Pesticide mixtures in the Swedish streams: Environmental risks, contributions of individual compounds and consequences of single-substance oriented risk mitigation

Highlights

• Detected pesticide mixtures regularly exceed the Swedish threshold concentrations.

• Analytical detection-limits for insecticides often exceeds concentration thresholds.

• 83 of 141 pesticides are needed to explain 95% of estimated risk across all samples.

• No change in risk from pesticide mixtures can be seen between 2002 and 2013.

• Single substance risk mitigation will not lower mixture risks to acceptable levels.

Abstract

This paper presents the ecotoxicological assessment and environmental risk evaluation of complex pesticide mixtures occurring in freshwater ecosystems in southern Sweden. The evaluation is based on exposure data collected between 2002 and 2013 by the Swedish pesticide monitoring program and includes 1308 individual samples, detecting mixtures of up to 53 pesticides (modal = 8). Pesticide mixture risks were evaluated using three different scenarios for non-detects (best-case, worst-case and using the Kaplan-Meier method). The risk of each scenario was analyzed using Swedish Water Quality Objectives (WQO) and trophic-level specific environmental thresholds.

Using the Kaplan-Meier method the environmental risk of 73% of the samples exceeded acceptable levels, based on an assessment using Concentration-Addition and WQOs for the individual pesticides. Algae were the most sensitive organism group. However, analytical detection limits, especially for insecticides, were insufficient to analyze concentrations at or near their WQO's. Thus, the risk of the analyzed pesticide mixtures to crustaceans and fish is systematically underestimated. Treating non-detects as being present at their individual limit of detection increased the estimated risk by a factor 100 or more, compared to the best-case or the Kaplan-Meier scenario.

Pesticide mixture risks are often driven by only 1–3 compounds. However, the risk-drivers (i.e., individual pesticides explaining the largest share of potential effects) differ substantially between sites and samples, and 83 of the 141 monitored pesticides need to be included in the assessment to account for 95% of the risk at all sites and years.

Single-substance oriented risk mitigation measures that would ensure that each individual pesticide is present at a maximum of 95% of its individual WQO, would also reduce the mixture risk, but only from a median risk quotient of 2.1 to a median risk quotient of 1.8. Also, acceptable total risk levels would still be exceeded in more than 70% of the samples.

Introduction

Multiple studies have demonstrated that complex pesticide mixtures are present in surface waters globally, in the US (e.g. Gilliom, 2001, Stone et al., 2014a, Stone et al., 2014b), Europe (e.g. Moschet et al., 2014, Schreiner et al., 2016, Ccanccapa et al., 2016) and elsewhere (e.g. in South America (Hunt et al., 2016), Australia (Allinson et al., 2015) and China (Zhang et al., 2011). Empirical evidence univocally shows that the combined toxic effects of such pesticide mixtures exceed the effect of each individual compound (e.g. Faust et al., 2001, Faust et al., 2003, Knauert et al., 2009 and Porsbring et al., 2010, see also reviews by Belden et al., 2007, Verbruggen and van den Brink, 2010 and Rodney et al., 2013).

Studies have repeatedly demonstrated that Concentration Addition (CA) describes the joint toxicity of pesticide mixtures well (reviewed by Belden et al., 2007, Rodney et al., 2013). This implies that all components contribute to the overall mixture toxicity, independently of whether they are present at concentrations above or below their individual No Observed Effect Concentration (NOEC) or Environmental Quality Standard (EQS). Mixtures might therefore cause toxic effects even if all components are present at concentrations below which an individual effect is detectable (e.g. Carvalho et al., 2014, Faust et al., 2001).Taken together, the available body of evidence thus clearly shows that pesticide mixtures warrant specific consideration during environmental risk assessment, monitoring and management.

Environmental risks of pesticides and pesticide mixtures are assessed in the European regulatory system from two perspectives. First, active ingredients and whole formulated pesticide products are evaluated for their environmental hazard, exposure and risk during market authorization (EFSA, 2013), following the legal frameworks that are laid down in Regulations EC 1107/2009 on the placing of plant protection products on the market (European Parliament, 2009a) and EC 546/2011 on uniform principles (European Commission, 2011a). However, ‘coincidental’ pesticide mixtures, i.e. mixtures of active ingredients that result from farmers applying different pesticide products in close proximity to a given water body or because of sequential spraying of different pesticides on the same field, are not considered in Regulation EC 1107/2009 nor in Directive 2009/128/EC (European Parliament, 2009b). However, it has been argued that the uniform principles in Regulation 546/2011/EC require authorization of plant protection products to be based on the “proposed conditions for use” and consequently – given common agricultural practice – to consider the environmental impact of the resulting pesticide mixtures (Frische et al., 2014).

