Monitoring pesticides in surface water and sediment porewater

Faculty of Science

Institution of Biodiversity and Ecosystem Dynamics

Monitoring pesticides in surface water and sediment

porewater

- An assessment of the Continuous Low-Level Aquatic Monitoring sampler and

the Silicone Rubber

sampler-Author: R.W. Prijden Supervisor: dr S.T.J. Droge Examinor: dr M.H. Kraak

A thesis submitted for the degree of MSc Earth Sciences

Acknowledgments

I would like to express my heartiest gratitude and sincere thanks to dr. Steven Droge for providing me valuable guidance and helpful feedback as a supervisor throughout my research. Thanks are due to him providing me with all equipment and knowledge, but most of all for the patience throughout the research process and for the time spent helping me with sample analysis and accompanying me during fieldwork.

Besides my supervisor, I would like to thank dr. Michiel Kraak, Associate Professor of Aquatic Ecotoxicology, Institute of Biodiversity and Ecosystem Dynamics, University of Amsterdam, for providing very useful feedback on presentations and written reports and for attending my presentations.

My sincere thanks also go out to Milo de Baat and Nienke Wieringa for answering my questions and guiding me in the laboratory.

Last but not least, I would like to thank my family and friends for all the unconditional love, support, and encouragement. And to my dearest, Björn, thanks for always being there. You have been incredibly patient and even joined me on fieldwork even though you still don’t want to know the subject of my thesis.

Table of content

Acknowledgments...2

List of figures, tables and appendices...5

Abstract...6

Chapter 1. Introduction...7

1.1 General introduction...7

1.2 Background...7

1.2.1 Pesticides...7

1.2.2 Environmental fate of pesticides...8

1.2.3 Grab sampling versus active/passive sampling...9

1.2.4 Continuous Low-Level Aquatic Monitoring (CLAM)...10

1.2.5 Silicone rubber as a passive sampler...10

1.2.6 Bioassessment with Daphnia magna...11

Chapter 2. Research objectives...12

2.1 Aim...12

2.2 Research questions...12

2.3 Hypothesis...13

Chapter 3. Materials and methods...14

3.1 Material description...14

3.1.1 CLAM...14

3.1.2 SR...14

3.2 Method overview...14

3.3 Methods for the laboratory study...15

3.3.1 Pesticide quantification with LC-MS/MS...15

3.3.2 Methods CLAM...16

3.3.3 Methods SR...18

3.4 Methods of the field study...20

3.4.1 Selection of sample locations...20

3.4.2 Sampling surface water with the CLAM...20

3.4.3 Sampling porewater with SR...20

3.4.4 Methods bioassay with Daphnia magna...21

Chapter 4. Results...22

4.1 Results laboratory study...22

4.1.1 Results CLAM...22

4.1.2 Results SR...27

4.2 Results field study...32

4.2.1 Results pumping efficiency...32

4.2.2 Sampling with the CLAM...32

4.2.3 Sampling with SR...33

4.2.4 Bioassay with Daphnia magna...34

Chapter 5. Discussion...36 5.1 Laboratory study...36 5.1.1 CLAM...36 5.1.2 SR...37 5.2 Field study...38 5.2.1 CLAM...38 5.2.2 SR...38

5.2.3 Pesticides detected in the field...39

5.2.4 Bioassessment with Daphnia magna...41

Chapter 6. Conclusions...43

Literature...44

Appendix...50

Appendix 1: Detailed pesticide list...50

Appendix 2: Field forms...61

List of figures, tables and appendices

Figure 1 Conceptual visualization of the environmental fate of pesticides in aquatic ecosystems...9

Figure 2 Continuous Low-Level Aquatic Monitoring Sampler...14

Figure 3 Altec AlteSil Industrial Sheet...14

Figure 4 Schematic overview of the experiments performed in this research...15

Figure 5 Pumping rate of the CLAM sampler...22

Figure 6 Aquarium water concentration during CLAM experiments...23

Figure 7 Amount of analyte remaining on the HLB1 filter after a second extraction...23

Figure 8 Amount of analyte retained by the glass pre-filter, expressed as the percentage retained by HLB1...24

Figure 9 Breakthrough on HLB2, expressed as the percentage retained by HLB1...25

Figure 10 Aquarium water concentration of all pesticides used in the CLAM experiment (pesticide mix 1)...26

Figure 11 Accuracy of HLB1, expressed as the percentage of the measured water concentration (CW)...27

Figure 12 Aquarium water concentration during SR experiments...28

Figure 13 Non-linear fit of the ratio between CSR and CW...29

Figure 14 Kinetic uptake parameters KSRw and ke...29

Figure 15 Calculated logKSRw of neutral pesticides in relation to the logKow...30

Figure 16 Results of non-linear regression analysis between logKow and logKSRw...31

Figure 17 Results of the non-linear regression analysis between logKow and logKSRw...32

Figure 18 Pumping efficiency of the CLAM sampler...32

Figure 19 Daphnia magna immobility...35

Table 1 Pesticides detected at the field locations in 2017...20

Table 2 Sampling period, number of counts and estimated sampling volume per field location...33

Table 3 Pesticides detected at the field locations with the CLAM...33

Table 4 Pesticides detected at the field locations with the SR...34

Table 5 Pesticides detected at the field locations in 2018...40

Table 6 Pesticides detected at the field locations between 2014-2016...41

Appendix 1 Detailed list of all pesticides targeted in this study...50

Appendix 2 Field form per location...61

Appendix 3 List of pesticide mixture 1...65

Abstract

This study aimed to evaluate the applicability of two relatively new sampling methods, active Continuous Low-Level Aquatic Monitoring (CLAM) and passive silicone rubber sampling, for respectively monitoring pesticides in surface water and sediment porewater. This was performed by investigating the efficiency of both samplers in a controlled laboratory setting and in the field. Subsequently, a toxicity test will be done to determine how well both samplers can estimate the toxic risk of the detected pesticides. First, a method was developed for the analysis of 92 pesticides in surface water and sediment porewater by respectively active and passive sampling and by liquid-chromatography/mass spectronomy. Method efficiency for the CLAM was determined in a controlled aquarium study by comparing the concentration measured by the samplers and the actual spiked water concentration. Also, the accuracy was assessed by determining the breakthrough of the filters, retention to the supposedly inert pre-filter and the efficiency of the applied extraction method. Based on the amount of pesticides retained on the sorbent filter and the recorded passing volume of water, it was possible to recover the pesticide concentration and to subsequently compare the concentration measured by the sampler and the actual spiked pesticide concentration. Method efficiency for the SR was determined in a controlled aquarium study by first determining the uptake kinetics for 68 neutral pesticides, and by subsequently fitting a linear predictive model. A particular focus was on the efficiency of both samplers to detect acidic pesticides. Second, the samplers that were applied in surface water and sediment from four field sites in the proximity of different agricultural uses were analyzed. Twenty-two pesticides were detected above detection limits, with concentrations ranging from 0.00398 ng L-1 _{to 0.261 µg L}-1_{. Third, the samplers’ extracts were used in} bioassays with Daphnia magna to test for potential toxicity of the studied environments. Although some samples showed significantly lower survival of the daphnids during 48 h exposure, this could not be explained by detected pesticides in the extracts.

Chapter 1. Introduction

1.1 General introduction

Monitoring toxic substances in water bodies is a major task of water quality assessment (Brack et al., 2019). The Water Directive Framework (WFD) aims to protect the chemical status of both surface-and groundwater bodies across Europe. The chemical status of a water body is based on the absence and presence of 45 Priority Substances and various sets of River Basin-Specific Pollutants (Brack et al., 2019). However, toxic events can often not be related to these substances (de Baat et al., 2018). Aquatic monitoring of other organic contaminants can, therefore, be used to determine which contaminants are responsible for deteriorated water quality (Ensminger et al., 2019).

