Introduction

Tropical forest canopies support a high diversity of organisms and life forms (Wardhaugh 2014) and may contain half of the diversity within a forest (Stork and Grimbacher 2006; Basset et al. 2012). Among the insects that occupy canopies, ants stand out, representing 20–40% of the arthropod biomass recorded in that habitat (Tobin 1995). Canopy ant species can display major dominance and territoriality on canopies (Majer 1972; Majer et al. 1994), resulting in well-defined territories of dominant and aggressive ant species (Dejean et al. 2015). These dominant ant species patrol and occupy areas without overlapping territories that, however, could be pervaded by subdominant species (Majer 1972; Majer and Delabie 1993; Majer et al. 1994; Dejean et al. 2007). Furthermore, epigeic ant species can also forage on vegetation (Rico-Gray and Oliveira 2007; Marques et al. 2017).

Ant species foraging patterns are associated with the distribution and the type of available resources (Gordon 2012; Klimes et al. 2015), related to vegetation traits, for instance, tree height (Campos et al. 2006; Dejean et al. 2008; Janda and Konečná 2011; Klimes et al. 2012) and canopy stratum (i.e., luminosity) (Blüthgen and Stork 2007; Ribeiro et al. 2013). Therefore, it is expected that taller trees may favor an increase in the number of microhabitats, as they have more space that can be used as nesting sites (Powell et al. 2011; Klimes et al. 2015). Furthermore, taller trees may also offer more sheltering sites and food (Campos et al. 2006; Tanaka et al. 2010; Yusah and Foster 2016), for instance, extrafloral nectaries (EFN), allowing the establishment of ant species related to the use of these feeding structures by the ants (Schoereder et al. 2010; Fagundes et al. 2016). Another important characteristic of the plant community in forests related to the ant fauna is the connectivity among treetops in the canopy, which increases the foraging area and potential territories for dominant ants (Majer and Delabie 1999; Yanoviak 2015).

The literature on canopy insect ecology has shown that any pattern in insect distribution in forests depends on the forest vertical size scale as well as the canopy strata (Basset et al. 2015). In a study of Amazonian understory, Falcão et al. (2015) found that the diversity of ant-plant interactions increased along different successional stages. In a study in Neotropical montane forest canopies, ant species dominance was more frequent in highly disturbed sites, and this pattern differed among tree crowns of the entire canopy (Lourenço et al. 2015). In a tropical forest in New Guinea, Klimes et al. (2012) found a greater species richness and a different ant species composition in late than in early successional stages. In summary, our knowledge about the relationships between canopy ants and successional stages lacks data, since previous studies have been restricted to the understory, to montane forest, or to evergreen forest. Furthermore, some studies have not considered the persistent ant foraging patterns on the canopy over time.

The lack of knowledge about canopy ants is especially evident in tropical seasonal environments, such as tropical dry forests (TDFs) (Castaño-Meneses 2014; Vergara-Torres et al. 2017). TDF plant communities present strong seasonality changes, shedding up almost of their leaves during the dry season (Pezzini et al. 2014). In addition, these forests are subject to prominent levels of anthropogenic perturbation (Arroyo-Mora et al. 2005), forming landscapes in different successional stages (Arroyo-Mora et al. 2005). Previous studies in tropical dry forests found changes in ant community compositions along different successional stages (Neves et al. 2010; Sousa-Souto et al. 2016; Marques et al. 2017) and regions (Silva et al. 2017), but were restricted to the tree trunks, the understory, or the ground level. Thus, further investigation is needed to understand the dynamics of canopy ant species, the trees they inhabit, and how they interact.

Ants interact with plants in different ways, such as protecting the plant from some herbivorous insects (Rico-Gray and Oliveira 2007), and the innumerable possibilities of these interactions can produce complex networks between ants and plants (Del-Claro et al. 2018). Based on studies of such networks, it is possible to assess properties at both the species level and in the community as a whole (Bascompte 2007; Antoniazzi et al. 2018). However, measuring the interaction networks in canopy habitats is not a trivial task (Lowman et al. 2012) due to the difficulty in accessing and sampling high above the ground. Although the knowledge of insect diversity, species distributions, and herbivory in the canopy has increased tremendously in the past years (Ribeiro et al. 2013; Neves et al. 2014; Basset et al. 2015; Leal et al. 2016; Nakamura et al. 2017), there have not yet been similar efforts to measure the interaction networks in canopy ant communities (DaRocha et al. 2016).

