Agricultural contaminants in amphibian breeding ponds: Occurrence, risk and correlation with agricultural land use

• Spatiotemporal screening of agrochemicals in freshwater ponds • Ecological risks based on reported ecotoxicological endpoints • Correlation assessment between agrochemical pollution and agricultural


H I G H L I G H T S
• Spatiotemporal screening of agrochemicals in freshwater ponds • Ecological risks based on reported ecotoxicological endpoints • Correlation assessment between agrochemical pollution and agricultural land use • Analytical methods to determine compounds within the European Directive 2008/105/EC • Identification of hazardous substances to include into the European watch list

G R A P H I C A L A B S T R A C T
a b s t r a c t a r t i c l e i n f o

Introduction
Agrochemical use has become an important component of global agricultural systems for decades, allowing for significant increases in crop yields and food production necessary to meet the exponential growth of the worlds' population (Carvalho, 2017). However, as a result of this extensive use, agrochemical residues have been spreading in the environment, posing a potential risk to terrestrial (Trapp et al., 2020), freshwater (Mirzaei et al., 2017) and marine ecosystems (Vanryckeghem et al., 2019). Today, pesticides are still among the most widely used agrochemicals worldwide, with a global usage of 2 million tons of active ingredients per year and with Belgium as one of the largest users in Europe (7.73 kg ha -1 ), next to the Netherlands (9.86 kg ha -1 ) (Sharma et al., 2019). Pesticides (i.e. fungicides, herbicides and insecticides) typically enter the aquatic environment by mechanisms of runoff, leaching and aerosol-mediated dispersion or drift, following their direct application to crops on land (Hasanuzzaman et al., 2019). Other types of agrochemicals, associated with veterinary practices, may indirectly enter the aquatic environment. Antimicrobial and antiparasitic drugs, such as anthelmintics and coccidiostats, are commonly used veterinary medicines, administered parenterally or orally to the individual animal, or procured as mass medication through feed or drinking water. As these compounds (and their in vivo metabolites) are excreted in feces and urine, they may enter surface waters through runoff from pastures or cropland amended with manure (Kim and Carlson, 2006).
Other important pollutants that may enter the aquatic environment are heavy metals resulting from natural and anthropogenic activities. Urbanization (e.g. surface runoff from paved roads and emissions from transportations) and industrialization (e.g. industrial emissions) are among the most important sources of heavy metal contamination. For copper and zinc, a major part also originates from agricultural practices, including the use of copper fungicides directly on land, and the supplementation of copper and zinc in pig feed to improve performance and prevent diarrhea at weaning. The use of copper and zinc feeding supplements may lead to contaminated manure that is applied to fertilize agricultural fields, and which provides an indirect source of metals reaching the aquatic environment through surface water runoff (Ali et al., 2016;Fontes et al., 1999).
Although scarcely reported, mycotoxins in the aquatic environment are associated with toxic effects (i.e. mortality) in aquatic organisms including amphibians and aquatic invertebrates (Arseculeratne et al., 1969;de Waart et al., 1972;Reiss, 1972). By definition, mycotoxins are secondary metabolites formed by filamentous fungi growing on organic material such as crops, known for their potential threat to human and animal health due to their toxic properties and their ability to increase vulnerability to several infectious diseases (Antonissen et al., 2014). Studies have shown that mycotoxins, produced directly on crops, may leach out of plant tissue to nearby waterbodies during rainfall (Gautam and Dill-Macky, 2012). Furthermore, surface water may also harbor aquatic fungi, that may produce mycotoxins, as shown by recent in vitro studies (Oliveira et al., 2013(Oliveira et al., , 2018. Finally, mycotoxin metabolites resulting from biotransformation in food-producing animals may indirectly enter surface waterbodies through surface runoff from manure-amended cropland (Biswas et al., 2017).
Surface waterbodies within agricultural landscapes are typically vulnerable to agrochemical contamination (Declerck et al., 2006). In this research focus was on lentic ecosystems, also known as still water ecosystems, which may be more subject to agricultural pollution than lotic ecosystems (i.e. flowing water ecosystem), attributed to the small volume of stagnant water that often functions as a reservoir for runoff from agricultural fields (Declerck et al., 2006;Williams et al., 2003). One example includes the detection of the organochlorine pesticide beta-benzene hexachloride (b-BHC) at higher concentrations in ponds versus rivers situated in an agricultural landscape (i.e. maximum concentrations of 2.72 versus 0.029 μg L -1 , respectively) (Teklu et al., 2016). However, ponds in agricultural areas frequently function as breeding habitat for amphibians, which has e.g. been shown for newt species (Triturus spp.) inhabiting the Pays de Herve, a rural area in Belgium (Denoël and Ficetola, 2008), as well as for a variety of amphibian species in small agricultural ponds in southeastern Minnesota (United States) and mid-western Entre Ríos Province (Argentina) (Knutson et al., 2004;Peltzer et al., 2006). Therefore, pond water contamination may affect amphibians during their most sensitive life stages, i.e. the embryonic and larval stages (Fryday and Thompson, 2017).
Based on the fact that chemical pollution of surface water presents a threat to the aquatic environment with effects such as acute and chronic toxicity to aquatic organisms, accumulation in the ecosystem and losses of habitats and biodiversity, the European Union has established environmental quality standards (EQS) for 33 priority hazardous substances, including 4 heavy metals (i.e. cadmium, lead, mercury and nickel) and 13 pesticides (i.e. alachlor, atrazine, chlorfenvinphos, chlorpyrifos, diuron, endosulfan, hexachlorobenzene, hexachlorocyclohexane, isoproturon, pentachlorophenol, simazine, trichlorobenzene and trifluralin) in natural surface waters. The maximum-allowableconcentration (MAC) EQS for pesticides and heavy metals in inland surface waters including rivers, lakes and related waterbodies such as ponds, ranges between 0.01 (i.e. endosulfan) and 4.00 μg L -1 (i.e. simazine) (European Parliament, 2008). Furthermore, Europe has established a watch list of substances including 5 antimicrobial drugs (i.e. amoxicillin, azithromycin, ciprofloxacin, clarithromycin and erythromycin) and additional pesticides for Union-wide surface water monitoring, to identify hazardous substances in the future (European Commission, 2018a). As a result of lacking field data and appropriate analytical methods, many other agrochemicals, known for exerting toxic effects to aquatic life at environmental concentrations, have not been included yet into the European legislation (Bird et al., 2018;Flemish Environmental Agency, 2019).
The agrochemical contamination of surface waterbodies within agricultural landscapes is typically most intense in the period between March and June due to the more frequent pesticide application and manure fertilization. As such, we hypothesize that agricultural contamination of amphibian breeding ponds within this period is positively correlated to the surrounding arable land (i.e. more contamination is found when ponds are surrounded by a higher percentage of arable land), and that several agricultural contaminants pose a substantial risk to aquatic organisms at the detected concentrations. Therefore, the aim of this work was to investigate the abundance (i.e. concentration and occurrence) and subsequent ecological risks, as well as the correlation with surrounding agricultural land use and monthly variation, of a wide variety of relevant agrochemicals in amphibian breeding ponds during the reproductive season of newts, i.e. March until June, when exposure to aquatic contaminants is most likely.

