An evolutionary modelling approach to understanding the factors behind plant invasiveness and community susceptibility to invasion


  • J. WARREN,

    1. Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Llanbadarn Fawr, Aberystwyth, Ceredigion, Wales, UK
    Search for more papers by this author
  • C. J. TOPPING,

    1. Department of Wildlife Ecology & Biodiversity, National Environmental Research Institute, Aarhus University, Kalø, Grenåvej, Rønde, Denmark
    Search for more papers by this author
  • P. JAMES

    1. Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Llanbadarn Fawr, Aberystwyth, Ceredigion, Wales, UK
    Search for more papers by this author

John Warren, Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Llanbadarn Fawr, Aberystwyth, Ceredigion SY23 3AL, Wales, UK. Tel.: +1 970 621637; fax: +1 970 611264; e-mail:


Ecologists have had limited success in understanding which introduced species may become invasive. An evolutionary model is used to investigate which traits are associated with invasiveness. Translocation experiments were simulated in which species were moved into similar but evolutionarily younger communities. The main findings were that species that had previously been the most abundant in their original communities have significantly higher rates of establishment than did species that had previously occurred at low abundance in their original community. However, if establishment did occur, previously abundant and previously low-abundant species were equally likely to become dominant and were equally likely to exclude other species from their new community. There was a suggestion that the species that were most likely to establish and exclude others were ‘genetically’ different. When species that had evolved in different simulations (but with identical environmental conditions) were transplanted into communities that had also evolved in different simulations of the same conditions, the outcomes were difficult to predict. Observed rates of establishment and subsequent competitive dominance were observed to be species- and community combination–specific. This evolutionary study represents a novel in silico attempt to tackle invasiveness in an experimental framework and may provide a new methodology for tackling these issues.


Introduced invasive species are regarded as one of the most significant threats to biodiversity and agricultural production around the globe (Williamson, 1996; Vitousek et al., 1997; Wilcove et al., 1998). The control of invasive species is problematic and economically costly (Pimentel, 2002; Pimentel et al., 2005). However, the vast majority of introduced species fail to establish in the wild outside their native range, and even when they do establish, most cause negligible damage (Williamson, 1996). For these reasons, ecologists have long tried to determine which introduced species have the potential to become invasive and which native communities are most susceptible to being invaded (Kolar & Lodge, 2001; Henderson et al., 2006).

Many ecological studies have focused on explaining why introduced species may have a fitness advantage over native species, and various mechanisms have been suggested. For instance, it is thought that introduced invasive species may thrive because they have escaped their natural enemies (Keane & Crawley, 2002), and there is evidence that invasive plants may experience lower levels of herbivore damage and pathogen loads in their naturalized range than in their native range (Fenner & Lee, 2001; Mitchell & Power, 2003). It has also been suggested (Blossey & Nötzold, 1995) that once released from their natural enemies by translocation, species may be able to reallocate resources from defence into growth, increasing their relative fitness and making them even more competitive, but evidence for this is not overwhelming (Bossdorf et al., 2004).

The relative fitness of introduced and native species is important, and it is only one factor that influences invasiveness. The overlap of niche requirements of species is also thought to be important in determining whether species can coexist or whether competitive exclusion of either species occurs (Adler et al., 2007). A recently developed theoretical framework for predicting invasiveness combines both relative fitness and the extent of niche overlap, with relative fitness becoming increasingly less important in determining competitive exclusion as niche requirements diverge (MacDougall et al., 2009). Successful establishment of an introduced species can result from either a higher relative fitness or a niche difference that allows it to colonize in spite of a lower average fitness (Chesson, 2000). This ecological framework developed further by Chesson & Kuang (2008) is built upon the assumption that average relative fitness can be considered as an ecological constant that differs from fitness in the context of evolutionary biology (MacDougall et al., 2009). However, others have pointed out that an important aspect of invasion ecology may involve small founder populations adapting to their new environment, with rapid genetic change being linked to increased fitness (Lee, 2002; Kanarek & Webb, 2010). It has been argued therefore that to understand invasiveness, ecologists need to explore evolutionary genetics (Lee, 2002). In an evolutionary sense, the relative fitness of a species may be better considered as a temporally dynamic state in an ongoing Red Queen arms race with its competitor species. In an ecological sense, fitness may be considered spatially dynamic, with fitness being highest when niche requirements are satisfied and declining as the niche becomes less ideal.

