Ordination techniques for analysing response of biological communities to toxic stress in experimental ecosystems RENI~ P . A . V A N W l J N G A A R D E N 1., P A U L J. V A N D E N B R I N K 1, JAN H. OUDE VOSHAAR 2 and PETER LEEUWANGH 1 1DLO Winand Staring Centrefor Integrated Land, Soil and Water Research (SC-DLO), PO Box 125, 6700 A C Wageningen, The Netherlands 2DLO Agricultural Mathematical Group (GL W-DLO), PO Box 100, 6700A C Wageningen, The Netherlands
Received 1 July 1993; revised and accepted 16 April 1994
The ordination techniques principal component analysis (PCA) and redundancy analysis (RDA) are considered to be useful tools for evaluating community responses in experimental ecotoxicology. Concepts and interpretation of these techniques are summarized. Application of PCA and RDA is illustrated in a case study. In this study, the effects of a single application of the insecticide Dursban ® 4E (a.i. chlorpyrifos) on an aquatic macroinvertebrate community in microcosms were analysed. Four treatment (nominal chlorpyrifos concentration: 35 /tg 1-1) and four control microcosms were used. PCA visualized a change in species composition with time. Immediately after treatment, a major shift in species composition occurred in treated microcosms. RDA demonstrated that this shift was due to the treatment. RDA also showed that non-arthropods were generally insusceptible to chlorpyrifos; most arthropods were affected. Dynamics of separate taxa were visualized, giving indications of possible primary and secondary effects for these taxa. A Monte Carlo permutation test was used to decide whether treatment had a significant effect on the species composition and to investigate the state of recovery in time. In general, the RDA results gave an adequate condensation of detailed information on abundance and effects obtained by more conventional univariate statistical analysis for some individual taxa of the community. In combination with toxicity and ecological data, ordination techniques can provide insight into effects of toxic substances in complex biological communities.
Keywords:chlorpyrifos; ordination; macroinvertebrates; microcosms; Monte Carlo permutation. Introduction Freshwater field tests are frequently carded out to evaluate the effects of micropollutants in natural aquatic ecosystems. Ecosystems are complex in their biotic and abiotic composition and communities are influenced not only by the chemical stress factors introduced. The complexity of both the system and the various interacting stress factors makes it difficult to estimate the explicit effects of a micropollutant on ecosystem structure and function. Nowadays, it is common practice in both terrestrial and aquatic ecology to deduce the main factors influencing species composition in complex communi*To whom correspondence should be addressed. 0963-9292 © 1995 Chapman & Hall
van Wijngaarden, van den Brink, Oude Voshaar and Leeuwangh
ties by means of multivariate techniques (Gauch 1982; Ter Braak 1987a; Ludwig and Reynolds 1988). These techniques reveal intrinsic patterns in the data matrices of collected species numbers and sampling units. The procedure of arranging samples on the basis of data on species composition is called ordination (Ter Braak 1987b). In spite of the ability of ordination techniques to visualize the response of a community to an experimental treatment, these techniques are rarely used in ecotoxicological studies using micro- and mesocosms. The aim of the present paper is to show that the linear response ordination techniques principal component analysis (PCA) and redundancy analysis (RDA) are useful tools for evaluating species and community responses to micropollutants in micro- and mesocosm studies. First, both techniques are introduced. The explanation of the techniques is based on Ter Braak (1987b). A more detailed introduction to ordination can be found in Jongman et al. (1987). Next, the application of PCA and RDA is demonstrated by analysing responses of macroinvertebrate populations in an indoor microcosm experiment following a single application of the insecticide chlorpyrifos. For the sake of simplifying the explanation, we chose the relatively restricted data set of this experiment. However, we have had good results in using the techniques for analysing very large data sets in longlasting ecotoxicological mesocosm studies.
