


BioSafety Journal, Volume 7, Paper 1 (BY02001) 2002 Christian Damgaard Department of Terrestrial Ecology, DMU, Vejlsøvej 25, 8600 Silkeborg, Denmark. Code Number: BY02001 Received: January 30^{th} 2002 An operative and quantitative methodology for assessing the ecological risk of a GMP invading a natural habitat is presented. The suggested method evaluates whether dispersal to a specific habitat may take place and calculates the predicted probabilities of establishment, using information on dispersal and Bayesian statistics on data from simple competition experiments. Additionally, the method allows evaluation of the probability that a native species goes locally extinct in the invasion process. Keywords: Genetically modified plants, GMP, GMO, Invasion probability, risk assessment. Introduction of new species or genotypes to an area may be an irreversible process. If a plant is introduced into a natural habitat it may be impossible in practice to remove it again. Therefore, it is important to make an assessment of possible ecological risks of genetically modified plants (GMP) before they are released for commercial growing. Here, the particular ecological risk of a GMP invading a natural habitat and possibly outcompeting a native plant species will be discussed. In risk assessment, a risk normally is defined as the probability that a specific event will occur multiplied by the negative impact of the event (Risk = Probability x Impact) (Vose, 2000, Damgaard and Løkke, 2001) . Thus, a risk assessment is separated in two parts; quantification of the probability that a specific event will occur, and quantification of the negative impact of the event. The negative impact of a specific ecological scenario, e.g. extinction of a native species, has to be quantified by the public or the decisionmakers who will experience the reduced biodiversity. Note that humans and not ecosystems perceive an ecological risk. If the public is not concerned whether a specific species (including its unique role in the functioning of the ecosystem) goes extinct, then the negative impact and consequently the risk is zero. Ecological systems are complex, so in the ecological risk assessment we have to concentrate on the most important factors that control the invasion of plants into a specific habitat. The term invasion is used in different ways in the ecological literature. Normally, the invasion process is thought of as a series of steps from the establishment phase (the species is present), to the naturalisation phase (the population can maintain itself), and finally the invasion phase (the species can spread to other areas) (e.g., Williamson, 1996). However, if we aim at quantifying the invasion process we have to be more precise than merely to speak of a capacity to spread to other areas, since the invasion probability will depend on the plant community structure at the area. It is useful to define the invasion process for specific habitats as a local increase in plant density from nonexistent to a density where the population can maintain itself (e.g., Bolker, et. al., 2000) . The probability that a GMP invades a specific natural habitat may be defined as the probability that the GMP migrates to the habitat times the probability that it will survive, reproduce and increase in numbers to some quasiequilibrium density (P_{invasion} = P_{immigration }X_{ }P_{local increase in density}). Here I will briefly discuss an operative methodology to quantify the invasion probability of a GMP using information on dispersal and data from simple competition experiments. Additionally, the method allows evaluation of the probability that a native species goes locally extinct in the invasion process. Plants depend on seed or vegetative dispersal in order to invade a new habitat. The new habitat may be regarded as either patchy or continuous on the scale of dispersal. If a plant should invade a discontinuous patchy habitat, it must rely on rare longrange seed dispersal events to get established. If a plant species invades a more or less continuous habitat, then the invasion front may be modelled by a diffusion equation approach using information of the dispersal kernel and the competitive interactions (Cruywagen, et. al., 1996, Shigesada and Kawasaki, 1997). The effect of secondary foci after rare longdistance dispersal events may also be included. Ideally, we would like to estimate the probability that a specific GMP immigrates into a specific habitat. However, since longrange dispersals are rare events, it is very difficult to make dispersal experiments such that the estimated dispersal kernel may be used to predict dispersal at long distances (Bullock and Clarke, 2000). It is only possible to conclude that less than a certain percentage of the seeds disperse more than a certain distance, but