Discussion:
By the end of the experiment, the initial state of the community had ceased to be a principal determinant of composition for all species (Fig. 3). A large contribution of initial state would indicate either that species frequencies had not changed, or that change was contingent on history, either through drift (for example, because of the increased risk of extinction when a species becomes rare) or sorting (for example, through facilitation, inhibition, or competitive intransitivity). The dominant process depended on the focal species, as the dynamics of Lm and Wc were governed predominantly by sorting, while those of Lt and Sp were governed largely by drift (Fig. 3). Overall, Lm benefited at the expense of Wc, while Lt and Sp remained largely static with a moderate amount of stochastic variation around their mean state. A modest contribution of sorting (and therefore a large relative contribution of drift) could be caused either by weak competition or because the equilibrium frequency of a species fell close to its average initial frequency over all community types (as was the case for Sp, whose initial and equilibrium relative abundances were both 0.25). Conversely, the dynamics of the species whose equilibrium frequencies are the furthest from their average initial frequencies (Lm and Wc) were most dominated by species sorting.
The strong effect of sorting and the absence of any strong effect of initial state suggests that the community tends towards an equilibrium composition, either through competitive exclusion or stable coexistence. Given a relatively constant and spatially homogeneous environment (May 1973, Chesson 2017), the distance from equilibrium which an actual community lies is determined by the balance of sorting and drift. Whether or not this equilibrium community involves the stable coexistence of several species can be determined by considering how the frequency of a species changes as a function of its current value (Fig. 5). The presence of a large negative correlation for all four species is evidence of negative frequency-dependent selection acting as a stable coexistence mechanism (Chesson 2000, Adler et al. 2007). The slope of this correlation indicates the strength of the frequency-dependence, and its intercept with zero change in frequency indicates the equilibrium frequency of that species. This negative frequency-dependence is probably a widespread coexistence mechanism in natural communities of floating freshwater macrophytes (Barrat-Segretain and Elger 2004, Gérard and Triest 2018, Armitage and Jones 2019, Hart et al. 2019) responsible for maintaining local diversity. Any real community will deviate from the ideal equilibrium composition through ecological drift.
The balance of sorting, drift and initial state shifted in a simple and predictable way over time. At the beginning of the experiment, since no change has yet occurred, all variation in community composition is due to initial state (Fig. 4). As time progresses, the relative contribution of initial state diminishes for all species, indicating the lessening contribution of initial state to community dynamics. Both species sorting and ecological drift increase in importance, and as the community nears equilibrium, roughly balance each other. This balance is however species-specific, due to the strength of competition. The further away a species’ frequency is from its equilibrium value, the stronger species sorting will act to bring it closer. Thus, both sorting and drift are acting on all species at all times, but how they combine depends on the species and will be determined by its competitive advantage and how far it is from its equilibrium frequency.
The standard approach to quantify stochastic effects on community structure is with the use of pairwise beta-diversity indices (eg. Bray-Curtis) (Anderson et al. 2011). Compositional dissimilarity is calculated for all pairwise combinations of sites in the same conditions and averaged to produce an estimate of among site (beta) diversity due to stochasticity, ranging from 0 to 1 (Gilbert and Levine 2017, Ron et al. 2018). Our approach of using an Anova framework allows us to partition the contribution stochastic processes into those due to ecological drift (variation in vital rates among individuals of the same species) and priority effects (stochastic variation in order of colonisation), and allows us to compare these numerically with the contribution from species sorting. Furthermore, obtaining estimates for each species separately can reveal how these fundamental processes may operate distinctly for different species in a community. The weakness of such an approach is that the contribution of sorting is highly sensitive to the initial frequencies (and their distances from the equilibrium frequencies). Consequently, estimates of the proportion of the total sum of squares due to stochasticity will also be sensitive to an arbitrary initial frequency. To work around this, it may also be informative to consider the contributions of stochastic processes as the raw within-group (drift) and among-group (initial state) sums of squares. We compared these measures to the more commonly used Bray-Curtis beta-diversity, calculated among replicate communities of each community type, at each time point, and found that both methods captured largely the same variability in species frequencies (Fig. 6). Equating this variability to drift of course assumes that the experiment is perfectly controlled, since any unintentional biotic and abiotic variation among replicate mesocosms contributes to our estimates of drift.
We conclude that the dynamics of our experimental communities shifted over time but were eventually dominated by species sorting, which resulted in the predictable and more or less deterministic shift towards an equilibrium state. The contribution of initial state declined consistently over time, but the balance between species sorting and ecological drift varied among species because, as in any community, some species were closer than others to their equilibrium frequency.
Acknowledgments: We thank Julia Kossakowski and Isabel Fernandez-McAuley, who helped with data collection. This experiment was supported by a Discovery Grant from the Natural Science and Engineering Research Council of Canada to GB and an Alexander Graham Bell Canada Graduate Scholarship from the Natural Science and Engineering Research Council of Canada to MDJ.