Study area and sampling
Data were collected for 26 amphibian breeding ponds in Flanders, Belgium. Site locations are presented in Fig. 1. Coordinates, surrounding dominant cultivation (Table S6) and pond characteristics (Table S7) are presented in Supplementary data. Pond selection was performed according to previously published work: local stakeholders were contacted to provide initial information on pond localization in Belgium and a geographical selection was made using QGIS 2.14 software, reducing ponds to those present in Flanders. Additionally, permanent dried up ponds and ponds with <1 m 2 in area or > 8 m in depth were excluded (Goessens et al., 2020b(Goessens et al., , 2020c(Goessens et al., , 2021. Finally, the presence of amphibians was assessed in March by placing 10 fish funnel traps in each pond for 24 h and only ponds that contained amphibian populations were retained for further investigation. As the vast majority of ponds were located on (sandy) loam soil rather than clay soil, the latter known for its higher sorption capacity of (organic) chemicals, minimal effects were expected of soil type on pond water concentrations (Table S7) (Doucette, 2003). Furthermore, based upon the average hardness of the pond waters included in this study, i.e. 395 mg L -1 CaCO 3 , 40°F (Table S7), it can be stated that most of the ponds were primarily fed by runoff water (Table S8) (Kenniscentrum Water, 2020). To avoid conclusions based on pulse exposures, agrochemicals were determined monthly, during March, April, May and June 2019 in each amphibian breeding pond. During the months May and June, respectively 2 (i.e. 103 and 121) and 4 ponds (i.e. 103, 121, 1518 and GER4) were excluded since they were completely dry, resulting in a total of 24 (May) and 22 sampled ponds (June). Along the banks of the ponds, three sampling sites were identified that were equidistant from each other. At two of these sites, surface water was collected at <1 m from the bank. At the third location surface water was collected at >1 m from shore ( Fig. S1). Grab water samples were collected with plastic buckets, pooled and filtered on site to eliminate most organic matter (Retsch® sieve, Novolab NV, Geraardsbergen, Belgium, 250 μm, 50 × 200 mm). Samples (1.5 L) for the analysis of anthelmintics, coccidiostats, mycotoxins and pesticides were poured in glass amber bottles. Samples (1.5 L) for antimicrobial drug analysis were collected in glass amber bottles rinsed with 1.0 M EDTA to prevent complexation of tetracyclines with Ca 2+ and Mg 2+ ions and residual metal ions (Jia et al., 2009). Samples (0.5 L) for heavy metal analysis were collected in plastic bottles. Following cooled transport, all samples were stored at 4°C before analysis within 96 h. Concerning the heavy metal analysis, all samples were acidified (pH 2) with HNO 3 prior to storage to avoid microbial activity, adsorption of heavy metals to the wall of the recipient, and oxidation and subsequent precipitation of heavy metals in the sample (Vanhoof et al., 2006).

Agrochemical multi-residue analysis methods
For all agrochemicals, instrumental parameters of analysis are presented in Table S9 to S13 of Supplementary data.

Antimicrobial drugs
Prior to sampling and analysis, 46 relevant antimicrobial drug residues (ADRs) were selected for analysis based on data collected by the Belgium Federal Agency for Medicines and Health Products (FAMHP) regarding veterinary antimicrobial drug consumption in Belgium in 2016 (FAMHP, 2016). ADRs were measured in pond water samples using a previously published and validated method (Goessens et al., 2020c). Briefly, 500 mL samples were filtered (Glass Microfiber Filters Whatman™, GE Healthcare Life Sciences, Buckinghamshire, United Kingdom, 0.45 μm, 90 × 90 mm) and extracted onto an Oasis® HLB solid-phase extraction (SPE) cartridge (6 mL, 500 mg, Waters, Zellik, Belgium). Elution was performed with 3.5 mL of methanol/acetonitrile/methyl tert-butyl ether (33%, v/v/v) acidified with 0.1% formic acid, following 3.5 mL of the same mixture alkalinized with 0.1% ammonia 25% solution before evaporation until dryness. Reconstitution was performed in 150 μL of 2/1 methanol/ultrapure water (v/v). Extracts were analyzed using an Ultimate 3000 XRS ultra-high performance liquid chromatograph (UHPLC) system coupled to a Q-Exactive™ benchtop high resolution mass spectrometer (HRMS). The HRMS was operated using heated electrospray ionization (ESI) in positive and negative mode and full-scan data were collected. The limit of quantification (LOQ) was 50 ng L -1 for all ADRs and the limit of detection (LOD) ranged from 10 to 50 ng L -1 . All sample and validation analyses were performed using matrix-matched calibration curves and appropriate internal standards (Goessens et al., 2020c).

