Predicting how ecological communities respond to environmental change is challenging, but highly relevant in this global change era. Ecologists commonly use current spatial relationships between species and environmental conditions to make predictions about the future. This assumes that species will track conditions by shifting their distributions. However, theory and experimental evidence suggest that species interactions prevent communities from predictably tracking temporal changes in environmental conditions, based on current spatial relationships between species and environmental gradients. We tested this hypothesis by assessing the dynamics of protist species in replicated two-patch microcosm landscapes that experienced different regimes of spatial and temporal environmental heterogeneity (light vs. dark). Populations were kept in monoculture or polyculture to assess the effect of species interactions. In monoculture, abundances were predictable based on current environmental conditions, regardless of whether they had experienced temporal environmental change. But, in polyculture, abundances depended also on the history of environmental conditions experienced. This suggests that, because of species interactions, communities should respond differently to spatial versus temporal environmental change. Thus, species interactions likely reduce the accuracy of predictions of future communities that are based on current spatial relationships between species and the environment. 

We used an aquatic microbial system (protist microcosms) to disentangle the effect of species interactions, dispersal, and temporal change in environmental conditions on local communities exposed to contrasting environments. This model system is well-established to test fundamental ecological questions, and especially well-suited to address effects of environmental change on community dynamics due to the high level of control and replicability (Warren 1996; Petchey et al. 1999; Altermatt et al. 2011; Jacquet et al. 2020). The environmental heterogeneity consisted of a light versus a dark treatment. Protist species have been shown to track these environmental conditions, and they can affect their behavior and population growth dynamics (Giometto et al. 2015, 2017). We measured how community assembly in heterogeneous two-patch metacommunities was shaped by the local conditions  and if and how communities could track a temporal change in this environmental state.

We tested the effect of species interactions by contrasting species responses to environmental change when they were present in polyculture, compared to in non-interacting single-species monocultures. We aggregated the data from these monoculture populations to simulate the composition of communities without interspecific interactions. We tested the effect of dispersal by repeated exchanges of a small fraction of the populations between the two local habitats mimicking dispersal (comparing to a no-dispersal control). We also contrasted the response of communities with and without temporal environmental change so that we could quantify how much of the changes in community composition were due to environmental change compared to temporal fluctuations in population size. Then, we assessed the degree to which the communities tracked local changes in environmental conditions based on how closely the composition in the communities that experienced environmental change matched the composition of the community that experienced the corresponding final environmental conditions for the duration of the experiment. We predicted that: 1) if species interactions prevent communities from tracking changes in abiotic conditions, then compositional tracking should be lower in the polyculture communities compared to the aggregated monoculture communities; 2) if dispersal reduces the degree to which species interactions prevent communities from tracking environmental change, then species interactions should reduce community tracking less in communities connected by dispersal.

Study organisms

We used seven freshwater protist species for the experiment: Colpidium striatum (Col), Euglena gracilis(Eug), Euplotes aediculatus (Eup), Loxocephalus sp. (Lox), Paramecium aurelia (Pau), Spirostomum teres(Spi), and Tetrahymena pyriformis (Tet). These protists were kept in a protist pellet (Carolina Biological Supply, Burlington) nutrient medium inoculated with the bacteria Serratia fonticola, Brevibacillus brevis, and Bacillus subtilis as a food source. Two of these species (Eug and Eup) are mixotrophs, and thus also able to photosynthesize. All species had been kept as monocultures and kept under clean laboratory conditions. Before the start of the experiment, we grew all seven protist species under optimal conditions to their carrying capacity. For details of the experimental system and handling procedures, see laboratory protocols published in Altermatt et al. (2015).

Experimental set-up

We used 6-well polystyrene multi-well plates (AxonLab), which were used to inoculate 2-patch metapopulations and metacommunities of the seven species respectively. The focal volume was 8 mL. We had single-species metapopulations of all species, as well as the respective seven-species metacommunity. Of each species’ stock culture, 1.143 mL was added to the respective patch at 1/7 of its carrying capacity. In the single species patches, we topped up the volume with protist medium to reach the focal volume. All experimental replicates were kept in a randomized order in incubators at 20 °C.

We manipulated the community context (seven metapopulations with 1 species, and the seven-species metacommunity), dispersal (the two patches of the 2-patch metapopulations and metacommunities connected by dispersal or not), and temporal variability of environmental conditions of the two patches (remaining constant or not). All single-species metapopulations were replicated 3-fold for each treatment combination, while the seven-species metacommunities were 6-fold replicated.

We conducted two dispersal events over the duration of the experiment, namely after two and three weeks (i.e., at the time point of the environmental temporal change treatment, and one week thereafter). The reciprocal rate of dispersal between the respective two patches was 5 % (0.4 mL). Dispersal was density-independent and passive, by pipetting the respective volume between the two well-mixed patches. In the no dispersal control, we applied the same mixing and pipetting treatment, but without spatial inference.

The environmental condition manipulated was light, which is used for photosynthesis by some of the species (Eug, Eup), and can trigger behavioral change in many of them. Of each 2-patch landscape, one patch was either fully illuminated (24 h, LED light, Ledoxon, 4.5 W, 445 lm luminous flux), or completely dark (respective well-plates wrapped in aluminum foil). The environmental condition of the two patches (dark vs. light) either remained constant throughout the experiment (“no temporal change”) or was reversed after two weeks of the start of the experiment (“temporal change”). By doing so, we could calculate how communities tracked environmental change (from dark to light or light to dark), and if and how this tracking was mediated by dispersal. 

