Understanding single-species and model ecosystem sensitivity: Data-based comparison

Risk assessments for compounds released to the environment typically rely on single-species toxicity studies to predict concentrations at which effects may be observed. These single-species toxicity studies are usually conducted with a few species, cultured under optimum conditions (diet, temperature, light, etc.) and tested in clean water with constant exposure to the compound of interest. Chronic toxicity data are then extrapolated to the ecosystem during risk assessments to predict concentrations that will not adversely impact the environment. Several approaches have been developed that apply statistical methods to estimate toxicant concentrations adversely affecting a small percentage of single species (e.g., 5%). There are several rarely stated, and infrequently tested, biological and statistical assumptions required to make this extrapolation. One test of the ability to use single-species toxicity data to protect ecosystems is to compare effects on single species with effects on experimental and natural ecosystems (e.g., microcosms, model ecosystems, field). Towards this end, we summarized the chronic single species and experimental ecosystem data on a variety of substances (n = 11), including heavy metals, pesticides, surfactants, and general organic and inorganic compounds. Single-species data were summarized as genus-specific geometric means using the NOEC or EC20 concentration. Genus mean values spanned a range of values with genera being affected at concentrations above and below those causing effects on model ecosystems. Geometric mean model ecosystem no effect concentrations corresponded to concentrations expected to exceed the NOEC of 10 to 52% of genera. This analysis suggests that laboratory-generated single species chronic studies can be used to establish concentrations protective of model ecosystem, and likely whole ecosystem, effects. Further, the use of the 5% of genera affected level is conservative relative to mean model ecosystem data but is a fairly good predictor of the lower 95% confidence interval on the mean model ecosystem NOEC.