number of chemicals present in water (Schwarzenbach et al., 2006; Villanueva et al., 2014), chemical analysis has been complemented for many years by in vitro bioassays which not only give information on the level of effect but also on the type of effect, i.e., the mode of action. Further, bioanalytical tools also integrate the mixture effects of chemicals that act according to the same mode of action in a concentrationadditive manner (Escher and Leusch, 2012). In vitro bioanalytical tools have also been recommended for the assessment of drinking water (van Wezel et al., 2010). Since 2008, the high-throughput in vitro toxicity program Tox21 of the National Institute of Health (NIH) and the US Environmental Protection Agency (EPA) has been and continues to provide quantitative information on the toxicity pathways of almost 10,000 organic compounds (Attene-Ramos et al., 2013b; Shukla et al., 2010). In vitro assays included in Tox21 are phenotypic assays (e.g., targeting mitochondrial toxicity (Attene-Ramos et al., 2013a)), pathway assays (e.g., targeting cell viability in specific cell lines) and nuclear receptor assays (e.g., targeting hormone receptors and metabolic pathways (Huang et al., 2011)) and various other endpoints (Tice et al., 2013). The chemicals and pathways screened for Tox21 are also relevant for water quality and some of the assays have been applied successfully to water quality testing together with more established in vitro assays (Escher et al., 2014). 1.3. Need for effect-based trigger values (EBT) Modern reporter gene assays and other cellular bioassays are so sensitive that even very clean waters often induce detectable
MCEQ MF MTEQ NIH P4EQ PTEQ PXR REF REP RfD RO SW TCDDEQ
metolachlor equivalent concentration membrane filtration malathion equivalent concentration National Institute of Health progesterone equivalent concentration parathion equivalent concentration pregnane X receptor relative enrichment factor relative effect potency reference dose reverse osmosis stormwater 2,3,7,8-tetrachlorodibenzodioxin equivalent concentration TDI tolerable daily intake TTEQ testosterone equivalent concentration ToxCast US EPA's program “Toxicity Forecaster” Tox21 NIH's program “Toxicology in the 21st Century” WHO World Health Organization
responses in vitro. High sensitivity is an advantage from the point of view of achieving the highest possible sensitivity and in the assessment of treatment efficacy (i.e., water treatment plants). In terms of water quality, however, a cellular response does not translate directly into a toxicological effect. A ‘simple’ yes or no answer will be defined by the sensitivity of the test and is not sufficient to evaluate if a water is “good” or “bad”. Therefore, trigger values are required that differentiate an acceptable response from an unacceptable response. Recent work that compared the bioassay response of chemicals detected in recycled water with the bioassay response of the entire water sample has shed light on knowledge gaps. For bioassays indicative of integrative effects (i.e., non-specific effects, reactive effects and adaptive stress responses) typically only a very small fraction (often less than 1% of the effect) can be explained by known and identified chemicals (Escher et al., 2013; Reungoat et al., 2012; Tang et al., 2013). Thus unidentified and unknown chemicals cause the majority of the biological response indicating the need to account for mixture effects in the derivation of EBTs. Also for these types of responses there is no clear-cut reference chemical available because many different chemicals can induce the effect. Therefore, we have previously proposed EBT-effect concentrations (EBT-EC) that account for mixture effects of these types of chemicals in non-specific bioassays (Escher et al., 2013; Tang et al., 2013). For example, for the derivation of the EBT for the bioluminescence inhibition test we assumed 1000 chemicals were present at 5% of their guideline value and applied a concentration-additive mixture toxicity model. The EBT-EC50 was expressed in relative enrichment factor units (REF), which describe how much the water sample was enriched by, e.g., solid phase extraction prior to the bioassay. The provisional EBT-EC50 for drinking and recycled water then came to 2.8 REF (Tang et al., 2013). In other words, if a sample that is roughly three times enriched (by solid phase extraction for example) shows 50% or more effect in this bioassay, then the EBT is exceeded; if on the other hand it induces less than 50% effect, then the water quality is acceptable. A similar approach for the oxidative stress response quantified with the AREc32 bioassay resulted in a provisional EBT-ECIR1.5 of 3 REF (Escher et al., 2013). Both provisional EBT-EC were able to suitably differentiate between water quality before and after advanced water treatment in recycling
B.I. Escher et al. / Water Research 81 (2015) 137e148
