The structure of ant–plant ecological networks: Is abundance enough?

. Knowledge of the mechanisms that shape biodiversity is essential to understand the ecological and evolutionary dynamics of interacting species. Recent studies posit that most of the organization of mutualistic networks is shaped by differences in species abundance among interacting species. In this study, we examined the mutualism involving plants with extraﬂoral nectaries and their associated ants. We show empirically that the difference in abundance among ants on vegetation partially explains the network structure of mutualistic interactions and that it is independent of ant species compositions: an ant species that is abundant usually interacts with more plant species. Moreover, nested networks are generated by simple variation in ant abundance on foliage. However, in ant–plant mutualistic networks, nestedness was higher than in networks describing the occurrence of ants on plants without a food resource. Additionally, the plant and ant species with the highest number of interactions within these networks interacted more among themselves than expected under the assumption of an abundance-based, random mixing of individuals. We hypothesize that the dominance of these ant species occurs because these ants are able to outcompete other ant species when feeding on extraﬂoral nectaries and because of the presence of ecophysiological adaptations to utilize liquid food.


INTRODUCTION
A central goal of ecology is to understand the mechanisms that determine the structure of ecological communities at different spatial and temporal scales (Turner 1990, Levin 1992. Recent studies have focused on the structure of ecological networks and have found nonrandom patterns of interaction in a wide range of ecosystems and involving multiple interacting species (Bascompte et al. 2003, Lewinsohn et al. 2006, Guimarães et al. 2007). Several factors have been proposed to explain the origin of structural patterns in ecological networks (reviewed by Va´zquez et al. 2009a).
One key factor to understand the structure of interaction networks is differences in relative abundance across interacting species (Krishna et al. 2008, Chamberlain et al. 2010, Verdu´and Valiente-Banuet. 2011). Abundant species should interact most frequently with each other and with less abundant species, forming a core of highly connected species. In contrast, less abundant species will rarely interact among themselves. As a consequence, it has been hypothesized that species abundance (or interaction frequency) sufficiently explains several struc-tural aspects of most mutualistic networks, including the observed heterogeneity in the number of interactions per species (Va´zquez et al. 2005), asymmetrical patterns of interaction (Va´zquez et al. 2007(Va´zquez et al. , 2009b and nestedness (Blu¨thgen et al. 2008, Krishna et al. 2008. Inferences on the role of species abundance in shaping network structure are often based on the comparison of the patterns of interaction of an empirical network with the predictions derived from null models. These null models often assume that interaction probabilities are determined by the random mixing of individuals, and their parameterization is based on independent data on species abundance or the use of interaction frequency as a surrogate for species abundance (Va´zquez et al. 2007, Blu¨thgen et al. 2008, Krishna et al. 2008. The null approach provides insight into the role of different factors shaping the patterns of interaction in mutualistic networks. Nevertheless, to understand the role of neutral, abundance-based processes in shaping network structure, one would ideally endeavor to combine evidence derived from null models with a comparison among empirical networks differing in the presence of the crucial resource on which the mutualism is based. Indeed, most mutualisms are based on a key resource, such as nectar and fruits, provided by one of the individuals to attract the partner. However, for most mutualisms, it is impossible to compare empirical mutualistic networks with an empirical benchmark in which the key resource is absent, e.g., recording Interactions between ants and EFN plants are ubiquitous in tropical forests and are well documented as a classical example of mutualism (Rico-Gray and Oliveira 2007). In these associations, plants produce a nutritious liquid to attract ants (Koptur et al. 1998), and, in exchange for food, the ants defend the plants against potential herbivores (Del-Claro et al. 1996, Rico-Gray andOliveira 2007). Because of ant diversity (Longino et al. 2002) and abundance (up to 94% of the arthropods and 86% of the biomass collected in canopies consists of ants [Majer 1990, Tobin 1995), it is extremely common to observe ants foraging on plants, even in the absence of extrafloral nectar (Andersen 1990, Blu¨thgen et al. 2000, Davidson et al. 2003. Accordingly, the diversity and abundance of ants create the opportunity to explore the organization of ant-plant interactions in the absence of the key resource for this interaction: the extrafloral nectar. Extrafloral nectar is a predictable, energy-rich resource, and several empirical studies focusing on a few species have demonstrated that different ant species compete to access and dominate such resources (e.g., Blu¨thgen et al. 2000, Dreisig 2000, Blu¨thgen and Fiedler 2004a. Moreover, only a few ant species possess adaptations for the acquisition, storage, and digestion of liquid resources (Fowler et al. 1991, Oliveira and Brandão 1991, Davidson et al. 2003, and ants with these traits should have a competitive advantage over species that do not have these adaptations but rather forage on vegetation for prey (Andersen 1990).
Here, we investigate the role of abundance in the shaping multiple networks formed by interactions among ants and EFN plants. In these networks, species are depicted as nodes, and their interactions are depicted by links describing the use of EFN plants by ant species. We compared these ant-plant interaction networks with paired networks describing the occurrence of ants on plants without EFNs. These networks provided a benchmark for studying how interaction networks would be organized if the main resource provided by a mutualistic partner (i.e., EFNs) did not affect the organization of the ant-plant interactions. The resources in plants without EFNs cannot be predicted, and ants forage randomly on plant individuals using the plant only as a substrate (Blu¨thgen et al. 2000). The presence of ants on such plants may be driven by neutral, nondeterministic factors, such as the available foraging area and species abundance. If specialization and competition for the liquid-energy food source of EFNs were important for structuring ant-plant networks, we would expect differences in the structure of the networks describing the interactions with EFN plants or non-EFN plants. Specifically, we expect that nectar-feeding ants monopolize the resource, interacting with plants more than would be expected from networks of the occurrence of ants on plants that do not offer any predictable type of food resource and forming the core of the highly interactive species observed in these networks.

