Ecological and statistical evaluation of effects of pesticides in freshwater model ecosystems

Aquatic risk assessment of pesticidesThe first tier in the aquatic risk assessment procedure consists of a comparison between a Predicted Environmental Concentration (PEC) with a No Effect Concentration (NEC). A requirement for registration is that the PEC should not exceed the NEC. The NEC is calculated from the toxicity of the pesticide for defined standard test species (viz. algae Daphnia , fish) and an assessment factor, which accounts for potential differences between standard test species and indigenous species. The assessment factors used are 100 (to be multiplied with the acute EC50 of Daphnia and fish) or 10 (to be multiplied with the chronic NOEC of fish or EC50 of algae). Because this approach lacks ecological realism, the first aim of the present thesis was to validate the assessment factors used in the first tier by evaluating three chemicals with different modes of action (insecticide, herbicide, fungicide) as benchmark compounds.We compared the No Observed Effect Concentrations (NOECs), resulting from microcosm and mesocosm experiments using these compounds, with the NECs as used for the risk assessment procedure. Table 1 summarises the standards calculated from the first tier criteria set by the Uniform Principles (UP-standard), as well as the NOEC ecosystem for acute and chronic exposure regimes for the three substances. In addition, Table 1 lists the Dutch water quality standards. The assessment factors seem to protect the tested aquatic ecosystem against acute and chronic exposure to the insecticide chlorpyrifos and against chronic exposure to the herbicide linuron and the fungicide carbendazim (Table 1; chapters 2, 3 and 4). Dutch water quality standards for these three compounds were lower than the UP-standards and thus also seem to protect the aquatic ecosystems tested when exposed to individual compounds.A comparison between the UP-standards and the Lowest Observed Effect Concentration at the ecosystem level (LOEC ecosystem ) indicates that when the NEC is exceeded by a factor of 10, effects cannot be excluded in the case of chronic exposure. In the case of a single application of the insecticide chlorpyrifos, however, the assessment factor can be considered overprotective; an assessment factor of 10 instead of 100 would also seem to suffice. Two extensive literature reviews on the impact of insecticides and herbicides on aquatic microcosms and mesocosms also demonstrate that the first tier criteria of the Uniform Principles are generally adequate to protect different aquatic ecosystems from pesticide stress (Lahr et al., 1998; Van Wijngaarden et al., 1998). For compounds such as fungicides, however, hardly any information could be found in the open literature, so that validation of the assessment factors for these types of pesticide needs further attention.Table 1:Derived UP-standards, Dutch water quality standards and NOEC ecosystem observed in semi-field studies for the insecticide chlorpyrifos, the herbicide linuron and the fungicide carbendazim (all concentrations in µg/L). UP-standards were calculated from criteria set by the first tier of aquatic risk assessment. For references to toxicity values see Table 3 in chapter 1 of this thesis. UP-standardDutch water quality standardNOEC ecosystem / LOEC ecosystemShort-termLong-termAcute exposureChronic exposureChlorpyrifos0.01 a0.01 c0.0030.1 / 0.9 (Chapter 2)0.01 d/ 0.1 eLinuron0.6 b*0.6 b*0.25- / -0.5 / 5 (Chapter 3)Carbendazim3.2 a1 c0.11- / -3.3 / 33 (Chapter 4)* Dutch standard would be 0.1 µg/L (0.1 x NOEC of the standard test algae; Anonymous, 1995); - No data available; a: 0.01 × LC50 Daphnia ; b: 0.1 × EC50 Algae; c: 0.1 × NOEC Daphnia; d: data from unpublished experiment, Van den Brink et al., in prep.; e: data from Van den Brink et al., 1995.Ecological effects and recoveryOne of the aims of the present thesis was to gain insight into long-term community responses and into the factors determining the recovery of affected populations after a single application of an insecticide in experimental ditches. As was expected from its mode of action, application of chlorpyrifos resulted in large adverse effects on arthropod taxa (chapter 2). Because this experiment was performed in relatively large, outdoor systems, the recovery of the affected populations could be investigated. The recovery of populations of individual species was highly dependent on their life-cycle characteristics, such as the number of generations per year, the presence of resistant life stages and the ability to migrate from one system to another. In chapter 2 this is illustrated by the responses of two mayflies, cladocerans and an amphipod. The mayflies Cloeon dipterum and Caenis horaria do not have life stages resistant to chlorpyrifos, but are able to migrate from one ditch to another. They are also almost equally susceptible to chlorpyrifos in the laboratory but showed a very different recovery pattern.The former species recovered within 12 weeks at the highest treatment level, whereas the latter species took 24 weeks to recover fully. This can be explained from the difference in the number of generations per year. C. dipterum has many generations per year and thus recolonises the ditch repeatedly, thus recovering as soon as the concentration of chlorpyrifos allows this. C. horaria , however, produces only one generation per year, so that recovery can only take place when the next generation recolonises the ditch. Unlike mayflies, Cladocerans are not able to migrate actively from one ditch to the other. They did, however, show a very fast recovery at the higher concentration (Chapter 