This study aimed to evaluate how canopy traits (i.e., tree height, tree connectivity, presence of EFNs, and successional stage) structure the ant assemblage in a TDF canopy. Hence, we postulated that taller trees, with more connected crowns, with EFNs, and located in advanced successional stages will have a higher ant species richness, as there are more available resources, a higher possibility of locomotion between tree crowns, and distinct kinds of microhabitats at the canopy. In addition, since arboreal-dwelling ants dominate and defend canopy resources, certain ant species could distribute in a set of trees, in territories across tree crowns; thus, we expected that interaction networks involving individual trees and ant species should be characterized by a low co-occurrence of certain ant species on a set of trees.

Material and methods

Study area

The study was conducted in the Parque Estadual da Mata Seca (PEMS; 14° 48′ 36″–14° 56′ 12″ S and 43° 55′ 12″–44° 04′ 12″ W; 493 m a.s.l.) in SE Brazil, in the transition zone between three important Brazilian biomes: Cerrado (Brazilian savanna), Caatinga (spiny-dry forest), and Mata Atlântica (Brazilian Atlantic Forest). Inserted in a mosaic of pristine (sensu undisturbed for a long time) and secondary forests (previously used for agriculture and livestock activities), the PEMS covers a total area of 15,466 ha (Madeira et al. 2009). The climate of the region is tropical semi-arid (Aw in Köppen’s classification) (Alvares et al. 2013), characterized by a severe dry season during the winter (May through September). The average annual temperature is 25.1 °C, with an average annual precipitation of 818 ± 242 mm (mean ± SD), concentrated between November and April (Pezzini et al. 2014).

The original vegetation in the area is a tropical dry forest on fertile soils and mostly flat terrains. In this region, the tropical dry forest (TDF) is dominated by deciduous vegetation, losing between 90 and 95% of its leaves during the dry season (Pezzini et al. 2014). We chose areas in two successional stages, henceforth early and late stage, with striking differences in their structure and land-use history (Madeira et al. 2009; Pezzini et al. 2014). The early stage was a pasture in recovery for about 15 years and composed of trees < 8 m in height, with an understory of herbs and grasses, without a distinguishable vertical stratification. The later stage had experienced little or no anthropogenic intervention for at least 60 years. There were emergent trees that exceeded 20 m in height, a distinct canopy and understory strata, with light input at the soil level only during the dry season. In this area, there were fewer new trees and lianas when compared with other adjacent areas at different levels of regeneration (for a detailed description of the study areas, see Madeira et al. (2009)).

Sample design

Along a 7-km transect, we established 10 rectangular plots of 20 × 50 m, at least 200-m apart, five in the early and five in the late successional stages (Madeira et al. 2009; Marques et al. 2017). In each corner and in the center of each plot, we selected three of the taller trees with a CBH (circumference at breast height) of at least 15 cm, forming five clusters. In each plot, we sampled 15 trees, with 75 trees in each successional stage, resulting in a total of 150 sampled trees.

Ant sampling

We performed the samplings during four consecutive periods, two in the dry season (September 2010 and September 2011) and two in the wet season (February 2011 and February 2012). To sample ants, we reached the canopy using a ladder for trees up to a height of 8 m; in trees taller than 8 m, we used the “single rope climbing technique” (Perry 1978). We used two complementary techniques for sampling the ants: arboreal pitfall trap and entomological beating technique (Campos et al. 2006). Arboreal pitfall traps consisted of plastic pots of 15 cm in diameter, filled with water and soap that we left active for 48 h. On each tree, we placed two arboreal pitfall traps tied to the trunk with a thin string, with 30 pitfalls per plot, 300 pitfalls per sample period, and a total of 1200 pitfalls. The entomological beating technique consisted of an entomological umbrella made by an inverted cloth funnel, with a surface of 1 × 1 m, and a plastic bag attached to the bottom. For ant sampling, we placed this apparatus under tree branches and beat the branches above (Campos et al. 2006). We beat the same trees on which pitfalls were placed, with 10 beatings on three branch groups, with the greatest possible distance between them on the treetop. We obtained one entomological beat sample per tree, with 75 samples per successional stage, 150 samples in each sample period, and a total of 600 samples. Throughout the four sampling periods (two dry and two rainy seasons simultaneously), we performed 1800 samplings among the arboreal pitfall traps and entomological beatings on the 150 trees of the early and late successional stages.

