Authors: Leo Posthuma, Dick de Zwart
Reviewers: Allan Burton, Ad Ragas
Learning objectives:
You should be able to:
Keywords: eco-epidemiology, mixture pollution, diagnosis, impact magnitude, probable causes, validation
Introduction
Approaches for environmental protection, assessment and management differ between ‘classical’ stressors (such as excess nutrients and pH) and chemical pollution. For the ‘classical’ environmental stress factors, ecologists use monitoring data to develop concepts and methods to prevent and reduce impacts. Although there are some clear-cut examples of chemical pollution impacts [e.g., the decline in vulture populations in South East Asia due to diclofenac (Oaks et al. 2004), and the suit of examples in the book ‘Silent Spring’ (Carson 1962)], ecotoxicologists commonly have assessed the stress from chemical pollution by evaluating exposures vis a vis laboratory toxicity data. Current pollution often consists of complex mixtures of chemicals, with highly variable patterns in space and time. This poses problems when one wants to evaluate whether observed impacts in ecosystems can be attributed to chemicals or their mixtures. Eco-epidemiological methods have been established to discern such pollution stress. These methods provide the diagnostic tools to identify the impact magnitude and key chemicals that cause impacts in ecosystems. The use of these methods is further relevant for validating the laboratory-based risk assessment approaches developed by ecotoxicology.
The origins of eco-epidemiology
Risk assessments of chemicals provide insights in expected exposures and impacts, commonly for separate chemicals. These are predictive outcomes with a high relevance for decision making on environmental protection and management. The validation of those risk assessments is key to avoid wrong protection and management decisions, but it is complex. It consists of comparing predicted risk levels to observed effects. This begs the question on how to discern effects of chemical pollution in the field. This question can be answered based on the principles of ecological bio-assessments combined with those of human epidemiology. A bio-assessment is a study of stressors and ecosystem attributes, made to delineate causes of impacts via (often) statistical associations between biotic responses and particular stressors. Epidemiology is defined as the study of the distribution and causation of health and disease conditions in specified populations. Applied epidemiology serves as a scientific basis to help counteracting the spreading of human health problems. Dr. John Snow is often referred to as the ‘father of epidemiology’. Based on observations on the incidence, locations and timings of the 1854 cholera outbreak in London, he attributed the disease to contaminated water taken from the Broad Street pump well, counteracting the prevailing idea that the disease was caused by transmission via air. His proposals to control the disease were effective. Likewise, eco-epidemiology – in its ecotoxicological context – has been defined as the study of the distribution and causation of impacts of multiple stressor exposures in ecosystems. In its applied form, it supports the reduction of ecological impacts of chemical pollution. Human-health eco-epidemiology is concerned with environment-mediated disease.
The first literature mention of eco-epidemiological analyses on chemical pollution stems from 1984 (Bro-Rasmussen and Løkke 1984). Those authors described eco-epidemiology as a discipline necessary to validate the risk assessment models and approaches of ecotoxicology. In its initial years, progress in eco-epidemiological research was slow due to practical constraints such as a lack of monitoring data, computational capacity and epidemiological techniques.
Current eco-epidemiology
Current eco-epidemiological studies in ecotoxicology aim to diagnose the impacts of chemical pollution in ecosystems, and utilize a combination of approaches in order to diagnose the role of chemical mixtures in causing ecological impacts in the field. The combination of approaches consists of:
1. Collection of monitoring data on abiotic characteristics and the occurrence and/or abundance of biotic species, for the environmental compartment under study;
2. If needed: data optimization, usually to align abiotic and biotic monitoring data, including the chemicals;
3. Statistical analysis of the data set using eco-epidemiological techniques to delineate impacts and probable causes, according to the approaches followed in ‘classical’ ecological bio-assessments;
4. Interpretation and use of the outcomes for either validation of ecotoxicological models and approaches, or for control of the impacts sensu Dr. Snow.