Ordination techniques Application of ordination in ecotoxicology seems attractive for several reasons. Ordination makes it possible to reveal environmental variables (e.g. treatment and time) affecting the species composition in the samples collected. Direct analysis of effects on separate species is often hampered by properties of the field data (low numbers, high variability, discontinuity), which is why more general patterns of coincidence of several species are sometimes of greater use in detecting relations between species and their environment. Ordination can give a more global description of effects at the community level and some of the techniques provide statistics to evaluate the significance of effects on the community. Ordination aims at producing plots in which sampling units with nearly identical species composition lie close together, while sampling units with very different species composition lie far apart. Linear versus unimodal
Ordination techniques are either based on a linear response model (principal components analysis, PCA) or on a unimodal response model (correspondence analysis, CA). In a linear response model, one assumes that the abundance of species either increases or decreases with the value of each environmental variable. This model is also the basis of multiple linear regression. In a unimodal response model, one assumes that the abundance of species rises and falls within a limited range of values of an environmental variable. In this model, the occurrence of each species follows a bell-shaped curve (a Gaussian curve). This curve is characterized by an optimum and a tolerance (t), indicating, respectively, the maximum and range of occurrence along a gradient of an environmental variable. Hence, t can be seen as a measure of the ecological amplitude of a species. In order to decide whether PCA or CA should be used, a third ordination technique,
Ordination in experimental ecotoxicology
Sampling range 1,5 SD
Sampling range 4 SD
Hypothetical gradient expressed in SD-units
Figure 1. Hypothetical diagram of the occurrence of species A-J over an environmental gradient. The length of the gradient is expressed in standard deviation units (SD units). Broken lines (A', C', H', J') describe fitted occurrences of species A, C, H and J respectively. If sampling takes place over a gradient range ~3 SD, occurrences of most species are best described by an unimodal model (H' and J').
detrended correspondence analysis (DCA) can be helpful. DCA allows the calculation of the length of the gradient of the environmental variables comprising all samples. In DCA, the mean square of the ts of the Gaussian curves of species is used as the unit for scaling the ordination axes. This unit is usually called the standard deviation (SD) (Hill and Gauch, 1980). Since a Gaussian curve with t = 1 rises and falls over a gradient length of approximately 4 SD, it can be expected that samples differing by 4 SD on an ordination axis hardly have any species in common (Fig. 1). If the length of the gradient is ~3 SD, species abundances are best described by an unimodal model (CA) (Fig. 2). Most species are then sampled in both the rising and the falling part of their abundance curves (Ter Braak and Prentice 1988) (Fig. 1). If the length of gradient lies in between 1.5 and 3 SD, two considerations can be helpful in choosing PCA over CA. In experimental ecotoxicology, experiments are generally done under similar environmental conditions. The only environmental variable showing major differences between experimental replicates will then be the micropollutant after treatment. Hence, PCA and derived techniques like R D A will most likely be the most appropriate ordination techniques in experimental ecotoxlcology. In addition, doseresponse relationships are expected to be monotonic in the case of primary effects of a
van Wijngaarden, van den Brink, Oude Voshaar and Leeuwangh
Linear ResponseModel ~
)uantified ~nvironmental ariables :ovariables
Figure 2. Outline of ordination techniques presented in this paper. DCA (detrended correspondence analysis) was applied for the determination of the length of gradient (LG). LG is important for choosing between ordination based on a linear or on an unimodal response model. Correspondence analysis (CA) is not considered any further because in the 'microcosm experiment discussed here LG was 20 gg l-i; A. aqualicus 96 h ECs0, 2.7 /~g 1-1 (Van Wijngaarden et al. 1993)). Proasellus coxalis is assumed to benefit from the decimation of the food competitors A. aqualicus and Gammarus pulex (Brock et al. 1992a,b). The juvenile Asellidae which could not be identified to species level were classified as one taxon. Classification above species level results in a loss of information at subordinate taxonomic levels. However, in the case of juvenile Asellidae, it can be deduced that the increase of this taxon in the treated microcosms was most probably due to P. coxalis, since A. aquaticus was no longer found in the treated microcosms (Brock et al. 1992a). Temporal variation in gastropod species (snails; hip com, phy acu, pop ant) and Tricladida (flatworms; dug lug) are also demonstrated by the R D A biplot (Fig. 5). To understand these shifts and relate them to possible (secondary) effects, one should look at the dynamics of the individual species. This information on the dynamics of species has to be placed in the context of available ecological knowledge of these species and has to be linked to available ecological information on other interacting taxa of the community. This has been done by Brock et al. (1992b) for this indoor microcosm study. In order to evaluate the ordination results, by way of example, we compared the R D A outcome with the results of three separate taxa described in Brock et al. (1992a,b). Projection of the centroids in the R D A biplot (Fig. 5) on the imaginary 'species line' of G. pulex (gam pul) indicates that the abundance of this species was more or less stable in the control microcosms throughout the entire experimental period. Abundance in the treatment microcosms after insecticide application was reduced for the entire posttreatment period. Table 2 shows that the number of G. pulex in the control microcosms was fairly stable over time, with highest numbers at 4 weeks post-treatment. Figure 5 indicates that the abundance of Tubificidae (tub dae) decreased with time. Table 2 confirms a decrease in the abundance of Tubificidae with time, the lowest numbers of these worms being found in the treated microcosms from 8 weeks post-treatment onwards. Potamopyrgus antipodarum (pop ant) increased with time, reaching its highest numbers in the control microcosms from week 6 onwards (Fig. 5). Table 2 confirms these highest numbers in the control microcosms. Hence, the ordinations gave a good condensation of the detailed information on the individual species in Table 2. The permutation test indicated that after 15 weeks the treated microcosms had not yet recovered (p-value still 0.02). However, the indoor microcosms are not very suitable for studying actual recovery, due to their isolated location. Conclusion
Gathering data on community structures can yield huge data sets, often characterized by great variability in the abundance of the taxa collected (see e.g. Table 1). Ordination techniques can produce an effective condensation of these data. PCA site plots are useful tools for visualizing differences in species composition between the samples collected. PCA species plots show species associated with the samples. ' With PCA, indirect inferences can be made about the influence of environmental variables - such as normal seasonal changes and treatment effects - on the species composition. A benefit of PCA is that it results in an overview and quantification (importance) of explained variation in samples. However, PCA does not specify the contribution of the treatment.