plants. In contrast to integrative assays, the majority of receptormediated effects (e.g., inhibition of enzymes or activation of hormone receptors) can typically be explained by known chemicals (Escher et al., 2011; Leusch et al., 2005, 2014b; Rutishauser et al., 2004; Tang and Escher, 2014). Effects can be expressed as socalled bioanalytical equivalent concentrations (BEQ), which represent the concentrations of a reference chemical that would elicit the same effect as the water sample of unknown composition. It has been suggested to apply concentration addition as the realistic worst-case scenario for mixture effects in risk assessment (Backhaus and Faust, 2012) and the BEQ concept is based on this mixture toxicity concept, which is valid for chemicals that have a common mode of action. As straightforward as it may seem, however, it would be wrong to simply compare the BEQ to the GV of the reference compound. While the BEQ of the water sample includes information about the mixture effect, the derivation of a single chemical GV is exclusively based on the effects of that single compound. 1.4. Framework for the application of EBT in water quality monitoring EBT are not meant to replace other monitoring tools but can be applied to complement chemical analysis and to decrease uncertainty in water quality assessment. A possible tiered approach to water quality assessment is shown in Fig. 1. In a first screening step, a limited number of indicator chemicals would be monitored and in parallel a limited number of indicator bioassays would be applied. Only if the measured chemical concentrations exceed one or more GV or the effects in the indicator bioassays are above the EBT, then a much larger list of chemicals with defined GV would need to be monitored to determine overall water quality. This is just one of many possible ways to implement EBTs in a water quality monitoring strategy. 1.5. Aims The aim of this study was to derive an algorithm for the derivation of EBT for receptor-mediated effects and to apply it as an illustrative example to Australian recycled water that was previously screened with 103 different bioassays (Escher et al., 2014). Eighteen of the 103 bioassays were selected for this exercise. The omitted bioassays were either not relevant for water quality (i.e., they had not shown any effects with water samples in the mentioned benchmarking study (Escher et al., 2014)), had no available data on reference chemicals or other regulated chemicals,
Fig. 1. Framework for the application of bioanalytical tools in water quality monitoring (C ¼ concentration, GV ¼ guideline value, EBT ¼ effect based trigger value).
139
or the BEQ concept could not be applied to these bioassays. We included bioassays without any direct connection to human health endpoints to test whether non-target bioassays could serve as purely (bio)analytical tools for the quantification of the presence of certain groups of chemicals. For example, the mixture effect of the herbicides included in a guideline can be evaluated using their effects on algae because this is a sensitive and quantitative indicator for this chemical group despite its lack of direct human health relevance. We applied the proposed EBT derivation method to GV from the Australian Guidelines for Water Recycling: Augmentation of Drinking Water Supplies (AGWR) (NRMMC & EPHC & NHMRC, 2008), which was adopted in the Public Health Regulation, Schedule 3B Standards for quality of recycled water supplied to augment a supply of drinking water in Queensland, Australia (Queensland Government, 2005). The water samples tested were all sampled in Queensland (Escher et al., 2014). The GV in the Australian Drinking Water Guidelines (ADWG) (NHMRC, 2011) are a subset of the GV in the AGWR thus the proposed EBTs should be applicable to both recycled and drinking water in Queensland and other States of the Commonwealth of Australia. Without much effort, the approach can be adapted to guidelines from other countries and regions, including the WHO Guidelines for DrinkingWater Quality (WHO, 2011).
1.6. Approach One previous study has derived bioassay trigger values for bioassays indicative of receptor-mediated effects in drinking water. Brand et al. (2013) proposed to derive effect-based trigger for the CALUX bioassays from the ADI of the reference compound using bioanalytical equivalents (BEQ) that account for the concentrationadditive effect of chemicals that act similarly but may have different toxicokinetics (Punt et al., 2013). The toxicokinetic differences between in vivo and in vitro were considered by accounting for oral bioavailability and binding to plasma proteins but not for metabolism (Brand et al., 2013). The derived toxicokinetic factors were used to estimate the acceptable target concentrations of the reference compound, which in turn were translated to acceptable target concentrations of another compound than the reference compound in the associated CALUX assay by the relative effect potency in vitro. Finally, the water consumption and an allocation factor were used to derive the trigger value in the various CALUX bioassays. An alternative approach would be to apply bioassays purely as an analytical tool and as a means to capture the mixture effects of chemicals in water. In this case one does not need to invoke any in vitro to in vivo extrapolation, which avoids the associated uncertainties. The algorithm we propose starts with existing sets of chemical GV and directly read across from the GV to the EBT. The only input data required are the GV and published effect concentrations (EC) of the regulated chemicals. Thus the proposed method is simple and can be easily adapted if water quality guidelines change or bioassays are applied in another region of the world because they can be easily transferred from one regulatory framework to another. Here we develop this read-across approach and illustrate it with examples for chemicals included in the AGWR. The approach is inspired by the work of Jarosova et al. (2014), who predicted safe concentrations of estrogenic activity in wastewater using similar in vitro bioassays as applied in the current study by reading directly across from the environmental predicted no effect concentrations that are derived from in vivo data.