Measurements

We recorded and analyzed the species composition, abundance and diversity in each community (monocultures and polycultures) at weekly intervals using highly resolved video analyses. To do so, we sampled 175 µl of each community immediately preceding the dispersal treatments. The sample was added to a counting chamber mounted on a glass-slide, and we recorded a video of 5 seconds (25 frames per second, 16x fold magnification, full light) with a digital Orca Flash 4.0 camera (C11440-22CU, Hamamatsu Photonics, Japan). The total volume analysed was 34.4 µL. We then used this video to identity and quantify the presence and abundance of all seven protist species, closely following the method and the R package BEMOVI developed and used by Pennekamp et al. (2017). The settings for the BEMOVI script were the following: Pixel size of 4.05 µm, difference lag of 10 frames, thresholds of 10 to 255 difference of pixel intensity, min particle size 5 pixels, max particle size 1000 pixels, link range 3 frames, displacement 16 pixels, detection frequency of 0.1 seconds, median step length of 3 pixels. This method links individuals across successive time steps in the videos and calculates a number of morphological and movement features for each one.

To identify the species, we used these morphological and movement features in a random forest algorithm. This algorithm is based on decision trees using binary thresholds to divide the observations into the most possible class at the end node (Pennekamp et al. 2017). The random forest algorithm was trained using the monoculture species (96.1% correct classification rate) and then applied to classify the identity of all individuals in the polyculture communities (Pennekamp et al. 2017). In four replicates, monocultures of E. gracilis reached population sizes that were too large for the BEMOVI software to properly link individuals across time steps in the videos, and thus it was not possible to obtain movement statistics. To estimate these populations, we used a linear model to estimate the relationship between the log transformed number of particles in a single video frame and the log transformed population sizes estimated as described above for all populations of E. gracilis (Figure S1). We then used this relationship to estimate the missing population sizes using the single frame output of particles from the BEMOVI analysis. All analyses were conducted in R v. 3.6.1 (R Development Core Team 2020). 

Dataset posted here

We have posted the following data files:

1. protist_light_dark_pops.csv
2. Density at start.csv

 

The first file contains the population sized from the experiment. It contains the following data columns: 

• temporal_change – values are yes or no which indicate whether the environment was changed after time step 2
• dispersal – values are yes or no which indicate whether the dispersal treatment was appliec
• time – the time step of the experiment
• predicted_species – the identity of the population based on the random forest algorithm
• environment_1 – values are light or dark. This is the initial environmental conditions experienced.
• community – values are monoculture or polyculture, which indicates whether the population was grown in isolation or together with the other species
• N – the size of the population

This file was produced using BEMOVI from the raw video files. The scripts that do this pre-processing are available on Zenodo – https://doi.org/10.5281/zenodo.4131142 as well as here. The raw video files are archived by Florian Altermatt (florian.altermatt@eawag.ch) and available upon request. 

The second file contains the estimate population sizes used to initialize the experiment. These estimates are based on replicate videos taken from the culture stock. The same volume of culture stock was then used to initialize each replicate in the experiment. This file contains the following data columns:

• file – the file name of the video script used to estimate the population size
• code – the species code for the population
• ind_per_volume – individuals in the volume used for the video. Not used in the analysis
• indviduals per mL – estimated number of individuals per mL
• mean density – estimate mean density. Not estimated for all replicate samples and not used in the analysis.

Associated scripts

We also provide the scripts used to produce this data:

• 1_analyze_protist_videos_Patrick_SH.R
• 2_merge_data.R
• 3_predict_protists.R
• 4_predict_missing_data.R
• 5_clean_up_data.R

The final script uses the posted data to run the analyses in the experiment and produce the figures:

• 6_protist_analysis.R

Works cited

Altermatt, F., E. A. Fronhofer, A. Garnier, A. Giometto, F. Hammes, J. Klecka, D. Legrand, et al. 2015. Big answers from small worlds: a user’s guide for protist microcosms as a model system in ecology and evolution. (M. Spencer, ed.)Methods in Ecology and Evolution 6:218–231.

Altermatt, F., S. Schreiber, and M. Holyoak. 2011. Interactive effects of disturbance and dispersal directionality on species richness and composition in metacommunities. Ecology 92:859–870.

Giometto, A., F. Altermatt, A. Maritan, R. Stocker, and A. Rinaldo. 2015. Generalized receptor law governs phototaxis in the phytoplankton Euglena gracilis. Proceedings of the National Academy of Sciences 112:7045–7050.

Giometto, A., F. Altermatt, and A. Rinaldo. 2017. Demographic stochasticity and resource autocorrelation control biological invasions in heterogeneous landscapes. Oikos 126:1554–1563.

Jacquet, C., I. Gounand, and F. Altermatt. 2020. How pulse disturbances shape size-abundance pyramids. Ecology Letters 23:1014–1023.

Pennekamp, F., J. I. Griffiths, E. A. Fronhofer, A. Garnier, M. Seymour, F. Altermatt, and O. L. Petchey. 2017. Dynamic species classification of microorganisms across time, abiotic and biotic environments—A sliding window approach. (Z.-K. Gao, ed.)PLoS ONE 12:e0176682.

Petchey, O. L., P. T. McPhearson, T. M. Casey, and P. J. Morin. 1999. Environmental warming alters food-web structure and ecosystem function. Nature 402:69–72.

R Development Core Team. 2020. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Warren, P. H. 1996. Dispersal and Destruction in a Multiple Habitat System: An Experimental Approach Using Protist Communities. Oikos 77:317–325.