Ordination in experimental ecotoxicology
Further analysis of the treatment effect and its contribution to the explained variation in the samples can be achieved with R D A techniques. R D A site plots yield ordination results visualizing differences in species composition between the samples, now constrained to treatment as environmental variables. R D A species plots show species associated with the samples. As regards RDA, biplots have the practical advantage of combining species and treatment (and/or site) data in one plot explicitly revealing relations between species and treatment. Statistical evaluation of R D A ordination results can be done by Monte Carlo permutation testing. R D A is very similar to a multivariate analysis of variance (MANOVA). The combination of R D A and Monte Carlo permutation testing has the advantage over M A N O V A of being distribution free and of having no restrictive upper limit on the number of taxa that can be analysed simultaneously (Verdonschot and Ter Braak 1994). Within RDA, covariables can be introduced to compensate for systematic effects. Time can also be regarded as a systematic effect. If time is introduced as a covariable, R D A becomes a whole-period analysis. In that case, however, the outcome of the R D A could be a mixture of possibly undetected primary effects, recovery and/or secondary effects. As a consequence, relative susceptibilities of individual taxa cannot be unequivocally determined by this whole-period analysis. In addition to the Monte Carlo permutation testing for the evaluation of ordination results, ordination techniques have also been found to be useful tools for the evaluation of community responses in combination with classical statistics. Since biological communities are often diverse, ordination techniques may reveal those taxa for which it is worthwhile to perform univariate statistics (e.g. ANOVA, t-tests, etc.) (Van Breukelen and Brock 1993). Microcosm studies show that PCA and R D A can provide a clear overview of temporal and treatment effects on the community structure of experimental freshwater ecosystems. In combination with available toxicity and ecological data, these techniques may reveal the important taxa that indicate pesticide stress in the often species-rich community.
Acknowledgements The study was supported by the Dutch Ministry of Education and Science's Program Committee for Toxicological Research (STO). Dursban ® 4E was provided by DowElanco Europe. We thank T.C.M. Brock for providing us the data of his microcosm study. We are indebted to C.J.F. ter Braak (DLO-Agricultural Mathematical Group, Wageningen), T.C.M. Brock (SC-DLO) and H. van Weerd (Agricultural University Wageningen) for their valuable discussions and comments on earlier versions of this manuscript.
References Brock, T.C.M., Crum, S.J.H., Van Wijngaarden, R., Budde, B.J., Tijink, J., Zupelli, A. and Leeuwangh, P. (1992a) Fate and effects of the insecticide Dursban ® 4E in indoor Elodeadominated and macrophyte-free freshwater model ecosystems. I. Fate and primary effects of the active ingredient chlorpyrifos. Arch. Environ. Contam. Toxicol. 23, 69-84.
