Trait divergence in plant community assembly is generated by environmental factor interactions

Prepared by Valério D. Pillar

Expected values of a trait (t1) in a set of simulated communities. The trait is selected by two random environmental factors (e1 and e2) according to specified linear functions. In (a), trait t1 is additively affected by factors e1 and e2, and in (b), it is determined by both factors plus their interaction, according to linear function slopes shown in the insets. The red trendline represents the response of trait t1 modelled a posteriori with the generated data, using only factor e1 as a predictor and ignoring factor e2 and the interaction (greyed in the insets), as if e2 were unknown. The effect of simulated values of factor e2 on trait t1 was nested within nesting factor e1. That is, larger units contained ten subunits that did not vary for e1 but did for e2. The corresponding variance of trait t1 within each of these nesting units and its observed response to factor e1 are shown in graphs c-d. (Credit: Valério Pillar, from the paper).

The prevailing explanation for trait divergence within communities is that competition between plants is avoided when assembled plants differ in traits relevant to their establishment and maintenance at the community site. However, this fails to explain why some plant communities, sharing the same regional species pool, i.e., in the same metacommunity, are more trait-diverse than others. Given that such variation is often related to environmental factors, I was puzzled as to whether environmental filtering, which selects plants with suitable trait values for local conditions and is usually thought to increase trait-convergence, could also generate trait-divergence. Note that competition and facilitation are often mediated by environmental factors that interacting plants modify at fine spatial and/or temporal scales, enhancing environmental heterogeneity. These altered environmental factors, generated through feedback mechanisms, act as selecting filters throughout the process of community assembly and might remain hidden and not readily observable, particularly when they affect community assembly at a finer scale than the resolution of the studied community units.

I demonstrate by community assembly simulation that trait divergence can be generated by filtering when the environmental factors interact in their selecting effects. Imagine expected values of a trait (t1) when species varying in the values of that trait are selected by two environmental factors (e1 and e2) and their interaction. The interaction modifies the strength and direction of one factor’s effect based on the other factor. Consequently, this affects the variation of expected trait values across communities that share similar environmental conditions given by factor e1 but vary regarding factor e2. Assume that only factor e1 can be measured and analysed, while factor e2 is hidden and not measurable. The result is an observable funnel-shaped pattern of expected trait values along factor e1, which I measured as environmentally-driven beta trait-divergence. If the interacting factors are spatially nested in their effects on the expected values of a trait, what is perceived as beta trait divergence at finer scales becomes trait divergence within communities (alpha trait divergence) when observed at coarser scales. That is, beta and alpha trait divergence are interconnected and scale-dependent when the interacting factors are spatially nested. Such trait divergence is not solely a consequence of environmental heterogeneity. Instead, it results from the interactive effects of environmental factors occurring in a heterogeneous environment.

I tested these ideas with a large number of simulated metacommunities, which were generated by an individual-based, spatially explicit simulation model considering spatially nested, feedback-generated environmental factors that filtered species by traits. For each metacommunity, the significance of beta and alpha trait divergence was tested by permutations under a proper null model.

Significant beta divergence was frequent with factor interactions incorporated in community assembly simulations, while it mostly remained within expected Type I error range when factor interactions were absent. Beta divergence was stronger than alpha divergence at a finer spatial resolution and weaker than alpha divergence when smaller community units were combined into larger units. Additionally, the method was applied successfully to grassland and soil data collected in plots across southern Brazil.

This is a plain language summary for the paper of Valério D. Pillar, published in the Journal of Vegetation Science (https://doi.org/10.1111/jvs.13259)