Agricultural run-off of pesticides is a significant contributor to deteriorating surface water quality in the European Union (Morrison, Luttberg & Belden, 2016). These pesticides may have a negative effect on the environment, causing acute and chronic toxicity to aquatic organisms, pollution of drinking water sources, and accumulation of toxic compounds in the human food chain (European Environment Agency, 2018; Xie et al., 2018). Pesticides can be present at low concentrations but effective and include organic compounds with a wide range of properties, which complicates sampling and analysis to monitor water quality (Martin et al., 2016). Pesticide monitoring is often performed by taking grab samples, which has been criticized in multiple studies because of analytical limitations and because it provides a snapshot in time (Ensminger et al., 2019; Van der Oost et al., 2017; Smedes, Bakker & de Weert, 2010).

The Netherlands has a national quality standard for approximately 700 pesticides, including several degradation products (Rijksinstituut voor Volksgezondheid en Milieu, 2014). Various governmental research institutes monitor a wide range of pesticides in surface water, and this monitoring data is available on the Bedrijdingsmiddelenatlas (Bestrijdingmiddelenatlas, 2019). However, there are two problems with the current monitoring program. Firstly, the current monitoring of surface waters in the Netherlands is based on grab sampling. This method has large disadvantages. Firstly because pesticide concentrations vary significantly over time and secondly because the majority of targeted substances are not detected in grab samples (Van der Oost et al., 2017). In most cases, this is because those pesticides were not used during or prior to the sampling period. However, in some cases, pesticides were not detected because they were present below detection limits (RIVM, 2014). A second problem is that Dutch water quality monitoring is mainly focused on monitoring the water compartment, while the sediment compartment is still largely ignored, even though sediment is considered to be an important contributor in poor water quality (Brils, 2008). Hence, there is limited knowledge on the chemical status of the sediment porewater, although this might contribute to the ecological degradation of aquatic ecosystems.

In response to these limitations, recent development in alternative sampling methods allowed (i) for lower detection limits, (ii) to better assess the diversity of pesticides in the environment in a more cost-effective way and (iii) to yield the freely dissolved concentration, which is more comparable to the in situ bioavailability of a pesticide in water and sediment porewater (Kim Tiam et al., 2016). In this study, two relatively new sampling methods, active Continuous Low-Level Aquatic Monitoring (CLAM) and passive Silicone Rubber sampling (SR), for respectively monitoring pesticides in surface water and sediment porewater were assessed. This was achieved by investigating the efficiency of both samplers in a controlled laboratory setting and in the field. Subsequently, a bioassay was performed to determine how well both samplers’ extracts can provide insight into the toxic risk of the detected pesticides.

1.2 Background

1.2.1 Pesticides

Pesticides include all chemicals that can be used to control or kill pests. This includes insecticides, herbicides, fungicides, rodenticides and nematicides (Kim, Kabir & Jahan, 2017). Pesticides may exert

different effects on the target species, including lethal effects, carcinogenesis, disruption of the endocrine system, inhibition of reproduction, and damage to DNA and cells (Gakuba et al., 2018). The effects can extend to other trophic levels in the ecosystem through bioconcentration, biomagnification, and changes in the food chain (Ongley, 1996). Over the past decades, many advances have been made in the detection and analysis of potentially harmful pesticides, in particular through the increased sensitivity of mass spectronomy coupled to liquid chromatography (LC-MS), which allows for particularly more polar toxicants to be analyzed in environmental samples. However, there are still many compounds present in aquatic systems that need to be identified and quantified in various environmental compartments such as water, sediment, and aquatic plants and organisms (Gavrilescu et al., 2015). These compounds include pesticides that were already used but could not be identified until new detection methods were developed (Gavrilescu et al., 2015; Kim Tiam et al., 2016). Another source of these compounds is the synthesis of new pesticides or changes in the use of the existing ones (Geissen et al., 2015). The ecotoxicological significance of many pesticides is often underestimated because the measured concentrations are relatively low or even below the detection limit, but their unintended bioactivity to non-target organisms may also occur at relatively low concentrations, particularly when present in mixtures of different pesticides. Besides, many monitored pesticides have slow persistency, and their degradation products may be consequently present in higher concentrations than the parent pesticide but are not included in standard monitoring campaigns (Houtman, 2010). Therefore, this study was performed to determine the efficiency of an active and a passive sampling method to detect and quantify pesticides present in surface water and sediment porewater.

1.2.2 Environmental fate of pesticides

Pesticides may leach into surface water via field runoff, drainage or spray-drift (Szöcs et al., 2017). The environmental fate of pesticides depends on the natural affinity of the chemicals for one of the environmental matrices: solid matter (particulate organic carbon and mineral matter), liquid (solubility in soil, ground and surface water), gaseous form (volatilization), and the biota (Ongley, 1996). The environmental fate of a pesticide is affected by factors such as the soil sorption coefficient (Koc), the degradation rate under environmental conditions, the solubility or Henry’s Constant (H), and the n-octanol/water partition coefficient (Kow) to represent the sorption into lipid tissues of biota (Carazo-Rojas et al., 2018).

Sediment is derived from weathering and erosion of minerals, organic material, and soils in upstream areas. As water flow velocity declines, the transported material is accumulated by sedimentation (Brils, 2008). Depending on the characteristics of the contaminant, sediments can act as both sink and source of contaminants in aquatic systems (Lydy et al., 2014). When acting as a reservoir, the sediment is continuously supplying the sediment porewater with a low pesticide concentration (Akerblom, 2007). Also, the degradation rate of pesticides in sediment is generally slower than in water, for example, due to the lack of oxygen.

Figure 1 illustrates the environmental fate of pesticides. After a pesticide is released into the water, it can volatilize into the atmosphere or partition into the suspended particular matter in the water or in the sediment particulates, and it can diffuse into deposited sediments (Gobas et al., 2018). Sediment porewater is considered the first partitioning phase in the dissolution/desorption of pesticides from sediment particulates (Chapman et al., 2002). When associated with sediment particulate matter, the pesticide is subject to settling or resuspension (Gobas et al., 2018). Pesticides may accumulate in sediment before being resuspended into the surface water after turbulence caused by heavy rainfall or storm. Therefore, pesticides that are not recently used can still pose a threat to aquatic ecosystems (Green, Chandler & Piegorsch, 1996; Nabhan et al., 2018). Loss of the pesticide can occur through sediment burial, chemical or bacterial degradation.

Hydrophobic pesticides (logKow>3) are relevant because they tend to sorb to sediment particulate matter due to the nonpolar matrix of the sediment or dissolve in the sediment porewater (Supowit et al., 2016). When freely dissolved, these pesticides may pose a risk to aquatic organisms

since the sediment is both habitat and food source for many benthic organisms (Niehus et al., 2018). Although some argue that sediment porewater testing is insufficient for sediment quality assessment, it can still provide useful information regarding pesticide contamination at a study location (Chapman et al., 2002).

Figure 1 Conceptual visualization of the environmental fate of pesticides in aquatic ecosystems.

