The difficult interpretation of species co-distribution
The taxonomic and functional biogeographies of phytoplankton and zooplankton communities across boreal lakes
Ecology is the study of the distribution of organisms in space and time and their interactions. As such, there is a tradition of studies relating abiotic environmental conditions to species distribution, while another one is concerned by the effects of consumers on the abundance of their resources. Interestingly, joining the dots appears more difficult than it would suggest: eluding the effect of species interactions on distribution remains one of the greatest challenges to elucidate nowadays (Kissling et al. 2012). Theory suggests that yes, species interactions such as predation and competition should influence range limits (Godsoe et al. 2017), but the common intuition among many biogeographers remains that over large areas such as regions and continents, environmental drivers like temperature and precipitation overwhelm their local effects. Answering this question is of primary importance in the context where species are moving around with climate warming. Inconsistencies in food web structure may arise with asynchronized movements of consumers and their resources, leading to a major disruption in regulation and potentially ecosystem functioning. Solving this problem, however, remains very challenging because we have to rely on observational data since experiments are hard to perform at the biogeographical scale.
The study of St-Gelais is an interesting step forward to solve this problem. Their main objective was to assess the strength of the association between phytoplankton and zooplankton communities at a large spatial scale, looking at the spatial covariation of both taxonomic and functional composition. To do so, they undertook a massive survey of more than 100 lakes across three regions of the boreal region of Québec. Species and functional composition were recorded, along with a set of abiotic variables. Classic community ecology at this point. The difficulty they faced was to disentangle the multiple causal relationships involved in the distribution of both trophic levels. Teasing apart bottom-up and top-down forces driving the assembly of plankton communities using observational data is not an easy task. On the one hand, both trophic levels could respond to variations in temperature, nutrient availability and dissolved organic carbon. The interpretation is fairly straightforward if the two levels respond to different factors, but the situation is much more complicated when they do respond similarly. There are potentially three possible underlying scenarios. First, the phyto and zooplankton communities may share the same environmental requirements, thereby generating a joint distribution over gradients such as temperature and nutrient availability. Second, the abiotic environment could drive the distribution of the phytoplankton community, which would then propagate up and influence the distribution of the zooplankton community. Alternatively, the abiotic environment could constrain the distribution of the zooplankton, which could then affect the one of phytoplankton. In addition to all of these factors, St-Gelais et al also consider that dispersal may limit the distribution, well aware of previous studies documenting stronger dispersal limitations for zooplankton communities.
Unfortunately, there is not a single statistical approach that could be taken from the shelf and used to elucidate drivers of co-distribution. Joint species distribution was once envisioned as a major step forward in this direction (Warton et al. 2015), but there are several limits preventing the direct interpretation that co-occurrence is linked to interactions (Blanchet et al. 2020). Rather, St-Gelais used a variety of multivariate statistics to reveal the structure in their observational data. First, using a Procrustes analysis (a method testing if the spatial variation of one community is correlated to the structure of another community), they found a significant correlation between phytoplankton and zooplankton communities, indicating a taxonomic coupling between the groups. Interestingly, this observation was maintained for functional composition only when interaction-related traits were considered. At this point, these results strongly suggest that interactions are involved in the correlation, but it's hard to decipher between bottom-up and top-down perspectives. A complementary analysis performed with a constrained ordination, per trophic level, provided complementary pieces of information. First observation was that only functional variation was found to be related to the different environmental variables, not taxonomic variation. Despite that trophic levels responded to water quality variables, spatial autocorrelation was more important for zooplankton communities and the two layers appear to respond to different variables.