van Wi]ngaarden, van den Brink, Oude Voshaar and Leeuwangh
Brock, T.C.M., Van den Bogaert, M., Bos, A.R., Van Breukelen, S.W.F., Reiche, R., Terwoert, J., Suykerbuyk, R.E.M. and Roijackers, R.M.M. (1992b) Fate and effects of the insecticide Dursban®4E in indoor Elodea-dominated and macrophyte-free freshwater model ecosystems. II. Secondary effects on community structure. Arch. Environ. Contain. Toxicol. 23,391-409. Gauch, H.G. (1982) Multivariate Analysis in Community Ecology. Cambridge: Cambridge University Press. Gittins, R. (1985) Canonical Analysis. A Review with Applications in Ecology. Berlin: SpringerVerlag. Hill, M.O. and Gauch, H.G., Jr (1980) Detrended correspondence analysis: an improved ordination technique. Vegetatio 42, 47-58. Jongrnan, R.G.H., Ter Braak, C.J.F. and Van Tongeren, O.F.R. (eds) (1987) Data Analysis in Community and Landscape Ecology. Wageningen: Pudoc. Leeuwangh, P. (1994) Comparison of chlorpyrifos fate and effects in outdoor micro- and microcosms of various scale and construction. In Hill, I.A., Heimbach, F., Leeuwangh, P. and Matthiesen, P. eds. Freshwater Field Tests for Hazard Assessment of Chemicals, pp. 217-48. Michigan: Lewis Publishers. Ludwig, J.A. and Reynolds, J.F. (1988) Statistical Ecology: A Primer on Methods and Computing. New York: John Wiley & Sons. Marshall, W.K. and Roberts, J.R. (1978) Ecotoxicology of Chlorpyrifos. Ottawa: National Research Council. Nelson, J.H. and Evans, E.S. (1973) Field Evaluations of the Larvicidal Effectiveness. Effects on Non-target Species and Environmental Residues of a Slow-release Polymer Formulation of Chlorpyrifos. Aberdeen Proving Ground, MD: US Army Environmental Hygiene Agency. Odenkirchen, E.W. and Eisler, R. (1988) Chlorpyrifos Hazards to Fish, Wildlife, and Invertebrates: A Synoptic Review. Laurel: US Fish. Wildl. Serv. Biol. Siefert, R.E., Brazner, J.C., Lozano, S.J., Knuth, M.L., Heinis, L.J., Jensen, D.A., Sargent, K.W., O'Halloran, S.L., Bertelsen, S.L., Tanner, D.K. and Kline, E.R. (1987) The Effects of Chlorpyrifos on a Natural Aquatic System: A Research Design for Littoral Enclosure Studies and Preliminary Data Report, Duluth, MN: US Environmental Protection Agency, Environmental Research Laboratory. Ter Braak, C.J.F. (1987a) Unimodal models to relate species to environment. Thesis, GLWAgricultural Mathematics Group, Wageningen. Ter Braak, C.J.F. (1987b) Ordination. In Jongman, R.G.H., Ter Braak, C.J.F. and Van Tongeren, O.F.R. eds. Data analysis in community and landscape ecology, pp. 91-173. Wageningen: Pudoc. Ter Braak, C.J.F. (1988) CANOCO - A FORTRAN Program for Canocical Community Ordination by [Partial] [Detrended] [Canocical] Correspondence Analysis, Principal Components Analysis and Redundancy Analysis (Version 2.1). Wageningen: Technical Report, LWA-8802, GLW-Agricultural Mathemathics Group. Ter Braak, C.J.F. (1990) Update Notes: CANOCO Version 3.10. Wageningen: GLW-Agriculturai Mathemathics Group. Ter Braak, C.J.F. (1994) Canonical community ordination. Part 1: Basic theory and linear methods. Ecoscience, 1 (2). Ter Braak, C.J.F. and Prentice, I.C. (1988) A theory of gradient analysis. Adv. Ecol. Res. 18, 271-317. Van Breukelen, S.W.F. and Brock, T.C.M. (1993) Response of a macroinvertebrate community to insecticide application in replicated freshwater microcosms with emphasis on the use of principal component analysis. Sci. Total Environ. Suppl. part 2, 1047-58. Van Wijngaarden, R., Leeuwangh, P., Lucassen, W.G.H., Romijn, K., Ronday, R., Van der Velde, R. and Willigenburg, W. (1993) Acute toxicity of chlorpyrifos to fish, a newt, and aquatic invertebrates. Bull. Environ. Contain. Toxicol. 51,716-23.
Ordination in experimental ecotoxicology
Verdonschot, P. and Ter Braak, C.J.F. (1994) An experimental manipulation of oligochaete communities in mesocosms treated with chlorpyrifos or nutrient additions: multivariate analysis with Monte Carlo permutation tests. Hydrobiologia, 278, 251-66.