Study sites
The study was carried out in sites located in the Doce River Basin, in Minas Gerais State, Brazil, in the upper and mid-basin regions (Fig. 1). This is an 83,400-km2 basin, the third largest in the State. We sampled in three State Parks: Rio Doce (PERD), Itacolomi (PEIT), and Serra de Ouro Branco (PESOB). The PERD (19°45′ S 42°38′ W) represents the largest remnant of semi-deciduous Atlantic Forest of Minas Gerais with an area of approximately 36,000 ha and altitude ranging between 200–500 m. The climate is tropical seasonal Aw according to Köppen (Alvares et al. 2014). Seasons are well defined, with wet season lasting for 7–8 months and dry season drought for 4–5 months, with average rainfall of 1,500 mm per year (Silva 2001).
The PEIT (20°22′30′′ S and 43°22′30′′ W) is located in the southern portion of the Espinhaço Range in the state of Minas Gerais. It has an approximate area of 7,000 ha, with altitude varying between 1400-1772m a.s.l. The predominant vegetation communities are distributed in quartzitic and ferruginous rock fields, formations of campo rupestre, montane forest and natural monodominant stands of Eremanthus erythropappus (Asteraceae), a pioneer tree (Fujaco et al. 2010). Average annual rainfall is 1,500 mm, summers are hot, rainy and last approximately 8 months and the dry season, lasts between 3 and 4 months. The climate is tropical humid type Cwb following the Köppen classification (Alvares et al. 2014).
The PESOB (20°31' S, 43°41' W) is situated in the southern boundary of the Espinhaço Range and has an approximate area of 7,500 ha. The altitude in the Park varies between 800 and 1,510 m a.s.l and vegetation is characterized by campo rupestre in highlands and, in the lower portions, by riparian forests and forest patches embedded inside a grassland ecosystem (Instituto Estadual de Florestas, 2015). The climate is tropical humid type Cwb following the Köppen classification and has similar annual average rainfall and seasons as PEIT (Alvares et al. 2014).
Considering the three areas sampled, this study was conducted along a gradient of resource availability, from a habitat of poor, shallow and sandy soils towards deep and nutrient-rich soils.
Plant species
The genus Byrsonima is one of the most important genera of the Malpighiaceae due to its large number of species, with 135 described species restricted to tropical and subtropical regions (Anderson 1977; Davis and Anderson 2010). The genus occurs throughout the Neotropical regions and in Brazil in the phytogeographical domains of the Amazon, Cerrado, Caatinga, Atlantic Forest and Pantanal (Mamede and Francener 2015). Byrsonima sericea DC. is a tree species with wide distribution (Mamede and Francener 2015). In our study system, we sampled this species in the forest habitat of the PERD, specifically in ecotonal habitats between forests and natural lakes, where its crowns grow branched in search of light. Byrsonima variabilis A. Juss. is a shrub species, endemic to the Brazilian mountains highlands (Mamede and Francener 2015). This species population in the PEIT was sampled in habitats of rocky outcrop in campo rupestre. Their crowns have cylindrical shapes with branches not far apart and small internodes. Byrsonima subterranea Brade & Markgr. is a sub-shrub species that occurs in campo rupestre Cerrado (latu sensu) and Amazonian Savanna (Mamede and Francener 2015). The individuals of this species in PESOB were studied in campo rupestre, in sandy quartzite soil, called white sand soil. They grow with subterranean stems and only the leaves and reproductive parts are aerial, hence not forming a typical crown.
Sampling design
Each species of Byrsonima was treated as a life form. We sampled three populations of B. sericea (tree) in PERD, three populations of B. variabilis (shrub) in PEIT and three populations of B. subterranea (sub-shrub) in PSOB. In each population we sampled 15 individuals, totalling nine populations and 135 individuals. All data were collected between January and March 2016, at the end of the rainy season, with the aim to quantify the accumulated herbivory during the last rainy season.