It is impossible with those results to formulate a strong conclusion about whether grazing influence the co-distribution of phytoplankton and zooplankton communities. That's the mere nature of observational data. While there is a strong spatial association between them, there are also diverging responses to the different environmental variables considered. But the contrast between taxonomic and functional composition is nonetheless informative and it seems that beyond the idiosyncrasies of species composition, trait distribution may be more informative and general. Perhaps the most original contribution of this study is the hierarchical approach to analyze the data, combined with the simultaneous analysis of taxonomic and functional distributions. Having access to a vast catalog of multivariate statistical techniques, a careful selection of analyses helps revealing key features in the data, rejecting some hypotheses and accepting others. Hopefully, we will see more and more of such multi-trophic approaches to distribution because it is now clear that the factors driving distribution are much more complicated than anticipated in more traditional analyses of community data. Biodiversity is more than a species list, it is also all of the interactions between them, influencing their distribution and abundance (Jordano 2016).
Blanchet FG, Cazelles K, Gravel D (2020) Co-occurrence is not evidence of ecological interactions. Ecology Letters, 23, 1050–1063. https://doi.org/10.1111/ele.13525
Godsoe W, Jankowski J, Holt RD, Gravel D (2017) Integrating Biogeography with Contemporary Niche Theory. Trends in Ecology & Evolution, 32, 488–499. https://doi.org/10.1016/j.tree.2017.03.008
Jordano P (2016) Chasing Ecological Interactions. PLOS Biology, 14, e1002559. https://doi.org/10.1371/journal.pbio.1002559
Kissling WD, Dormann CF, Groeneveld J, Hickler T, Kühn I, McInerny GJ, Montoya JM, Römermann C, Schiffers K, Schurr FM, Singer A, Svenning J-C, Zimmermann NE, O’Hara RB (2012) Towards novel approaches to modelling biotic interactions in multispecies assemblages at large spatial extents. Journal of Biogeography, 39, 2163–2178. https://doi.org/10.1111/j.1365-2699.2011.02663.x
St-Gelais NF, Vogt RJ, Giorgio PA del, Beisner BE (2021) The taxonomic and functional biogeographies of phytoplankton and zooplankton communities across boreal lakes. bioRxiv, 373332, ver. 4 peer-reviewed and recommended by Peer community in Ecology. https://doi.org/10.1101/373332
Warton DI, Blanchet FG, O’Hara RB, Ovaskainen O, Taskinen S, Walker SC, Hui FKC (2015) So Many Variables: Joint Modeling in Community Ecology. Trends in Ecology & Evolution, 30, 766–779. https://doi.org/10.1016/j.tree.2015.09.007
Wisz MS, Pottier J, Kissling WD, Pellissier L, Lenoir J, Damgaard CF, Dormann CF, Forchhammer MC, Grytnes J-A, Guisan A, Heikkinen RK, Høye TT, Kühn I, Luoto M, Maiorano L, Nilsson M-C, Normand S, Öckinger E, Schmidt NM, Termansen M, Timmermann A, Wardle DA, Aastrup P, Svenning J-C (2013) The role of biotic interactions in shaping distributions and realised assemblages of species: implications for species distribution modelling. Biological Reviews, 88, 15–30. https://doi.org/10.1111/j.1469-185X.2012.00235.x
Dominique Gravel (2021) The difficult interpretation of species co-distribution. Peer Community in Ecology, 100082. https://doi.org/10.24072/pci.ecology.100082
Evaluation round #2
DOI or URL of the preprint: 10.1101/373332
Version of the preprint: 2
Decision by Dominique Gravel, 02 Dec 2019
The answers to the different comments by the two reviewers were satisfactory and I am positive about the preprint. The main reason I find it interesting is that the study of functional traits could be a much more conclusive approach to investigate co-distribution between consumers and resources than the morecommon pairwise approach. It is hard to detect the spatial association between a given consumer and its resource if both are involved in multiple interactions. Traits may be a more relevant way to quantify the total amount of interactions a species experiences than the presence of a single resource or predator.
That said, even though the manuscript could be publishable somewhere, I am not ready to write a recommendation with my name associated to it. Unfortunately, answers to my comments were the weakest of the rebutal. The reason is simple, I find that the predictions are not clear enough and as a result, conclusions are fairly limited.