Antiparasitic drugs: coccidiostats and anthelmintics
Twelve coccidiostats registered as feed additive or veterinary medicine (i.e. toltrazuril) in Europe (European commission, 2019) and three regularly used anthelmintics in Flanders (BCFI Vet (Belgian Centre for Farmacotherapeutic Information), 2018; Van De Steene et al., 2010) were selected for multi-residue analysis. Compounds were measured in pond water samples using a previously published and validated method (Goessens et al., 2020b). Similar to the extraction of ADRs, 500 mL water samples were filtered using a Glass Microfiber Filter (0.45 μm, 90 × 90 mm, Whatman™) and extracted onto an Oasis® HLB SPE cartridge. Cartridges were eluted subsequently with 5 mL of methanol and 5 mL of methanol acidified with 0.1% formic acid and 200 μL of the combined eluate was transferred to an UHPLC autosampler vial containing 50 μL of ultrapure water. Extracts were analyzed on an Acquity H-Class UHPLC system coupled to a Waters Xevo® TQ-XS triple quadrupole mass spectrometer (MS/MS) equipped with an electrospray ionization (ESI) interface, operated in positive and negative mode. Data were collected in selected reaction monitoring (SRM) mode. LOQs and LODs ranged between 2.5 and 250 ng L -1 and 0.7 and 40 ng L -1 , respectively. All sample and validation analyses were performed using matrixmatched calibration curves and the appropriate internal standards (Goessens et al., 2020a).

Heavy metals
Based on data collected by the Flanders Environment Agency regarding source of heavy metal emissions and occurrence of heavy metals in surface waters in Flanders (Flemish Environment Agency, 2017), eight major heavy metals were selected for the multi-residue analysis, i.e. arsenic, cadmium, chromium, copper, lead, mercury, nickel and zinc. Sample analysis was performed at the Technology and Food Science Unit of the Research Institute for Agriculture, Fisheries and Food (ILVO, Melle, Belgium). Before analysis, 500 mL samples were filtered (Macherey-Nagel™, Hoerdt, France, 7 μm, 0.2 × 90 mm), acidified by adding nitric acid (HNO 3 ) to reach a pH 1-2, and an aliquot of 250 μL was subjected to analysis, which was performed by inductively-coupled plasma optical emission spectrometry (ICP-OES) (Agilent 5110 VDV, Santa Clara, CA, United States). A method validation analysis was conducted according to the International Conference on Harmonisation (ICH) guideline for the validation of analytical procedures, by analyzing spiked pond water samples (ICH, 2005). Linearity was evaluated using the coefficient of determination (R 2 ) and a good linearity (i.e. R 2 > 0.99) was observed for all compounds within a range of 0.1-1000 μg L -1 . The LODs (i.e. 0.06-1.83 μg L -1 ) and LOQs (i.e. 0.21-6.09 μg L -1 ) were determined by dividing the standard deviation of the response, which was based on the standard deviation of ten blanks, by the slope of the calibration curve, multiplied by 3.3 and 10, respectively. Apparent recovery and precision were estimated by analyzing 10 blank samples spiked at three different concentrations, i.e. 10, 100 and 150 μg L -1 . Obtained results were satisfactory according to the previously reported guideline (ICH, 2005) with overall apparent recovery and precision values ranging between 92 and 109% and ≤6%, respectively. All sample and validation analyses were performed using a matrix-matched calibration curve and the appropriate internal standards. The method validation results are presented in Table S11 of Supplementary data.

Mycotoxins
A total of 20 mycotoxins were selected for multi-residue analysis based on recent monitoring studies for mycotoxins in food and feed in Belgium (Fraeyman et al., 2017;Royal Association of Belgian Grinders, 2017) and Europe (Gruber-Dorninger et al., 2019), comprising both mycotoxins regulated by the European Union and major metabolites, as well as mycotoxins for which legislation is currently lacking (i.e. emerging mycotoxins). Mycotoxins were measured in pond water samples using a previously published and validated method (Goessens et al., 2021). Filtration and extraction of 500 mL samples was performed similar to the method of coccidiostats and anthelmintics using a Glass Microfiber Filter, Oasis® HLB SPE cartridge and ditto elution solvents. Extracts were analyzed on an Acquity H-Class UHPLC system coupled to a Waters Xevo® TQ-XS triple quadrupole mass spectrometer equipped with an ESI interface, operated in positive and negative mode. Data were collected in SRM mode. LOD values ranged from 0.04 to 27.91 ng L -1 and LOQ values were between 1 and 40 ng L -1 (Goessens et al., 2021). All sample and validation analyses were performed using a matrix-matched calibration curve and appropriate internal standards.