Soil nutrients
For soil characterization, one sample was collected at every three individual plants, totalling five samples per population. We chose to collect the soil at 10 cm depth due to its relationship to vegetation characteristics, as suggested by Ruggiero et al. (2002). The soil parameters analysed were pH, organic matter (OM), and concentrations of nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and aluminium (Al). All nutrient analyses were performed according to Defelipo and Ribeiro (1997), Embrapa (1999), and Raij et al. (2001), at the Laboratory of Plant Analysis of the Department of Soils of the Agricultural Sciences Centre of the Federal University of Viçosa, Minas Gerais.
Functional traits and Herbivory
Plant Architecture
Due to structural differences among species, we adjusted the architectural complexity measurements for the three species, allowing for direct comparisons between their different life forms. The following parameters were measured: total height (cm), trunk diameter below the first branch (cm), number of growth units, and mean leaf number (see Pérez-Harguindeguy et al. 2013). Growth units were defined as the set of terminal branches of each ramification from which the leaves originated (see Bell et al. 1999). In each individual plant we counted all the growth units.
Leaf nutrients and sclerophylly
To quantify leaf nutrient content, a set of 20 mature, fully expanded leaves of each individual were collected and oven dried to quantify N, P, K, Ca and Mg content. Leaf sclerophylly was measured as leaf thickness (Choong et al. 1992), using a Digimess digital micrometer with an accuracy of 0.001 mm. This method is correlated with other methods of measuring sclerophylly, such as dry weight, fibre content and protein content (Choong et al. 1992; Pérez-Harguindeguy et al. 2013).
Leaf area, leaf density and herbivory
Leaf samples were taken using a random delimited volume within each crown. For such, we used three hollow cubes, thus combining the canopy cylinder method (Ribeiro and Basset 2007) with the subplot frame model (Shaw et al. 2006) to sample leaves and quantify herbivory levels. This method was selected to avoid biased sampling, commonly documented in herbivory studies (Kozlov et al. 2015). The cube volume was chosen to control differences in the size of each species and to fit the approximately area occupied by the foliage of a growth unit. In B. variabilis, three cubes with 20 cm length, 20 cm width, and 20 cm height were used, with a total volume of 0.06 m³. In B. sericea three cubes, 30 cm long, 30 cm wide and 30 cm high, with a total volume of 0.09 m³ were used for sampling. Each cube was positioned in three different branches (chosen at a distance of 5 meters from the plant to avoid bias) of the crown of each individual and all leaves inside the cube were collected (Fig. 2). Due to its small size, the cube was not used in B. subterranea and all leaves of each plant were collected.
To quantify herbivory, leaf density and leaf area, all mature and fully expanded leaves per individual (collected as described above) were counted and digitized. Leaf density was considered as the total number of leaves collected. Average leaf area and the percentage of leaf damage by chewing insects (in cm2) were calculated through Image J♥ 1.6.0 software.
Statistical analyses
We tested whether each site differed for soil nutrients using an Analyses of Variance (ANOVA) for each element, considering each environment as a level of a simple fixed factor. The functional traits were explored to collinearity, and in cases when variables were positively correlated (Pearson > 0.70), only the variable with the greatest biological meaning for the hypotheses tested was used. Normality was tested for the models constructed based on error distribution. Non-significant explanatory variables were eliminated from analyses to obtain an adequate minimum model (Crawley 2013).
To test the hypothesis that the functional traits and herbivory of plants responded to the resource availability gradient, a multiple regression analysis was performed. The functional traits and herbivory were used with response variables and soil nutrients with explanatory variable. To test whether functional traits affect herbivory levels in Byrsonima sp., an analysis of covariance (ANCOVA) was performed with error distribution adjusted for quasibinomial, with leaf area weighted by leaf size. Species and environment were considered as fixed factors, populations as nested randomized blocks, and architectural measures, leaf nutrients, and sclerophylly as covariates. The average leaf area lost by each individual was the response variable.
Statistical analyses were performed in R Environment v. 3.2.0 (R Development Core Team 2015) and ANOVA, multiple regression analysis and ANCOVA were completed using generalized linear models (GLM).