Introductions of invasive species have been regarded by many ecologists as a series of large-scale unplanned ecological experiments (Sax et al., 2007). But as an experimental system, the release of introduced species is far from ideal, because the introduction events are rarely planned or subsequently scientifically monitored. The methods of introduction and numbers of individuals released vary between species, and where introductions were intentional, several sequential release events are common, with release number being confounded with apparent initial success of establishment (Veltman et al., 1996). The greatest limitation with this form of data is simply the unknown number of unsuccessful introduction events, as species are constantly being moved around the world, but only those which establish are ever effectively recorded.

Preston et al. (2002) identified the 100 species that had shown the greatest relative increase in occurrence in the UK between 1930–1969 and 1987–1999. Of these 100 species, about one-third are native species and very few are problematic invasive species. It is clear therefore that the ecology of establishment within a community/range expansion and subsequent competitive dominance are different. Invasive species must have both the ability to establish in a new community and subsequently become dominant.

The aim of this study was to try and identify the life-history traits and other factors that enable a species to successfully establish in a new community and which traits subsequently enable it to increase in abundance at the expense of the original species. To avoid the ethical considerations of experimental release and complications that are associated with field data having different and poorly documented histories, we use an evolutionary ecological model of plant competition to investigate the probability of establishment and subsequent competitive exclusion when translocating species between different plant communities. This theoretical approach has several advantages; it allows the manipulations of host communities, it can be used to simulate long-term experiments, and it facilitates replication.

The model

The Evolve model is an agent-based system that simulates plant growth, reproduction and death in a simplified 3D arena. Previous publications (Warren & Topping, 1999, 2004; Warren et al., 2009) have demonstrated that the ‘Evolve’ vegetation model can robustly simulate short-term community change and ecological interactions in grasslands. It has also been used to investigate longer-term evolutionary processes (Warren & Topping, 2001; Warren et al., 2009). This model has been previously described (Warren & Topping, 1999, 2001, 2004; Warren et al., 2009) but has developed significantly over this period, and therefore, a full technical description has been compiled and updated from these references and is provided in the Appendix S1. In addition, full documentation is provided using the ODdox protocol (Topping et al., 2010) at The ODdox is a formal presentation of the model design and components together with fully hyperlinked documentation of the source code. This should be considered the primary resource for Evolve documentation.

Individual plant phenotypes and hence ecological interactions are regulated by the interaction of their life-history parameters, ‘genes’ and environment, both of which are controlled by a series of parameters detailed in Table 1 and the Appendix S1. The main environmental parameters used as variables in the virtual experiments reported here are gap creation and community age, as these are considered to be key factors in affecting a community’s susceptibility to invasion. Individual plants grow month by month by capturing resources from the environment. Resource capture may be limited by the phenotype of the individual, the time of year, the environment or the presence of other individuals. Thus, in the short term, differences in competitive ability result in changes in species relative abundance. In the longer term, allowing life-history parameters (genes) to mutate allows evolutionary change to occur and new species to evolve.

Table 1.   Life-history and environmental parameters used in Evolve; initial values are those used in the initial simulations that generate the communities for subsequent simulations. Life-history parameters that are optionally mutable are indicated. Values in bold act as variables in subsequent simulations.
RefLife-history parameterMutableInitial valueRefEnvironmental parameterInitial value
lh_1Initial number of propagules100 per sp.env_1Proportion of area grazed0.1 Apr–Nov
lh_2Ramet life span25 monthsenv_2Max. and min. grazer defoliation1–1
lh_3Max. heightY1env_3Number of cuts per year1
lh_4Max. width4env_4Month(s) in which cutting occursAug
lh_5Shade toleranceY0env_5Height to which vegetation is cut6
lh_6Below-ground extraction efficiencyY0.001–0.25env_6Max. & min. below-ground resources50–50
lh_7Extent of monthly winter die-back0.02env_7Size of initial nutrient grid10x10
lh_8pH tolerance range (max & min)7–7env_8pH range7–7
lh_9Reproduction vegetative or seedYBothenv_9pH patch size8
lh_10Month of reproductionY6env_10Max. and Min. disturbance events0
lh_11Month of germinationY7env_11Size of disturbance events0
lh_12Propagule sizeY1env_12Occurrence of seed rain0
lh_13Resources allocated to growing tallY0env_13Amount of seed rain0
lh_14Resources allocated to growing wideY0.7env_14Mutation rate10 000
lh_15Resources allocated to reproductionY0.3env_15Length of SimulationFixed by mutation No
lh_16Resources allocated to herbivore defenceY0   
lh_17Growth direction priorityYBoth   

