Community dynamics are governed by two opposed processes: species sorting, which produces deterministic dynamics leading to an equilibrium state, and ecological drift, which produces stochastic dynamics. Despite a great deal of theoretical and empirical work aiming to demonstrate the predominance of one or the other of these processes, the importance of drift in structuring communities and maintaining species diversity remains contested. Here we present the results of a basic community dynamics experiment using floating aquatic plants, designed to measure the relative contributions of species sorting and ecological drift to community change over about a dozen generations. We found that species sorting became overwhelmingly dominant as the experiment progressed, and directed communities towards a stable equilibrium state maintained by negative frequency-dependent selection. The dynamics of any particular species depended on how far its initial frequency was from its equilibrium frequency, however, and consequently the balance of sorting and drift varied among species.

Source community

The source plant community was isolated from a eutrophic pond adjacent to fallowed agricultural fields on the Macdonald campus of McGill University, Quebec, Canada (45° 42’ N, 73° 94’ W). The pond sustains a diverse community of floating macrophytes, the four most abundant being Lemna minor (Lm), Lemna trisulca (Lt), Spirodela polyrhiza (Sp) and Wolffia columbiana (Wc), all in the family Lemnaceae. Large samples consisting of hundreds of thousands of individuals were taken in June 2020 and manually separated into the constituent species.

Experimental design

The experiment was conducted at the LEAP (Large Experimental Array of Ponds) facility at  Gault Nature Reserve of McGill University in Quebec, Canada (45° 32’ N, 73° 08’ W), (Fugère et al. 2020) (Fig. 1). 48 large mesocosms (surface area=2.43m²) were filled each with 500L of water piped from Lac Hertel, a mesotrophic lake on the reserve, 1km upstream of the experiment. The water was sieved to remove fish, tadpoles, macroinvertebrates, and macrophytes, but contained intact communities of zooplankton and phytoplankton. The removal of these larger organisms was to decrease unwanted variation due to sampling. Material from the source community was used to assemble four community types defined by the initial relative abundance of each species (10%, 20%, 30% or 40%) (Table 1). Relative abundance was calculated as mass-weighted frequencies using an average value of individual mass for each species. Each community was seeded with a total of 1g wet mass of community biomass, which works out to between ~2,000–3,000 individuals, depending on the community type (Table 1). Abundances of the larger species were determined by manual counting, while Wc, only ~0.5mm wide, was weighed and added in bulk, using an estimate of mean frond mass. Initial community densities translated to roughly 5–10% surface cover.  Communities were mixed after inoculation to remove any initial spatial variation.  Each of the four community types was replicated in 12 mesocosms (total number of mesocosms = 48) which were arranged in six blocks of eight mesocosms, each block containing two replicates of each community type, with community type randomized within block. All mesocosms received a one-time initial addition of inorganic Nitrogen and Phosphorus (KNO₃ and H₂KPO₄), to obtain initial dissolved concentrations of these nutrients in the mesocosms comparable to those of the pond from which the source community was taken (800 µgL-1 N and 40 µgL-1 P).  The mesocosms were covered with 70% shade cloth to mimic canopy cover. Although this minimized the input of wind-carried debris like leaf litter, rainwater could pass through the mesh cloth, which roughly balanced water lost due to evaporation. Communities were then left to grow for 12 weeks, from the beginning of July to the end of September, ending shortly before the first frost. All mesocosms were randomly sampled every two weeks to estimate species relative abundances. This was done by first mixing the communities to break up species clustering (Hart et al. 2019, Jewell and Bell 2022), then removing a fixed percentage of the surface area (~5%) with a net. Although mixing eliminated communities’ spatial structure, it allowed us to efficiently obtain representative samples. These samples were exhaustively counted before being returned to the mesocosm.

Statistical analysis

The main goal of this experiment was to estimate the contributions of species sorting, ecological drift, and initial state to community change. Overall variation in final species composition among communities can be broken into these three components, whose contributions to variation can be partitioned using an Anova framework (Travisano et al. 1995, Bell 2013). If Y_ij is the final frequency of the focal species in community type i and replicate j, then its deviation from that species’ mean initial frequency, Y_initial, can be partitioned into three additive components representing the three sources of variation:

Y_ij  - Y_initial = (Y_ij - Y_i) + (Y_i - Y) + (Y - Y_initial)  

where Y_i is the mean final frequency of the focal species in community type i, and Y is the grand mean final frequency of the focal species across all community types and replicates. For n community types (communities with different initial species composition) each replicated m times, the total variation attributable to sorting, drift and initial state can be calculated as follows:

• nm S (Y - Y_initial)², the shift in grand mean representing an overall convergence to an equilibrium composition (sorting),
• m S (Y_i - Y)², the variance among community types around the grand mean representing the influence of a community’s initial state, and
• S S (Y_ij - Y_i)², the variance among replicates of same community type representing neutral variation (drift).

Such a partition was done for each species at the end of the experiment to estimate the overall contributions of sorting, drift and initial state, as well as for each intermediate census to assess how the contributions changed over time.

We calculated a normalized value of the change in relative abundance of each species at each census in each mesocosm as the difference between the relative abundance of that species in the current and immediately preceding census, divided by its relative abundance in the preceding census. For each species, we used the regression of this normalized change in relative abundance on its preceding relative abundance to determine whether species dynamics were frequency-dependent. We used the X-intercept (at which change in relative abundance is zero) as an estimate of the equilibrium abundance of that species in a stable community.