The experiment was set up in the experimental garden of the Hortus Botanicus in Leiden, The Netherlands and was conducted in 48 65 L polyethylene tubs, hereafter called mesocosms. To mimic natural pond conditions and to prevent the mesocosms from unnatural heating, tubs were placed 30 cm into the soil. The 8 cm of the mesocosms that was raised above the ground was covered in reflective tape in order to reflect sunlight to prevent heating. Additionally, mesocosms were tightly covered with 50% shadow cloth nets to prevent colonization, escape and excessive heating. In order to provide the animals with oxygen and to homogenize the water column, each mesocosm was aerated using a 5 watt aquarium pump.

In order to set up an adequate sediment layer, a combination of finely ground hay and fine-grained, ignited quartz sand as mineral substrate (grain size: 0.1–0.5 mm) was used. First, 500 g of finely cut dry hay (5 cm) was pre-soaked and mixed with 12.5 kg moist sand, which was placed at the bottom of the mesocosms. After placement of the sediment layer, ~36 L of copper-free tap water was added. Copper removal was conducted by filtering tap water with active carbon for 14 days. Subsequently, each mesocosm received 4l of filtered ditch water (mesh width 0.5 mm) that was mixed first in a large volume and was added to inoculate the mesocosms with a natural species composition of unicellular organisms. The water column and sediment layer were left for 1 week in order to allow the microbial community to establish.

The food web composition of the species in the experiment consisted of 10 commonly co-occurring species in natural and agricultural freshwater ecosystems in the Netherlands (Ieromina et al., 2016) of a variety of feeding types; (I) grazer, (II) shredder, (III) filter-collector, (IV) predator, and (V) gatherer/collector (Ieromina et al., 2016). Species were assigned to trophic groups based on literature. This resulted in the following three trophic groups: (1) herbivorous diet: Cyclops sp., Cloeon dipterum and Daphnia magna (Ahlgren et al., 1990; Santer and Van Den Bosch, 1994) (2) Detritivore diet: Tubifex sp., Chironomus riparius, Culex pipiens, and Asellus aquaticus (Graça et al., 1993; David et al., 2000), (3) Mixed diet: Sigara striata, Lymnaea stagnalis (Klecka and Boukal, 2012; Groendahl and Fink, 2016), (4) Predator group consisting of a single species (Notonecta glauca). Notonecta species are generalist predators that can potentially have large impacts on pond community structure (Cockrell, 1984; Wilbur and Dec, 1997; Klecka and Boukal, 2012) and are known to prey on all the species in the introduced community except for Tubifex sp. and only very rarely on L. stagnalis.

Species were either grown in lab culture or were collected from non-polluted field sites. To be able to trace the change in stable isotope levels during the experiment we added individuals of most species as juveniles. A full description of species and their numbers is presented in Table 1. Seven days after setting up the sediments, non-predacious animals were placed in the mesocosms. One day later, we added the predators. Herbicide, insecticide and fertilizer treatments were installed 1 day after predator additions.

The following nominal concentrations were used: 3.5 μg/L terbuthylazine and 4.0 μg/L imidacloprid, which were based on realistic concentrations in Dutch surface waters (Pesticide Atlas, 2015). The substances were dissolved in demineralized water in glass bottles and mixed by magnetic stirring. In addition, we added 6.16 mL of soluble plant feed in nutrient treatments (Nitrogen: Phosphorus: Potassium 7:4:7 (Pokon Naturado), in order to reach nominal concentrations of soluble phosphorus of 3.8 mg/L, which are realistic for Dutch surface waters and were previously shown to lead to high algal production (Ieromina et al., 2016). A second addition of insecticides, herbicides and nutrients, in equal concentrations to the first application, was conducted halfway through the experiment (t = 18, t = time in days), thus mimicking a biweekly application of substances. In order to further promote homogenization of the water column (next to continuous aerating of the water), all substances were added to the mesocosms in large volumes. Presence of either terbuthylazine (Omegam Laboratoria BV (Amsterdam, The Netherlands) and imidacloprid (Wageningen University, Wageningen, The Netherlands) was measured using liquid chromatography-tandem mass spectrometry (LC-MS/MS) on a weekly basis. The measured concentrations showed a gradual decrease in pesticide concentrations, confirming expectations (Electronic Appendix Figure S1). The time weighted average for terbuthylazine was 2.04 μg l⁻¹ and for imidacloprid 2.9μg l⁻¹. Also for nutrients we found that actual concentrations were lower than nominal concentrations, indicating uptake of nutrients.

