Authors: Leo Posthuma, Dick de Zwart
Reviewers: Ad Ragas, Keith Solomon
Learning objectives:
You should be able to:
Keywords: Species Sensitivity Distribution (SSD), benchmark concentration, Potentially Affected Fraction of species (PAF)
Introduction
The relationship between dose or concentration (X) and response (Y) is key in risk assessment of chemicals (see section on Concentration-response relationships). Such relationships are often determined in laboratory toxicity tests; a selected species is exposed under controlled conditions to a series of increasing concentrations to determine endpoints such as the No Observed Effect Concentration (NOEC), the EC50 (the Effect Concentration causing 50% effect on a studied endpoint such as growth or reproduction), or the LC50 (the Effect Concentration causing 50% lethal effects). For ecological risk assessment, multiple species are typically tested to characterise the (variation in) sensitivities across species or taxonomic groups within the ecosystem. In the mid-1980s it had been observed that–like many natural phenomena–a set of ecotoxicity endpoint data, representing effect concentrations for various species, follows a bell-shaped statistical distribution. The cumulative distribution of these data is a sigmoid (S-shaped) curve. It was recognized, that this distribution had particular utility for assessing, managing and protecting environmental quality regarding chemicals. The bell-shaped distribution was thereupon named a Species Sensitivity Distribution (SSD). Since then, the use of SSD models has grown steadily. Currently, the model is used for various purposes, providing important information for decision-making.
Below, the dual utility of SSD models for environmental protection, assessment and management are shown first. Thereupon, the derivation and use of SSD models are elaborated in a stepwise sequence.
The dual utility of SSD models
A species sensitivity distribution (SSD) is a distribution describing the variance in sensitivity of multiple species exposed to a hazardous compound. The statistical distribution is often plotted using a log-scaled concentration axis (X), and a cumulative probability axis (Y, varying from 0 – 1; Figure 1).
Figure 1. An species-sensitivity distribution (SSD) model, its data, and its dual use (from YàX, and from XàY). Dots represent the ecotoxicity endpoints (e.g., NOECs, EC50s, etc.) of different species.
Figure 1 shows that different species (here the dots represent 3 test data for algal species, 2 data for invertebrate species and 2 data fish species) have different sensitivities to the studied chemical. First, the ecotoxicity data are collected, and log10-transformed. Second, the data set can be visually inspected by plotting the bell-shaped distribution of the log-transformed data; deviations of the expected bell-shape can be visually identified in this step. They may originate from causes such as a low number of data points or be indicative for a selective mode of action of the toxicant, such as a high sensitivity of insects to insecticides. Third, common statistical software for deriving the two parameters of the log-normal model (the mean and the standard deviation of the ecotoxicity data) can be applied, or the SSD can be described with a dedicated software tool such as ETX (see below), including a formal evaluation of the ‘goodness of fit’ of the model to the data. With the estimated parameters, the fitted model can be plotted, and this is often done in the intuitively attractive form of the S-shaped cumulative distribution. This curve then serves two purposes. First, the curve can be used to derive a so-called Hazardous Concentration on the X-axis: a benchmark concentration that can be used as regulatory criterion to protect the environment (YàX). That is, chemicals with different toxicities have different SSDs, with the more hazardous compounds plotted to the left of the less hazardous compounds. By selecting a protection level on the Y-axis–representing a certain fraction of species affected, e.g. 5%–one derives the compound-specific concentration standards. Second, one can derive the fraction of tested species probably affected at an ambient concentration (XàY), which can be measured or modelled. Both uses are popular in contemporary environmental protection, risk assessment, and management.
Step 1: Ecotoxicity data for the derivation of an SSD model
The SSD model for a chemical and an environmental compartment (e.g., surface water, soil or sediment) is derived based on pertinent ecotoxicity data. Those are typically extracted from scientific literature or ecotoxicity databases. Examples of such databases are the U.S. EPA’s Ecotox database, the European REACH data sets and the EnviroTox database which contains quality-evaluated studies. The researcher selects the chemical and the compartment of interest, and subsequently extracts all test data for the appropriate endpoint (e.g., ECx-values). The set of test data is tabulated and ranked from most to least sensitive. Multiple data for the same species are assessed for quality and only the best data are used. If there is > 1 toxicity value for a species after the selection process, the geometric mean value is commonly derived and used. A species should only be represented once in the SSD. Data are often available for frequently tested species, representing different taxonomic and/or trophic levels. A well-known triplet of species frequently tested is “Algae, Daphnids and Fish”, as this triplet is a requested minimum set for various regulations in the realm of chemical safety assessment (see section on Regulatory frameworks). For various compounds, the number of test data can be more than hundred, whilst for most compounds few data of acceptable quality may be available.