The simulations

Initial simulations

Twenty-five sets of initial simulations were carried out to investigate the invasive potential of species that had evolved as part of a grassland community when translocated into a different community. The initial simulations were only used to generate the introduced species and plant communities for the subsequent translocation simulations. The simulated communities were based on a grazed grassland habitat with low soil fertility, because the model has previously been shown to successfully simulate this type of vegetation and because grasslands are known to be prone to invasion (MacDougall & Turkington, 2005). Each simulation started with 100 propagules of each of six species, the parameter values of which are presented in Table 1. These values were deliberately not pre-adapted to the environment, so that older communities would be expected to contain better-adapted species than younger ones. The 25 sets of simulations were based on combinations of two variables: a, the number of gaps created in the vegetation per month [because disturbance events are thought to encourage invasion (Leishman & Thomson, 2004)], and b, the evolutionary history of the community. [Although controversial, it has long been considered that evolutionary older communities are more diverse, stable and less prone to invasion (McCann, 2000).] There were five different gap frequencies used: 0, 25, 50, 75 and 100 (with gap locations occurring randomly across the arena), and there were five evolutionary history treatments, based on simulation of the communities for: 10 000, 20 000, 30 000, 40 000 and 50 000 mutations per simulation (see Table 2). These were generated as separate simulations, rather than stopping single simulations at different time. Mutations allow randomly selected trait parameters to vary by ±5%; the rate of mutation was fixed at one event every 10 000 propagules produced. Previous unpublished simulations suggest that higher mutation rates occurring over shorter time periods produce equivalent evolutionary outcomes as lower mutation rates applying over longer time periods. Evolutionary history was limited by mutation events rather than time, allowing treatments to be more directly comparable. All possible combinations of gap frequencies and evolutionary histories represent the 25 different initial simulation types.

Table 2.   Schematic of the 25 initial simulations, each of which was replicated ten times. After these were completed, the translocation simulations were then carried out by transplanting species that had evolved in the five simulations on the right (shaded) into each of the nonshaded communities on the same row. For instance, species that evolved in the first replicate of initial simulation 5 were transplanted into the communities that had evolved in the first replicates of initial simulations 1–4. Thumbnail image of

The 25 different initial simulations were each replicated ten times. At the end of the five simulation sets with the longest evolutionary histories (50 000 mutations), the most abundant and least abundant species present were identified. Abundance was defined by the number of individuals, but this was always synonymous with biomass. These species then became the introduced species in the subsequent translocation simulations.

Translocation simulations

The basic concept behind the translocation simulations was that previously abundant and nonabundant species from the initial simulations were translocated into other evolutionary younger communities that had evolved under identical environmental parameters to those from which the translocated species were derived. The justification for this is that alien invasive species tend to colonize similar communities from which they were originally derived, i.e. they invade habitats with similar edaphic conditions to which they are pre-adapted (Sakai et al., 2001).

Three separate sets of translocation simulations were carried out. Each of these three sets of simulations involved 20 (5 gap × 4 age) different treatment combinations, and these are represented in Table 2. The first set involved the translocation of the most abundant species that evolved in each of the ten replicate initial simulations with 50 000 mutations (for each of the five gap frequencies). Each of these most abundant species was transplanted into each similar (but younger) community that had evolved under the same conditions, but for less time (fewer mutations). For example, the most abundant species that evolved in the first replicate of initial simulation 5 (gaps = 0 and mutations = 50 000) was subsequently transplanted into four different younger communities all derived from the first replicate, with gaps = 0 and mutations = 10 000, 20 000, 30 000 and 40 000 (initial simulations 1–4). The most abundant species from initial simulation 10 was translocated into communities six to nine and so on. This gives a total of 40 simulations just for the first set of translocation simulations (see Table 2). In the second set of translocation simulations, the above-mentioned process was repeated, but this time, the least abundant species were translocated. The final set of translocations was identical to the first set, but this time the below-ground fertility of the arena was elevated by 5% (at the time of translocation) to represent an environmental change from the conditions under which the species evolved.