Pesticides
A total of 89 relevant pesticides were selected for multi-residue analysis based on data collected by the Flanders Environment Agency regarding the occurrence of pesticides in surface water in Flanders in 2016 (Flemish Environmental Agency, 2016). An extraction and detection method was adopted from a previously published and validated method regarding the multi-residue analysis of pesticides in river water (Deknock et al., 2019;Houbraken et al., 2016) and validated for pond water. Prior to extraction, 500 mL water samples were filtered using a Glass Microfiber Filter (0.45 μm, 90 × 90 mm, Whatman™) and subsequently extracted using a reversed phase (RP) SPE. Samples were pumped through a Sep-Pak®C18 classic cartridge (0.85 mL, 360 mg, Waters, Zellik, Belgium). Samples were eluted using 10 mL of acetonitrile (ACN) and the eluate was divided into two 5 mL subsamples. Subsamples were evaporated under vacuum until dryness using a rotavapor (Büchi Rotavapor R-200, Flawil, Switzerland) and reconstituted using 2 mL of hexane (subsample for gas chromatography with electron capture detection or GC-ECD) or 2 mL of ACN/water (10/ 90, v/v) (subsample for LC-MS/MS) before transferring into the appropriate vials. Vials were stored at -21°C until analysis within 24 h. The method validation analysis was conducted according to SANTE (Pihlström et al., 2019) and ICH guidelines (ICH, 2005) by analyzing spiked blank pond water samples. Pesticides were quantified by use of an external calibration curve and single standard addition, and detected concentrations were corrected for extraction recovery. Good linearity was obtained (i.e. R 2 ≥ 0.99) for all compounds within a range of 8-800 ng L -1 and 80-800 ng L -1 for LC-MS/MS and GC-ECD, respectively. The LOQs (i.e. 8-80 ng L -1 ) were determined by analyzing the lowest concentration at which the method was validated within the limits of apparent recovery (i.e. 70 -120%) and precision (≤ 20%) according to the guideline described above (Pihlström et al., 2019). The LOD of the compounds (i.e. 0.6-36 ng L -1 ), were determined by dividing the standard deviation of the response, which was based on a linear calibration curve, by the slope of the calibration curve, multiplied by 3.3. Furthermore, apparent recovery and precision were estimated by analyzing 8 blank samples spiked at 400 ng L -1 . Recoveries were adequate for all compounds except for chlorpyrifos which showed a higher value, i.e. 131%. Furthermore, good precision was obtained for all compounds except for chlorpyrifos, ethoprophos, hexachlorobenzene, metribuzin and spiroxamine exerting values >20%, i.e. 31, 30, 26, 35 and 25%, respectively. Results could be attributed to the well-known limitation of multi-residue methodologies, where conditions are optimized for the group of compounds rather than the individual analyte, resulting in a compromise on the final analytical performance. During sample analysis, samples with measurements outside of the linear range of the calibration curve were diluted to fit the linear range of the plot and concentrations were adjusted accordingly. The method validation results are presented in Table S13 of Supplementary data.

Agricultural land use
Contaminants in the water column may be present in two forms: dissolved or bound to suspended particulate matter (SPM), e.g. suspended particles, bacteria and algae. For heavy metal analysis, both dissolved and SPM-bound fractions were determined using a cellulose filter (7 μm pore size), followed by acidification of the sample according to the international standard organization guideline (ISO 11885:2007) (ISO, 2007. For all other contaminants, the dissolved fraction, as well as the bound fraction to SPM with a diameter ≤ 0.45 μm was determined by using a glass fiber filter with a 0.45 μm pore size (Batt et al., 2008).
Following the approach of Johnson (2018) (Johnson, 2018), concentrations below the LOQ were recalculated as LOQ/2 and included as such, to avoid false negatives and include trace concentrations. Information on agricultural land use was gathered through the Geopunt database (Grootschalig Referentiebestand 2019, GRBgis, shapefile; Biologische Waarderingskaart 2019, BWK, shapefile) and calculations were performed using R, version 4.0.2 for Windows (raster, gdistance and lwgeom packages). For each pond, the percentage of arable land within a 200 m radius was determined, ranging between 0 and 43.6% (Table S6). A 200 m radius was selected according to a previous study in which 75% of amphibians were situated within a 200 m of the pond as well as a previously published method, in which amphibian abundance and agricultural land use intensification within a 200 m radius surrounding the pond, was negatively correlated (Beja and Alcazar, 2003;Kovar et al., 2009;Petranka et al., 2004).

Data analysis
Statistical analyses were conducted using R version 3.6.2 software.

Impact of percentage arable land within a 200 m radius on pollution
The goal of the analysis was to find out whether the percentage of arable land within a 200 m radius had an impact on (1) the concentration of specific agrochemicals per pond (i.e. 4-epioxytetracycline, levamisole, zinc, copper, enniatin B and terbuthylazine), which were selected based upon the highest C max and detection frequency within each agrochemical group, and (2) the number of detected compounds per pond. A regression model was fit to the data with percentage agricultural land (percAgric) as explanatory variable (either as linear predictor or as a smooth function) (Wood, 2020;Wood et al., 2016). As the data points were not independent because several samples were obtained in a single pond (location) and at the same moment (month), a mixed model was selected as regression model with, besides percAgric as a fixed factor, location and month as a random factor, with the restriction that month was a longitudinal, correlated factor (Wood, 2017(Wood, , 2020Wood et al., 2016).
Several models were fit in order to check whether the percAgric had an impact, and if so, whether that relation was linear or non-linear. The final model was selected on the basis of the Akaike Information criterion (AIC) (Wood, 2017).
As most of the data were non-normal distributed and contained many zero's, it was opted to model the number of compounds as a poisson model and the concentrations as a tweedie model. These combinations were possible via generalized additive models (Wood, 2017). The function gam from the R package mgcv was used to fit these models and a p-value <0.05 was considered significant.
Finally, in order to obtain a more global overview of the impact of surrounding arable land on the number of compounds per pond, a multivariate analysis was carried out using a multidimensional scaling (MDS) of a Manhattan distance matrix based on all (rescaled) measurements, obtained via the R function metaMDS from the library vegan (Wood, 2017).