The take-home is nicely summarized at L381 : "Because no coupling between phytoplankton and zooplankton was observed after controlling for either water quality, morphometry or space, it suggests that one of these is the main driver of the observed coupling." I don't think this observation of no residual covariation is sufficient to make any strong conclusion about the effect of interactions on co-distribution. The authors know the story better than I do with seasonal fluctuations in phyto an zoo composition. It's well-known there is a shift in body size distribution over the season because of an initial bloom of nutrients and primary productivity, followed by a delayed response of grazers. Both bottom-up and top-town regulation are driving this covariation over the season. If one simply models the composition of both groups with time as the main predictor, we would not expect much residual covariation between them. In more general terms, if a given abiotic variable is driving the variation of a group and this one is clearly impacting the distribution of the other group, then both of them will respond to the abiotic variable and there won't be much residual covariation.
I am sure there is enough in the Procust analysis to conclude if proper predictions were made, but they are not provided at the moment.
-- As a side, I was not suggesting to use JSDM with traits as covariates with the HMSC package. I was simply suggesting to use JSDM to model simultaneously all of the species (both phyto and zoo), using the group as a dummy trait variable. This way, the residual covariation would be properly represented and all of the operations performed with a singleanalysis. I am not strong about this suggestion though, I understand the extra amount of work it represents. It would only make the analysis simpler, and incidently, more elegant.
Evaluation round #1
DOI or URL of the preprint: 10.1101/373332
Version of the preprint: 1
Decision by Dominique Gravel, 26 Sep 2018
The manuscript reports an investigation of the co-distribution of phytoplankton and zooplankton communities in boreal lakes of Northern Québec, Canada. It aims at testing the hypothesis that trophic regulation by zooplankton should impact the distribution of phytoplankton, the main prediction being a correlation between community compositions. This is a big problem in biogeography right now, explored in many systems with a wide variety of approaches, ranging from species pairs analysis with sophisticated distribution models to the exploration of food web beta-diversity. The originality of the study is the simultaneous analysis of taxonomic and functional co-distribution between groups. The observation that both aspects of community structure are correlated among trophic levels, but that this correlation disappears once the effect of the abiotic environment is taken out, is interpreted as evidence that trophic interactions do not matter at this spatial scale for community distribution. Both reviewers and myself agree that there is a lot of potential with the manuscript and would be happy to provide a positive recommendation after appropriate corrections. The reviewers are constructive and provide several specific ways to improve the manuscript. I invite the authors to consider each of them in their reply, but more specifically, I would like the authors to consider explicitly the limitation of the analysis of correlation between trophic levels (a point that is common to both reviews). I personally think that this point would be best addressed with a better explicit review of theory in the introduction, in order to formulate specific and discriminant predictions to test (and relate them to the statistical analysis). In particular, I would like the authors to compare the expected co-distributions in situation of a bottom-up assembly of food webs (which I think should lead to a positive correlation between traits) and a top-down assembly (leading to a negative correlation). Statements such as "It is also important to acknowledge that the coupling between plankton groups that we attribute to environmental factors could mask the effect of trophic interactions. " are vague and likely the result of inadequate predictions. In addition to the different comments of the reviewers, I would encourage the authors to explore the possibility of using Joint Species Distribution Models such as the one described in Ovaskainen et al. 2017 in Ecology Letters. It is mentioned in the introduction that "We expected that using functional traits characterizing the trophic interaction would improve our ability to detect joint distributions". This is very precisely the objective JSDMs, and a much more powerful approach to describe multivariate (community) data than RDAs. Such models aims at representing the covariance among species and groups once controlling for the environment. Such analysis use latent variables to deal with missing predictors (a point raised by one of the reviewers) and allows a much more detailed representation of the data structure. The "spatial" model also needs revision because it is way too simple to either use euclidean distance among sites or lat/long coordinates as predictors. There are much more flexible algorithms that could be used to represent spatial autocorrelation. Further, the underlying hypothesis must be better explained and the interpretation of "space = dispersal limitations" taken with caution. That was a standard approach 15 years ago, but we now have a much better understanding of the potential drivers of spatial autocorrelation. I would also like to personally thank the authors for encouraging the new publishing model proposed by PCI and wish them success with the communication of their study.