Key examples of chemical effects in nature
Although impacts of chemicals in the environment were known before 1962, Rachel Carson’s book Silent Spring (see Section on the history of Environmental toxicology) can be seen as early and comprehensive eco-epidemiological study that synthesized the available information of impacts of chemicals in ecosystems. She considered effects of chemicals a novel force in natural selection when she wrote: “If Darwin were alive today the insect world would delight and astound him with its impressive verification of his theories of survival of the fittest. Under the stress of intensive chemical spraying the weaker members of the insect populations are being weeded out.”
Clear examples of chemical impacts on species are still reported. Amongst the best-known examples is a study on vultures. The population of Indian vultures declined more than 95% due to diclofenac exposure which was used intensively as a veterinary drug (Oaks et al. 2004). The analysis of chemical impacts in nature becomes however more complex over time. The diversity of chemicals produced and used has vastly increased, and environmental samples contain thousands of chemicals at often low concentrations. Hence, contemporary eco-epidemiology is complex. Nonetheless, various studies demonstrated that contemporary mixture exposures affect species assemblages. Starting from large-scale monitoring data and following the four steps mentioned above, De Zwart et al. (2006) were able to show that effects on fish species assemblages could be attributed to both habitat characteristics and chemical mixtures. Kapo and Burton Jr (2006) showed the impacts of multiple stressors and chemical mixtures in aquatic species assemblages with similar types of data, but slightly different techniques. Eco-epidemiological studies of the effects of chemicals and their mixtures currently represent different geographies, species groups, stressors and chemicals/mixtures that are considered. The potential utility eco-epidemiological studies was reviewed by Posthuma et al. (2016). The review showed that mixture impacts occur, and that they can be separated from natural variability and multiple-stressor impacts. That means that water managers can develop management plans to counteract stressor impacts. Thereby, the study outcomes are used to prioritize management to sites that are most affected, and to chemicals that contribute most to those effects. Based on sophisticated statistical analyses, Berger et al. (2016) suggested chemicals can induce effects in the environment at concentrations much lower than expected based on laboratory experiments. Schäfer et al. (2016) argued that eco-epidemiological studies that cover both mixtures and other stressors are essential for environmental quality assessment and management. In practice, however, the analysis of the potential impacts of chemical mixtures is often still separate from the analysis of impacts of other stressors.
Steps in eco-epidemiological analysis
Various regulations require collection of monitoring data, followed by bio-assessment, such as the EU Water Framework Directive (see section on the Water Framework Directive). Therefore, monitoring data sets are increasingly available. The data set is subsequently curated and/or optimized for the analyses. Data curation and management steps imply amongst others that taxonomic names of species are harmonized, and that metrics for abiotic and biotic variables represent the conditions for the same place and time as much as possible. Next, the data set is expanded with novel variables, e.g. a metric for the toxic pressure exerted by chemical mixtures. An example of such a metric is the multi-substance Potentially Affected Fraction of species (msPAF). This metric transfers measured or predicted concentrations into the Potentially Affected Fraction of species (PAF), the values of which are then aggregated for a total mixture (De Zwart and Posthuma 2005). This is crucial, as adding each chemical of interest as a separate variable implies an increasingly expanding number of required sampling sites to maintain statistical power to diagnose impacts and probable causation.
The interpretation of the outcomes of the statistical analyses of the data set is the final step. Here, it must be acknowledged that statistical association is not equal to causation, and that care must be taken to explain the findings as indicative for mixture effects. Depending on the context of the study, this may then trigger a refined assessment, or alignment with other methods to collect evidence, or a direct use in an environmental management program.