All the sampled materials were transported to the laboratory for sorting, mounting, and identification to the lowest taxonomic level. For ant species identification, we used the following resources: Baccaro et al. (2015), “AntWeb” (Fisher 2002), “AntCat” (Bolton 2012), “Ants of Costa Rica” (Longino 2007), and determination by experts. All materials are deposited in the Laboratório de Ecologia de Insetos (LEI), Universidade Federal de Minas Gerais (UFMG), Brazil.

Tree structure

Trees were identified to morphospecies, or species when possible, with the help of experts from the Universidade Estadual de Montes Claros (Unimontes). For each of the 150 sampled trees, 75 per successional stage, we estimated the tree height, using a ladder or ropes as length reference, and measured the connectivity among treetops. Connectivity among treetops was measured as the total number of times that a tree was connected to another tree either by its own structures (leaves, branches, trunk) or by structures associated to the tree (epiphytes and lianas). To determine the presence of extrafloral nectaries (EFNs) on the tree species, we used the guide “World List of Angiosperm Species with Extrafloral Nectaries” (Weber et al. 2015) and published data (Rivera 2000; Gonzalez 2011, 2013) for those trees we could identify (27 were not identified).

Statistical analysis

Comparisons of different successional stages could be potentially biased by variation in the habitat structure. To circumvent this issue, we developed ant species accumulation curves to determine whether the species was well represented by our collection effort, using rarefaction and extrapolation curves (the number of samples was extrapolated to double the sample size, as suggested by Chao et al. (2014). We implemented these methods within the package iNEXT (Hsieh et al. 2016), using the R software (R Core Team 2017).

To assess the relationship among ant species richness and canopy traits, we used the generalized linear mixed model (GLMM) procedure (Bolker et al. 2009), with tree height, tree connectivity, presence/absence EFNs, and successional stage as the fixed effect variables. Due to hierarchical sampling, we used the plot within the successional stage (p) and the tree within the plot within the successional stage (t) as random effects (1|successional_stage/p/t; see Crawley (2013)). Additionally, we performed a model for each successional stage separately, which we tested for seasonal effect (dry and wet seasons), considering the presence of EFNs, in relation to ant activity. Thus, we avoided mistakes when presenting the results of the relationship between EFN and ant activity in the trees among different seasons, because the samples of each season of the year (two dry and two rainy seasons) were initially combined. For the GLMMs, we first performed the complete model and then excluded the non-significant terms from the model until we obtained the simplest model with minimal model adequacy (Crawley 2013), with only the interaction between tree height and successional stages. The GLMM was made using the packages nlme (Pinheiro et al. 2014), lme4 (Bates et al. 2015), and hnp (Moral et al. 2017) in the statistical program R 3.1.1 (R Core Team 2017).

To examine the variation in species composition, we used the permutation multivariate analysis of variance (PERMANOVA) to test for significant clustering of successional stages and the height of the trees. We computed 999 permutations to account for successional stage and tree height effects. Subsequently, we used non-metric multidimensional scaling (NMDS), using presence/absence data and Jaccard’s dissimilarity index, to examine the extent to which there were successional stage and tree height in the ordination.

To study the interaction patterns between ants and trees, we initially built two A weighted matrices (for both early and late successional stages), in which Aij represents an interaction between a tree (individual) i and an ant species j (Bascompte et al. 2003) from 0 (absent) to 4 (ant species occurrences accumulated over the four sampling periods). Then, we calculated the level of specialization in the networks (H2′), using a metric derived from the Shannon diversity index, and performed 10,000 randomizations in comparison with random matrices. This index is based on the probability deviation expected from the distribution of interactions and varies from 0 (low degree of specialization) to 1 (high degree of specialization); it is a quantitative index and extremely robust to variations in sampling intensity and in the number of interacting agents (Blüthgen et al. 2006). For these analyses, we used the package bipartite (Dormann et al. 2008) in the statistical program R 3.1.1 (R Core Team 2017).

To determine if the ant species settle into a consistent pattern with low co-occurrence patterns, we used the co-occurrence index (C-score) (Stone and Roberts 1990) for each successional stage, which is based on the average number of interactions for each unique pair of species and varies from 0 to 1. High C-score values indicate that the interactions between species are highly exclusive and not randomly distributed within the network, while low C-score values indicate that the two species involved co-occur more frequently than expected at random (Gotelli 2000). We generated null models by applying Patefield’s algorithm (Patefield 1981) according to the “r2dtable” method, using 10,000 simulations and confronted with the index obtained by the data matrix. The C-score and the bipartite network graphs were generated using the packages bipartite (Dormann et al. 2008) and vegan (Oksanen et al. 2015) in the statistical program R 3.1.1 (R Core Team 2017).