140
Table 1 Bioassays, reference chemicals and summary of EBT-BEQs for all endpoints; m ¼ available number of effect data points; n ¼ number of data EC values included in the distributions. Bioassay
References bioassay method
Literature source of EC value collection
Reference chemical for EBT
Ab-brevi-ation
EC (reference chemical) (M)
1 2 3
PXR-cisFACTORIAL PXR-transFACTORIAL HG5LN PXR
(Romanov et al., 2008) (Romanov et al., 2008) (Lemaire et al., 2006)
Metolachlor Metolachlor Metolachlor
MC MC MC
ECIR1.5 ECIR1.5 EC50
4 5
PPARg-transFACTORIAL AhR-CAFLUX
(Romanov et al., 2008) (Nagy et al., 2002)
(Martin et al., 2010) (Martin et al., 2010) (Creusot et al., 2010; Lemaire et al., 2006) (Martin et al., 2010) (Kruger et al., 2008; US EPA, 2008)
MT TCDD
6
H4IIEluc
(Jarosova et al., 2012; Sanderson et al., 1996)
(Lee et al., 2013)
7 8
(Romanov et al., 2008) (Escher et al., 2008)
(Martin et al., 2010) (Tang and Escher, 2014)
(Ellman et al., 1961)
11 12
AR-GeneBLAzer ERa-GeneBLAzer
(Huang et al., 2011) (Huang et al., 2011)
13
ER-CALUX
(Sonneveld et al., 2005)
14
E-SCREEN
(Soto et al., 1995)
15
YES
(Routledge and Sumpter, 1996)
16 17
hERa-HeLa-9903 PR-CALUX
(OECD, 2009) (Sonneveld et al., 2005)
18
GR-CALUX
(Sonneveld et al., 2005)
(Mishra et al., 2002; Neale and Escher, 2013) (Houtman et al., 2009; Leusch et al., 2014b; Sonneveld et al., 2006) (Huang et al., 2011) (Huang et al., 2011), this study (Escher et al., 2014; Houtman et al., 2009, Houtman et al. 2006; Legler et al., 2002; Leusch et al., 2010, Leusch et al. 2014b; Schenk et al., 2010; Schreurs et al., 2005; Sonneveld et al., 2005, Sonneveld et al. 2006; van der Burg et al., 2010) (Behnisch et al., 2001; Escher et al., 2014; € rner et al., 2001; Ko Leusch et al., 2010; Soto et al., 1995) (Escher et al., 2014; Leusch et al., 2010; Rutishauser et al., 2004; Sanseverino et al., 2005; Vinggaard et al., 2000) (Takeyoshi, 2006) (Houtman et al., 2009; Leusch et al., 2014b) (Houtman et al., 2009)
Parathion
10
AhR-cisFACTORIAL Algae photosynthesis inhibition Acetylcholinesterase inhibition AR-CALUX
Malathion 2,3,7,8Tetrachlorodibenzo-dioxin 2,3,7,8Tetrachlorodibenzo-dioxin Carbaryl Diuron
9
(Sonneveld et al., 2005)
m
n
EBT-BEQ
2.4 106 2.3 105 1.2 106
65 32 18
18 4 1
MCEQ MCEQ MCEQ
ECIR1.5 EC50
1.2 106 1.5 1011
37 5
2 1
N.D. N.D.
TCDD
EC50
1.2 1011
6
0
N.D.
Carb D
ECIR1.5 EC50
1.2 106 1.5 108
17 11
3 7
CarbEQ DEQ
18 0.6
mg/L mg/L
PT
EC50
1.9 107
PTEQ
26
mg/L
10
5
10
3
1
59 N.D. N.D.
mg/L
Testosterone
TT
EC50
9.1 10
Testosterone 17b-Estradiol
TT E2
EC50 EC50
1.6 109 6.5 1011
6 22
4 6
TTEQ EEQ
14 1.8
ng/L ng/L
17b-Estradiol
E2
EC50
6.4 1012
28
7
EEQ
0.2
ng/L
17b-Estradiol
E2
EC50
7.1 1012
16
6
EEQ
0.9
ng/L
17b-Estradiol
E2
EC50
3.2 1010
14
5
EEQ
12
ng/L
17b-Estradiol Progesterone
E2 P4
EC50 EC50
8.2 1012 7.0 1010
8 1
7 0
EEQ P4EQ
0.6 N.D.
ng/L
Progesterone Dexa-methasone
P4* Dex
EC50
7.2 107 3.9 1010
1
1
P4EQ* DexEQ
230 150
mg/L
N.D.
ng/L
N.D. not defined. *Because the conventional reference compound for GR reporter gene assays (dexamethasone) does not have a drinking water guideline value, progesterone (P4, which is biologically active in both PR and GR reporter gene assays) was used for the GR-CALUX.
