Interactive – not only additive – models predict trait-by-environment effects on fitness

Nuria Pistón and colleagues recently had their paper on functional traits and fitness published in Journal of Ecology. Nuria tells us more about the study below. 


Trait-based approaches are extensively used in ecology to examine species interactions and species response to environmental change. However, although the link between traits and fitness is essential for the definition of functional traits, evidence linking traits and fitness proxies remains scarce and generally weak. Does this mean that traits do not influence fitness? Not necessarily. Rather, we think that we are not looking at this problem in the best possible way.

In our paper we tested whether considering multiple traits that represent different ecological strategies can help us to uncover the relationship between traits and demographic components underlying fitness. This also allowed us to examine the existence of a single optimum (i.e. a single peak with maximum performance in the fitness landscape) vs. alternative designs (i.e. more than one peaks). Despite the great variability in plant forms that is usually observed in natural communities, the exploration of alternative designs and whether they could improve the predictability of trait-fitness relationships within an environment has not been tapped. This is not surprising, since this very task is challenging, requiring both functional trait and demographic data from the same species/sites under variable environmental conditions, and a robust methodological approach to account for multiple trait interactions. Here, we used boosted regression trees (BRT) to test hypotheses on trait combinations compared to standard linear models. We used two published datasets for comparisons: i) an observational dataset with five traits, and elasticities of three vital rates (survival, growth, and reproduction) of 222 plant species worldwide (Adler et al., 2014), and ii) a botanical garden dataset with 11 traits and, spontaneous seed and vegetative reproduction of 951 plant species from the collection of native plants of central European Flora (Herben et al., 2012).

Using the observational datasets, we found that BRTs outperform linear models in predicting the role of traits, and their interactions, on the relative importance of survival, growth, and reproduction for population growth rate (Fig 1). In order to make a fair comparison between methods we needed to remove species for which we were lacking information for the considered traits. This is because linear models fully disregard observations when there are missing values for any of the predictors, while BRTs can deal with missing values, which are so frequent in trait information. Therefore, if we hadn’t had to remove species, the remarkable increases in predictive ability when using BRTs would likely even have been bigger than we show. Another important advantage of BRTs is their ability to automatically fit interactions between predictors as well as non-linear responses in a way that cannot be achieved with linear models.

Figure 1

Figure 1. Boosted Regression Trees (BRT) models are better predictors of the relationship between traits and fitness than linear models alone. We obtained observed vs. predicted values for both types of models using data from three vital rates elasticities (survival, growth, and reproduction) of plant species worldwide (observational dataset; Adler et al. 2014). For BRTs (orange lines), we calculated one model for each of the three vital rate elasticities using SLA, leaf N content and seed mass using tree complexity (tc) optimum and selected parameters (See Table S3 in Pistón N et al., 2019). For linear models (LM; green lines), we also performed a standard major axis regression for each of the three vital rate elasticities including the same traits, their pairwise interactions, and the quadratic terms of the traits. We show Pearson correlation coefficient on the top-left.

Using the botanical garden dataset, we found that traits shape fitness through their interactive effects. For instance, in 83% of the cases, the values of vegetative reproduction were better explained by models including interactions among traits than with models without interactions (Fig 2). Our results highlight the crucial importance of considering interactions among traits that depict different ecological dimensions to reliably assess trait-by-environment effects on fitness.

Fig2

Figure 2. Vegetative reproduction depends on interactions between aboveground, bud banks, and clonal traits. The ranked relative influence of multiple traits was calculated by the mean of 100 simulations for a tree complexity (tc) optimum value using Boosted Regression Trees (BRT) (See Fig. 2 in Pistón N et al., 2019) and the botanical garden dataset. White colour indicates aboveground traits while grey colour indicates bud bank and clonal growth traits.

While traits interact in complex ways, resulting in different combinations of traits that can yield equivalent fitness values, we only found alternative designs in one (meadow) of the six habitats we evaluated from the botanical garden dataset (Herben et al., 2012). Both high and low SLA values combined with high lateral spread resulted in similarly high vegetative reproduction (Fig 3). As SLA is a proxy of individual growth rate, our findings suggest that both slow and fast-growing clonal plants that are able to spread further away from the ramets are successful in the meadow habitat.

Figure 3

Figure 3. Partial figure from Pistón et al. (2019). Fitness landscapes of the first ranked pairwise interactions using the optimum values of model parameters (See Fig 2 in Pistón N et al., 2019). The strength of two-way interactions between determinants of vegetative reproduction is shown above. LS: lateral spread; SLA: specific leaf area.

Our study demonstrates that including complex trait interactions is necessary to adequately quantify the multiple dimensions of plant traits and how these shape their fitness. The analysis of trait combinations, and corresponding alternative designs via BRTs represent a promising approach for understanding and managing functional changes in vegetation composition through measurement of suites of relatively easily measurable traits.

Nuria Pistón, Federal University of Rio de Janeiro – UFRJ, Brazil


Read the full paper: Multidimensional ecological analyses demonstrate how interactions between functional traits shape fitness and life history strategies

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s