As with all simulations, in all cases, the translocation simulations started with 100 propagules of each species present scattered randomly across the arena. The average number of species at the start of the translocation simulation was around 15. Although defining the species concept at a genetic level is problematic (Rubinoff, 2006), for simplicity we defined a ‘species’ as any individual differing from another; however, in most cases, ‘species’ differed from each other across several life-history parameters. Transplant simulations were run for 100 years to allow for a lag phase in the establishment of the introduced species. Although genetic change may be rapid during range expansion (Burton et al., 2010), further mutations were not allowed during the translocation simulations as these were regarded as short-term ecological experiments rather than evolutionary ones. At the end of each set of ten replicate simulations, the following outcomes were recorded: (i) in how many of the simulations had the introduced species established (remained alive for 100 years); (ii) in how many of the replicates had the introduced species become the most abundant species; and (iii) what was the mean number of original species that had been competitively excluded by the introduction of the new species, in comparison with an identical control simulation in which the introduced species was absent. In most cases, the outcomes were very clear and fell into discrete classes: failed to establish, established but failed to suppress other species, or established and became dominant at the expense of other species.

Stochasticity simulations

The ‘Evolve’ model simulations include a stochastic element that governs where seeds land, where gaps are created and up to what extent grazing occurs. Thus, with identical starting conditions, in some simulations, an introduced species may die out, and in others, the same species may become dominant. To establish the importance of this stochasticity, more replicate simulations were carried out for one of the above translocation types. This involved the fully reciprocal translocation of the most abundant species derived from each of the ten replicates of initial simulation 5, transplanted individually into each of the ten replicate communities from initial simulation 4 (see Table 2). That is, for each of the ten (50 000 mutation + no gap) simulations, the most abundant species were transplanted individually into each of the ten (40 000 mutation + no gap communities), and each of these 100 combinations were replicated ten times, giving a total 1000 individual translocation simulations. Thus, in this scenario, the only difference between the communities from which the potential invader came and those being invaded was their evolutionary history. This is justified because invading species tend to be derived from similar communities to those they invade. As before, these simulations were run for 100 years, and the success of the introduced species was recorded. To allow subsequent statistical analysis, once these 1000 simulations had been completed, the whole process was repeated, giving two ‘blocks’ of 1000 replicates each. Only the translocation element of the simulations was replicated, not the evolution of the communities, so that the translocated species and host communities were identical across the two blocks. Having two replicate blocks of simulations allows the subsequent analysis of variance to determine whether there was an interaction between introduced species and host communities, even when these are derived from replicate evolutionary events.

To try and quantify the extent of stochasticity in the process of establishment, a randomly selected abundant species from the 50 000 mutation + no gap simulations was competed against multiple copies of itself. One hundred replicate simulations were performed with each of, 5, 10, 15, 20 and 25 identical copies of the same species. In each of these simulations, genetically identical individuals were given unique species identifiers within the model. At the start of each simulation, there were 100 propagules of each unique identifier. Thus, the sowing rate in the five copy replicates was 500 propagules per arena. Each of these simulations was run for 100 years at the end of which the establishment rate of individually identified species was recorded.


The outcome of the translocation simulations was analysed using a general linear model (GLM Minitab 14.1, Coventry, UK) (Table 3) and plotted in Figs 1–3, to investigate the factors affecting the invasiveness of the introduced species and the community susceptibility to invasion. The first set of translocation results are presented in Figs 1–3a, the second set in Figs 1–3b and the final set in Figs 1–3c. Each of the ten replicate simulations per species × community combination was used to generate one datum i.e. the number of times out of the ten in which translocated species: (i) remained extant after 100 years; (ii) became the most abundant species; or c, resulted in the extinction of all the other species. The resulting scores were tested for normality and did not require transformation before GLM analysis.

Table 3.    (Translocation simulations) Effects of recipient community age (defined by number of mutation), the number of gaps and the species type (most abundant ± N addition or least abundant) on the number of replicate occasions the species a, rate of establishment b, frequency of becoming dominant and c, the number of times invasion resulted in the competitive exclusion of all the other species.
SourcedfMean SquareF-ratioP
 Community age31.3780.740.532
 Species type232.31717.39<0.001
 Community age324.1946.17<0.001
 Species type213.6173.470.039
 Community age35.2091.410.251
 Species type22.1670.590.560
Figure 1.