since a single seed is sufficient to initiate establishment of the species, this information is not sufficient for calculating the immigration probability. In some cases when only limited vegetative dispersal is possible, or if it has been observed that an invading plant species cannot invade an island or an isolated habitat, it may be possible to deduce a maximum dispersal distance. Such information on a maximum dispersal distance may be used to exclude the possibility of immigration to a specific habitat. However, in most cases, a GMP will be grown over a large area and be subjected to massive humanaided dispersal via agricultural machines etc., so that the GMP most likely will be capable of dispersing to most natural habitats. Thus in our attempt to quantify the immigration probability of a specific habitat we are most often left with only two possibilities; either to assume that immigration will take place at some point in the future (P_{immigration} = 1), or to reject the possibility that a specific GMP immigrates into a specific habitat (P_{immigration} = 0). In some case it may be possible to make valuable quantitative accounts on the probability of immigration as a function of the time since the first release of the GMP. When a GMP immigrates into a new habitat it will not be alone. Typically, the habitat will be occupied by a plant community that may or may not be in successional equilibrium. The probability of a local increase in density of a GMP in the habitat depends on abiotic factors (soil and nutritional conditions, water availability, climate) and biotic factors, such as demographic stochasticity (Lande, 1993), the competitive interaction between the GMP and the other plants in the plant community (Goldberg and Barton, 1992, Rees, et. al., 1996, Rees, et. al., 2001), as well as the spatial setting of the individual plants (Bolker and Pacala, 1999). The probability of a local increase in density may be calculated by estimating mortality and fecundity of GMP plants in a natural habitat (Crawley, et. al., 1993). However, the probability of germination, establishment, and reaching reproductive age in a natural habitat is a critical and variable factor in predicting population growth. Crawley et al. (1993) compared estimates of population growth of transgenic rapeseed (Brassica napus) in natural habitats obtained by either using independent information on seed germination, mortality, and fecundity, or measuring population growth directly by the difference in the number of seedlings from year to year. They obtained qualitatively different estimates of population growth by the two approaches. When the independently obtained information on mortality and fecundity was used, the rapeseed population was predicted to increase in density. However, the directly obtained estimate of population growth was less than one, i.e. the rapeseed population was by this method found to decrease in density. The probability of germination, establishment, and reaching reproductive age is expected to vary from year to year and from habitat to habitat. Thus, in order to estimate population growth of a GMP in a natural habitat directly, we need observations over more years in several habitats. Alternatively, and as a first step in the ecological risk assessment let us assume that the ecological success of an immigrant transgene GMP depends on the interspecific competitive interactions between other species, and furthermore, that P_{local increase in density} in a given abiotic environment mainly is determined by the competitive interactions with one or two key plant species in the plant community. The key plant species has to be selected based on prior knowledge of the ecosystem and may belong to the same functional group (Weiher, et. al., 1998, Symstad, 2000) as the GMP. To simplify the risk assessment, it is assumed that the immigrating GMP is above a threshold density so that extinction due to demographic stochasticity is excluded, and that the plants are randomly distributed in the environment (these last two assumptions can be loosened if relevant, e.g., Lande, 1993, Cruywagen, et al., 1996, Bolker and Pacala, 1999). Alternatively, if it is difficult to select one or two key species in the habitat, the competitive ability of the transgenic GMP may be compared to an untransformed genotype. Under these strong assumptions P_{local increase in density} may be predicted from simple two or three species competition experiments using a discrete hyperbolic competition model and Bayesian statistics (Damgaard, 1998). When two species compete against each other there are four possible ecological scenarios: 1) the two species coexist, 2) the first species outcompete the second species, 3) the second species outcompete the first species, and 4) either of the two species