Do an ecosystem engineer and environmental gradient act independently or in concert to shape juvenile plant communities? Tests with the leaf-cutter ant Atta laevigata in a Neotropical savanna

Study site and system

We conducted our study at Panga Ecological Station (19°10′45″S, 48°23′44″W), a 404 ha reserve (Bruna et al., 2010) administered by the Universidade Federal de Uberlândia (UFU). The climate at Panga is highly seasonal, with mean annual temperature of ∼23° and most of the ∼1,600 mm of annual precipitation between October-April (UFU Santa Mônica Climate Station). Most of the major Cerrado vegetation types can be found at Panga Station, including the two known as cerrado ralo and cerrado denso (Cardoso et al., 2009; Appendix A). Cerrado ralo has a dense layer of grasses and herbs and sparsely distributed shrubs and trees typically <3 m tall; the average canopy cover in cerrado ralo is ∼30%. Cerrado denso has less grass cover and more abundant trees that can reach a height of ca. 6 m; average canopy cover in cerrado denso is ∼60%. There is large variation in the canopy cover of both vegetation types, however, so there can be strong gradients in canopy cover in landscapes where they abut. At Panga Station, for instance, the canopy cover gradient in the Cerrado ralo/Cerrado denso mosaic ranges from 0–95% (Mean = 52% ± 33.1 SD; Fig. 1A).

Our focal ecosystem engineer is Atta laevigata, whose large nest mounds are formed by workers depositing excavated soil around the main entrance to the nest. Atta laevigata is the most common Atta species in both cerrado ralo and cerrado denso (Costa & Vieira-Neto, 2016); A. sexdens is also found at Panga Station, but primarily in closed-canopy forest. In 2010 we haphazardly selected 10 active A. laevigata nests in each vegetation type (surface area of the N = 20 nests: 7–37 m², mean = 16.7 m² ± 6.7 SD). We then established three 1 ×2 m plots around each nest—one on the center of the nest mound, one immediately adjacent to the mound, and one 10 m from the mound edge (Appendix A). In each plot, we recorded the identity and height of all woody plants ≤1.2 m tall as well as data on environmental conditions (described below). Although there are some abandoned nests in our site, we restrict our analyses to active colonies because the effects of time-since-abandonment on environmental variables is unknown.

Environmental data

Litter biomass in each plot was measured by collecting all litter from a randomly selected half of each plot once during the 2010–2011 rainy season, drying it at 50 °C for 72 h, and weighing it with a microbalance. Similarly, we dried and weighed all grasses from a randomly selected half of each plot to estimate above-ground grass biomass. Canopy cover above each plot was estimated using photos analyzed with Adobe Photoshop (Adobe Systems Inc., San Jose, California, USA). In our analyses we used the average canopy cover in two photos taken during the same rainy season. Photos were taken with a Nikon Coolpix 950 from a height of 50 cm (modified from Engelbrecht & Herz (2001) in either the early morning (6h) or early evening (18h).

At the end of the 2011 dry season we estimated surface soil moisture content in plots by collecting a sample of the top 20 cm of soil from two points separated by 100 cm. These samples were bulked, weighed, dried at 50 °C for 96 h, then weighed again to estimate percent moisture content. As a proxy for soil compaction we used soil penetrability: we dropped a 1 m long × 5 mm diameter iron rod vertically from a height of 50 cm at three haphazardly selected points in each plot, then measured the depth to which the rod penetrated the soil at each point. We used the average of these values in our analyses; these data were recorded at the end of the 2010–2011 rainy season. Finally, at the end of the 2010–2011 rainy season we also counted all woody and herbaceous plants ≤ 120 cm tall in each plot and identified them with the help of local specialists and comparison with the collections of the UFU herbarium (HUFU). Of the 1,827 stems recorded 25% could only be identified to genus and morphospecies.

During the 2011 rainy season we selected N = 5 nests in Cerrado ralo and N = 5 nests in Cerrado denso nests for analyses of soil chemistry. For each nest we collected soils in the plots on nest mounds and the plots 10m from nests. We collected five soil samples of ∼100 g samples each from each plot: one from the plot center and one from each corner. The five samples from each plot were bulked into a single sample and taken to the Soils Analysis Lab of the Federal University of Uberlândia (UFU), where pH, P, K⁺, Ca²⁺, Mg ²⁺, Al³⁺, and total organic matter were measured using their standard protocols (EMBRAPA, 1997).

Statistical analyses: does Atta laevigata modify the gradient in canopy cover?

To test for an effect of plot proximity to A. laevigata nests on logit-transformed canopy cover (Warton & Hui, 2011) we used using Generalized Linear Mixed Models (GLMMs Bolker et al., 2009). The significance of plot proximity was assessed by comparing the model including only the random effect of nest identity with models including this random effect, plot proximity to ant nests, nest mound area as a covariate, and plot location × covariate interactions. All models used a Gaussian distribution with an identity function; nest mound area was not included as a covariate because preliminary analyses indicated it did not improve the fit of models.