 (Translocation simulations) The mean number of translocated species that were able to establish in the 20 different recipient communities, from two sets of ten replicate simulations. (a) The species translocated were the most abundant to evolve in the initial simulations; (b) the translocated species were the least abundant species from the initial simulations; (c) the simulations were identical to those in fig. 1a, except the soil fertility was increased marginally.

Figure 2.

 (Translocation simulations) The mean number of translocated species that were able to become the most abundant species in the 20 different recipient communities from two sets of ten replicates. (a) The species translocated were the most abundant to evolve in the initial simulations; (b) the translocated species were the least abundant species from the initial simulations; (c) the simulations were identical to those in Fig. 2a, except the soil fertility was increased marginally.

Figure 3.

 (Translocation simulations) The mean number of species excluded from the community following the introduction of the translocated species. Results are based on two sets of ten replicate simulations. The translocation was performed over 20 different recipient community types. The numbers of excluded species presented are relative to those lost by chance from control simulation in which the translocated species were not included. Hence, it is possible to have negative values for simulations in which the addition of the translocated species was associated with fewer species being lost than in the control. (a) The species translocated were the most abundant to evolve in the initial simulations; (b) the translocated species were the least abundant species from the initial simulations; (c) the simulations were identical to those in Fig. 3a, except the soil fertility was increased marginally.

To determine whether successful invaders within the translocation simulations were genetically different from nonsuccessful invaders, their life-history parameter values were compared using principal components analysis (PCA). Species were classified according to their performance in the end of the initial simulations as the most abundant or least abundant species. The species were then further classified according to their performance in the translocation simulations as failed to establish or established at the expense of the original species. This classification identifies extremes of success and failure and thus provides a better chance of identifying genetic differences between them. The PCA was performed on the life-history parameter values obtained for all these classes of species. The first and second principal components were plotted (Fig. 4a), but the results reveal no separation of successful or unsuccessful species.

Figure 4.

 (a) (Translocation simulations) Principal components analysis of the ‘genomes’ of successful and unsuccessful introduced species. Both most abundant and least abundant species translocated into initial simulations 4 (open symbols) and 19 (closed symbols), (the most extreme examples of completive exclusion observed). A = most abundant species (initially), L = least abundant species (initially), F = failed to establish. S = successfully established and excluded all other species. 4 or 19 = recipient community derived from initial community 4 or 19 (given in Table 2). The first PC explains 46% of the variation and the second PC 14%. Arrows indicate the contribution of the most significant life-history traits. A = nutrient extraction efficacy, B = max height, C = max width, D = seed resources, E = width resources, F = reproduction time, G = propagule size. (b) (Stochasticity simulations) Principal components analysis of the ‘genomes’ of introduced species translocated from into initial simulations 5 into 4. Species ranked according to mean establishment rate with higher number being most invasive. The first PC explains 40% of the variation and the second PC 30%. Arrows indicate the contribution of the most significant life-history traits. A = nutrient extraction efficacy, B = max height, C = max width, D = seed resources, E = width resources, F = germination time, G = reproduction time.

The stochasticity simulations of the most abundant species derived from initial simulation 5 transplanted into initial simulation 4 were also analysed using a general linear model (GLM Minitab 14.1, Coventry, UK) (Table 4). Again, the raw data were the number of occasions out of ten replicate simulations in which translocated species became established, became the most abundant species or resulted in the extinction of all the other species. As before, to try and identify whether any particular traits are associated with the likelihood of introduced species becoming invasive or not, the life-history parameter values of all ten introduced species were compared using principal components analysis. The species were ranked in terms of their mean establishment rate, and then the first and second principal components were plotted for these rankings (Fig. 4b). The rates of establishment and rates of total exclusion of other species between the two replicate blocks of 100 transplant experiments were compared using a paired sample t test.