1.2.3 Grab sampling versus active/passive sampling

Current monitoring of surface waters in the Netherlands is based on grab sampling, which provides a non-representative overview of the chemical status (Van der Oost et al., 2017). There are three main limitations involved in monitoring the chemical composition of surface water and sediment porewater with grab sampling. One limitation is that grab sampling provides only a snapshot in time, at the moment of sampling, of the chemical composition in the (pore) water (Booij et al., 2016). Because the sample does not reflect the dynamic changes in the water body, episodic events can be missed (de Baat et al., 2018; Monteyne, Roose & Janssen, 2013). This may lead to either an under or overestimation of the concentration of the present chemicals. A second limitation is that the small sample volumes of grab samples may not allow for low-level detection of chemicals (Ensminger et al., 2017; Monteyne et al., 2013). And thirdly, grab samples only yield the total dissolved and particulate concentration of a chemical in a sample, rather than the freely dissolved concentration (Booij, et al., 2016).

Active and passive samplers were developed as an alternative to grab sampling. Both samplers are environmental monitoring techniques that use a collecting medium to accumulate contaminants from the (pore) water into a sorbent over time. Passive sampling yields the concentration of (freely) dissolved compounds, while in situ active sampling (i.e. in the field during a certain period) can measure the time-average total and dissolved concentrations of a compound (Booij et al., 2016; Supowit et al., 2016). The freely dissolved concentration (Cfree) that can be obtained with passive sampling is considered to be a good proxy of sediment toxicity and bioaccumulation in benthic organisms (Booij et al, 2016; Lydy et al., 2014). Cfree in porewater can be compared to quality criteria determined for surface/groundwater when assessing the overall quality of a water body (Lydy et al., 2014). Active and passive sampling are less time consuming and more efficient because fewer field trips are required to collect a time-averaged sample. Moreover, the extracts of passive samplers are relatively free from colloidal material and thus require less clean-up steps prior to analysis (Smedes, 2012). Consequently, the extracts of passive/active samplers are considered to be promising to apprehend the toxicity of pesticide mixtures in potentially contaminated field locations (Kim Tiam et al., 2016). In this study, an active and a passive sampling method were used to assess the presence of pesticide in four small water bodies in Flevoland,

respectively the Continuous Low-Level Aquatic Monitoring sampler as an active in situ sampler for surface water, and silicone rubber as a passive sampler for sediment porewater.

1.2.4 Continuous Low-Level Aquatic Monitoring (CLAM)

The Continuous Low-level Aquatic Monitoring (referred to as CLAM) actively filters water at a low rate, while capturing total and dissolved organic compounds (Aqualytical, nd). According to Ensminger et al. (2017), the CLAM sampler can be used to monitor a wide spectrum of organic contaminants in surface water bodies. Because of the automated sampling, it eliminates the collection of large water samples (Supowit et al., 2016). In traditional sampling campaigns, shipping and logistical challenges include coolers full of bottles, require packing and ice, which is heavy and thus expensive to transport, and may present loss of samples due to breakage. In contrast, the CLAM sampler is lightweight and readily stored in sealed system directly after collection at the field site. The CLAM analysis can be split into several subsets to achieve standard laboratory methods and low detection levels (Aqualytical, n.d.). According to Supowit (2016), the CLAM can achieve detection limits in the low ng L-1 _{range for several (non-)polar organic compounds in water, such as pesticides.} The CLAM sampler can continually extract and concentrate organic contaminants from water onto a solid phase extraction (SPE) disk (Aqualytical, nd). Although different SPE disks can be used, the HLB Media Disk with a hydrophobic/lipophilic balance (HLB) sorbent is the most effective for extracting a wide range of pesticides, as it is also commonly used in solid-phase extraction during water sample processing (Hladik, Smalling & Kuivila, 2008). A glass fiber pre-filtration disk can be added on top of the HLB media disk to limit clogging of the HLB filter and to ensure that only the freely dissolved concentration of pesticides reaches the HLB sorbent (Aqualytical, n.d.).

A limitation of this method might be the maximum binding capacity (MBC) of the amount of sorbent used. When the MBC is exceeded, breakthrough of substances can occur. A breakthrough study by using a second HLB Media Disk can be easily done in the CLAM set-up to assess the MCB capacity of the disk. Another limitation could be that not all substances have the same affinity for the HLB sorbent. Highly polar chemicals and particularly ionizable acids and bases are likely to have a relatively low binding affinity for HLB compared to being dissolved in water. The affinity of substances can be assessed in a residue study in which water concentrations and residue concentrations are compared (personal communication Droge, 2019).

1.2.5 Silicone rubber as a passive sampler

Silicone rubbers (referred to as SR) are used as a passive sampling tool for hydrophobic organic contaminants. A SR sampler is a piece of silicon-based organic polymer polydimethylsiloxane (PDMS) (Smedes & Booij, 2010). Despite the hydrophobic characteristics, PDMS can absorb organic compounds with a wide range of polarities. Hydrophobic compounds (logKow>4) are absorbed to a higher extent in SR than hydrophilic compounds, which improves the detection limits for hydrophobic compounds (Martin et al., 2016). Equilibrium can be achieved for compounds with a logKow > 2-3 when exposure remains constant for long enough (Smedes & Booij, 2010). Equilibrium times are generally faster at higher water flow rates and higher area/volume ratio of the sampler (Booij et al., 2016). Equilibrium can be reached for SR in porewaters and lipid-rich phases, rather than for sampling in the water column, where equilibrium for very hydrophobic compounds is often not observed due to small sampling rates (Booij et al., 2016).

The following equation can be used to calculate the silicone rubber-water partition coefficient/L/kg (KSRw), of organic contaminants in water:

K

=

C

C

(

1)

In which CSR is the concentration in the silicone rubber (e.g. in μg kg-1 _{SR), and Cw is the concentration} of a pesticide in water (e.g. in μg L-1_{), giving the KSRw the unit L kg}-1 _{SR. This means that the Cw can be} calculated when KSRw and CSR are known (Yates et al, 2013). Previous studies found that for organic contaminants, the KSRw of PDMS correlates with the Kow when measured at equilibrium (Martin et al., 2016; Sun et al., 2019)

1.2.6 Bioassessment with Daphnia magna

Bioassays are effect-based tools that enable the detection, identification, and quantification of interactions between chemicals and the environment (Brack et al., 2019). An acute toxicity test with the water flea Daphnia magna is a widely used method to determine the acute toxicity of water (samples) (Persoone et al., 2009). Moreover, bioassays with daphnids are considered to be an effective and sensitive method to evaluate the toxicity of pesticides in surface water samples, but also in porewater collected from sediment (Wernersson et al., 2015). These tests are typically conducted for 24-48 hours over different concentrations of contaminants in toxicity studies. Brack et al, (2019) recommended a 48-hour Daphnia magna immobilization test as bioassay, for detecting the effects of pesticides in an aquatic ecosystem.

Chapter 2. Research objectives

2.1 Aim

This research aimed to assess alternative sampling methods for monitoring pesticides in freshwater aquatic systems. The objectives for this study were: (1) to assess the efficiency of Continuous Low-Level Aquatic Monitoring as an active sampler of pesticides in surface water, (2) to assess the efficiency of Silicone Rubber as a passive sampler of pesticides in sediment porewater, and (3) to estimate the toxicity of the water and sediment porewater at four locations with expected emissions of pesticides.

2.2 Research questions

To meet the objectives, the following research question will be answered: “How efficient are the

Continuous Low-level Aquatic Monitoring sampler and the passive silicone rubber sampler to respectively measure pesticide concentrations in surface water and in sediment porewater?”. It is

expected that with the use of these samplers, it will be possible to detect present pesticides and to provide an approximation of the concentration of a wide range of pesticides in surface water and sediment porewater, but to certain limitations of both the chemical domain of the pesticide and the influence of test conditions. To assess the efficiency, the following sub-questions will be answered:

How efficient is the CLAM sampler in taking up pesticides from an aquarium spiked with pesticides?

 What is the pumping rate of the CLAM sampler in clean water?  Does the applied extraction method extract all pesticides?