Statistical analyses: do environmental conditions that influence plant establishment vary with canopy cover or proximity to A. laevigata nests?

We used Principal Components Analyses (PCA) to summarize environmental conditions in each plot because many of the biophysical variables we measured were highly correlated (Appendix B). The complete suite of environmental data was only collect in a subset of N = 10 nests, so we conducted two separate PCAs. The first was conducted using the environmental data collected in all N = 60 plots (i.e., plots on, adjacent to, and far from all N = 20 nests). These variables were: total grass biomass, total litter biomass, soil penetrability, surface soil moisture content, and percent canopy cover. The second was conducting using the subset of N = 10 nests for which we also collected data on soil chemistry; it was therefore the most comprehensive with respect to the environmental variables included, but was more limited in nest number and plot location because it only included plots on and far from the N = 10 nests. We hereafter refer to these PCAs as “PCA-1” and “PCA-2”, respectively. Both were conducted with correlation matrices because of the scales of each variable were different.

Each plot’s PCA scores are new variables that summarize local environmental conditions after controlling for correlation among the variables measured, and can therefore be used as dependent or independent variables in subsequent analyses (sensu Baiser et al., 2012). To determine if environmental conditions in a plot vary with canopy cover or proximity to A. laevigata nests we used a plot’s score on the 1st Principal Component Axis as the dependent variable in Generalized Linear Mixed Models with Gaussian errors. Plot location (i.e., on, adjacent to, or far from the nest) was a main effect with canopy cover included as a covariate. Although the effects of large colonies could potentially extend further from the nest boundary than those of smaller ones (Costa et al., 2008), we did not include nest area as a covariate because preliminary analysis indicated it did not improve the fit of models including just canopy cover. However, we did include nest identity as a random effect. The resulting models were ranked with Akaike Information Criteria (Burnham & Anderson, 2002) to determine which model best fit the observed data.

Statistical analyses: do A. laevigata and canopy cover act independently or in concert to influence plant abundance and species richness?

We used two sets of Generalized Linear Mixed Models with Poisson error distributions to determine if plant abundance and species richness in plots were best explained by ant-related factors (proximity to leaf-cutter ant nests, environmental conditions in plots), canopy cover, or a combination of both. The two analyses that were identical except for the PCA scores used to summarize local environmental conditions: the first group of models used the axis scores from ‘PCA-1’ i.e., all nests and plots but no data on soils), while the second used the axis scores from ‘PCA-2’ all environmental variables but fewer nests and plot locations). Juvenile plant abundance or richness were the dependent variables. Once again preliminary analyses indicated including colony area did not improve the fit of models. Nest identity was again included as a random effect; because of significant overdispersion the models for plant abundance also included a random per-observation term.

All analyses were conducted using the R statistical programming language (R Core Development Team, 2014). For the GLMMs we used the lme4 package (Bates et al., 2015), while PCAs were conducted with package ggbiplot (Vu, 2015).

Does Atta laevigata modify the canopy cover gradient?

The model that best fit the data on the amount of canopy cover over a plot is the one including only the random effect of nest identity (Table 1). This indicates A. laevigata colonies alter canopy cover around their nests, but not in a systematic way related to nest size, and that there is no predictable change in canopy cover as a function of proximity to ant nests (Fig. 1B).

Do environmental conditions that influence woody plant establishment vary with canopy cover or proximity to Atta laevigata nests?

Plots on nest edges and those far from nests overlapped in ordination space, indicating they had very similar environmental conditions (Fig. 2A). However, there was almost no overlap in ordination space between either of these locations and the plots located in the middle of A. laevigata nest mounds (Fig. 2A), even when the number of nests was reduced to include soil data (Fig. 2B). In ‘PCA-1’ the first axis explained 45.6% of the variance and was positively correlated with litter biomass and soil moisture content. The second axis explained an additional 29.6% of the variance; it was negatively correlated with grass biomass and soil penetrability (Table 2). In ‘PCA-2’ the first axis explained 42.9% of the variance and was positively correlated with litter and grass biomass, soil moisture content, and soil P, Al³⁺, and organic material (Table 3). The second axis explained 21.4% of the variance and was positively correlated with all other environmental variables measured. In light of these results we used the scores from the first principal components as the dependent variable in subsequent analyses.