Table 4.    (Stochasticity simulations) Effects of introduced replicate species into replicate communities on (a) the likelihood of establishment; (b) the likelihood of dominance; and (c) the amount of resulting competitive exclusion. The ten species and communities used were derived from replicates of the same simulations.
SourcedfMean squareF-ratioP
 Species type942.10219.95<0.001
 Community ID920.0559.50<0.001
 S × C8110.9095.17<0.001
 Species type938.90323.37<0.001
 Community ID914.8588.92<0.001
 S × C816.6614.00<0.001
 Species type935.87423.84<0.001
 Community ID911.0087.31<0.001
 S × C817.1474.75<0.001


The key results to emerge from this are as follows:

  • 1 Previously abundant species when translocated are more likely to establish than less abundant species (Fig. 1a,b), but subsequently they are no more likely to become abundant (Fig. 2a,b) or exclude other species (Fig. 3a,b).
  • 2 The occurrence of disturbance within a community in the form of gaps is likely to encourage introduced species to become established (Fig. 1a–c) and abundant (Fig. 2a–c) and result in the exclusion of original species (Fig. 3a–c).
  • 3 Community disturbance in terms of elevated soil fertility appears to increase the likelihood of introduced species becoming abundant (Fig. 2c).
  • 4 Introduced species are more likely to become abundant in younger communities than in communities derived after more mutations (Fig. 2a–c).

The stochasticity simulations reveal highly significant differences between replicate species in their rates of establishment, ability to become abundant and competitive exclusion of other species. Similarly, highly significant differences were found between replicate recipient communities in their susceptibility to being invaded. However, these observations were inconsistent. Some introduced species that were on average very successful invaders failed to establish in some recipient communities, whereas some invasion resistant communities were found to be successfully colonized by some species. This complexity was reflected in the highly significant species x community interaction in terms of rate of establishment, dominance of the introduced species and extent of competitive exclusion (Table 4). The paired sample t tests revealed that the two replicate blocks of these 100 translocation experiments did not differ significantly (= 0.273 and 0.554) in either the rates of establishment or subsequent total competitive exclusion.

When identical copies of a single species were competed against each other, all 100 simulations succeeded in establishing for 5, 10, 15 and 20 species, but at 25 species (well above the number of species used in the scenarios here), there was a 25% failure to establish. Hence, stochasticity in establishment in suitable communities did not seem to play a role unless the proportion of invaders was small.

The two species with the highest rates of establishment (Fig. 4b) were found to have the highest first principal component scores, and these were associated with the parameters regulating plant width (growing wider), reproductive strategy (seed production) and timing of seed production (reproducing earlier).

Finally, from the stochasticity simulations, mean susceptibility of the ten replicate communities to invasion was plotted against community species richness (Fig. 5). Although the significant relationship found was not tight, more species-rich communities were less susceptible to invasion than were more species-poor ones. The figure presents successful invasion defined as the introduced species becoming the most abundant species with 100 years of introduction, but similar results were also seen for rates of establishment.

Figure 5.

 (Stochasticity simulations) The relationship between the number of species in a recipient community and the mean number of times the community was successfully invaded by the introduced species, = 0.11. Successful invasion was defined as the introduced species becoming the most abundant species with 100 years of introduction.


At a superficial level, the results of the various simulations presented here are reassuringly unoriginal. They show that more abundant species are more likely to become established than are less abundant species, that perturbing the environment either by the elevation of soil fertility or by the creation of gaps makes recipient communities more prone to invasion and that recently evolved and species-poor communities are more prone to invasion (Figs 1 and 5). The fact that these observations are well documented in nature (Elton, 1958; Stachowicz & Tilman, 2005) gives us more confidence that our simulations meaningfully represent real life.

Delving below the surface however, our results reveal a level of complexity that is mirrored by the repeated failure of ecologists to provide a definitive answer to why only a minority of introduced species become invasive (Williamson, 1996). This is seen most clearly in the stochasticity simulations (Table 4 and Fig. 4b) and in our associated failure to achieve our original aims. Although there is some suggestion that at the extreme, the most invasive species are ‘genetically’ distinct (Fig. 4b), the highly significant species x community interactions demonstrate that even successful invaders may not always be so, whereas our results show that apparently benign species can invade certain communities that are otherwise resistant to invasion. This observation exposes the naivety of Chesson & Kuang’s (2008) approach that is built on the assumption that relative fitness is an ecological constant. The implication of this is that predicting potentially invasive species is always likely to be problematic because fitness and invasive potential are a function of both introduced species and host community. This could also explain why we and other studies (Williamson, 1996) have struggled to find ‘genes’ associated with invasiveness, because life-history traits that promote invasion in one community appeared not to enhance fitness in very similar communities.