We explored which ant species have the most interactions with individuals of trees, as proposed previously (Dáttilo et al. 2013), defining ant species (network components) as either core or peripheral for each of the successional stages. This classification is based on Gc = (ki − Kmean)/σk, in which ki = “average number of links for a tree/ant species,” kmean = “average number of links for all tree individuals/ants,” and σk = “standard deviation of number of links for tree individuals/ant species.” Species with a value of Gc > 1 are represented by a higher number of interactions in relation to other species of the same trophic level and are therefore considered as species that constitute a generalist core. On the other hand, Gc < 1 classifies species with fewer interactions in relation to other species of the same trophic level, and these species are therefore considered as species that constitute a periphery of networks. It should be noted that, for this analysis, we considered that the interaction was counted for each ant worker; thus, the ant individual presence was the replicate level of relevance, regardless of the colony of origin. Considering that most of the ant canopy community is based on extremely large dominant colonies, to assume those as replicates would basically flatten the results. Moreover, ant individuals are the actual source of interactive links, and regardless of belonging to common territories and groups, they are the interaction component that makes ecological sense.

Results

The ant fauna

We found 78 ant species belonging to 26 genera and eight subfamilies in the tropical forest canopies studied (Table 1). The sampling effort was sufficient to characterize the ant fauna sampled for the early successional stage, with a sampling coverage (SC) of 96.08%, and for the late successional stage (SC = 96.22%). Rarefaction analysis showed that ant species diversity was similar in the early and successional stages, since the 95% confidence intervals overlapped (Fig. 1).

Table 1 Sampled ant species in the tropical dry forest, Parque Estadual da Mata Seca (PEMS), Minas Gerais, southeast Brazil. “Species code” is an abbreviation used in the graphs of Fig. 4. “Number of Interactions” refers to the number of interactions of each ant species on each one of the successional stages (“Early,” “Late”) and both together (“Both”). The dominant ant species are marked with an asterisk
Full size table
Fig. 1
figure 1

Sample-size-based rarefaction (solid lines) and extrapolation (dashed lines, up to double the sample size) curves of ant species diversity for the early (black lines) and late (gray lines) successional stages of a tropical dry forest. Gray-shaded regions represent the 95% confidence intervals. Sample sizes are denoted by a solid circle (early successional stage) and a solid triangle (late successional stage)

Full size image

The trees

Of the 150 sampled trees, 123 were identified to the genus or species level, with a total of 27 morphospecies distributed in nine families. The richest arboreal families in terms of species were Fabaceae (10 species), Bignoniaceae (four species), Anacardiaceae, and Apocinaceae (three species each; Table 2). Although Fabaceae had the highest richness, only 20% of the sampled trees belonged to this family, while Bignoniaceae represented 30% and Anacardiaceae 18% of the total sampled trees. These three families together represented 68% of the total trees sampled. Out of all trees, 27 individuals were Handroanthus chrysotrichus (Bignoniaceae), 25 were Myracrodruon urundeuva (Anacardiaceae), 15 were Handroanthus ochraceus (Bignoniaceae), and 11 were Senegalia polyphylla (Fabaceae). We verified that 11 tree species sampled and identified (47% of the total) had EFNs (Table 2).

Table 2 Tree species sampled in the tropical dry forest of Parque Estadual da Mata Seca (PEMS), Minas Gerais, Southeast Brazil. “Abundance” represents the number of individual trees for each species in each successional stage and in both; “EFN” indicates species that possess extrafloral nectaries; “Freq.” represents the frequency (%) of tree species sampled in relation to total trees sampled (n = 150)
Full size table

Ant species richness and composition related to canopy traits

The subfamily Myrmicinae had the highest species richness (34 species), followed by Formicinae and Pseudomyrmecinae (13 species), Dolichoderinae (10 species), Ectatomminae (three species), Dorylinae and Ponerinae (two species each), and Heteroponerinae (one species; Table 1). We found 13 species of the genus Pseudomyrmex, 10 species of the genus Camponotus and Cephalotes, seven Pheidole species, and six Crematogaster species (Table 1).

Although the mean species richness did not differ between the early (mean ± SE = 57 ± 2.21) and late (60 ± 2.30) successional stages, we verified that tree height affects ant species richness; however, this effect interacted with the successional stage (Table 3). Thus, tree height had a negative effect on the ant species richness in trees in the early successional stage and a positive effect on ant species richness in trees in the late successional stage (Fig. 2). Treetop connectivity and EFN-bearing plants did not affect ant species richness on trees, and the EFN-bearing plants were not related to the seasonality (Table 3). Moreover, we found distinct ant species compositions in relation to tree height (NMDS, Fig. 3; PERMANOVA F1,147 = 9.21, P < 0.001), and between successional stages (NMDS, Fig. 3, PERMANOVA F1,147 = 26.31, P < 0.001).