 To what extent does the outer glass pre-filtration disk retain pesticides?  To what extent do pesticides break through the HLB1 filter?

 Does the CLAM provide an accurate estimation of the actual pesticide concentration in the water (within a factor of 2), after sampling 1, 3 and 7 liters of spiked tap water?

How efficient is the SR sampler in taking up pesticides from an aquarium spiked with pesticides?

 Can the uptake kinetics for all neutral pesticides be determined after three weeks deployment in a stirred 18 L aquarium with a nominal pesticide concentration of ~10 µg L-1_?

 How long does it take to equilibrate different neutral pesticides from the water into the SR sampler?

 Can the partition-coefficient between silicone rubber and water at equilibrium (KSRw) be predicted for each pesticide?

 Is there a relation between the logKow and the logKSRw?

 Can we use the relation between the logKow and logKSRw to predict the logKSRw for pesticides that were not included (within a factor of 2)?

Are the CLAM sampler and SR sampler effective for accurate field measurements of pesticides in small agricultural water bodies in Flevoland, the Netherlands?

 What is the pumping efficiency of the CLAM sampler in field water?

 Which pesticides are detected with the CLAM in the surface water and what are the measured concentrations?

 Which pesticides are detected with the SR sampler in the sediment porewater and what are the measured concentrations?

 Are there significant indications of toxicity at the field locations, based on a 48-hour immobility bioassay with Daphnia magna?

 Can these results be explained by the presence of pesticides, measured by the CLAM and/or SR?

2.3 Hypothesis

The following assumptions have been made for the CLAM experiments: It is expected that some of the pesticides will be poorly retained by the first filter and will be partially retained by the second filter. This applies to acids that are present in the system mostly as dissociated anions and for polar substances with a relatively low logKow (e.g. logKow<1). Also, it is expected that some of the pesticides with relatively high a logKow (e.g. logKow>4) will partially be retained by the glass fiber pre-filter. Overall, it is expected that the CLAM is able to detect all pesticides present in the aquarium water when spiked at 5 µg L-1_{. For natural pesticides with a moderate hydrophobicity the CLAM is expected} to give an accurate (<10% difference) measurement of the pesticide concentration. However, measured concentrations may deviate from actual concentrations in the aquarium because of the factors described above (Coes et al., 2014). Therefore, this study established the cut-off logKow values at which the CLAM method deviates by more than a factor 2.

The following assumptions have been made for the SR experiments: It is expected that after three weeks, all pesticides have reached an equilibrium between sampler and water, because diffusion in the silicone polymer material is relatively fast, comparable to water. Pesticides with a higher Kow are expected to take longer to equilibrate because a larger concentration difference between the water and the silicon material needs to be achieved through diffusion processes (personal communication Droge, 2019). Anionic pesticides are not expected to accumulate significantly in SR when spiked at 10 µg L-1_{. Therefore, it is expected that the KSRw can only be calculated for neutral pesticides (Smedes et} al, 2010). Yates et al (2013) found that the partitioning of pesticides into silicone rubber is determined by hydrophobicity. Therefore, it is expected that there is a correlation between Kow and KSRw and that subsequently, logKow is a good predictor of logKSRw.

The following assumptions have been made for the field experiments: It is expected that the CLAM sampler will operate without trouble in the field because of the removal of DOC by the pre-filter. The pumping speed is expected to be similar to the observed speed in the laboratory within a 24-hour timeframe because the pre-filter avoids clogging of the HLB filter. Ensminger (2018) concluded that the CLAM provides lower detection limits than grab sampling for different chemical substances in water. Based on these findings, it is expected that the CLAM might identify pesticides that could not be detected during previous monitoring events.

Chapter 3. Materials and methods

3.1 Material description

3.1.1 CLAM

In this research, the CLAM was used to measure the dissolved concentration of pesticides in the water. The CLAM was provided by the University of Amsterdam, purchased from Aqualytical ®, Louisville, KY, USA. A picture of the sampler is given in figure 2. The sampler consists of a polycarbonate housing, a small display, a micropump, a counter, and a USB charging port. The rechargeable battery allows for continuous pumping for 36 hours and can sample a volume up to 100 L (Aqualytical, n.d.). The CLAM sampler was equipped with two HLB filters and one glass fiber disk during the laboratory experiments. The glass filter was used to assess whether hydrophobic pesticides are retained by the glass fiber membrane, before being able to reach the HLB sorbent in the first HLB filter. The second HLB filter was used to assess if there was any breakthrough between filters.

Figure 2 Continuous Low-Level Aquatic Monitoring

Sampler (Aqualytical, n.d.).

3.1.2 SR

In this research, SR was used to calculate the KSRw and subsequently measure the freely dissolved concentration of pesticides in the porewater of sediment samples. SR sheets were purchased from Altec, St Austell, UK. In this research, AlteSilTM industrial Silicone Sheet with a thickness of 0.5 mm was used (figure 3).

Figure 3 Altec AlteSil Industrial Sheet (Altec, n.d.).

3.2 Method overview

Figure 4 shows a schematic overview of the experiments that were performed in this research. First, a laboratory study was performed. During this phase, the CLAM and SR sampler were deployed in an 18 L aquarium. A series of experiments were performed with both samplers. Then, the field study was performed. During this phase, four locations in Flevoland, the Netherlands were studied. After, a

series of bioassessments was performed with the field samples. Below, the methodology of the experiments with both samplers is described.

Figure 4 Schematic overview of the experiments performed in this research.

3.3 Methods for the laboratory study

3.3.1 Pesticide quantification with LC-MS/MS

Liquid Chromatography-Mass Spectrometry (LC-MS) was used for quantitative monitoring of the pesticides in the water and in the sampler extracts. LC-MS allows for the detection of individual compounds in a mixture with high molecular specificity and high detection sensitivity. This method was used because it was the most sensitive method for the detection of 500 pre-selected pesticides in water samples (Alder et al., 2006). For this research, a method to analyze and quantify a mixture of

92 pesticides was readily available at the UvA. These pesticides were selected based on previous detection in Dutch surface water. A detailed list of these 92 pesticides can be found in appendix 1.

Separation and analysis were performed on a Shimadzu Prominence XR liquid chromatograph, coupled to tandem mass spectrometry (LC-MS/MS) with electrospray ionization (ESI). The LC was equipped with a binary LC-20AD XR gradient pumping system, SIL-20AC autosampler, CTO-20AC column oven, and CBM-20A controller. For the analysis, a C18 stationary phase (Acquity C18(2), 1.7 μm, 130 Å, 150 × 2.1 mm ID, Phenomenex) was used. For the quantification, two transitions were measured in the positive and negative ion mode with multiple reaction monitoring (MRM).

Positive mode MS method used initially 70% of solvent A (MQ-water + 0.5% acetic acid) and 30% B (acetonitrile) at a flow rate of 0.3 mL/min. After the first 5 minutes at this eluent composition, which should elute inorganic impurities that go to the waste for the first 3 minutes, but still retain the most polar pesticides. Eluent B was increased stepwise up to 100% within 15 minutes, kept at 100% for 10 minutes and then switched back to 30% B for the final 4 minutes (29 minutes total run time). In between the samples, the HPLC equilibrated for 7 minutes at 30% B. The negative mode MS method initiated with 50% A and B, and increased to 95% B between minute 5-10 and remained at 95% until minute 13, after which the eluent switched back to 50% B, and equilibrated for 7 minutes until the next injection.