Second, the impact of mixtures of pesticides (and other hazardous chemicals) on the ecological status of an aquatic system is assessed from the perspective of the Water Framework Directive (WFD) (European Parliament, 2000). In order to be classified as having a good ecological status, a water body also needs to have good chemical status, which requires that the concentrations of each of 45 priority pollutants, which are currently listed in Directive 2013/39/EC (European Parliament, 2013), do not exceed European-wide thresholds, so-called Environmental Quality Standards (EQS). In addition, in order to track progress towards the national goal of a “non-toxic environment” (adopted in 1999), Sweden also developed national Water Quality Objective(s) (WQO) for pesticides, defined as concentrations which are not expected to cause any adverse effects in the aquatic environment (Norberg, 2004, Lindström and Kreuger, 2015). These values are similar to EQS values and serve as a tool to evaluate surface water quality based on monitoring results, but are not legally binding. WQO's are derived using a method that closely follows the REACH approach for deriving Predicted No Effect Concentrations (PNEC) values, based on single species data and assessment factors between 10 and 1000, depending on the underlying ecotoxicological endpoints (Andersson et al., 2009, Andersson and Kreuger, 2011, KEMI, 2008).

Risk assessment of chemical mixtures is routinely performed using CA (Kortenkamp et al., 2009, Bopp et al., 2015). CA has also been suggested specifically for the assessment of pesticide mixtures (EFSA, 2013) and it is the recommended approach for setting EQS values for chemical mixtures within the context of the WFD (European Commission, 2011b).

According to CA the risk quotient (RQ) of a mixture, RQ_CA, is defined as:

$${\mathrm{RQ}}_{\mathrm{CA}}=\frac{{\mathrm{c}}_{\mathrm{mix}}}{{\mathrm{ECx}}_{\mathrm{mix}}}={\sum }_{i=1}^n=\frac{{\mathrm{c}}_{\mathrm{i}}}{{\mathrm{ECx}}_{\mathrm{i}}}={\sum }_{i=1}^n{\mathit{TU}}_i$$

where c_mix is the total concentration of the mixture, ECx_Mix is the mixture concentration causing x% effect, while c_i and ECx_i denote the corresponding concentrations of substance i. The ratio c_i/ECx_i provides a dimensionless measure of the toxicity contribution of compound i usually termed a Toxic Unit (TU). Although the concept is rooted in the idea of the mixture components sharing the same mode of action, as well as not taking possible synergistic (or antagonistic) effects into account (Cedergreen, 2014), CA has been successfully used for the risk assessment of heterogeneous mixtures (Belden et al., 2007, Kortenkamp et al., 2009, Verbruggen and van den Brink, 2010, Rodney et al., 2013, Bopp et al., 2015). The toxicity estimates in Eq. (1) (ECx_mix and ECx_i) in principle refer to the same ecotoxicological endpoint recorded for the same species under identical exposure conditions. However, in practice CA is often applied in a broader setting, e.g. by using data from different algal species in order to predict the toxicity to algae in general.

In the present paper, we have applied CA in order to separately calculate the risks for algae, crustaceans and fish. The corresponding CA-based mixture RQs are termed RQ_Algae, RQ_Crust and RQ_Fish. Moreover, by substituting the ECx_i with the WQO_i and c_i with the Measured Environmental Concentration (MEC_i) we determined ecosystem-wide RQ_WQO values as the sum of the individual MEC/WQO ratios, following the rational outlined by (Backhaus & Faust,

$${\mathrm{RQ}}_{\mathrm{WQO}}={\sum }_{i=1}^n\frac{{\mathrm{MEC}}_{\mathrm{i}}}{{\mathrm{WQO}}_{\mathrm{i}}}={\sum }_{i=1}^n\frac{{\mathrm{MEC}}_{\mathrm{i}}}{min\left({\text{Toxicity data}}_{\mathrm{i}}\right){{{}^{\ast }\mathrm{AF}}_{\mathrm{i}}}^{-}}$$

Comparing the trophic-level specific RQs with RQ_WQO is difficult, as the latter is calculated using assessment factors to account for the different amounts of data available for each compound, while RQ_Algae, RQ_Crust and RQ_Fish are calculated without using any assessment factors. In order to bridge these two approaches, we therefore also calculated a mixture RQ for the most sensitive trophic level (RQ_MST), defined as:

$${\mathrm{RQ}}_{\mathrm{MST}}=\sum _{\mathrm{i}=1}^{\mathrm{n}}\frac{{\mathrm{MEC}}_{\mathrm{i}}}{min\left({\mathrm{EC}50}_{\text{Algae}}{,\mathrm{EC}50}_{\text{Crustaceens}}{,\mathrm{EC}50}_{\text{Fish}}\right)}$$

RQ_MST provides a measure for the risk across trophic levels, but is calculated without using any assessment factors. It thus takes an interim position and bridges the trophic-level specific RQs (RQ_Algae, RQ_Crust or RQ_Fish) to the ecosystem-wide RQ_WQO. The RQ_MST is conceptually identical to the point of departure index (PODI), frequently used in human toxicology (Wilkinson et al., 2000).