When using the results of ‘PCA-1’, canopy cover over a plot was positively correlated with a plot’s PCA1 score (ρ = 0.44, t = 3.77, df = 58, p < 0.001), suggesting an association between canopy cover and local environmental conditions. However, the best-fit model included both canopy cover and plot location. This indicates leaf-cutter ants also influenced environmental conditions, but that the magnitude of the effect varied with plot proximity to nests (Table 4, Fig. 3A). When data on soils were included in the PCA, however, there was no longer a correlation between canopy cover over a plot and that plot’s score on the 1st PCA axis (ρ = 0.30, t = 1.35, df = 18, p = 0.20). Furthermore, the model that best fit the data on environmental conditions in a plot only included the proximity of a plots to ant nests and the random effect of nest identity (Table 5, Fig. 3B). In other words, the impact of ants on environmental conditions influencing establishment far outweighs that of canopy cover, but this is only revealed once data on soil properties are included in the analyses.

Do A. laevigata and canopy cover act independently or in concert to influence woody plant abundance and species richness?

We found 20.95 ± 18.14 SD recruits (range = 0–85) in each 2 m² study plot. However, the mean number of recruits plot⁻¹ decreased as one moved closer to the center of nests: plots far from nests had on average 29.55 ± 19.39 SD recruits in them vs. 24.9 ± 15.62 SD recruits plot⁻¹ on nest margins and 8.4 ± 11.93 SD recruits plot⁻¹ in the center of nest mounds. The mean number of species per plot was also lowest in plots on the center of nests (3.2 ± 2.9 SD) with almost four-fold higher species richness in plots on nest margins (10 ± 4.4 SD) and 10 m from nests (11.8 ± 4.6 SD).

Plant abundance in plots appears to be primarily influenced by ant-related factors (Figs. 4A, 4B). The best fit models include the proximity of plots to ant nests and environmental conditions in plots (Tables 6 and 7), though note d AIC values for the model including canopy cover was <2. Species richness in plots, however, appears to be shaped by both canopy cover and Atta-related effects (Figs. 4C, 4D). These results were consistent whether the PCA used to summarize environmental conditions in plots included data on soils or not (Tables 6 and 7). The significant effect of nest identity also indicates that some nests exert larger or smaller effects on local the abundance and diversity of recruits than others of similar size.

Implications for Cerrado plant communities

In Paleotropical savannas, canopy cover and herbivores interact in complex ways to influence soil properties and vegetation dynamics (Holdo & Mack, 2014; Rugemalila, Anderson & Holdo, 2016). In contrast to these ecosystems, however, the density and diversity of large mammalian herbivores in the Cerrado is very low (Marinho-Filho, Rodrigues & Juarez, 2002). This has led many to conclude that plant population and communities in this biome are largely structured by edaphic factors (reviewed in Hoffmann & Moreira, 2002; Mistry, 1998; Ruggiero et al., 2002) and that the influence of herbivores is negligible (e.g., Gardner, 2006). Although the key role of physical factors in plant recruitment in the Cerrado is indisputable (Hoffmann, 1996; Hoffmann, 2000; Salazar et al., 2012a; Salazar et al., 2012b), studies evaluating the impacts of herbivores are rare (Ferreira, Bruna & Vasconcelos, 2011; Mundim et al., 2012), especially those simultaneously assessing the effects of herbivores and edaphic conditions (e.g., Klink, 1996). Our study supports Costa et al.’s (2008) hypothesis that plant-herbivore interactions can herbivores play a dominant role in Cerrado plant demography. Furthermore, our results provide compelling evidence that leaf-cutter ants do so both directly by harvesting seedlings and as ecosystem engineers modifying the conditions influencing plant recruitment, growth, and survival. As such, failing to consider the myriad impacts of these keystone species will undermine attempts to develop general theory for vegetation dynamics in this biome (e.g., Gardner, 2006), as well as conservation and restoration efforts.

Future directions

How the spatio-temporal impacts of engineers are influenced by disturbance type, frequency, and intensity is conceptually critical (Crain & Bertness, 2006) but conspicuously understudied (Hastings et al., 2007). Fire is one of the most important forms of disturbance in savannas worldwide, where it can strongly influence seedling establishment (e.g., Iacona, Kirkman & Bruna, 2010), environmental gradients, and the foraging of leaf-cutter ants (Lopes & Vasconcelos, 2011). However, the density and abundance of Atta nests can also influence how fire spreads (Carvalho et al., 2012) and post-fire nutrient availability (Sousa-Souto, Schoereder & Schaefer, 2007). We suggest future studies in this system should focus on how Atta’s engineering of seedling communities is influenced by fire and fire-Atta feedbacks. Understanding these disturbance-engineer interactions is especially important given how deforestation, road creation, and other human activities lead to more frequent fires (Nepstad et al., 1999) and elevated Atta abundance (Cameron & Bayne, 2009; Vasconcelos et al., 2006; Vieira-Neto, Vasconcelos & Bruna, 2016). The ecological and economic footprint of these engineers is therefore likely to increase dramatically in coming decades in ways that remain underappreciated and poorly understood.