The results of the LC-MS/MS analysis were gathered with the Analyst Software. The analyte concentrations in the samples were calculated from these results by using pre-made calibration standards. For each analyte, a calibration curve was constructed using 10 standards with known concentrations (333 – 0.017 µg L-1_{), with signals detected for specific fragments from the parent} compounds (using MS/MS). Using a weighting factor of 1/peak area2 _{on the orders of magnitude} range of the detection signal, a linear curve was then used to calculate the analyte concentrations in the sample. The calibration curves were plotted using least-square regression. Each analyte showed sufficient linearity for the LC-MS/MS analysis in the studied working range, with correlation coefficients (R2_{) greater than 0.950. Subsequently, the measured concentrations in each sample were} converted into measured concentrations per sampler in Microsoft Excel by accounting for dilution steps and extraction volumes. Statistical analysis was performed in GraphPad Prism 8.

3.3.2 Methods CLAM

Before use, each CLAM filter was cleaned by pre-extraction with ~30 mL of methanol, ~30 mL of dichloromethane (DCM), and ~30 mL of demineralized water with a plastic syringe. After each cleaning step, 100 mL of air was pushed through the filters to eliminate solvent residues in the filters. The effluent was captured in a glass bottle and was deposed of carefully. Before use, the clean filters were stored in aluminum bags, in a -18 °C freezer.

3.3.2.2 Determining the pumping rate

First, the pumping rate of the CLAM sampler in clean water was determined. This was done by running the CLAM for 90 minutes in a clean aquarium with ~10 L of tap water. After 5, 10, 15, 30, 60 and 90 minutes, the volume of the effluent (mL) was weighted, while writing down the counter value. This allowed for the calculation of the pumping rate (mL/count). This was calculated with the following equation:

Pumpingrate=

Sampled volume

Counts

3.3.2.3 Spiking the aquarium water

To spike the aquarium water, pesticide mixture 1 was prepared. This mixture consisted of 40 herbicides and 15 insecticides. A complete list of pesticide mixture 1 can be found in appendix 3. The mixture was made by pipetting 0.5 mL of each pesticide from a pre-made stock solution (1 gr L-1_{) into} a 50 mL volumetric flask. The total volume of pesticides was 27.5 mL. 22.5 mL of methanol was added to create a final volume of 50 mL with an estimated concentration of ~10 µg mL-1_{. The aquarium was} thoroughly cleaned with water and acetone before filling it with 18 L of tap water. Next, 9 mL of pesticide mixture 1was added to achieve an approximate concentration of ~5 µg L-1 _{in the aquarium,} with a methanol fraction of <0.05%. The aquarium was placed in a fume hood, standing on two magnetic stirring plates. Two Teflon based stir bars ensured that the water was properly mixed during the research period.

3.3.2.4 Pesticide concentration calculation after CLAM Sampling

To test the accuracy of the CLAM, it was deployed in the spiked 18L aquarium for 1, 3 and 7 hours. These different sampling periods were chosen to determine whether (i) a larger sample volume leads to a more accurate estimation of the pesticide concentrations in the water, and (ii) whether there is more breakthrough between filters with a larger sample volume.

Each sampling experiment was performed with a clean set of filters and the CLAM battery was charged between the experiments. At the end of each experiment, the counter value on the CLAM was noted and a 5 mL water sample was taken to measure the actual pesticide concentration in the water (CW). The water samples were prepared for the LC-MS/MS analysis by first diluting the sample two times with 5 mL acetonitrile and by then taking 0.5 mL of that mixture and diluting that with 0.5 mL ultra-pure water to obtain a sample with 25% acetonitrile, which corresponded to the eluent composition during the injection in the HPLC. The aquarium water sampled by the CLAM was routed to a waste collection tank and was disposed of carefully.

The filters were extracted by pressing ~100 mL of acetonitrile through each separate filter, over a clean 300 mL round bottom flask with a clean plastic syringe. The LC-MS/MS samples were prepared by taking 0.5 mL of each extract and then diluting it three times with 1 mL ultra-pure water, to obtain a sample with ~33% acetonitrile. For each filter, the recovered concentrations in the water (µg L-1_{) were calculated with equations 3 and 4. The quantification results (calculated within the} Analyst software) are in the same unit as the calibration standards (µg L-1_{). The exact extract volumes} (L) can be found in the supplementary materials. The volume sampled by the CLAM was calculated with equation 2.

Total amount of pesticide on filter (ug)=

¿

Quantification result∗dilution factor∗extract volume

(3)

¿

¿

Recovered pesticide concentration filter=

¿

Total amount of pesticide∈extract

Water volume sampled by CLAM

(

4)

¿ ¿

3.3.2.5 Accuracy assessment of the CLAM

The relative accuracy of the CLAM was calculated by comparing the recovered pesticide concentration from HLB1 (µg L-1_{) with the actual pesticide concentration in the water (CW) (µg L}-1_{). The} accuracy was considered ‘excellent’ if the recovered pesticide concentration was ± 10% of the actual pesticide concentration for at least 75% of the pesticides. The accuracy was considered ‘acceptable’ if

the recovered pesticide concentration was within a factor 2 (50% - 200%) of the actual pesticide concentration for at least 75% of the pesticides.

Recovered pesticide concentration

Relative a ccuracy (%)=

¿

HLB

Actual measured pesticide concentrationC

∗100

¿

(5)

¿

In addition, it was determined whether breakthrough had occurred during the sampling period. Therefore, the measured amount of pesticide on the second filter was expressed as a percentage of the measured amount of pesticide on the first filter. The breakthrough was considered ‘excellent’ if the amount on HLB2 was <1% of that on HLB1 for at least 75% of the pesticides. The breakthrough was considered ‘acceptable’ if the amount on HLB2 was <10% of that on HLB 1 for at least 75% of the pesticides.

Relativeb reakthrough (%)=

Amount on HLB

Amount on HLB

∗100

(6)

The retention of pesticides on HLB1 after extraction was calculated with the same equation, but with the Amount on HLB1 after cleaning as the numerator. The extraction method was considered

‘excellent’ if the amount on the filter was <1% for at least 75% of the pesticides. The extraction method was considered ‘acceptable’ if the amount on the filter after extraction was <2% for at least 75% of the pesticides. This assessment was performed at the beginning of the test series, with HLB1 after 1-hour pumping, to make sure that the extraction method for all other experiments would be sufficient. If the amount on the filter after extraction would not be less than 1% for at least 75% of the pesticides, the extraction method should have been altered.

The retention of pesticides on the outer glass filter was also calculated with the same equation, but with the Amount on the glass filter as the numerator. The permeability was considered ‘excellent’ if the amount on the filter was <1% for at least 75% of the pesticides. The permeability was considered ‘acceptable’ if the amount on the glass pre-filter was <2 % for at least 75% of the pesticides.

3.3.3 Methods SR

3.3.3.1 Preparing the samplers

One individual SR sampler consisted of one piece of silicone of approximately 1 gram (~20 cm2_{) and} was attached to the aquarium with steel wire. Before use, the SR sampler was cleaned by pre-extraction by subsequently 24 hours in 100 mL acetonitrile, 24 hours in 100 mL methanol and 3 hours in 300 mL ultra-pure water, while continuously moving on a rolling belt.