outcompete the other depending upon the initial conditions (the rare species is outcompeted). Since P_{local increase in density} was defined as the probability that an initially rare species will increase in numbers to some quasiequilibrium frequency, this is equal to the probability of coexistence plus the probability that the invading species outcompetes the resident species. However, if the immigration rate is high, e.g. when the habitat is next to a field with the GMP, or if there is significant withinspecies spatial covariance , then the GMP may not be assumed to be initially rare. In such cases the GMP may also invade the habitat under the fourth ecological scenario where the rare species is outcompeted. Note that the negative impact of the invasion process probably is regarded as higher if a native species goes locally extinct, and the risk assessment can then be differentiated in the cases where the GMP coexist with the native plant species and where the GMP outcompetes the native plant species. The Bayesian posterior probabilities of the four different ecological scenarios calculated from an Avena fatua and Avena barbata competition experiment are summarised in Table 1. The predicted most likely longterm ecological scenario was that Avena fatua would outcompete Avena barbata. The two scenarios, that the two species would either coexist, or that the most common species would outcompete the rarer species, were almost equally likely to occur based on the competition experiment, whereas it was predicted to be unlikely that Avena barbata would outcompete Avena fatua. Note that the calculated posterior probabilities are relatively independent of the probability of germination, establishment, and reaching reproductive age as long as they have the same relative magnitude (Table 1). Consequently, the predictions based on the competitive interactions between the GMP and a key species may therefore be less variable from year to year than estimating the population growth of the GMP directly. Table 1. Bayesian posterior probabilities of the four different ecological scenarios when Avena fatua and Avena barbata compete against each other . A discrete hyperbolic competition model may be used to model the competitive interaction between the two plant species: Fig.2 and Fig.3, where Y_{i} is the fecundity of species i, X_{i} is the density of species i, 1/a_{i} is a measure of the fecundity of species i at low density, 1/b_{i} is a measure of the fecundity of species i per area at high density in a monoculture, and c_{i} is the estimated competition coefficient of species i . The Bayesian posterior probabilities of the four possible ecological scenarios may be calculated from a competition experiment after a parameter transformation. The competition coefficient, c_{i}, was expressed by (b_{i}(p_{j}  a_{j})/b_{j}(p_{i}  a_{i})) + ð_{i}, where the index j signifies the alternate species, p_{i} is the probability that a seed germinates and establishes at the site where it lands. The sign of ð_{i }determines the longterm ecological scenario. In the present case, the probability that a seed germinates at the site where it lands, p_{i}, was assumed to be equal for the two species. The method of calculating the posterior probabilities is explained in more detail in Damgaard (1998).
The species distribution of Avena fatua and Avena barbata was observed in a number of natural plant communities in two regions in California (Marshall and Jain, 1969). Interestingly, the predicted probabilities agree quite well with the observed distribution of the two plant species in the Mediterranean warm summer region, whereas the observed competitive interactions do not explain the observed plant distribution in the Mediterranean cool summer region (Fig. 1). Fig. 1 Number of plant communities shown as a function of the relative abundance of A. barbata in Mediterranean warm summer region (black) and Mediterranean cool summer region (grey). It is an open question how well the predicted establishment probability obtained from a competition experiment will reflect an actual invasion process in a natural plant community (Kareiva, et. al., 1996) and especially so for perennial GMP. The realism will to a large extent be determined by the design of the competition experiment, since the predicted probability will never be better than the competition experiment. If a key factor determining the competitive output between two plant species is not included in the experiment, then the predicted probabilities most likely will be erratic and misleading. Depending upon the predicted probabilities integrated with "ecological common sense" into an expert opinion and the expected impact of the possible invasion the GMP may either be placed on the market or rejected. If the described first step in the ecological