B.I. Escher et al. / Water Research 81 (2015) 137e148
#
B.I. Escher et al. / Water Research 81 (2015) 137e148
141
2. Material and methods 2.1. Collection of literature data The effect concentrations of chemicals listed in the AGWR for the selected 18 bioassays were collected from the literature. All bioassays included are described in detail in a previous paper (Escher et al., 2014, Table 1). In a few cases, review papers were used but mostly original publications were selected. If more than one EC value was available, the median was used. All median EC values are listed in the Supplementary Data, Table S1 and S2. Additional experiments were performed according to the experimental methods described in Escher et al. (2014). With exception of the FACTORIAL assays, all EC values refer to EC50. The FACTORIAL assays are biosensors with multiplexed library of transcription factors that allow the simultaneous quantification of response of 25 nuclear receptors and 48 transcription factor response elements to chemicals (Martin et al., 2010) or water samples (Escher et al., 2014). Only a few endpoints of the FACTORIAL assays were active when exposed to water samples and we included these in the analysis (Table 1, bioassays # 1,2,4,7). To enable comparison with water samples for which the ECIR1.5 were available, the raw concentration-effect data for FACTORIAL assays given in Martin et al. (2010) were reevaluated using induction ratios (IR) according to Escher et al. (2014), and the effect concentrations causing an IR of 1.5 (i.e. 50% over the control), ECIR1.5, were calculated for better comparability with the water samples' ECIR1.5.
2.2. Bioanalytical equivalent concentrations of water samples Ten water samples had been profiled with 103 bioassays in a previous study (Escher et al., 2014). The water data for 18 of these bioassays were selected for the current study to illustrate the application of the EBT-BEQ approach. Samples were taken at two recycling water plants that use different technologies: Secondary treated effluent from a conventional wastewater treatment plant (Eff-1) was the input to the first plant; samples were taken after membrane filtration (MF), after reverse osmosis (RO) and after advanced oxidation (AO). The treatment train of the second plant was based on oxidation of secondary effluent (Eff-2), and the recycled water was obtained after ozonation and biologically activated carbon filtration (O3/BAC). For comparison samples from a drinking water plant inlet (river) and outlet (DW), as well as stormwater (SW) were tested. The sample “blank” refers to a laboratory ultrapure water sample that underwent the same extraction and enrichment process as the samples. The EC values were previously published (Escher et al., 2014) and are summarized for convenience in the Supplementary Data, Table S3 (reprinted with permission from Escher et al., Benchmarking organic micropollutants in wastewater, recycled water and drinking water with in vitro bioassays. Environ Sci Technol. 48:1940e1956; 2014. Copyright 2014 American Chemical Society). The EC10 and ECIR1.5 values of the water samples have units of relative enrichment factors (REF), which have the net units of Lwater sample/Lbioassay, (Eq. (1)).
REF ¼
water volume equivalent transfered to biosassay total volume of bioassay
(1)
The EC10 and ECIR1.5 values were converted to BEQ using the same reference compounds as used to derive the EBT-BEQ. The BEQ was calculated from the EC of the reference compound (in mol/ Lbioassay) divided by the EC of the sample, which resulted in BEQs with the units mol/Lwater sample (Eq. (2)).
Fig. 2. Approach to derive EBT-BEQs for cell-based bioassays (GV ¼ guideline value, EC ¼ effect concentration, BEQ ¼ bioanalytical equivalent concentration, REP ¼ relative effect potency, n ¼ number of chemicals i, m ¼ number of available EC values per bioassay j, ref ¼ reference compound).