2). This is possible because they have a short generation time and resistant life stages in the form of ephyppia. If a taxon is not able to recolonise an impacted system and does not have resistant life stages, the species can become extinct in isolated systems like the experimental ditches. This applies for the amphipod Gammarus pulex , which became extinct at the two highest concentrations and did not recover within the 55 week experimental period. No significant effects on the invertebrate community, with the exception of Gammarus, were found from week 24 after insecticide application onwards, suggesting recovery.As part of the third aim of the thesis, the long-term responses in ecosystem structure and functioning after chronic exposure to a herbicide and fungicide were studied in aquatic microcosms. The higher concentration of the photosynthesis-inhibiting herbicide linuron resulted in a decreased biomass of the macrophyte Elodea nuttallii and decreased abundance of most algal taxa (chapter 3). The dissolved oxygen and pH levels also decreased at lower pesticide concentrations as a consequence of inhibited photosynthesis. Although a decrease in the abundance of most algal taxa was observed after to the herbicide application, a net increase in chlorophyll-a was found for the phytoplankton, periphyton and neuston. This increase was completely caused by the green alga Chlamydomonas sp., which appeared to be relatively tolerant to linuron and also had the ability to develop a tolerance to relatively high concentrations within a week. As a result of this tolerance and the reduced competition for nutrients with macrophytes, the community in the microcosms shifted from macrophyte-dominated to an algae-dominated state, especially at the highest treatment level (150 µg/L). The Copepoda and Cladocera benefited from this increased food supply and showed elevated abundance values at the higher treatment levels. Some macrophyte-associated invertebrates decreased in abundance as a result of the decline of their habitat.The fungicide carbendazim, which belongs to the bendimidazoles, is known to adversely affect microorganisms and worms. This property explains its effects on the "worm-like" taxa of the Turbellaria and Oligochaeta, but could not explain its effects on invertebrate groups like Amphipoda, Gastropoda and Cladocera (chapter 4). Unlike the direct effects of chlorpyrifos and linuron, therefore those of carbendazim on freshwater populations could not be completely deduced from the latter's taxonomic relation with the pest organisms, carbendazim it is supposed to control. The fungicide appeared to have the mode of action of a biocide rather than a chemical with a specific mode of action. Due to the decline of many invertebrates and the concomitant reduction in grazing pressure, the chlorophyll-a level and the abundance values of some phytoplankton taxa increased at the two highest concentrations (330 and 1000 µg/L).The "eutrophication-like" consequences of insecticide contamination have also often been reported and discussed in the literature (e.g. DeNoyelles et al., 1994, Cuppen et al., 1995). The increased abundance of algae due to a decrease in susceptible herbivores is a commonly reported consequence of insecticide contamination (Van Wijngaarden et al., 1998).In the present thesis, the occurrence of herbicides in the aquatic ecosystem is regarded as an undesirable side effect of its use on land. However, herbicides are also deliberately released into aquatic ecosystems for the control of nuisance aquatic vegetation (Pieterse and Murphy, 1990). Aquatic weeds are most commonly removed using compounds with a mode of action specific to macrophytes. Since algae are relatively tolerant to these chemicals (Lahr et al., 1998), they may increase their biomass due to reduced competition for nutrients (Kobriae and Whyte, 1996). Terrestrial weeds are, in the Netherlands, usually controlled by means of photosynthesis-inhibiting herbicides (NEFYTO, 1996). Although their mechanism is different, chapter 3 shows that prolonged exposure to the photosynthesis-inhibiting herbicide linuron may also result in a shift from macrophyte dominance to plankton dominance. The review published by Lahr et al. (1998) shows that this may be true for photosynthesis-inhibiting herbicides in general.The effects of fungicides are largely unstudied, but chapter 4 indicates that fungicide contamination can also cause elevated algal densities. This means that all three pesticides can contribute to "eutrophication-like" effects, though the mechanisms differ. The significance of realistic concentrations of pesticides in causing symptoms of eutrophication in surface waters, however, largely remains to be investigated.Tools to evaluate microcosm and mesocosm experimentsSemi-field experiments are usually evaluated at the taxon level. Since many species normally have low abundance values and/or show high variability (Van Wijngaarden et al., 1996), this approach has the great disadvantage that only a limited number of species can be properly analysed. This means that a substantial part of the information gathered is not used for the evaluation. This thesis presents a new multivariate tool for the analysis of treatment effects at the community level. Multivariate techniques have already been used for a long time in ecology to analyse the relation between communities and their environment. The most commonly used ordination technique is correspondence analysis, which is based on the bell-shaped unimodal model. This model fits in with the theory of the rise and fall in a preference of a species along an environmental gradient, described by their optimum and tolerance.Chapter 