The experiment was harvested after 35 days. To retrieve all macrofauna, the entire mesocosm was filtered through two sieves, one for the larger specimen (mesh size 2 mm) and one for the smaller specimen (mesh size 0.106 mm). All animals except for D. magna and Cyclops sp. were collected by hand with forceps and pipettes. Individuals of a single species were pooled per mesocosm in a single sample tube. In addition, a 20 × 10 cm plot of periphyton was scraped from the side of the mesocosm interior that received the most sun hours per day. Algal quantity in the mesocosm was insufficient for stable isotope analyses.

To investigate effects of the three different pressures on food web properties, we used carbon and nitrogen stable isotopes. Stable isotopes of carbon are excellent tools to study such changes in food web organization because there is a difference between the δ¹³C/δ¹²C of C3 detritus and freshwater algae (Hamilton and Lewis, 1992). Changes in δ¹³C/δ¹²C thus reflect changes in the utilization of different carbon sources by the respective trophic groups and taxa. Shifts from algal feeding to detritus feeding are expected to result in an increase in δC¹³ levels. Heavy nitrogen isotopes (δ¹⁵N) accumulate in the trophic chain (Hobson and Welch, 1992; Post, 2002), and can thus inform about effects of treatments on trophic structure and position. A change in δ¹⁵N can expected to happen when there is a shift in N source to the base of the food web, for example when algae start using different N sources or as a result of shift in carnivore feeding preference, which could for instance result from weakened prey/intermediate trophic levels.

After the harvesting the experiment, all sampling tubes were stored at −20°C prior to further processing. For the largest species (N. glauca and L. stagnalis), we used only part of the body for stable isotope analysis. For N. glauca, we used the pronotum, which was separated from the remainder of the body manually, and for L. stagnalis, we used only the foot of the snail. We expected that, as a result of growth, these parts would most readily reflect treatment-induced changes in carbon and nitrogen isotopes. Samples were freeze-dried, ground and weighed in tin cups prior to stable isotope analysis. Large samples of N. glauca, L. stagnalis, and S. striata were ground for 5 min. Smaller samples of C. dipterum, A. aquaticus, and periphyton were ground for 1.5 min, using a ball mill (Retsch Technology GmbH, Haan, Germany). Samples of the remaining species were small enough to be used entirely. We used 1.0 (±0.2) mg of each sample of a single species from a single mesocosm, which were weighed into 5 × 8 mm tin capsules. The stable isotopes were measured using an elemental analyser (NC2500; ThermoQuest Italia, Rodana, Italy) coupled with an isotope ratio mass spectrometer (Delta Plus; Thermo-Quest Finnigan, Bremen, Germany) at the VU University Amsterdam. For calibration, USGS 40 and USGS 41 were used. The reproducibility of the δ¹³C and δ¹⁵N analysis determined by repeated analysis of an internal standard (Bovine liver, NIST 1577c) was within 0·15 ‰ (n = 3). Stable isotope abundance was expressed using the δ notation:

where X represents the ¹³C content and R_sample and R_standard denote the δ¹³C/δ¹²C ratios of the sample and the standard Vienna PeeDee Belemnite (VPDB), respectively. For N, X represents the δ¹⁵N content and R_sample and R_standard denote the δ¹⁵N/δ¹⁴N ratios of the sample and the standard air, respectively.