BEQ ¼
EC50 ðreference compoundÞ EC50 ðsampleÞ
(2)
The EC50 of the reference compounds were converted to EC10 values assuming a slope of the log-logistic concentration-effect curve of one, which is a reasonable assumption for most in vitro assay data in our experience. While EC50 values had been used for the EBT-BEQ derivation the experimental samples had low effects and often did not reach the 50% effect level. To avoid extrapolations, the 10% effect level was reported for the water samples in the previous study (Escher et al., 2014). However, in principle, the BEQ are independent on the effect levels they were derived from, provided that the concentration-effect curves of reference compound and samples have the same (or at least a similar) slope (Villeneuve et al., 2000). The minimum detectable BEQ was calculated from the highest tested REF of the water samples. 3. Theory/calculation 3.1. General approach to derive EBT-BEQs For each bioassay j, a separate EBT-BEQ must be derived (Fig. 2). As a first step, effect concentrations were collated from literature for each bioassay and for as many chemicals as possible for which
142
B.I. Escher et al. / Water Research 81 (2015) 137e148
GV are defined (Fig. 2, “Input data”). The GV are not associated with specific modes of action but generally protective of human health. In vitro bioassays quantify one step in the toxicity pathway but not necessarily the one that is most toxicologically relevant for a given chemical. The effects observed in the in vitro bioassays thus often do not match the GV. Therefore, as the second step, it is important to filter the available number of effect data points (m) to reduce to the number of chemicals i (n) for which the bioassays are indicative of the corresponding GVi (Fig. 2, “Filtering”). EC values were excluded if they were smaller or larger than the GV by an order of magnitude, because if they were included the BEQ distribution would be skewed (as discussed in detail in the next sections below). In the next step, the relative effect potencies REPi were calculated in each bioassay j for each chemical I (Fig. 2, “Bioanalytical equivalents”). Each GVi was then converted to the corresponding BEQi by multiplication with the corresponding REPi. Finally, a cumulative distribution of BEQi was plotted and the EBT-BEQ derived from the fifth percentile of this distribution (Fig. 2, “Cumulative distribution”). The 5th percentile was selected as this will be protective for the majority of chemicals and will account for the mixture effects that chemicals may have in the given bioassays.
3.2. Filtering: selection of appropriate bioassay data As the cell-based bioassays are used as analytical tools, they must be sufficiently responsive to the chemicals selected from the guideline for the derivation of EBT-BEQ. The ECi of the m chemicals, for which EC values were available in the literature, were plotted i against the GVi (Fig. 2). Those chemicals outside a band of 0.1 GV ECi 10 were excluded from the analysis, resulting in n chemicals going forward. This band was chosen to ensure that only chemicals that target the mode of action or receptor of relevance in the bioassay were included. The point of departure for the derivation of the GVi does not necessarily need to be consistent with the endpoint of the bioassay but with the filtering step, we ensure that only those chemicals that are responsive in the given bioassay are selected. If GVi < 0.1 ECi, the bioassay is not sensitive enough to capture the effects caused at the GV-concentrations in the bioassay. While there are EC available, these chemicals are not potent enough to trigger the target mode of action of the bioassays at the GVi concentrations and are thus not suitable for inclusion into the derivation of that specific EBT-BEQ. If GVi > 10 ECi then the bioassay would be sensitive enough to capture the effect at the GV-concentrations in the given bioassay but chemical i may dominate the overall effect. This could be due to the chemical being more potent in vitro than in vivo or the GVi being based on a different mode of action. Therefore these values were also excluded to assure a fairly symmetric distribution of GVi/ECi ratios around 1, which also resulted in a more robust BEQi distribution. The filtering step has the advantage that it recognizes multiple modes of action of a chemical and accounts for a range of intrinsic potencies. The disadvantage is that it is possible that highly potent chemicals, even well-known reference chemicals, can be excluded in the filtering step if they have no relationship to the point of departure for the GVi.
REPi ¼
ECref ECi
(3)
Typically used reference chemicals for bioassays were only used if these reference chemicals were also included in the selected guideline. Where these were not included, instead a potent (but not necessarily the most potent) chemical was used as the reference chemical. If several bioassays were available for the one endpoint, a consensus reference chemical was selected that had available EC values in all bioassays for this endpoint. The BEQi for each chemical i were then calculated by multiplying the GVi with the REPi (Eq. (4)).
BEQ i ¼ REPi GVi
(4)
The choice of the reference compound is crucial. In relation to bioassay performance a reference compound needs to fulfill a number of performance criteria such as selectivity, specificity and have a clear mode of action, while as a trigger value it should also have chemical and biological relevance. 3.4. Distribution of BEQ and derivation of EBT-BEQ Biological data are typically log-normally distributed (Limpert et al., 2001) and therefore cumulative distributions of the log BEQi were used to derive the EBT-BEQ. The distributions were calculated according to the US EPA SSD generator (http://www.epa. gov/caddis/da_software_ssdmacro.html). The calculated bioanalytical equivalents BEQi were ranked and then the rank converted to proportions with Eq. (5). The logBEQi were plotted against the proportioni (Eq. (6)). The 5th percentile (Eq. (7)) was defined as EBT-BEQ and was calculated from a linear regression through zero.