Temporal analysis
To assess if there was a monthly variation between the number of detected compounds per pond (within each compound group), mixed models with a poisson error distribution were applied, with month as a fixed factor and pond as random grouping factor (Wood, 2017). The comparison via the log likelihood of a model with month as predictor, provides a p-value for the factor month. A post-hoc Tukey test was carried out on the pairs of months and the corrected p-values <0.05 were considered significant.
Finally, as for the impact analysis of percentage arable land mentioned above, a multivariate analysis was carried out using a multidimensional scaling of a Manhattan distance matrix based on all (rescaled) measurements, obtained via the R function metaMDS from the library vegan, to obtain a more global overview of the monthly impact on the number of compounds per pond (Wood, 2017).

Occurrence and concentration of agrochemicals
An overview of the occurrence (n = number of ponds), concentration (mean concentration C mean and maximum concentration C max , μg L -1 ), and standard deviation (SD, μg L -1 ) of the detected agrochemicals in the amphibian breeding ponds sampled in March, April, May and June 2019, is presented in Table 1.
During the sampling period, a total of 18 different ADRs were found, with C max ranging between 0.025 and 0.422 μg L -1 and highest values obtained for 4-epioxytetracycline (0.422 μg L -1 ), penicillin G (0.390 μg L -1 ), doxycycline (0.371 μg L -1 ) and 4-epitetracycline (0.336 μg L -1 ). Additionally, doxycycline was the most frequently detected ADR with a prevalence in 10 out of 22 ponds found in June. Besides ADRs, four antiparasitic drugs were found as well, with levamisole being the most abundant, i.e. found in all 26 ponds in April, with a C max of 0.550 μg L -1 . Regarding heavy metal occurrence, a total of eight heavy metals were found during the sampling period with the highest C max obtained for zinc, i.e. 333.1 μg L -1 . Additionally, zinc was most frequently detected during the sampling period with detections in all ponds in April, May and June. Eight mycotoxins were found over time, with C max ranging between 0.001 (i.e. aflatoxin B1, AFB1) and 0.013 μg L -1 (i.e. 3-acetyldeoxynivalenol, 3-ADON). Remarkably, enniatins and more particular enniatin B (ENNB) was the most Table 1 Concentration (C mean and C max , μg L -1 ), standard deviation (SD, μg L -1 ) and occurrence (occ., n = number of ponds) of the detected agrochemicals in 26 amphibian breeding ponds sampled in March, April, May and June 2019.

March
April frequently found mycotoxin in March, April, May and June, in 22, 21, 23 and 21 ponds, respectively. Finally, a total of 42 different pesticides were found during sampling, with C max ranging between 0.004 and 38.7 μg L -1 and highest concentrations for terbuthylazine, cypermethrin, epoxiconazole and propiconazole, i.e. C max of 38.7, 6.23, 3.37 and 3.33 μg L -1 , respectively. Overall, prosulfocarb, tebuconazole and terbuthylazine were the most frequently found pesticides during the entire sampling period with detections in all ponds in May (i.e. prosulfocarb and terbuthylazine) and June (i.e. tebuconazole).
Finally, when comparing obtained C max values with the EQS proposed by the European Union, only concentrations of mercury (detected in 1 pond in May) and hexachlorobenzene (detected in 2 ponds in March and May) exceeded the MAC-EQS values, implying the need for further follow-up rather than taking reducing measures based on the low detection frequency (European Parliament, 2008).

Impact of percentage arable land within a 200 m radius on the concentration of agrochemicals
The percentage of agricultural land in the vicinity of the ponds (200 m radius) was throughout the sampling not significantly linked (p-value <0.05) to any of the concentrations of 4-epioxytetracycline, copper, enniatin B, levamisole, terbuthylazine and zinc (Table S14).

Number of detected compounds per pond
The number of detected compounds per pond, for different percentages of arable land within a 200 m radius, during the months of March, April, May and June 2019, is presented in Table 2. Values ranged between 0 and 5 (ADRs), 0 -2 (coccidiostats and anthelmintics), 1 -7 (heavy metals), 0 -4 (mycotoxins), and 0 -12 (pesticides) detected compounds per pond, across the four months.

Impact of percentage arable land within a 200 m radius on the number of detected agrochemicals per pond
Similar to the impact on detected concentrations, the percentage of agricultural land in the vicinity of the ponds (200 m radius) was throughout the sampling not significantly linked (p-value <0.05) to the number of detected compounds per pond for any of the agrochemical groups. A MDS plot of the data did not show any clustering based on a difference in agricultural percentage (Fig. 2).

Temporal analysis
No monthly difference in number of detected compounds per pond could be demonstrated for ADRs, coccidiostats and anthelmintics, and  (Table S15). For heavy metals, number of compounds per pond was significantly higher in March versus April, May and June, and also significantly higher in April versus May (Table 3). Furthermore, the number of pesticides per pond was significantly higher in April, May and June versus March, and significantly higher in May and June versus April (Table 3). A MDS plot of the data indicates a clustering and gradient from March to June indicating a shift in observed pollution over time (Fig. 2).