Eco-epidemiological methods
A very basic eco-epidemiological method is quantile regression. Whereas common regression methods explore the magnitude of the change of the mean of the response variable (e.g., biodiversity) in relation to a predictor variable (e.g., pollutant stress), the quantile regression looks at the tails of the distributions of the response variable. How this principle operates is illustrated in Figure 1. When a monitoring data set contains one stressor variable at different levels (i.e., a gradient of data), the observations typically take the shape of a common stressor-response relationship (see section on Concentration-effect relationships). If the monitoring sites are affected by an extra stressor, the maximum-performance under the first stressor cannot be reached, so that the area under the curve contains the XY-points for this situation. Further addition of stressor variables and levels fills this space under the curve. When the raw data plotted as XY show an ‘empty area’ lacking XY-points, e.g. in the upper right corner, it is likely that the stressor variable can be identified as a stressor that limits the response variable, for example: chemicals limit biodiversity. The quantile regression calculates an upper percentiles (e.g., the 95th percentile) of the Y-values in assigned subgroups of X-values (“bins”). Such a procedure yields a picture such as Figure 1.
Figure 1. The principle of quantile regression in identification of a predictor variable (= stressor) that acts as a limiting factor to a response variable (= performance). It is common to derive e.g. the 95th percentile of the Y values in a ‘bin’ of X values to derive a stressor-impact curve. As illustration, the 95th percentile is marked only for the first bin of X values, with the blackened star.
More complex methods for analysis of (bio)monitoring data have been developed and applied. The methods are closely associated to those developed for, and utilized in, applied ecology. Well-known examples are ‘species distribution models’ (SDM), which are used to describe the abundance or presence of species as a function of multiple environmental variables. A well-known SDM is the bell-shaped curve relating species abundances to water pH: numbers of individuals of a species are commonly low at low and high pH, and the SDM is characterized as an optimum model for species abundance (Y) versus pH (X). Statistical models can also describe species abundance, presence or biodiversity, as a function of multiple stressors, for example via Generalized Linear Models. These have the general shape of:
Log(Abundance)= (a. pH + a’ pH2) + (b. OM + b’ OM2) + …… + e,
with a, a’, b and b’ being estimated from fitting the model to the data, whilst pH and OM are the abiotic stressor variables (acidity and Organic Matter, respectively); the quadratic terms are added to allow for optimum and minimum shaped relationships. When SSD models (see Section on Species Sensitivity Distribution) are used to predict the multi-substance Potentially Affected Fraction of species, the resulting mixture stress proxy can be analysed together with the other stressor variables. Data analyses from monitoring data from the United States and the Netherlands have, for example, shown that the abundance of >60% of the taxa is co-affected by mixtures of chemicals. An example study is provided by Posthuma et al. (2016).
Prospective mixture impact assessments
In addition to the retrospective analysis of monitoring data, in search of chemical impacts, recent studies also show examples of prospective studies of effects of mixtures. Different land uses imply different chemical use patterns, summarized as ‘signatures’. That is, agricultural land use will yield intermittent emissions of crop-specific plant protection products, aligning with the growing season. Emissions from populated areas will show continuous emission of household chemicals and discontinuous emissions of chemicals in street run-off associated to heavy rain events. The application of emission, fate and ecotoxicity models showed that aquatic ecosystems are subject to the ‘signatures’, with associated predicted impact magnitudes (Holmes et al. 2018; Posthuma et al. 2018). Although such prospective assessments did not yet prove ecological impacts, they can assist in avoiding impacts by preventing the emission ‘signatures’ that are identified as potentially most hazardous.
The use of eco-epidemiological output
Eco-epidemiological analysis outputs serve two purposes, closely related to prospective and retrospective risk assessment of chemical pollution:
1. Validation of ecotoxicological models and approaches;
2. Derivation of control measures, to reduce impacts of diagnosed probable causes of impacts.
If needed, multiple lines of evidence can be combined, such as in the Triad approach (see section on TRIAD) or approaches that consider more than three lines of evidence (Chapman and Hollert, 2006). The higher the importance of a good diagnosis, the better the user may rely on multiple lines of evidence.
First, the validation of ecotoxicological models and approaches is crucial, to avoid that important environmental protection, assessment and management activities rely on approaches that have limited relationship to field effects. Eco-epidemiological analyses have, for example, been used to validate the protective benchmarks used in the chemical-oriented environmental policies.