proportioni ¼
rank 0:5 n
propotioni ¼ slope logBEQ i logEBT BEQ ¼
0:05 slope
(5) (6) (7)
In an ideal world, if the given bioassay was fully representative of the mode of action for which the health-based GV was derived, the BEQi would be equal for all chemicals i. This is not the case because the bioassay endpoints do not directly match the ultimate health effects underlying the derivation of GV even if they are related after the filtering step in Section 3.2. In reality, as will be discussed in the results section, the BEQi varied by a factor 10 to 100 between different chemicals for a given bioassay. As a realistic measure of the mixture effect one could use the 50th percentile (or median) of the distribution of EBT-BEQ. However, in a precautionary approach and to account for the possibility that in a realistic mixture chemicals with a higher potency contribute more to the mixture than low-potency chemicals, we chose the lower 5th percentile as EBT-BEQ. This aligns with the HC5 concept of species sensitivity distributions where HC5 stands for hazardous concentration to 5% of the species (Posthuma et al., 2002). 4. Results and discussion
3.3. Calculation of bioanalytical equivalent concentrations
4.1. Development of EBT-BEQs
Each ECi was converted to relative effect potencies REPi in relation to a reference compound with ECref (Eq. (3)).
Of the 103 bioassays applied to the ten water samples, only 65 bioassays targeted receptor-mediated effects. The other 38
B.I. Escher et al. / Water Research 81 (2015) 137e148
143
Fig. 3. Steps in the derivation of EBT-BEQ on the example of the E-SCREEN assays. A. All available experimental EC50 values; B. EC50 values after the filtering step; C. Cumulative distribution of estradiol equivalent concentrations (EEQi).
bioassays targeted reactive MOAs, adaptive stress response pathways and cytotoxicity, for which no EBT-BEQs can be derived, although EBT-ECs have been derived (Escher et al., 2013; Tang et al., 2013). For 18 out of the 65 receptor-mediated bioassays, we found experimental data for chemicals included in the AGWR (Table 1 and individual EC-values in Supplementary Data, Tables S1 and S2). Between one and 65 EC values were found per bioassay (median ¼ 13). In particular the assays that were included in ToxCast screening of large datasets had many matching data, e.g., 65 EC values were available for the PXR-cisFACTORIAL assay (Martin et al., 2010). The process of derivation of the EBT-BEQ is illustrated on the example of the E-SCREEN assay in Fig. 3 and depicted for all 18 bioassays in the Supplementary Data, Figure S1. 4.1.1. Filtering: selection of appropriate bioassay data After the filtering step, only 0 to 18 (median ¼ 5) EC values remained (Table 1). Large numbers of chemicals were removed by filtering for the bioassays indicative of metabolism (Table 1, Supplementary Data, Figure S1): for the pregnane X receptor (PXR), 65, 32 and 18 EC values were available in the PXRcisFACTORIAL, PXRtransFACTORIAL and HG5LN PXR assays, respectively. As the majority of assays were not sensitive to these chemicals in relation to the GV, only 18, 4 and 1 chemicals, respectively, remained after filtering. This is consistent with the general observation that PXR is a highly promiscuous receptor (Ihunnah et al., 2011). A similar observation was made for the peroxisome proliferatoractivated receptor gamma (PPARg), where only 2 out of 39 chemicals were within one order of magnitude of the GV. Thus no EBTBEQ could be derived for PPARg. The aryl hydrocarbon receptor (AhR) is mainly activated by dioxin-like chemicals but also by many more water-soluble chemicals with typically lower affinity (Denison and Nagy, 2003), but most chemicals in the AGWR had too low GVi to be included in the BEQ derivation. A different picture was observed for the algal photosynthesis inhibition test. This bioassay was included despite the fact that it has no direct relevance to human health, because eleven herbicides are included in the guideline. As we suggest using bioassays as bioanalytical indicators for the presence of mixtures of chemicals it is legitimate to apply a sensitive indicator bioassay as (bio)analytical tool even if it is not clearly related to a human health endpoint. Since the GV was derived for human health but the alga test targets photosynthesis, the bioassay was often more sensitive than the GV. Non-herbicides are also active in this bioassay but at low potency, therefore they were omitted upfront (Tang and Escher, 2014).