3.3.3.2 Pesticide concentration calculation after SR sampling

One SR sampler of 1 gram was deployed in the spiked 18 L aquarium for ~3 weeks. The aquarium was filled and spiked with pesticide mixture 1, to an approximate concentration of ~10 µg L-1_{, with a} methanol fraction of <0.05%. A constant movement was created in the aquarium by two magnetic PTFE stirring bars to increase uptake rates, as recommended by Ripszam (2015). About 0.1 g of SR was cut off and split into triplicate pieces of ~0.035 g after 1, 5, 24, 48, 144, 312 and 552 hours of exposure to measure CSR. Each piece was weighed individually. The extraction was performed in a 1.5 mL vial, by adding 1 mL of acetonitrile to each triplicate. The LC-MS/MS samples were then prepared and diluted similarly to the samples from the CLAM experiments. Simultaneously, a 5 mL water

sample was taken from the aquarium to measure the actual pesticide concentration CW(µg L-1). The water samples were prepared and diluted similarly to the water samples from the CLAM experiment. For each piece of SR, the amount of pesticide (µg gram SR-1_{) was calculated with equations 7 and 8.} The quantification results (calculated with the Analyst software) are in the same unit as the calibration standards (µg L-1_{). The extract volume was 1 mL for each piece of SR. The mass of each} piece (g) can be found in the supplementary materials.

Total amount of pesticide∈SR extract (ug )=

¿

Quantification result∗dilution factor∗extract volume

(7)

Amount of pesticide per gramof SR=Total amount of pesticide∈SR extract Massof SR

(

)

3.3.3.3 Predicting the uptake kinetics for neutral pesticides

To obtain the uptake kinetics for all neutral pesticides that were included in the available LC-MS/MS method, the experiment was repeated with a second pesticide mixture containing different pesticides. Pesticide mixture 2 was made by pipetting x mL from the pre-made stock solutions (~1 g L -1_{) into a 50 mL volumetric flask, to obtain a nominal concentration of exactly 20 µg L}-1 _{of each} pesticide. The list can be found in appendix 4. This approach resulted in a more nominal pesticide concentration in the aquarium than with pesticide mixture 1. This was done because the pesticide concentration in mixture 1 was not nominal (see results §4.1.2.1). A clean aquarium was filled with 18 L of tap water and spiked to a concentration of 10 µg L-1 _{with a methanol fraction of <0.05 %. The} set-up of this experiment and the calculation of the uptake kinetics was similar to the experiment with mixture 1, however, only six sampling moments were performed instead of seven, due to time limitations. The uptake kinetics were calculated by combining the results of the two experiments.

In this research, it was assessed whether all pesticides would reach equilibrium within the three weeks deployment period. With the obtained actual Cw and the amount of pesticides on CSR (µg gram-1_{), the uptake kinetics could be predicted. A one-phase exponential association equation was} used to predict the KSRw at equilibrium and the kinetic constant ke (equation 9). In this equation, Y starts at CSR/Cw at one hour after deployment (t=1) and increases to a maximum equal to Ymax. Ymax represents KSRw which was defined as the ratio between the concentration of the analyte in the silicone and the concentration in water in equilibrium conditions (Yates et al, 2013). KSRw was used to back-calculate the pesticide concentration in sediment porewater from the amount of pesticides absorbed by the SR. The kinetic equilibration rate constant ke can be used to calculate the equilibrium rate (equation 10).

Y =Y

∗(1−e

₎

(9)

(

)

(

)

(

)

The method was considered ‘acceptable’ if 95% of the equilibrium was reached within 168 hours (1 week) for at least 75% of the pesticides.

3.3.3.4 Correlation between logKow and logKSRw and simple linear regression

Then, it was tested whether there was a correlation between the calculated KSRw values and the logKow valuesretrieved from literature. The KSRw values were first transformed to log10 and a Shapiro-Wilk test for normality was performed, before the correlation analysis was conducted. Finally, a log-linear curve was fitted, using simple linear regression. If acceptable, this model could then be used to predict logKSRw values for pesticides that were not included in this research. The model was considered ‘acceptable’ if the predicted values were within a factor 2 (± 0.3 log units) of the observed logKSRw for 75% of the pesticides. Regardless of the efficiency, this model was used to predict the logKSRw when prediction with uptake kinetics was not successful (§3.3.3.3).

3.4 Methods of the field study

3.4.1 Selection of sample locations

A field study was conducted at four agricultural ditches in Flevoland, the Netherlands, which were all characterized as (former) official monitoring locations (Bestrijdingsmiddelenatlas, 2019). The location choice was based on recent detection of at least ten of the target pesticides listed in appendix 1. The most recent data originated from 2017. An overview of the locations and detected pesticides is shown in table 1. For all locations, a comprehensive field observation was performed including land use, vegetation, geomorphology and present flora and fauna in water and sediment samples. The locations were characterized by different land-use and each location showed signs of recent agricultural activity. The sample site description can be found in appendix 2. At each location, two ~300 mL water samples were taken. The water samples were then stored in a 4 °C fridge before use.

Table 1 Pesticides detected at the field locations in 2017 (Bestrijdingsmiddelenatlas, 2019).

Location 1 Location 2 Location 3 Location 4

Kottertocht Gruttotocht 1 Gruttotocht 2 Rassenbeektocht

2017 Bentazon Chloorprofam Chloridazon 4,6-dinitro-o-cresol Ethofomesaat Imidacloprid* Linuron MCPA Metamitron Metazachlor Metolachlor-s Pentachlorfenol Propoxur Prosulfocarb Thiacloprid Bentazon Chloorprofam Chloridazon Dimethenamide-p Ethofumesaat Fluoxastrobin Imidacloprid Lenacil MCPA Metazachlor Metolachlor-s Metribuzine Metsulfuron-methyl Prosulfocarb Bentazon Chloorprofam Chloridazon Dimethenamide-p** 4,6-dinitro-o-cresol Ethofumesaat Fluoxastrobin Lenacil MCPA Mesotrione Metamitron Metazachlor Metolachlor-s Metribuzine Prosulfocarb Bentazon Chloorprofram Chloridazon Dimethenamide-p* Ethofumesaat MCPA Mecoprop Metamitron Metazachlor Metolachlor-s Prosulfocarb Thiacloprid

* _{Above yearly average norm (JG-MKN).} ** _{Five times above the yearly average norm (JG-MKN).}

3.4.2 Sampling surface water with the CLAM

First, the CLAM was deployed in a ditch at Science Park Amsterdam to determine the pumping efficiency in surface water under natural conditions. The CLAM was equipped with one glass fiber disk and one HLB filter, as intended for field deployment and to save research costs. At the end of

deployment, the counter value was noted. The volume of water pumped (mL) was back-calculated with equation 2. The results of the pumping efficiency of CLAM in the aquarium and in the ditch were compared.

At each location, the CLAM was deployed for at least three hours. This decision was based on the pumping performance of the CLAM in a natural surface water setting. The CLAM sampler was equipped with a pretreated HLB filter and a pretreated glass fiber disk, as described in §3.3.2.1. At the end of deployment, the counter value was noted. The sampled volume was back-calculated with equation 2.

3.4.3 Sampling porewater with SR

At each location, a sediment sample of at ~3 kg was taken with an Ekman sediment grab. This was done after the CLAM had been retrieved to prevent resuspension of sediment particles into the water that could disturb the CLAM sampling. The wet sediment was sieved over a 5 mm sieve and stored in a clean glass beaker. ~1.5 kg of wet sieved sediment was transferred to a 2 L glass jar and ~1 L of demi water was added to facilitate the contact between the SR and the dissolved pesticides (Ghosh et al., 2014). One pretreated sheet of ~1 gram of SR was exposed to the sediment sample for one week in the laboratory. During the experimental period, the jars were continuously moving on a rolling belt. This ex-situ passive sampling of sediment allowed for compounds to reach equilibrium within a reasonable time scale (Booij et al, 2016). This method was therefore preferred over in situ sampling because static in situ equilibration may take longer, and because the water concentration (CW)can be directly calculated from the measured CSR and the predicted KSRw, determined in the aquarium study (equation 11) (Ghosh et al., 2014).