It seems likely that the significant variation between replicate species and replicate communities observed in the stochasticity simulations is a function of the fact that separate evolutionary simulations were never identical. This concept was popularized by Gould (1991). Thus, although throughout we have used the word ‘replicate’ to describe repeat simulations, evolutionary repeats are unlike other forms of replication. A translocated species that evolved in one replicate and a recipient community that evolved under the same conditions but in a different replicate simulation can therefore not be considered as being co-evolved. Experience with the model has shown us that species that genuinely co-evolve within a simulation interact as a community and diversity is maintained, whereas a group of species assembled from replicate simulations of the same environment fail to coexist for any period of time.

Table 3 reveals that the ecology of establishment (which is significantly affected by gap frequency and species type) differs from the species’ subsequent ability to increase in abundance (which was significantly affected by community age and gap frequency but not species type). Thus, a species ability to establish is not a good predictor of its subsequent ability to increase in abundance. Similarly, the ecology of dominance differed from that of ecological impact as measured by competitive exclusion (which was only significantly affected by gap frequency). D’Antonio & Hobbie (2005) have already recognized that these processes are independent in nature, which again gives us more confidence in our simulated results.

The fact that the outcome of replicate simulations was not always identical suggests an important role for chance in the processes of establishment and competitive dominance. Chance was involved in determining whether establishment occurred or not on a particular occasion. However, translocation of species identical to the recipient community was always successful at the translocation rates used for scenarios here. This suggests that within the model, establishment is primarily determined by niche compatibility with the species life-history traits. Over replicate simulations, the probability of establishment of a particular species establishing in a particular community over several simulations was remarkably constant. This was demonstrated by the fact that the overall outcome of the two replicate blocks of 100 transplant experiments was remarkable similar, and a paired sample t test showed that they did not differ significantly. Overall, this indicates that apparent stochasticity of establishment was in fact an emergent property of the evolution of the recipient community.

The most difficult question raised by our results remains why are introduced species sometimes invasive and sometimes not, as is argued above invasiveness does not appear to have a simple genetic control or to be a simple function of general fitness. The results presented here (Table 4) show that invasiveness is a function of both the species (genotype) and community in combination, rather than being a property of either individually. It is not therefore the result of environmental stochasticity within simulations (although we have already seen that the stochastic occurrence of gaps will encourage establishment). A clue to what might be happening is provided by Fig. 4b where the two most invasive species were found to differ from the other species in their timing of seed production. This trait has previously been found to be important in determining competitive ability within the model, with its competitive advantage being regulated by cyclic Red Queen dynamics (Warren & Topping, 2001). If the majority of individuals within a community germinate late in the year, then a species that germinates earlier will be at a competitive advantage and spread/become invasive. Consequently, over evolutionary time, the community will switch to being primarily populated by early-germinating species. At this stage in the cycle, the community would be prone to being invaded by late-germinating species. Thus, depending on where a community is in such a cycle, genes for reproducing either early or late may promote invasiveness. Similar phenomena with other traits are reported in nature (Callaway & Aschehoug, 2000; Lee, 2002). If this mechanism does operate, then it implies that the view held by some that most invasive species can be identified by traits alone (Kolar & Lodge, 2001) may need some refinement.

This suggestion that genetic change is important in determining whether an introduced species is invasive or not is not new (Lee, 2002); however, it does differ from the current theoretical framework (MacDougall et al., 2009) that considers a species average fitness to be a constant.

What is clear from both the literature and from this study is that the question of what regulates plant invasiveness is complex, and as such our predictive abilities are likely to be poor. Progress is hampered by the fact that, for obvious reasons, there are considerable problems involved in running controlled replicated experiments of species introductions, and this paper therefore represents an opportunity to do this, albeit ‘in silico’. The model used here, although much simpler than the real-world model, indicates a number of mechanisms that may be in play. Results largely, although not entirely, confirm received wisdom also providing confidence in the general approach. Because there is clearly great scope for improvement in the model and its application, this study represents only the first steps towards development of simulation approaches to address these issues and increase our predictive ability.


We thank two anonymous referees and Fay Newbury, Richard Kipling and Rosie Shoosmith for advice and help with the manuscript. C. T. was supported by the Centre for Integrated Population Ecology.

Data deposited at Dryad: doi: 10.5061/dryad.db17m