A RQ provides a yardstick for assessing the need to act. Values of RQ_WQO exceeding 1 indicate the need for either a more advanced mixture risk assessment, and/or for the implementation of risk mitigation measures. We defined the corresponding critical values for RQ_Algae, RQ_Crust, RQ_Fish as 0.1, 0.01 and 0.01, respectively, following the assessment strategy for individual pesticides (EFSA, 2013). Defining a critical value for RQ_MST is not feasible at the moment, as no strategy has been suggested yet on how an overall assessment factor should be calculated that reflects the overall uncertainty in Eq. (2). The RQ_MST will always be higher than any of the organism-group specific RQs (Backhaus and Faust, 2012) and, because no assessment factors are applied, lower than the RQ_WQO.

The ratio between the total RQ of a mixture and the maximum RQ of its components has been termed the maximum cumulative ratio (MCR, Price and Han, 2011). That is,

$$\mathrm{MCR}=\frac{{\mathrm{RQ}}_{\mathrm{CA}}}{\underset{\mathrm{i}=1\dots \mathrm{n}}{max\left({\mathrm{RQ}}_{\mathrm{i}}\right)}}$$

If all components of a mixture are contributing equally to the predicted mixture risk, the MCR equals the number of compounds in the mixture. In a mixture whose TU distribution is dominated by one compound, the MCR approaches 1. Therefore, the MCR has been suggested as a tool to assess the value of performing mixture toxicity assessments (Price and Han, 2011).

Chemical risk assessment is in general based on comparing relevant exposure estimates (measured or modeled) with hazard estimates, such as NOEC's, EC50’s and EQS values. Such estimates are straightforward to calculate on the basis of monitoring results, as long as detected environmental concentrations are quantified, either above the chemical-analytical limit of quantification (LOQ) or the limit of detection (LOD). However, sometimes when the detection is below the LOQ but still above the LOD the concentration is not quantified (only given as ‘trace’) in order to save time in the laboratory. Nevertheless, reasonable assumptions on the trace concentrations present can be made using (LOQ + LOD) / 2 as a surrogate for unquantified detections between the LOQ and the LOD, as long as these two parameters are stated in the analytical protocol.

However, the situation becomes problematic if a monitored chemical is not detected. Such a result does not prove that the compound is not present, it only shows that the concentration is somewhere between zero and the LOD. Assuming a zero concentration for all non-detects will therefore underestimate the total risk, if no additional knowledge about e.g. emission or use pattern is available.

On the other hand, assuming that all non-detected compounds are present just below their LOD – the worst-case scenario that is still compatible with the recorded values – is also unrealistic. Such an approach immediately leads to the logical inconsistency that the estimated risk becomes simply dependent on the number of compounds analyzed. The same is true for setting the concentration used for the risk assessment a priori to any other value above zero.

Parametric and non-parametric statistical methods are available for data with “less-than” values, i.e. findings of concentrations < LOD. They allow the estimation of the likely contribution of non-detects to the total RQ. In this paper we used the non-parametric Kaplan-Meier (KM) method (Helsel, 2010, Helsel, 2012, Bolks et al., 2014), because it is not possible to ensure that the distributional assumptions of parametric alternatives are fulfilled in the analyzed data. A KM-adjusted sum of RQs lies between the sum of RQs that result from substituting all non-detects with their respective LOD and the sum of RQ that results from substituting all non-detects with zero.

The KM-method ignores the potential risk contribution of a compound, if its potential RQ exceeds the maximum of the RQs that are based on a quantified concentration. For such compounds, better analytical data are required for a reliable quantification of their risk contribution.

The southern part of Sweden is an area of intense agricultural activity and pesticide residues have been systematically monitored at six sites since 2002 (Lindström et al., 2015, Lindström and Kreuger, 2015). In this paper, we applied CA-based risk assessment approaches in order to estimate and characterize the environmental risks from the detected pesticide mixtures, using RQ_Algae, RQ_Crust, RQ_Fish, RQ_MST and RQ_WQO. The results will then be used for a broader discussion on the impact of non-detects on component-based mixture risk assessments. Finally, we explore the consequences of a single-substance oriented risk management, i.e. assuming that risk mitigation measures ensure that all individual concentrations are below their corresponding WQO's.

In order to explore how the different possibilities to incorporate (or ignore) concentrations below the LOD influence the final mixture risk estimates, we calculated all RQs for three different exposure scenarios (Table 1). Scenario 1 and 2 assumes that non-detects are present at a concentration equal to their LOD or at zero, respectively. Scenario 3 uses the KM-adjustment for compounds present < LOD.