| Front page | | Contents | | Previous | | Next |

Mesocosm experiments in the approval procedure for pesticides

4 Database/mesocosm data

First, an extensive search in databases for literature about the effects of pesticides on mesocosms, streams and lakes was undertaken. Having eliminated all non-relevant papers, 1744 papers remained of which the number was reduced to comprise only papers containing available data on specific pesticides. Having scrutinised the abstracts, the papers on the selected pesticides were obtained. The papers were then thoroughly examined, and effect concentrations etc. were entered into a database provided that they met certain criteria for documentation and quality, which included:

	Effect concentrations were already established or could be estimated from tables and figures.
	True replicates were included in the experiment to allow evaluation of variability.
	Besides the control systems, two or more pesticide concentrations were analysed.
	The experimental conditions were described in detail.
	Pesticide exposure was described, including solvents, active or formulated products.
	In stagnant water mesocosms analysis of pesticide concentration were carried out at the minimum at the start of experiment.
	The mesocosms encompassed a complex ecosystem (at least two different functional groups present were present, e.g. phytoplankton and zooplankton).
	Type 1 Errors (e.g. spurious differences between treatments and controls) were not included in the data base (see below).

Several papers and reports in the scientific literature originate from the same experiment and may be based on the same data set. To avoid duplication a particular study was identified by the pesticide administration, dates of start and end, location (latitude & longitude, name of facility) etc during the evaluation process. In most studies the deviation in a response parameter (e.g. abundance) is referred to a corresponding control and tested for being significantly different. At increasing number of observations (dates and taxonomic groups) the risk for Type 1 Error increases accordingly, e.g. at a significance level of 5% every 20^th observation by chance will be different from the corresponding control. Prior to entering data into the data base we eliminated Type 1 Errors by a Bonforroni adjustment. In case a p-value in a study was not explicitly given, we assumed it to be 0.05. A measured difference was only considered significant if the number of observed significant differences was above a minimum number given by:

(0.05/total number of tests) > (p-value)^MASF, where

total number of tests represent the total number of tests (i.e. number of observation days*number of taxonomic groups monitored), p-value given in the study and MASF the minimum number of significant differences recorded for a taxonomic group.

The generated database is based on 112 publications and includes 91 experiments covering 3,635 effect concentrations for 31 different pesticides. Of a total of 3,635 effect concentrations 410 focus on flow-through systems. The majority of the effect concentrations are on zooplankton (1,644 values), followed by effects on macroinvertebrates (1,191 values), phytoplankton (558 values) and periphyton (145 values) (see Fig. 1).

Figure 1.
Number of effect concentrations in data base distributed among different taxonomic groups.

Abundance is the dominant effect parameter with 3,114 values out of a total of 3,635 followed by mortality with only 177 values (Fig. 2). Functional effect parameters such as production and growth have seldom been measured and are therefore represented only at a limited scale. The sensitivity of the different effect parameters is discussed in Chapter 6 (primarily covering macroinvertebrates).

Figure 2.
Number of different effect parameters contained in the data base.

The database encompassed mesocosm studies with 8 different herbicides (2,4-D, Alachlor, Atrazine, Glufosinate-ammonium, Glyphosate, Hexazinone, Linuron, Triclopyr ester), 22 insecticides (Aminocarb, Azinphos-methyl, Bifenthrin, Carbaryl, Carbofuran, Chlorpyrifos, Cyfluthrin, Deltamethrin, Diazinon, Diflubenzuron, Dimethoate, Endosulfan, Esfenvalerate, Fenitrothion, Fenvalerate, Lambda-cyhalothrin, Lindane, Methoxychlor, Mexacarbate, Permethrin, Tebufenozide, Tralomethrin) and 1 fungicide: Propiconazole.

The dominant end points in the studies of the database are NOEC and LOEC values (2,046 and 1,512 values, respectively). The database only includes 67 EC50 values and 11 NEC values (Fig. 3). For the majority of experiments both NOEC and LOEC values were stored in the database.

Figure 3.
Number of different end-points contained in the data base.

In the mesocosm data analysed 81 out of 90 experiments had multiple replicates allowing to identify LOEC’s and NOEC’s (i.e. Hypothesis test). More than 80% of the statistical tests were carried out using ANOVA. Eighteen experiments had a sufficient range of concentrations to calculate EC50 or NEC’s (i.e Point estimate tests), while 9 experiments could be characterised as hybrid tests with calculated values of EC50 (NEC) in addition to NOEC and LOEC. For further details, see Chapter 5 and 7.

The mesocosm investigations typically lasted several months (Fig. 4). More than 75% of the effect values included measurements made more than 2 weeks after the first (or only) addition of pesticides. Generally, samplings within 1-2 weeks refer to experiments where effects on phyto- and zooplankton were studied while sampling schemes in excess of 1-2 months also included organisms with long life cycles (macroinvertebrates and macrophytes).

The size of the mesocosms varied widely (Fig. 5). Systems with a volume from 0.003 m³ (3 litres) up to 1,100 m³ (1,100,000 litres) are included in the database, investigations in small systems typically having been undertaken in flow-through environments.