Discussion
To the authors' knowledge, this is the first spatiotemporal screening study comprising 178 different contaminants in freshwater ponds in Europe. Compared to two similar recent studies determining pesticides in pond water, this research covers more compounds (i.e. 178 versus 42 and 67 compounds) in multiple ponds (i.e. 26 ponds versus 1 pond) at similar analytical sensitivity (Brodeur et al., 2021;Le Cor et al., 2021). The study of Brodeur  Note -= no data available due to dried up ponds.   (2021) demonstrated the presence of a high number of pesticides (i.e. 38 different pesticides and pesticide transformation products) at varying concentrations (0.009-23.9 μg L -1 ), as well as a monthly variation in detected concentrations within the studied pond, which is similar to findings within the current study. Six pesticides were found in both studies, either at higher (i.e. bentazone, flufenacet, boscalid and tebuconazole), similar (i.e. imidacloprid) or lower concentrations than those measured in the current study (i.e. terbuthylazine). In the study of Le Cor et al. (2021), only 3 pesticides (i.e. atrazine, glyphosate and its metabolite aminomethylphosphonic acid) were detected, at concentrations ranging between 12 and 212 μg L -1 , using a continuous flow-based automatic water sampling -instead of grab samples, which allows the assessment of chemical pollution over a large period of time and the impact of various environmental changes (Le Cor et al., 2021). As the use of automatic sampling may be of concern in shallow ponds, in which stagnant water and high presence of organic matter may be able to clog up the sorbent, hence reducing the uptake rate, future research will be needed to assess its broad applicability (de Weert et al., 2020). Results of the current and few previous studies indicate that ponds are subject to varying pollution with many different, site-related compounds at a wide concentration range, having a varying impact on pond water ecosystems and pointing towards specific attention for these small waterbodies.
A total of 80 different compounds, resulting from five agrochemical groups were found in amphibian breeding ponds situated across Flanders, Belgium. As we demonstrated that the detected concentrations of certain heavy metals (i.e. cadmium, copper, mercury and zinc) and pesticides (i.e. bifenthrin, cypermethrin, hexachlorobenzene and terbuthylazine) exceeded the reported ecotoxicological endpoints for the aquatic invertebrate D. magna and the gray treefrog H. versicolor (i.e. copper), posing a potential risk, it is suggested that measures to reduce pond water contamination primarily focus on these compounds. Additionally, concerned compounds would be excellent candidates to include into the European watch list for monitoring of environmental waters, ultimately establishing EQS. Furthermore, although the relatively high detected concentrations of the fungicides epoxiconazole and propiconazole, as well as the concentrations of the frequently detected tebuconazole, do not exceed reported ecotoxicological endpoints, the presence of azole fungicides in pond water raises concerns about their potential sub-lethal effects, as these compounds have been previously associated with mortality in daphnids and mortality, endocrine perturbations and limb deformities in amphibians (Belden et al., 2010;Bernabò et al., 2016;Chambers et al., 2016;Svanholm et al., 2021). Apart from reducing the applied amount, reductive measures for pesticide pollution include the use of mixed, less concentrated products, crop rotation and mixed intercropping cultivation, nonapplication, riperian buffer zones and erosion rills to redirect surface runoff to the appropriate organs, such as containers or treatment systems (Zubrod et al., 2019). Measures to reduce heavy metals in the environment are somewhat less straightforward, owing to the diverse origin of contamination, i.e. natural, industrial and/or urban sources. However, pond water contamination may be highly influenced by surface runoff from fields treated with copper fungicides, and copper and zinc contaminated swine manure. Apart from the use of erosion rills, zinc and copper in pond waters may be reduced by the use of a manure application protocol applying a minimal amount of zinc-and copperrich swine manure (Hsu and Lo, 2001). Because of its environmental risk and association with antimicrobial resistance through co-selection mechanisms (i.e. metal resistance promotes antimicrobial resistance by linkage of resistance genes), Europe has reduced the regulated amounts of zinc (European Commission, 2010) and copper (European Commission, 2018b) in animal feed over the past five years, and is planning on banning zinc completely in animal feed by the year 2022 (European Commission, 2017).
The relatively low concentrations of ADRs, coccidiostats and anthelmintics and mycotoxins in these pond waters are expected to be related to their indirect pathway (i.e. variable uptake, metabolisation and excretion before ending up in manure sprayed on land and subsequent surface runoff), physicochemical parameters (i.e. solubility, degradability, lipophilicity and organic carbon-water partition coefficient) and/or variable use (Goessens et al., 2020a(Goessens et al., , 2020c(Goessens et al., , 2021. To assess the overall toxicity of our pond water samples to various aquatic organisms, the use of bioassays Daphnia feeding inhibition assays would have been a helpful tool and is recommended in future research (Barata et al., 2008). Such tests cover a broad range of toxicity mechanisms in diverse organisms, and can account for risks posed by target and non-target compounds, as well as mixtures (Di Paolo et al., 2016). In the aquatic environment, organisms are exposed to numerous pollutants simultaneously, which was demonstrated in the current study as well, i.e. up to 12 different pesticides were found within one pond (i.e. BRA 3). In the study of Godoy et al. (2019), it was found that even statistically significant non-effect concentrations of the pharmaceuticals metformin, bisoprolol, ranitidine and sotalol could nonetheless add up to elicit significant mixture responses in D. similis and D. rerio (Godoy et al., 2019). Other studies have also shown synergistic and potentiating toxic effects on aquatic organisms of pesticides applied in a mixture (Godoy et al., 2019;Hernández et al., 2017;Laetz et al., 2009;Sanches et al., 2017). Mixture effects are commonly assessed by means of mixture models, such as the quantitative structure-activity relationship (QSAR) model and index-isobologram model (Altenburger et al., 2003). Taking mixture toxicology into account, ecotoxicological risk assessment based on single toxic effects of chemicals can lead to an underestimation of the real impact of these compounds to the aquatic ecosystems (Godoy et al., 2019). To improve the accurate prediction of mixture models, additional (acute and chronic) ecotoxicity data regarding mixture effects of environmental contaminants are needed (Godoy et al., 2019).