Statistical analyses were performed with Rstudio, version 0.99.902 (R version 3.3.0) and Statistica 10 (Stat Soft Inc., Tulsa, USA). To illustrate food web relationships between single species, linear least square regressions were carried out for both C and N isotopes. A multiple regression analysis of the nitrogen isotopes was carried out to detect the likely food sources of A. aquaticus in the various treatments (species within 3‰ N from A. aquaticus: S. striata, C. dipterum, periphyton). For the analysis of changes in stable isotope composition within green and brown parts of the food web, estimations for species were averaged within trophic groups, by taking the average of the species in each of the trophic groups. We tested the single and combined effects of the agricultural chemicals on nitrogen and carbon isotopes of each species and of the entire trophic group by means of factorial ANOVA (type-II sum of squares). The factors tested were imidacloprid (2 levels; yes/no), terbuthylazine (2 levels; yes/no), nutrients (2 levels; yes/no) and all possible combinations of these factors. Stable isotopes of the different species/trophic groups were always (and only) used as response variables in these analyses. Assumptions for homogeneity of variances as well as normality of residuals were met.

All species survived until the end of the experiment, with the exception of dipteran larvae C. pipiens, and C. riparius, most likely as a result of predation by N. glauca (Klecka and Boukal, 2012). These are therefore excluded from the remainder of the results and discussion.

As expected, we found that, across all treatments, δ¹³C/δ¹²C ratios differ markedly between species, thus indicating different food preference between species (Figure 1). Similarly, we find different δ¹⁵N/ δ¹⁴N ratios between species, indicating different trophic levels (Figure 1). For both N and C stable isotopes, positive correlations were observed between a number of species (C. dipterum, D. magna, Cyclops sp., L. stagnalis, and periphyton) (Tables 2, 3), suggesting a similar reaction to changing environmental conditions or trophic interactions. In contrast, a negative correlation was found between S. striata and N. glauca, for both nitrogen and carbon isotopes (Tables 2, 3), thus indicating possible competition between these species.

The nutrients that were added to the treatments had a δ¹⁵N/δ¹⁴N ratio that was >3‰ lower than the periphyton, indicating that the periphyton used this food source. It was not possible to determine the stable carbon isotope ratio of added inorganic nutrients, because they contained only trace amounts of carbon. Periphyton δ¹⁵N/δ¹⁴N was positively affected by insecticide additions [F_{(1, 26)}4.3; P = 0.05], and herbicide additions [F_{(1, 26)}9.2; P = 0.005] and was negatively affected by nutrient additions [F_{(1, 26)}8.5; P = 0.007]. Carbon isotope ratios for periphyton were positively affected by the nutrient treatment [F_{(1, 26)} 9.7, P = 0.004].

When averaged over all species, herbivores exhibited a decrease in δ¹⁵N/δ¹⁴N ratio when exposed to nutrients [F_{(1, 40)}25.6, P < 0.0001; Figure 2A]. The other treatments (herbicides, insecticides) had no significant main effect on the δ¹⁵N/δ¹⁴N ratio of herbivores. Carbon isotopes for the herbivores showed a significant main effect of nutrients [F_{(1, 40)} 16.4 P < 0.001] and a significant interaction effect of herbicides and nutrients [F_{(1, 40)} 7.6, P = 0.008; Figure 2B]. Insecticides had a marginally significant positive effect on the carbon isotope ratio of the herbivore channel [F_{(1, 40)} 3.2, P = 0.08]. Herbivore species exhibited changes in nitrogen isotopes as a result of herbicide and nutrient additions. For Cyclops sp. and D. magna, nutrient additions lead to significantly decreased δ¹⁵N/δ¹⁴N ratios [Cyclops: F_{(1, 38)} 10.0; P = 0.003; D. magna: F_{(1, 28)} 5.2, P = 0.03]. Carbon isotope ratios of herbivore species also changed in response to the various treatments. For Cyclops sp, we found a significantly negative effect of nutrients on δ¹³C/δ¹²C ratio [F_{(1, 38)}9.8, P = 0.003] and an interaction effect between herbicides and nutrients [F_{(1, 38)} 9.8, P = 0.004], where the effect of nutrients was larger when herbicides were added [F_{(1, 38)} 4.0, P = 0.05]. For D. magna, no effect of any single stressor was found, only an interaction effect between herbicides and nutrients [F_{(1, 27)} 5.2, P = 0.03], as well as an interaction effect between nutrients, herbicides and insecticides [F_{(1, 27)} 6.1, P = 0.02].