7 indicates why clustering and ordination based on correspondence analysis are not suitable for the analysis of the ecotoxicological data sets presented in this thesis. It argues that species normally have no optimum along the environmental axis of a stressor such as pesticides. Their response is more accurately described by a linear method; expected direct effects will increase with the concentration. On the basis of laboratory tests, this relation between the endpoint and the concentration of stressor is assumed to be sigmoid, and it is argued that a linear response model is a good approximation of this.Chapters 2 and 3 use Redundancy Analysis (RDA) to elucidate the effects of pesticides at the community level. RDA is the constrained version of the well-known ordination technique Principal Component Analysis (PCA) and is based on a linear response model (Jongman et al., 1995). In chapters 2 and 3 the analysis is constrained to the variance explained by treatment, time and their interaction. It was concluded that RDA successfully summarised the effects of a pesticide on a community in a single diagram, and is very useful especially when combined with Monte Carlo permutation tests for the determination of the significance of treatment effects. Kersting and Van den Brink (1997), however, found that output from RDA can sometimes result in very cluttered diagrams.Chapter 5 presents a new method, termed the Principal Response Curves, which overcomes this problem. PRC is based on RDA and extracts the first principal component from the treatment variance, by excluding from the analysis the variance explained by time as well as differences between replicates. It results in an easy-to-read diagram, showing the deviations of all treatments from the control in time. In contrast to most other techniques, it also allows a quantitative interpretation down to the species level. Chapter 6 introduces the rank 2 model of PRC, this means that after the extraction of the first basic response pattern, a second pattern is extracted, which expresses the most important deviation from the first response present in the data set. The second pattern is of particular importance if no single dominant response pattern is present in a data set but several sub-dominant ones occur. In chapter 6 this is illustrated by an analysis of the invertebrate and phytoplankton data sets of a microcosm experiment with two stressors, the insecticide chlorpyrifos and nutrient additions. This example shows that PRC is also able to summarise several different response patterns in two diagrams.Microcosm and mesocosm experiments are often said to be of limited value due to ecological variability and noise. From the experiments and statistical tools as described in this thesis we can conclude that despite the noise clear response patterns are revealed, if experiments are properly designed and analysed. Chapters 2, 3 and 4 illustrate that, even with a limited number of replicates, an ecological threshold level (e.g. NOEC ecosystem ) and an effect-chain covering different trophic levels can be obtained.Suggestions for future researchIn normal agricultural practice, protection of crops from pest organisms is not achieved by the application of a single compound; usually, several different compounds with different target organisms are used. Some pesticides are also administered repeatedly. The effects of combinations of pesticides on freshwater ecosystems are, however, largely unstudied (Hartgers et al., 1998). Therefore, it is important to develop criteria for the ecological risk assessment of mixtures of compounds, using realistic pesticide treatment regimes for particular crops.The problem of combination toxicity becomes even more complex when other substances used in agricultural areas, such as fertilisers, are taken into account. The combined effects of eutrophication and contaminant stress are largely unknown. It can be expected, however, that the trophic status of an ecosystem will alter the effects of pesticides (Chapter 6; Kramer et al., 1997).The ecological effect chain resulting from the experiments with the herbicide linuron and fungicide carbendazim demonstrated that microcosm and mesocosm experiments with pesticides as stressors can be very useful tools to investigate trophic interactions in aquatic ecosystems. The results of these experiments are currently being used to build a food-web model (Traas et al., 1998). Such models are considered to hold great promise for an improved understanding of ecosystem functioning and may eventually provide the ability to predict effects of contaminants at ecosystem level (Health Council of the Netherlands, 1997). The greatest obstacles that have to be overcome are the lack of solid data on parameter values (data on for instance maximum growth rate) and the lack of validation. This means that the further development of food web models require not only laboratory research on parameters values but also semi-field research for the collection of validation data sets (Health Council of the Netherlands, 1997).The modeling of direct effects and recovery patterns at the population level can be of great use for an assessment of the risks and a ranking of the effects of pesticides. For the future, modeling treatment effects and recovery patterns may be of great value as a research tool but also as a predictive tool. Models have the advantage that they allow integration of ecological and ecotoxicological knowledge, something that was largely absent from ecotoxicology until a few years ago. Development of these models will allow to a better evaluation of microcosm and mesocosm experiments performed for scientific or registration purposes.