Filtering removed only 4 chemicals. For acetylcholinesterase inhibition, a similar observation as for herbicidal activity was made. Evidently this endpoint was not the basis of the derivation of the GV, therefore 5 out of 10 EC50 values had to be removed either as a result of too high or too low potency. For the endocrine active compounds, the available EC50 dataset had to be reduced during the filtering step but not to an as great extent as for the xenobiotic metabolism endpoints. For example, in the E-SCREEN assay (Fig. 3), 16 available EC50 values were reduced to 6 after filtering. Even 17b-estradiol had to be excluded because it fell into the category of GVi > 10 ECi. Most likely, this is because the GV of 17b-estradiol in the AGWR is based on its carcinogenicity not on estrogenic effects. However, 17b-estradiol was still used as reference compound to derived 17b-estradiol equivalents (EEQ). After filtering, no EC50 value remained for H4IIEluc and PRCALUX and those bioassays were not further considered. This does not mean that the endpoints of AhR activation and progesterone receptor response were considered to be irrelevant, only that the current potency dataset is insufficient to derive an EBT-BEQ value. 4.1.2. Distribution of BEQ and derivation of EBT-BEQ The lack of available data meant that EBT-BEQ could not be calculated for 7 bioassays. For the remaining 11 bioassays, linear regressions of high quality were obtained after plotting the log BEQ versus the cumulative proportions (Fig. 3 and Supplementary Data, Figure S1 and Table S4). The high correlation coefficients indicate that the BEQs were indeed log-normally distributed as had been anticipated from theoretical considerations. The EBT-BEQs were derived from the lower fifth percentile of the distribution and are listed in Table 1. 4.2. Modes of action 4.2.1. Xenobiotic metabolism 4.2.1.1. Pregnane-X receptor (PXR). Metolachlor (MC) served as the reference chemical for induction of the PXR. It was chosen as the one compound for which EC50 values were available in all three assays. Due to limited availability of experimental data a valid EBTMCEQ of 59 mg/L could only be derived for PXRcisFACTORIAL (Supplementary Data, Table S4 and Table 1). 4.2.1.2. Peroxisome proliferator-activated receptor (PPARg). For PPARg, malathion (MT) was chosen as reference compound because it was the chemical closest to the 1:1 line in the filtering plot. This is certainly not the prototype PPARg activator and in the
144
B.I. Escher et al. / Water Research 81 (2015) 137e148
future, fingerprinting of larger lists of chemicals should come up with a potent reference compound that is also included in the guideline. We recognize the relevance of this endpoint but conclude that at present it is premature to define an EBT for the PPARg-transFACTORIAL assay. 4.2.1.3. Aryl hydrocarbon receptor (AhR). After filtering only one data point remained for the AhR-CAFLUX assay (4-nonylphenol, Table 1). This value could theoretically be converted to 2,3,7,8-tetrachlorodibenzodioxin equivalents TCDDEQ, and used as tentative EBT-TCCDEQ. Clearly, this is unsatisfactory and more data on watersoluble compounds would need to be generated to obtain a reliable EBT-TCCDEQ. Data on activation of the AhR by waterborne chemicals in the AhR-CAFLUX and H4IIEluc remained too scarce to be able to derive an EBT at this stage. Further research on ligands relevant in the aqueous environment would be required before an EBT can be derived. However, the AhR-cisFACTORIAL was tested for a large number of pesticides (Martin et al., 2010), 17 of which overlapped with our guideline. After the filtering step only three chemicals remained. The most potent of the three, carbaryl (Carb), was chosen as reference compound and the EBT-CarbEQ was 18 mg/L, which is a preliminary EBT due to the low input data quality. 4.2.2. Specific receptor-mediated modes of action 4.2.2.1. Herbicidal activity. As discussed above, it is difficult to compare drinking water GVs, which are derived for human health protection with specific assays such as these based on photosynthetic organisms, algae. While it would of course be preposterous to use inhibition of photosynthesis as a relevant endpoint purely based on human health considerations, our premise was to apply bioassays as analytical tools to assess the mixture effects of chemicals relevant for water pollution. This approach renders this assay relevant as herbicides constitute a major threat to drinking water sources and are included in drinking and recycled water guidelines. We have previously evaluated the activity of herbicides and non-herbicides on photosynthesis inhibition and concluded that the mixture effects in realistic water samples are dominated by the herbicides (Tang and Escher, 2014). Therefore we included only the 11 herbicides from the AGWR in our analysis. While in most bioassays filtering removed chemicals that had too low potency in the bioassays, here the number of chemicals included was reduced from 11 to 7 by removing four herbicides whose GV was too high in relation to their activity in the bioassay. The remaining seven herbicides yielded a high quality distribution (Supplementary Data, Table S4). With the reference compound diuron (D) (EC50 15 nM, Supplementary Data, Table S1) a
Fig. 4. Comparison of BEQ distributions for the estrogenicity bioassays.