Figure 4.
Distribution of recorded effect concentrations after initial dosing of pesticide.

Figure 5.
Distribution of system volume in mesocosm experiments contained in data base.

Depending on the objectives of the study mesocosm experiments are designed at various levels of complexity. Generally, systems should be as natural as possible and should contain different taxonomic groups at various trophic levels. Therefore, seen on this background it is highly conspicuous that the number of major groups (defined as macroalgae, phytoplankton, epibenthic microalgae, vascular plants, zooplankton, macroinvertebrates and vertebrates) monitored and included in the database mesocosm experiments is extremely low (Fig. 6). Thus, in more than 40% of the experiments only data from one major group is included and in approx. 30% of the experiments only two major groups are included. However, based on descriptions of experiments, mesocosms generally contained more major groups than were actually monitored during the study.

Figure 6.
Number of major groups monitored (NOEC or LOEC) for ef-fects of pesticides in mesocosm studies contained in data base. In 31 experiments only one major group (e.g. zooplankton) were followed.

The database contains a system description of the mesocosm experiments (i.e. size, time table, addition of pesticides, solvents in which the pesticide may have been dissolved, statistical design, construction and selection of materials for mesocosm systems, water quality, etc.) and of the effects (effect concentrations at various points of time, taxonomic groups, species, functional groups, life stages, effect parameters, end points, measurement methods, statistical tests, measurement programmes, etc.). Parameters for water quality (e.g. the concentration of inorganic nutrients) have only been briefly described and are therefore not combined in the analyses. Likewise, in mesocosms including sediment, properties of the sediment (e.g. grain size, organic content) are only described in sufficient detail in about 40% of the studies.

Most mesocosm experiments contained in the database included a sediment compartment (87 %), while macrophytes were present in at least 28 experiments (explicitly noted) and absent in 37 experiments. Based on the information given in analogous experiments (same laboratory and mesocosm set-up) we have assumed presence and absence of macrophytes in an additional 8 experiments. For every species and higher taxonomic groups a functional grouping according to habitat and feeding mode was conducted for the zooplankton and macroinvertebrates, provided that this information was available. The grouping was based on information obtained from various sources (Friberg, pers. comm.)

Single species data

The single species data encompassed by the database have primarily been gathered from the US-EPA (US- Environmental Protection Agency) ’Aquire’ database. For some pesticides, data have been obtained either from the Danish EPA or the Pesticide Manual (1998). From the US-EPA database, only data with complete or moderate experimental documentation have been used (see Table 1). In Annex 0 is shown all primary toxicity data used for extrapolation

Table 1.
Overview of pesticides contained in the mesocosm data base including calculated hazard concentrations. Primary data used for calculation of Hazard Concentrations is shown in Annex A.

*F: Fish, I: Insect, C: Crustacea, A: Algae, M: macrophyte

** only 1 record for algae

The data collected are limited to freshwater organisms. As a minimum, we have endeavoured to procure toxicity data for fish (LC50-96h), crustaceans (LC50-48h) and algae (EC50-96h). For invertebrates both mortality and immobilisation were accepted as effect parameters. For a number of insecticides, (azinphos-methyl, mexacarbate, diazinon, lindane, chlorpyrifos, diflubenzuron, fenvalerate, tralomethrin, bifenthrin), however, we have not been able to obtain toxicity values for algae. However, most of these insecticides are probably of low toxicity to algae.

Hazard concentrations (HC_5,50%) for ecosystems have been calculated from distribution-based extrapolation of single species toxicity data (EC50, LC50) as described in Miljøprojekt Nr. 250 (Miljøstyrelsen, 1994) and a statistical method (Wagner & Løkke, 1991). HC_5,50% denotes the pesticide concentration that with a 50% probability protects 95% of the ecosystem organisms. We have chosen the 50% probability level instead of e.g. 95 % level for statistical reasons, as the uncertainty of the estimate will increase both at higher and lower probabilities (see Fig. 7).

Figure 7:
Principles of probabilistic extrapolation methods adopted from Smith and Cairns (1993).

A: a denotes the hazardous concentration to be estimated as a fraction of a log-normal distribution (Wagner and Løkke 1991) or a logistic distribution (Aldenberg and Slob 1993) of LOEC’s estimated for different species. The fraction of species to the right of a is supposed to be protected.

B: As the density function in figure A represents a sample of toxi-city data a must be considered as an estimate with an associated error. Consequently a is assumed to follow the distribution superimposed on the log- normal or the logistic distribution. a therefore denotes the pesticide concentration that with a 50% probability protects 95% of the ecosystem organisms .

C: Based on the distribution of a a hazardous concentration providing a probability of exceeding a might be provided. The HC_5,95 concentration thus denotes a concentration protecting 95 % of the species with a probability of 95%; in this fictive example HC_5,95 is 2 orders of magnitude lower than HC_5,50.

Physico-chemical properties

Solubility in water, distribution coefficients between octanol and water (logK_ow), sorption coefficients and degradation half-life are all database parameters. Data have mainly been obtained from the Pesticide Manual (1998).

| Front page | | Contents | | Previous | | Next | | Top |