Pesticideconcentration∈ pore water=

Concentration∈SR

K

(11)

3.4.4 Methods bioassay with Daphnia magna

A bioassay was performed to assess whether the samplers’ extracts could indicate toxicity in the field. The 48-hour immobility test with Daphnia magna was performed after all field samples were collected. The experiment was divided into two because of insufficient daphnia availability. The test was slightly modified after guideline OECD 202 (OECD, 2004) to accommodate small volumes. All tests were performed in triplicate, in a 10 mL test volume, with 10 juvenile daphnia (<48 hours old). The dilution medium in this experiment was Artificial Daphnia Medium (ADaM), which was produced and provided by the UvA. During the test period, the test tubes were constantly aerated with a controlled flow of compressed air using glass pipettes. The test was performed in a climate chamber.

 Water samples - All water samples were sieved over a 100 um sieve and aerated for at least 24 hours to maintain sufficient dissolved O2. 10 mL was then transferred into a 10 mL test tube. A control was performed in triplicate with 10 daphnids in 10 mL of Daphnia medium ADaM.

 CLAM – 1 mL of HLB extract was transferred into a 10 mL test tube. This volume was reduced to ~0.1 mL under nitrogen inspissation. 0.1 mL of methanol was added to ensure the dissolving of the pesticides in the test medium. Then, 9.9 mL of ADaM was added to the test tube to create a higher concentration by a factor of 10 than the actual concentration of the filter extract. 10 daphnids were added to each test tube. A control in triplicate was performed in triplicate with 0.1 mL ACN, 0.1 mL of methanol and ~9.8 mL ADaM.

 SR – For each location, three pieces of SR (~0.2 grams each) were cut off the 1 gram strip that had been exposed to the sediment. The pieces were cleaned with a paper towel to remove organic matter. The individual pieces were transferred to a 10 mL test tube with

shaking the test tubes for three days on a rolling plate. After, 10 daphnids were added to each test tube, while the SR remained present.

3.4.5 Methods LC-MS

To obtain the pesticide concentrations measured by the CLAM and the SR, samples were analyzed with the LC-MS/MS. The samples were prepared as described above. The grab samples were not analyzed because the presence of organic matter in the water samples might have disturbed the LC-MS/MS signal and could have caused significant analytical challenges.

Chapter 4. Results

4.1 Results laboratory study

4.1.1 Results CLAM

4.1.1.1 Results pumping rate

Figure 5 shows the measured volume of water per recorded count at several time points. The average pumping rate was derived from the slope of the linear regression line that was fit on this data and was determined to be 0.057 mL/count.

Figure 5 Pumping rate of the CLAM sampler expressed as a linear fit of the counts and the measured water volume that was

pumped.

4.1.1.2 Results pesticide concentrations in the water

50 of the 55 pesticides were detected in the aquarium water samples. Due to analytical challenges, these following pesticides were excluded from this research: bromofenoxim, chlorantrilipole, desmedifam, fenmedifam, and pentanochlor. These substances were not detected in the water samples, extracts or in the standards. For the remaining 50 pesticides, reliable calibration curves were fitted (R2 _{>0.95). The retention times for ametryne, desmetryne, ethofumesaat, pyridaat and} triflusulfuron-methyl were adjusted because initially, there was no detection peak within the retention window. Therefore, the first run was repeated.

The results of CW after 7 hours were considered nonrepresentative due to unrealistic high concentrations (>10 times higher than the spiked concentration)(figure 6a). Instead, the average between CW after 1 hour and CW after 3 hours was used as the pesticide concentration in the water to calculate accuracy and breakthrough. Overall, the pesticide concentration remained relatively constant during the 1 hour and 3-hour experiment (figure 6a). Pyridaat showed a remarkable high

concentration compared to the other pesticides in both samples. This may indicate a particular sensitivity of the LC-MS/MS or perhaps a pipetting failure. Figure 6b shows the relative change between the 1 and 3-hour experiments. The ethofumesaat concentration decreased by a factor two. The dimethametryne and triflusulfuron concentrations increased >2 times, which indicates uncertainty in the quantification by the LC-MS/MS. Because no replicates were taken at the beginning of the study, the increase could not be explained. Because of changing pesticide concentrations over time, it was decided to spike the aquarium with a biocide, NaN3, at 1 g L-1 _{prior to the SR experiments,} to inhibit microbial degradation.

Figure 6 Aquarium pesticide concentrations during CLAM experiments. Figure 6a shows the concentration in µg L-1 _during

the 1, 3 and 7-hour experiments. The horizontal line represents mean value. Figure 6b shows the percent change in pesticide concentration of the aquarium water samples between the 1 and 3-hour experiments.

4.1.1.3 Results extraction method

To assess the sample efficiency of the CLAM, the following assessments were made. First, the accuracy of the applied extraction method was assessed (figure 7). To do this, it was assessed whether the extraction was sufficient (<1% retention on the filter after the first extraction). After extraction with ~100 mL ACN , the HLB1 filter retained <1% for 37 of 49 pesticides and for 46 of 49 <2%. Pyridaat was identified as an outlier and not included in this analysis because of analytical challenges. For two hydrophobic pesticides (logKow>4), the amount of pesticide residue after extraction was >2% (teflubenzuron [logKow=4.56] and flufenoxuron [logKow=6.16]). For one hydrophilic pesticide, the amount of residue after extraction was >30% (metribuzin [logKow=1.70]). A Shapiro-Wilk’s test showed that the retention was not normally distributed. Spearman correlation was used to determine the relationship between retention after extraction and logKow, showing there was a significant but weak positive correlation between the two variables (r=0.3707, p=0.0087).

Overall, 75% of the pesticides were extracted for more than 99%. Therefore, the extraction method met the requirement and was considered ‘excellent’.

Figure 7 Amount of analyte remaining on the HLB1 filter after a second extraction, expressed as the percentage of amount on HLB1 after initial extraction, plotted against the logKow of the pesticides.

4.1.1.4 Results glass filter retention

After, it was assessed to what extent the pesticides were retained by the glass pre-filter (figure 8). After pumping 7 hours, the glass filter retained <1% for 32 of the 50 pesticides and for 41 of 50 <2%. Two hydrophobic pesticides were retained >10% (pyridaat [logKow = 5.73] and flufenoxuron [logKow = 6.16). A Shapiro-Wilk’s test showed that the retention was not normally distributed. Spearman correlation was used to determine the relationship between retention after extraction and logKow, showing there was a significant but weak positive correlation between the two variables (r=0.4847, p=0.0004).

Overall, 82% of the pesticides were retained less than 2% by the glass pre-filter. Therefore, the glass pre-filter retention met the requirement and was considered ‘acceptable’.

Figure 8 Amount of analyte retained by the glass pre-filter, expressed as the percentage retained by HLB1, plotted against the logKow of the pesticides.

4.1.1.5 Results breakthrough

Figure 9 shows the percentage of breakthrough expressed as the ratio between HLB2 and HLB1 in relation to the logKow. The HLB2 filter contained between 1 - 100 % of the concentration on HLB1 for all three sample volumes. After 1-hour pumping, the breakthrough was <1% for 5 out of 50 pesticides and <10% for 14 out of 50 pesticides. After 3 hours of pumping, the breakthrough was <1% for 4 out of 50 pesticides and <10% for 27 out of 50 pesticides. After 7 hours the breakthrough was <1% for none of the 50 pesticides and <10% for 4 out of 50 pesticides. The pesticides that were present by <1% were: ametryne, desmetryne, fosalon, hexazinone, metribuzin, and pyridaat. For these pesticides, there was no amount found on HLB2 or the concentrations were present below the detection limit.