provisional EBT-DEQ of 0.6 mg/L was obtained (Table 1). 4.2.2.2. Acetylcholinesterase (AChE) inhibition. Many insecticides are inhibitors of the AChE but humans are not necessarily sensitive due to extensive detoxification, and the human health GV is therefore not necessarily closely related to the mode of action of AChE inhibition. Consequently only a small set of five insecticides could be included in the BEQ distribution. The EC50 of the reference compound parathion (PT) was 0.2 mM (Table 1) and the provisional EBT-PTEQ resulting from a high-quality linear regression was 26 mg/ L (Supplementary Data, Table S4 and Table 1). 4.2.3. Endocrine disruption 4.2.3.1. Androgenic activity. While the experimental database of the AR-CALUX assays was too small to derive a BEQ distribution, four chemicals from the AGWR remained after filtering in the ARGeneBLAzer assay allowing the calculation of an EBT-BEQ for that assay. The EC50 of the reference compound testosterone (TT) was 1.6 nM (Table 1) and the EBT-TTEQ was 14 ng/L (Supplementary Data, Table S4 and Table 1). 4.2.3.2. Estrogenic activity. Five bioassays for estrogenicity were included. The EC50 of the reference compound 17b-estradiol ranged over two orders of magnitude from 3 to 7 pM for ER-CALUX, ESCREEN and hERa-HeLa-9903 over 65 pM for ERa-GeneBLAzer to 320 pM for the YES assay (Table 1). The EEQ-distributions were comparable between ER-CALUX, E-SCREEN, hERa-HeLa-9903 and ERa-GeneBLAzer (Fig. 4 and Supplementary Data Table S4). The BEQ distributions for the YES assay were at the highest concentrations, resulting in an EBT-EEQ of 12 ng/L, which was almost an order of magnitude higher than the EBT-EEQ of the four other estrogenic assays, which ranged from 0.2 to 1.8 ng/L. As different numbers and types of chemicals were available, the chemicals included in the distribution varied, so that a direct comparison is limited. To assess the true differences in the BEQ distributions one would have had to measure EC values of the same chemicals. The comparison we have done is only valid if the experimental EC represented random samples out of the true distribution and this is less likely the fewer the number of EC values are available. In the ER-CALUX, we obtained an EBT-EEQ of 0.2 ng/L. For comparison, the independent derivation based on an entirely different rationale in Brand et al. (2013) resulted in a trigger value of 3.8 ng/L EEQ. A difference of a factor of 20 appears large but one needs to account for the fact that entirely different assumptions are underlying the two methods for EBT derivation. Safe concentration of estrogenic equivalents in wastewater determined from environmental predicted no-effect concentration and extrapolated to cell-based bioassays ranged from 0.2 to 0.4 ng/L EEQ for ER-CALUX, 0.1e0.3 ng/L for E-SCREEN and 0.1e0.2 ng/L EEQ for the YES (Jarosova et al., 2014), i.e., they cover very similar concentration ranges as the EBT-EEQs derived here. The protection target for these “safe concentration of estrogenic equivalents” is fish reproduction but since indirect potable reuse schemes rely on residence in an environmental buffer such as a reservoir or lake prior to reuse, it is interesting to note that the proposed EBT-EEQ proposed here would also be protective of ecological endpoints. 4.2.3.3. Progesterone and glucocorticoid receptor activity. Typically dexamethasone (Dex) is used as reference compound for glucocorticoid activity (Schriks et al., 2013). Since this chemical was not included in the AGWR, we used progesterone (P4, EC50 0.7 mM, Table 1) as the reference compound, as it is active in both the PR and GR reporter gene assays (Houtman et al., 2009). No distribution could be made because this was also the only EC50 available that
B.I. Escher et al. / Water Research 81 (2015) 137e148
145
Table 2 Summary of BEQs of the bioassays in relation to EBT-BEQs. Definition of sample abbreviation in Section 2.3. The abbreviations of the BEQ for the different bioassays are given in Table 1. The recycled water samples are in bold font. #
1 7 8 9 11 12 13 14 15 16 18
Abbreviation
PXR-cisFACTORIAL AhR-cisFACTORIAL Algae photosynthesis inhibition Acetylcholinesterase Inhibition AR-GeneBLAzer ER-GeneBLAzer ER-CALUX E-SCREEN YES hERa-HeLa-9903 GR-CALUX
BEQ
Units
EBT-BEQ
jddRecycling plant 1ddj
jddRecycling plant 2dj
jddReference samplesddj
Eff-1
MF
RO
AO
Eff-2
O3/BAC
River
DW
SW
Blank
2000 17 0.2
1900 19 0.2