Table 3 Analyses of the deviance of the minimal adequate models showing the effects of tree traits and successional stage on ant richness; seasonality on ant species richness; and tree specialization on tree species, for the tropical dry forest of Parque Estadual da Mata Seca (PEMS), Minas Gerais, Southeast Brazil
Full size table
Fig. 2
figure 2

Ant richness as a function of tree height in early (black circles) and late (gray circles) successional stages (Table 2) of the tropical dry forest of the Parque Estadual da Mata Seca (PEMS), northern Minas Gerais, Southeast Brazil

Full size image
Fig. 3
figure 3

NMDS plot of species composition according to tree height (point sizes) and early (dark gray) and late (light gray) successional stages in a tropical dry forest, Southeast Brazil

Full size image

Ant-tree interactions

We found 786 independent interactions among the 57 ant species and the 75 individual trees in the early stage throughout the four sampling periods (Fig. 4). For the late stage, there were 914 independent interactions among the 60 ant species and the 75 individual trees. We found a low complementarity of interactions at the community level of early (H2′ = 0.15, P < 0.05) and late (H2′ = 0.17, P < 0.0001) successional stages. Furthermore, we found that the interactions between ant species were not exclusive and non-randomly distributed in the trees of early (C-score = 0.77, P < 0.001) and late successional stages (C-score = 0.76, P = 0.002), indicating that a high degree of interspecific interactions determined the species distribution through the sampled trees. Additionally, we found no relationship between the tree specialization index (d′) and tree species both in the early and late successional stages (Table 3).

Fig. 4
figure 4

Tree-ant interaction networks in different successional stages, early (left) and late (right), in a tropical dry forest, Southeast Brazil. The boxes on the left-hand side represent individual trees, while the boxes on the right-hand side represent the ant species (nodes). The thickness of the boxes represents the centrality per degree of each node and the core ant species are in gray

Full size image

Based on the core-peripheral species analyses, we found seven core ant species for the early stage: Camponotus blandus, Camponotus renggeri, Cephalotes pusillus, Crematogaster goeldii, Camponotus melanoticus, Pseudomyrmex gracilis, and Pseudomyrmex pallidus (Table 1; Fig. 4, left). For the late successional stage, the core ant species were Crematogaster goeldii, Camponotus arboreus, C. melanoticus, C. vittatus, C. atriceps, Pseudomyrmex schuppi, and Cephalotes atratus (Table 1, Fig. 4, right). Only C. goeldii and C. melanoticus were core species in both successional stages (Table 1, Fig. 4), following the remarkable differences in species composition between the two successional stages presented above.

Discussion

Based on our results, tree height is an important driver for ant species richness in the tropical dry forest canopies studied, and this effect is dependent upon the successional stage. Furthermore, we found that core ant species in the ant species–tree networks have low spatial co-occurrence, which means that these ant species establish in sets of trees. Increasing tree height decreased the ant species richness in trees in the early successional stage, and we found a greater ant species richness in the smallest trees within this stage. This is most likely due to the foraging habitat of the ant species. In a TDF in Mexico, Vergara-Torres et al. (2017) found epigeic ant species foraging on trees, which has also been reported by Marques et al. (2017) in our study area. We found a particular ant fauna in each successional stage, in which only trees of the early successional stage were mainly used by epigeic ants for foraging. Examples of this include the typically predatory epigeic species Ectatomma muticum and Odontomachus bauri, some generalist myrmicines (e.g., Pheidole spp., Solenopsis spp., Wasmannia spp., Rogeria spp.), army ants such as Labidus praedator and Eciton sp1 (Brandão et al. 2012), and a typical fungus-grower ant genus that nests in the soil, represented by the ant species Cyphomyrmex sp1 (Mackay et al. 2014).