One-way ANOVA indicated that there was a significant effect of the sampled volume on the sample accuracy at the p<.05 level [F(2,49)=17.02, p<0.0001]. Post hoc comparisons using the Turkey HSD test indicated that the mean score for the 7 L sampled volume (M=28.59, SD=14.88) was significantly higher than the mean score for the 1 L sampled volume (M=16.16, SD=13.89) and the 3 L sampled volume (M=18.68, SD=24.08). However, the 1 L sampled volume was not significantly different from the 3 L sampled volume.

Figure 9 Breakthrough on HLB2, expressed as the percentage retained by HLB1, plotted against the

logKow of the pesticides.

Figure 9 shows an obvious tendency between sample accuracy and logKow. A Shapiro-Wilk test showed that the sampling accuracy was not normally distributed for all three sampled volumes. Spearman correlation was used to determine the relationship between the sample accuracy and logKow, showing there was a significant but weak negative correlation between logKow and 1 L sampled volume (r=-0.3103, p=0.0284), and a strong negative correlation between logKow and 3 L (r=-0.6280, p<0.0001) and 7 L volumes (r=-0.7394, p<0.0001). This indicates that breakthrough tends to decrease with logKow.

Overall, the breakthrough was less than 10% for 28% of the pesticides for the 1-hour experiment, 54% for the 3-hour experiment and 8% for the 7-hour experiment. In no cases, it met the requirements and therefore, the breakthrough was considered ‘unacceptable’ for all sample volumes below 7 L.

4.1.1.6 Results accuracy of the recovery of pesticide concentrations by the CLAM To assess the ability of the CLAM to provide an accurate estimation of the pesticide concentrations in the water, two analyses were performed. First, the recovered concentrations were compared to the actual concentrations in the aquarium. Second, it was assessed whether the recovery was related to hydrophobicity.

Figure 10 Pesticide concentrations measured in the aquarium (pesticide mixture 1).

Figure 10 shows the average pesticide concentration in the aquarium water measured after 1, 3 and 7 hours sampling (CW). The figure also shows the recovered pesticide concentrations from the first HLB filter (HLB1) after pumping 1, 3 and 7 hours. Figure 10 shows that for some pesticides, the CLAM provides a substantial underestimation of the actual pesticide concentration in the water (azinofos-ethyl, chlortoluron, cycloxydim, fipronil, imidacloprid, pyridaat). For other pesticides, the CLAM provides a substantial overestimation (methoxyfenozide, triflusulfuron). Desmetryne and diuron were excluded from the analysis because of missing water concentrations (CW = 0).

After pumping 1, 3 and 7 hours, most recovered concentrations were between 2% - 200% accurate. After 1-hour pumping, the recovered concentration was ±10% for 6 out of 48 pesticides and ±50% for 36 out of 48 pesticides. After 3 hours of pumping, the recovered concentration was ±10% for 4 out of 48 pesticides and ±50% for 31 out of 48 pesticides. After 7 hours of pumping, the recovered concentration was ±10% for 4 out of 48 pesticides and ±50% for 26 out of 48 pesticides.

The recovery of four pesticides was <10% for 1, 3 and 7 L sampled volume: imidacloprid, mesotrione, chloortuloron and pyridaat. An additional 8 pesticides were recovered <50% for 1, 3 and 7 L sampled volume: aldicarb, bentazon, cycloxydim, fipronil, irgarol, mesotrione, sulcotrion, and teflubenzuron.

One-way ANOVA indicated that there was a significant effect of the sample volume on the sampling accuracy at the p<.05 level [F(2,47)=39.58, p<0.0001]. Post hoc comparisons using the Turkey HSD test indicated that the mean score for the 1 L sampled volume (M=71.39, SD=35.39) was significantly higher than the mean score for the 3 L sampled volume (M=61.07, SD=35.95) and the 7 L sampled volume (M=49.50, SD=24.89).

Figure 11 shows the sample accuracy in relation to the logKow. Yet, no obvious relationship was observed between accuracy and logKow. A Shapiro-Wilk test (p>0.5) showed that the sampling

accuracy was normally distributed for all three sample volumes. Pearson correlation was used to determine the relationship between the sampling accuracy and logKow, showing there was no significant correlation between logKow and 1 L (r=0.0248, p=0.8529), 3 L (r=0.07095, p=0.00503) and 7 L (r=0.1595, p=0.02545) sampled volume.

Figure 11 Accuracy of HLB1, expressed as the percentage of the measured water concentration (CW), plotted against the

logKow of the pesticides.

Overall, the relative accuracy was within a factor 2 for 75% of the pesticides for the 1-hour experiment, for 65% for the 3-hour experiment and for 54% for the 7-hour experiment. The relative accuracy only met the requirements in the one hour experiment and therefore, the accuracy was only considered ‘acceptable’ for small sample volumes (~1 L).

4.1.2 Results SR

4.1.2.1 Results water concentrations

The uptake kinetics were determined by performing two similar experiments with two different pesticide mixtures. Water samples were taken from the aquarium at the start of the experiment and after retrieval of the SR sampler. Pesticide mixture 1 consisted of 55 pesticides including three acidic pesticides that were excluded from the experiments with SR (4,6-dinitro-o-cresol, sulcotrion, triflusulfuron). Due to analytical challenges, six pesticides could not be detected in the aquarium water samples: bromofenoxim, chlorantrilipole, desmedifam, fenmedifam, pentanochlor, and pyridaat. These substances were not detected in the water samples, extracts or in the standards. For the remaining 48 pesticides, the concentration was calculated with calibration curves. Pesticide mixture 2 consisted of 37 pesticides among which were 12 acidic pesticides that were excluded from the experiments with SR. All pesticides were detected in the water samples.

Figure 12a shows the mean pesticide concentration in the aquarium during the two experiments. During the first experiment, analytical difficulties occurred when analyzing the water samples at 24, 48 and 144 hours. Pesticide concentrations were ±10 times higher for all pesticides (~50 µg L-1 _{instead of 5 µg L}-1_{). Repeated analysis with the same sample provided similar results.} However, no replicates were taken, thus, it remains uncertain whether these deviating results were representative for the pesticide concentration in the aquarium. Therefore, t=three, four and five from experiment 1 were excluded from the analysis. The second experiment was performed with the second pesticide mixture with more uniform concentrations (~10 µg L-1_{). This can be seen in figure} 12b, which shows the concentration after spiking with mixture 1 and mixture 2. The intended concentration of 10 µg L-1 _{was nearly achieved with mixture 2. Overall, the concentrations remained} relatively stable throughout the experiments. However, a subtle decline in concentration was observed which is substantiated by the uptake of pesticides by the SR.

As can be seen in figure 11b, trisulfuron, methoxyfenozine, imidacloprid, chloortuloron, and azinofos-ethyl showed relatively high concentrations after spiking with mixture 1. This corresponds with the water concentrations during the CLAM experiments (figure 10). Pesticide mixture 2 was prepared with accurate measurements to achieve a nominal concentration, which resulted in less variation between the pesticides. Only fenuron showed relatively high concentration which may be a result of a pipetting failure.

Figure 12 Pesticide concentration in the aquarium water during the SR experiments. Figure 12a shows the average water

concentration at time X for pesticide mixture 1 (black) and for pesticide mixture 2 (pink). Figure 12b shows the actual concentration of all pesticides in the aquarium for mixture 1 (black) and mixture 2 (pink). Straight-line represents the mean value.

4.1.2.2 Results uptake kinetics

The uptake kinetics were determined by merging the results from the first and second experiments with the SR. In total, 68 neutral pesticides were used to determine the uptake kinetics. Figure 13 shows an example of the best-fit uptake kinetics of 7 pesticides with different logKow. The hypothetical equilibrium was reached when the line leveled off (KSRw).