Second, the outcomes of an eco-epidemiological analysis can be used to control causes of impacts to ecosystems. Some studies have, for example, identified a statistical association between observed impacts (species expected but absent) and pollution of surface waters with mixtures of metals. Though local experts first doubted this association due to lack of industrial activities with metals in the area, they later found the association relevant given the presence of old spoil heaps from past mining activities. Metals appeared to leach into the surface waters at low rates, but the leached mixtures appeared to co-vary with species missing (De Zwart et al. 2006).
References
Berger, E., Haase, P., Oetken, M., Sundermann, A. (2016). Field data reveal low critical chemical concentrations for river benthic invertebrates. Science of The Total Environment 544, 864-873.
Bro-Rasmussen, F., Løkke, H. (1984). Ecoepidemiology - a casuistic discipline describing ecological disturbances and damages in relation to their specific causes; exemplified by chlorinated phenols and chlorophenoxy acids. Regulatory Toxicology and Pharmacology 4, 391-399.
Carson, R. (1962). Silent spring. Boston, Houghton Mifflin.
Chapman, P.M., Hollert, H. (2006). Should the sediment quality triad become a tetrad, a pentad, or possibly even a hexad? Journal of Soils and Sediments 6, 4-8.
De Zwart, D., Dyer, S.D., Posthuma, L., Hawkins, C.P. (2006). Predictive models attribute effects on fish assemblages to toxicity and habitat alteration. Ecological Applications 16, 1295-1310.
De Zwart, D., Dyer, S.D., Posthuma, L., Hawkins, C.P. (2006). Use of predictive models to attribute potential effects of mixture toxicity and habitat alteration on the biological condition of fish assemblages. Ecological Applications 16, 1295-1310.
De Zwart, D., Posthuma, L. (2005). Complex mixture toxicity for single and multiple species: Proposed methodologies. Environmental Toxicology and Chemistry 24,: 2665-2676.
Holmes, C.M., Brown, C.D., Hamer, M., Jones, R., Maltby, L., Posthuma, L., Silberhorn, E., Teeter, J.S., Warne, M.S.J., Weltje, L. (2018). Prospective aquatic risk assessment for chemical mixtures in agricultural landscapes. Environmental Toxicology and Chemistry 37, 674-689.
Kapo, K.E., Burton Jr, G.A. (2006). A geographic information systems-based, weights-of-evidence approach for diagnosing aquatic ecosystem impairment. Environmental Toxicology and Chemistry 25, 2237-2249.
Oaks, J.L., Gilbert, M., Virani, M.Z., Watson, R.T., Meteyer, C.U., Rideout, B.A., Shivaprasad, H.L., Ahmed, S., Chaudhry, M.J., Arshad, M., Mahmood, S., Ali, A., Khan, A.A. (2004). Diclofenac residues as the cause of vulture population decline in Pakistan. Nature 427(6975), 630-633.
Posthuma, L., Brown, C.D., de Zwart, D., Diamond, J., Dyer, S.D., Holmes, C.M., Marshall, S., Burton, G.A. (2018). Prospective mixture risk assessment and management prioritizations for river catchments with diverse land uses. Environmental Toxicology and Chemistry 37, 715-728.
Posthuma, L., De Zwart, D., Keijzers, R., Postma, J. (2016). Water systems analysis with the ecological key factor 'toxicity'. Part 2. Calibration. Toxic pressure and ecological effects on macrofauna in the Netherlands. Amersfoort, the Netherlands, STOWA.
Posthuma, L., Dyer, S.D., de Zwart, D., Kapo, K., Holmes, C.M., Burton Jr, G.A. (2016). Eco-epidemiology of aquatic ecosystems: Separating chemicals from multiple stressors. Science of The Total Environment 573, 1303-1319.
Posthuma, L., Suter, II, G.W., Traas, T.P. (Eds.) (2002). Species Sensitivity Distributions in Ecotoxicology. Boca Raton, FL, USA, Lewis Publishers.
Schäfer, R.B., Kühn, B., Malaj, E., König, A., Gergs, R. (2016). Contribution of organic toxicants to multiple stress in river ecosystems. Freshwater Biology 61, 2116–2128