As micropollutants in the aquatic environment can undergo different biotic (i.e. microbial processes) and abiotic transformation processes (e.g. hydrolysis and photochemical reactions) which often leads to the formation of (toxic) transformation products, it is important to include these compounds in the screening of pond waters. Beyond transformation products related to environmental transformation processes, additional transformation products present in the aquatic environment are metabolites resulting from parent drug metabolisation in the animal (e.g. phase I and II biotransformation), ending up in the aquatic environment by surface runoff from manure amended land. One example is the formation of amoxicillin-diketopiperazine-2',5'dione from amoxicillin (Hirte et al., 2016;Maté et al., 2017). However, to date, the chemical identities and toxicity of most transformation products in the aquatic environment remain still unknown (Menz et al., 2017). For this reason, only several well-known transformation products were included in the targeted pond water analysis, including amoxicillin-diketopiperazine-2',5'-dione and 3 tetracycline degradation products. Furthermore, 8 mycotoxin metabolites, i.e. 3-ADON, AFM1, DOM-1, ZAN, α-ZAL, α-ZEL, β-ZAL and β-ZEL, resulting from the metabolisation of commonly found parent mycotoxins in food and feed samples in Europe and Belgium (Gruber-Dorninger et al., 2019;Knutsen et al., 2017;Royal Association of Belgian Grinders, 2017;Schrenk et al., 2020), were included as well. Additionally, by using the suspect screening mode of the UHPLC-HRMS method for ADRs quantification, it was possible to screen for additional transformation products, including 19 antimicrobial transformation products (e.g. 4hydroxysulfadiazine, apo-oxytetracycline and amoxicillin penicilloic acid) and 4 pesticide transformation products (i.e. deisopropylatrazine, 2-hydroxyatrazine, deethylatrazine and desethylterbuthylazine). The authors believe the latter method may be further applied to identify potentially environmentally relevant compounds by matching fullspectrum data acquisitions to home-made or online mass spectra libraries such as Massbank (Hernández et al., 2014).
As many different pesticides can be used for each type of crop, and the choice remains at the discretion of the farmer, it is difficult to relate pesticides detected within a pond to the dominant surrounding cultivation. However, in the case of potatoes and fruit trees, most registered products on the Belgian market contain cymoxanil, mancozeb and clomazon for the control of Phytophthora fungi and Dicotyledonae weeds. For grains such as wheats and oats, many products contain prothioconazole, tebuconazole, and deltamethrin for the control of a.o. Fusarium fungi, and aphids, respectively, whilst deltamethrin is the major component of products used to repel aphids in corn (Federal Public Service Health, 2021). As such, the frequent detection of tebuconazole in the 26 amphibian breeding ponds included in this study may stem from fungicides used in grain cultivation.
The fact that the number of detected compounds per pond, as well as the detected concentrations of 4-epioxytetracycline, levamisole, zinc, copper, enniatin B and terbuthylazine, could not be linked to the percentage of arable land within a 200 m radius, within each month, seems surprising, but may be related to different catchment variables surrounding each pond within 200 m, such as the presence of buffer strips and erosion rills, and the slope of land, which are known to influence surface runoff and were not studied nor accounted for. As such, future research should provide a detailed study of catchment variables in addition to land use within the drainage area of each pond to provide conclusions on pond water contamination related to agricultural land use (Dabrowski et al., 2002;Sliva and Williams, 2001). Apart from runoff from nearby pesticide-and manure-contaminated meadows, various non-agricultural sources of pond water pollution may have an influence as well. With regards to ADR pollution, it should be noted that antimicrobials (i.e. natural antibiotics) can occur naturally in the environment, although it is generally stated that the anthropogenic source of pollution is predominant (Felis et al., 2020). In the study of Asagbra et al. (2005), it was found that Streptomyces spp., which occur naturally in agricultural soil (Dhanasekaran et al., 2011), were able to produce tetracyclines using agricultural waste such as corncobs as a substrate, reaching up to 13.18 mg tetracycline per g waste within 3 days (Asagbra et al., 2005). As these agricultural wastes are present on fields subject to rain and irrigation events, it is plausible to expect tetracycline in water runoff ending up in the aquatic environment. Additionally, some studies have isolated aquatic Streptomyces spp. from lakes and rivers, able to produce ADRs as well (Eskandari et al., 2020;Saadoun et al., 1999). Heavy metal pollution in surface water may also result from natural and/or anthropogenic (non-agricultural) activities. Natural release of heavy metals in the aquatic environment occurs during the oxidation or reduction of minerals, and during natural desalination processes, and has been described for arsenic, chromium, copper, lead, nickel and zinc (Cao et al., 2003). Furthermore, anthropogenic, non-agricultural sources of heavy metal pollution include metal industry emissions, with volatilization, atmospheric transportation and deposition of particles (Flemish Environmental Agency, 2013;Hutton, 1983), metal leaching from asphalted roads, constructed using dust waste from metal industry (Flemish Environmental Agency, 2013), and surface runoff from metallic rooftops (De Buyck et al., 2021;Md Meftaul et al., 2020). In Flanders, copper and arsenic contamination of surface and ground waters in some natural areas are related to the historic pollution by metal industry dating from more than 50 years ago, e.g. arsenic and copper industry in Bocholt (Belgium) (Flemish Environmental Agency, 2016). In the current study, industrial sources of heavy metal pollution is expected to be of major influence, as all ponds situated in Brakel, Lierde, Zottegem and Zwalm were located nearby local heavy metal industry at a minimum distance ranging between approximately 1 (i.e. ponds in Brakel) and 14 km (i.e. ponds in Zottegem) (Table S14). In fact, for chromium, lead and mercury, only non-agricultural sources have been reported in general. Nonetheless, agricultural practices such as the current use of copper fungicides, the supplementation of copper and zinc to animal feed and their residues in manure (on average 476 and 1010 mg kg -1 for copper and zinc, respectively, in pig manure), the use of artificial fertilizers contaminated with cadmium and nickel, as well as the historic use of arsenic pesticides, may add to the heavy metal pollution of ponds (Flemish Environmental Agency, 2013;Lenssinck, 2018).