Step 2. The derivation and evaluation of an SSD model
Standard statistical software (a spreadsheet program) or a dedicated software model such as ETX can be used to derive an SSD from available data. Commonly, the fit of the model to the data set is checked to avoid misinterpretation. Misfit may be shown using common statistical testing (Goodness of Fit tests) or by visual inspection and ecological interpretation of the data points. That is, when a chemical specifically affects one group of species (e.g., insects having a high sensitivity for insecticides), the user may decide the derive an SSD model for specific groups of species. In doing so, the outcome will consist of two or more SSDs for a single compound (e.g., an SSDInsect and an SSDOther when the compound is an insecticide, whilst the SSDOther might be split further if appropriate). These may show a better goodness of fit of the model to the data, but – more importantly – they reflect the use of key knowledge of mode of action and biology prior to ‘blindly’ applying the model fit procedure.
Step 3a. The SSD model used for environmental protection
The oldest use of the SSD model is the derivation of reference levels such as the PNEC (YàX). That is, given the policy goal to fully protect ecosystems against adverse effects of chemical exposures (see Section on Ecosystem services and protection goals), the protective use is as follows. First, the user defines which ecotoxicity data are used. In the context of environmental protection, these have often been NOECs or low-effect levels (ECx, with low x, such as EC10) from chronic tests. This yields an SSD-NOEC or SSD-ECx. Then, the user selects a level of Y, that is: the maximum fraction of species for which the defined ecotoxicity endpoint (NOEC or ECx) may be exceeded, e.g., 0.05 (a fraction of 0.05 equals 5% of the species). Next, the user derives the Hazardous Concentration for 5% of the species (YàX). At the HC5, 5% of the species are exposed to concentrations greater than their NOEC, but–which is the obverse–95% of the species are exposed to concentration less than their NOEC. It is often assumed that the structural and functional integrity of ecosystems is sufficiently protected at the HC5 level if the SSD is based on NOECs. Therefore, many authorities use this level to derive regulatory PNECs (Predicted No Effect Concentration) or Environmental Quality Standards (EQS). The latter concepts are used as official reference levels in risk assessment, the first is the preferred abbreviation in the context of prospective chemical safety assessments, and the second is used in retrospective environmental quality assessment. Sometimes an extra assessment factor varying between 1 and 5 is applied to the HC5 to account for remaining uncertainties. Using SSDs for a set of compounds yields a set of HC5 values, which–in fact–represent a relative ranking of the chemicals by their potential to cause harm.
Step 3b.The SSD model used for environmental quality assessment
The SSD model also can be used to explore how much damage is caused by environmental pollution. In this case, a predicted or measured ambient concentration is used to derive a Potentially Affected Fraction of species (PAF). The fraction ranges from 0–1 but, in practice, it is often expressed as a percentage (e.g., “24% of the species is likely affected”). According to this approach, users often have monitored or modelled exposure data from various water bodies, or soil or sediment samples, so that they can evaluate whether any of the studied samples contain a concentration higher than the regulatory reference level (previous section) and, if so how many species are affected. Evidently, the user must clearly express what type of damage is quantified, as damage estimates based on an SSDNOEC or an SSDEC50 quantify the fractions of species affected beyond the no effect level and at the 50% effect level, respectively. This use of SSDs for a set of environmental samples yields a set of PAF values, which, in fact, represent a relative ranking of the pollution levels at the different sites in their potential to cause harm.
Practical uses of using SSD model outcomes
SSD model outcomes are used in various regulatory and practical contexts.
Today, these three forms of use of SSD models have an important role in the practice of environmental protection, assessment and management on the global scale, which relates to their intuitive meaning, their ease of use, and the availability of a vast number of ecotoxicity data in the global databases.