In the late successional stage, the effect of tree height on ant species richness followed the expected pattern, i.e., increased height resulted in greater ant species richness. A positive correlation between tree height and ant species richness has also been reported in other studies in the Brazilian savanna (Costa et al. 2011; Koch et al. 2016) and semi-deciduous and wet forests (Campos et al. 2006; Lourenço et al. 2015; Yusah and Foster 2016). In our study, the genus Azteca occurred exclusively in the late successional stage. It is a common arboreal ant genus (Brandão et al. 2012), especially Azteca instabilis, forming patches around shade trees (Philpott 2006). Taller trees are associated with the increase in available microhabitats or nest sites (Klimes et al. 2012), creating more opportunities for shelter and resource use (Campos et al. 2006; Klimes et al. 2012). In our study area, trees with similar heights of both successional stages were subjected to different environmental conditions. Specifically, while trees of the early successional stage tended to be more exposed to solar radiation, trees of the late successional stage with similar height were mostly located in the understory, with limited radiation. Therefore, to control sun exposition on the trees, we sampled ants on trees with different tree heights in both successional stages. This approach allowed us to infer about our specific initial hypothesis, focused on the importance of tree height under similar sun exposition levels, for the maintenance of ant diversity and the interactions in an individual-based network approach in the two successional stages. Our results raised main issues related to the importance of the successional stage and tree height to canopy ant diversity, a key issue for conservation and ecology.

In many systems, the connectivity among treetops is a determining factor for arboreal ant fauna, as it can promote movement and increase available resources (Majer and Delabie 1999; Powell et al. 2011; Yanoviak 2015; Adams et al. 2017). In our study, however, this factor did not explain ant species richness. Although the trees were in contact with each other via their branches, leaves, and lianas, our findings may be related to the paucity of lianas in the late successional stage in the tropical dry forest studied (Madeira et al. 2009). Besides, we found that ant species richness was not related to the distribution of the EFNs, in accordance with studies previously performed in the low woody vegetation, in which the authors found that EFNs have a limited effect on ant communities (Camarota et al. 2015; Sendoya et al. 2016). Most likely, this is because EFN secretion is seasonally restricted, with a significant emission of exudates being associated with the production of new buds and leaves, increasing ant foraging activity (Blüthgen and Reifenrath 2003; Melo et al. 2010; Schoereder et al. 2010; Lange et al. 2013). Indeed, environmental seasonality influences plant phenology (Pezzini et al. 2014) and insect diversity (Neves F de et al. 2013; Novais et al. 2016). In addition, as expected in a species-area relationship, in tropical forests, higher trees should have greater connectivity with neighboring trees and a greater amount of EFNs (Villamil et al. 2013; Dáttilo et al. 2014). Although we measured only the vertical height of the trees, we consider that the explanatory variables that we used (i.e., tree height, tree connectivity, the presence of EFNs) were sufficient to support our hypothesis. Nevertheless, we acknowledge that other canopy traits, such as cavities or other spaces formed by suspense gardens that accumulate litter, e.g., epiphytes (Powell et al. 2011; DaRocha et al. 2016; Klimes 2017), crown size, number of ramifications, growth units, and foliage (Campos et al. 2006; Blüthgen and Stork 2007; Lourenço et al. 2015), or tree density (Ribas et al. 2003), may contribute to ant occurrence and distribution and thereby to the network interactions found in this study.

We found a low co-occurrence (as expressed by the C-score index) between ant species and trees in each successional stage, suggesting that particular ant assemblages are associated with specific sets of trees (individuals). Additionally, in our ant-plant co-occurrence networks, individual trees were not influenced by tree species in terms of specialization (given by d′). Although we did not find an influence of the tree species in our networks, other factors related to habitat requirements could explain our results, such as food and shelter availability (Philpott and Foster 2005; Houadria et al. 2015), which were not measured in our study. Thus, in addition to the habitat requirement, we propose that biotic forces may be structuring the ant communities in our study, for example, species competition. According to this, in a study of ant-tree relationships in the Brazilian savanna (Camarota et al. 2016), the authors found that competition better explained co-occurrence patterns than habitat requirements. More detailed studies evaluating competition or habitat requirements could definitively conclude which factor most interferes with tree foraging in tropical dry forest canopies. Another exciting avenue of research is related to the physiological tolerance of ant species and the implications related to the persistence of these ant species in such environments (Bestelmeyer 2008; Spicer et al. 2017).

We conclude that the effects of tree height on ant species richness are dependent on the successional stage and may be related to the ant species composition and their functional role. We evidenced a high number of some arboreal ant species in the canopies, since the generalist core ant species with a low spatial-temporal co-occurrence established specific territories persistent over time (i.e., four samples over the course of 2 years analyzed together). Based on this, and because the ant-plant interactions were non-randomly distributed, our findings strongly suggest that ants play a key role in the structure of the associated fauna on the studied tropical dry forest canopies.