risk assessment leaves us uncertain about the invasion probability, i.e., the probability was not either sufficiently high nor sufficiently low to decide on the invasion probability with reasonable certainty, then the next step in the ecological risk assessment should be a full lifecycle analysis in relevant natural habitats (Kjær, et. al., 1999). It will be foolish to claim that the predicted invasion probabilities obtained from a simple shortterm competition experiment will closely mimic the future dynamics of a plant community. Ecological predictions will always have to be taken with some scepticism. In the words of Kareiva et al. (1996): "Indeed, we have so little faith in models and shortterm experiments regarding predictions about invasions, that we advocate extensive monitoring of any introduced genetically engineered organisms with ecologically relevant traits (such as disease resistance, herbivore tolerance, and so forth)". Nevertheless, I believe that quantitative ecological predictions are useful in the assessment of the risk of invasion by a GMP. The most invasive GMP will probably be revealed by such an ecological risk assessment, and actively using models and experiments in the ecological risk assessment will allow us to explore the limitations of our methodology and improve it. Acknowledgement Thanks to Marianne Erneberg, Gösta Kjellsson and Morten Strandberg for commenting on an earlier version of the manuscript. References Bolker B.M. and Pacala S.W. 1999. Spatial moment equations for plant competition:Understanding spatial strategies and the advantages of short dispersal. American Naturalist 153: 575602 Bolker B.M., Pacala S.W. and Levin S.A. 2000. Moment methods for ecological processes in continuous space. In The geometry of ecological interactions: Simplifying spatial complexity, ed. U Dieckmann, R Law, JAJ Metz. Cambridge: Cambridge University Press Bullock J.M. and Clarke R.T. 2000. Long distance seed dispersal by wind: measuring and modelling the tail of the curve. Oecologia 124: 50621 Crawley M., Hails R.S., Rees M., Kohn D. and Buxton J. 1993. Ecology of transgenic oilseed rape in natural habitats. Nature 363: 6202 Cruywagen G.C., Caregiver P., Lewis M.A. and Murray J.D. 1996. Competition in a spatially heterogeneous environment: Modelling the risk of spread of a genetically engineered population. Theoretical Population Biology 49: 138 Damgaard C. 1998. Plant competition experiments: Testing hypotheses and estimating the probability of coexistence. Ecology 79: 17607 Damgaard C. and Løkke H. 2001. A critique of the "concept of familiarity" as used in the ecological risk assessment of genetically modified plants. BioSafety Volume 6: Paper 1 (BY01001) Online Journal  URL: http://www.bioline.org.br/by Firbank L.G. and Watkinson A.R. 1985. On the analysis of competition within twospecies mixtures of plants. Journal of Applied Ecology 22: 50317 Goldberg D.E. and Barton A.M. 1992. Patterns and consequences of interspecific competition in natural communities: A review of field experiments with plants. American Naturalist 139: 771801 Kareiva P., Parker I.M. and Pascual M. 1996. Can we use experiments and models in predicting the invasiveness of genetically engineered organisms? Ecology 77: 16705 Kjær C., Damgaard C., Kjellsson G., Strandberg B. and Strandberg M. 1999. Ecological risk assessment of genetically modified higher plants (GMHP)  Identification of data needs, NERI Technical Report. Rep. 303, National Environmental Research Institute, Denmark. Lande R. 1993. Risks of Population Extinction from Demographic and Environmental Stochasticity and Random Catastrophes. American Naturalist 142: 91127 Marshall D.R. and Jain S.K. 1969. Interference in pure and mixed populations of Avena fatua and A. barbata. Journal of Ecology 57: 25170 Rees M., Condit R., Crawley M., Pacala S. and Tilman D. 2001. Longterm studies of vegetation dynamics. Science 293: 6505 Rees M., Grubb P.J. and Kelly D. 1996. Quantifying the impact of competition and spatial heterogeneity on the structure and dynamics of a fourspecies guild of winter annuals. American Naturalist 147: 132 Shigesada N. and Kawasaki K. 1997. Biological invasions: Theory and practice. Oxford: Oxford University Press Symstad A.J. 2000. A test of the effect of functional group richness and composition on grassland invasibility. Ecology 81: 99109 Vose D. 2000. Risk analysis  a quantitative guide. Chichester: John Wiley & Sons Weiher E., Clarke G.D.P. and Keddy P.A. 1998. Community assembly rules, morphological dispersion, and the coexistence of plant species. Oikos 81: 30922 Williamson M. 1996. Biological invasions. London: Chapman and Hall Copyright belongs with the author.
The following images related to this document are available:Photo images[by02001b.jpg] [by02001c.jpg] [by02001a.jpg] 