The δ¹⁵N/¹⁴N ratio of the entire group of detritivores group was significantly affected by the insecticide ^* nutrients treatment [F_{(1, 40)}5.6290, P = 0.02; Figure 2C]; where insecticides increase the δ¹⁵N/δ¹⁴N ratio, but only in absence of nutrients. Carbon isotopes in the detritivore group were only affected by insecticide presence, where δ¹³C/δ¹²C levels decreased when insecticides were present [F_{(1, 35)} 4.7, P = 0.03; Figure 2D].

Within the detritivore guild, nitrogen isotope ratios were only significantly affected by the treatments for L. stagnalis, whereas A. aquaticus and Tubifex sp. showed no significant response. A significant, negative main effect of nutrients on the δ¹⁵N/δ¹⁴N ratio of L. stagnalis was found [F_{(1, 40)}20.3, P < 0.0001] whereas insecticide additions had a positive effect [F_{(1, 40)}3.5820; P = 0.05] on δ¹⁵N/δ¹⁴N-levels of L. stagnalis. Trophic position was also affected by the interaction between herbicides, insecticides and nutrients [F_{(1, 40)}8.1, P = 0.007].

Overall, detritivore species had 3.3‰ higher δ¹³C/δ¹²C ratios than herbivore species (Figure 1), thus confirming that food preferences were different between these functional groups. However, this was not the case for A. aquaticus; carbon isotopes for this species were significantly lower and the δ¹⁵N/ δ¹⁴ N ratio of A. aquaticus was almost an entire trophic level [2.6–3.1‰; one trophic level generally amounts to 3.4‰ (Post, 2002)] higher than the other primary consumers (Figure 1). This indicates that this species may have consumed food sources other than the hay debris. A multiple regression analysis of the nitrogen isotopes of the likely food sources of A. aquaticus (species within 3‰ from A. aquaticus: S. striata, C. dipterum, periphyton) indicated that C. dipterum may have composed a likely food source for this species [Electronic Appendix, Table 1; F_(1.21) 9.4 P 0.02]. A pairwise correlation between A. aquaticus and C. dipterum showed a significant positive relationship between C. dipterum and A. aquaticus for both nitrogen isotopes and carbon isotopes (Figures 3A,B), thus suggesting that A. aquaticus might have had a diet partly composed of dead individuals of C. dipterum.

We found no significant effect of nutrient and pesticide treatment on the average δ¹⁵N/δ¹⁴ N level of mixed feeders. Within this guild, nutrients had a negative effect on carbon isotope ratio of C. dipterum [F_{(1, 18)}8.9; P = 0.008], whereas herbicides had a positive effect [F_{(1, 18)}5.0; P = 0.04]. Moreover, we found an interaction effect between herbicides and nutrients [F_{(1, 18)} 6.7, P = 0.02], where the effect of nutrients was larger when herbicides were added. Nutrient additions led to a significantly decreased δ¹⁵N/δ¹⁴N ratio in C. dipterum [F_{(1, 18)}10.084; P = 0.005], but we found no effect of insecticide or herbicide additions on nitrogen isotope ratio.

As indicated by the high δ¹⁵N/ δ¹⁴ N levels, N. glauca was the top predator (Figure 1). We found no significant effect of any of the treatments on both Hemiptera species (S. striata, N. glauca) (Tables 1, 2).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer, VG, and handling Editor declared their shared affiliation.