Regarding mycotoxin pollution, certain land fungi, such as Aspergillus fumigatus, A. niger and Purpureocillium lilacinum naturally occur in aquatic matrices, and are able to produce mycotoxins such as AFB1, aflatoxin B2 (AFB2), fumonisin B3 (FB3) and ochratoxin A (OTA), at concentrations up to 35 ng L -1 (Oliveira et al., 2018). In addition, the fact that pesticide residues are detected even in remote aquatic ecosystems, far from any agricultural activity, can be attributed to the sequence of volatilization from relatively warm source locations, atmospheric transport, and deposition upon encountering decreased air temperatures, as previously demonstrated for chlorpyrifos, dieldrin, endosulfan, fenpropimorph and hexachlorobenzene (Anyusheva et al., 2012;Hageman et al., 2006). Historic pollution with persistent pesticides such as boscalid, dichlorodiphenyldichloroethylene (DDE), epoxiconazole and tebuconazole (i.e. degradation half-lives >1 year in water and/or sediment) may provide an additional explanation for the detection of these compounds in ponds regardless of current agricultural influences (Agriculture and Environment Research Unit, 2020;Hathaway-Jenkins et al., 2010). Finally, pesticide drift and/or runoff from lawns and gardens can be considered as non-agricultural, urban-related sources of pond water pollution (Fevery et al., 2016). In Belgium, about 250,000 kg plant protection products for nonprofessional use were used per year in 2010-2012. As such, we can conclude that agrochemical contamination resulting from crop production activities covers only one aspect of aquatic environmental pollution in an agricultural landscape (Fevery et al., 2016). Finally, detected concentrations may as well be influenced by pond size variability as surface area and pond depth ranged between 23.4 and 283 m 2 (± 59.5 m 2 on average) and 0.36 and 1.38 m (± 0.25 m on average), respectively, resulting in higher detected concentrations in smaller water bodies and vice versa.
The monthly pattern observed for heavy metal and pesticide detections should be interpreted with caution, as results were based upon a single grab sample per month, being subject to variable weathering conditions, farm and industrial practices, catchment variables, etc. In general, it might be expected that the application of metal-contaminated manure or artificial fertilizers before sowing, variable leakage from metal industry sites and application of pesticides during the plants growing season may attribute to the observed monthly variation of heavy metals and pesticides in the sampled ponds (De Boer et al., 2011;Flemish Government, 2019;Hammink, 2008).
Apart from the five agrochemical groups studied in this paper (ADRs, coccidiostats and anthelmintics, heavy metals, mycotoxins and pesticides), an important part of agricultural pollution is also derived from fertilizers such as nitrogen and phosphorus (i.e. total phosphorus, total nitrogen, ammonium, nitrate and nitrite), which are present in animal manure and commercial fertilizers used on land, reaching nearby surface waters through surface runoff (Céréghino et al., 2007). Excessive nutrient loads, and resulting euthrophication, leads to a decrease in dissolved oxygen and subsequent mortality of aerobic aquatic organisms, and has been previously associated with reduced survival, altered feeding and swimming activity, decreased growth and the emergence of disease in amphibians (Johnson et al., 2007;Peltzer et al., 2008). As such, future research is warranted to study the occurrence and risk of nutrients in these amphibian breeding ponds as well, for which mechanistic models might lead to more specific sources of pollution and solutions in the context of sustainable development (Cambien et al., 2020;Forio and Goethals, 2020).
As agrochemicals are able to exert adverse effects on amphibians and aquatic invertebrates, both playing a pivotal role in the food web and potentially amphibian disease dynamics, this study exemplifies the need for continuous monitoring campaigns following up the occurrence of agrochemical residues in ponds and taking timely interventions if necessary (Deknock et al., 2020;Schmeller et al., 2014;Seda and Petrusek, 2011). Finally, this research identifies potential hazardous substances which may be added to the European watch list in the future (European Commission, 2018a).

Conclusions
The results of this study show that amphibian breeding ponds across Flanders (Belgium) are frequently contaminated, with bifenthrin, cadmium, copper, hexachlorobenzene, mercury, cypermethrin, terbuthylazine and zinc posing a substantial ecological risk at the measured concentration levels. Furthermore, the number of detected compounds per pond, as well as the detected concentrations of 4-epioxytetracycline, copper, levamisole, zinc, enniatin B and terbuthylazine, did not depend on the percentage of surrounding arable land, suggesting the influence of various other natural and/or anthropogenic sources of agrochemical pollution in amphibian breeding ponds such as the natural occurrence of ADRs, heavy metals and mycotoxins, industrial heavy metal pollution, historic heavy metal and pesticide pollution, pesticide drift from remote areas, lawns and gardens, and metal contaminated runoff from metallic rooftops. Finally, the monthly pattern observed for heavy metal and pesticide detections, should be interpreted with caution, as results were based upon a single grab sample per month, being subject to variable weathering conditions, farm and industrial practices and catchment variables. Based upon this research, the authors would propose to add bifenthrin, copper, cypermethrin, terbuthylazine and zinc, for which currently no EQS have been established, to the European watch list for Union-wide monitoring in water policy.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.