Study site and system
The study was conducted at the São Nicolau Farm (9848 0 S and 58815 0 W, elevation 254 m) in the municipality of Cotriguac¸u, northern Mato Grosso, Brazil. The reserve area covers 7000 ha of continuous forest and is characterized as a tropical rainforest within the southern Brazilian Amazon not subject to seasonal inundation. Canopy trees reach heights of approximately 40 m, with a few emergent canopy trees reaching up 50 m ). The understory is relatively open and composed mainly by the palm Orbignya phalerata Mart. (Arecaceae). The terrain is moderately undulating, with a maximum elevation difference of approximately 50 m. The region's climate is wet tropical (Am, according to Ko¨ppen's climate classification), with annual means of temperature and rainfall of approximately 248C and 2300 mm, respectively (Ko¨ppen classification; Camargo et al. 2010), and is divided into two distinct seasons: a November-April rainy season and a May-October dry season.

Data collection
For sampling, we used a module system managed by the Brazilian Research Program in Biodiversity (PPBio) on São Nicolau Farm. The module is composed of sampling points uniformly distributed in two parallel east-west 5-km trails, 1 km apart (5 km 2 ). There is one sampling plot of 250 3 25 m (6250 m 2 ) every 1 km along each trail, for a total of 12 sampling points. The central trail in each sampling point has a constant elevation to minimize variations of soil type and elevation (Magnusson et al. 2005). We considered each of the 12 sampling points as an independent sample of ants and plants, generating 12 different ant-plant mutualistic networks and 12 networks that describe the occurrence of ants on plants without EFNs. In the ant-plant mutualistic network, the nodes represent species, and the links depict interactions in which ants are visiting plants with EFNs. In the second set of networks, a link represents the occurrence of a given ant species on a given plant species. We assumed that the ants are sessile organisms (Fourcassie´et al. 2003) and that the distance between sampling points was sufficient to guarantee that the organisms of a given plot would never interact with the organisms in another plot. We collected ants and plants in December 2010 and January 2011, always between 09:00 and 15:00. For each of the 12 sampling points, we recorded all the ants collecting nectar from EFN plants that were accessible to the collector (from 0.5 m to 3 m high). No plants with homopterous or any other visible liquid-energy resources were included in our sampling.
We only recorded the interactions between free-living ant species and EFN plants. Furthermore, we did not explore ant-myrmecophyte interactions because these intimate interactions are organized by particular evolutionary and ecological processes that produce different interaction patterns than interactions among ants and EFN plants (Guimarães et al. 2007, Da´ttilo 2012. To minimize the possibility of collecting individuals from the same ant colony foraging on different plants, all the plants recorded within each sampling point were at least 10 m apart. We used the entomological umbrella method to estimate the interactions in both networks by collecting interacting ants from EFN plants and ants that occurred on plants without EFNs. By sampling both networks in the same way in a paired design, we were able to test the effect of the presence of a resource provided by a mutualistic partner (EFN plants) and lacking in the other network (plants without EFNs). We observed and collected additional ants, particularly more secretive species, before and after performing this procedure to record all the ant species feeding on EFNs. The ant species collected on the EFN plants were always observed feeding on EFNs. Additional details of the sampling procedure can be found in Appendix A.

Species abundance and patterns of interaction
We used three different approaches to explore the role of abundance in shaping the organization of mutualistic networks of ants and plants with EFNs. First, we evaluated whether differences in abundance among ant species affect the patterns of interaction within the networks of ants and plants with EFNs. We used the records of ants in networks including plants without EFNs as a measure of ant abundance on the vegetation. We assumed that plants without EFNs offer no predictable food resource to the ants and that the frequency of ants on these plants reflects the spatial abundance of these species on the vegetation without the aggregation caused by the attraction to a food resource (Blu¨thgen et al. 2000). We performed a Spearman correlation to test the hypothesis that the number of interactions increases with the abundance of ant species.
If the difference in abundance is the only factor shaping networks, we should expect that the networks describing the interactions between ants and plants with EFNs would be structurally similar to the networks formed by the occurrence of ants on plants without EFNs. To test this hypothesis, we contrasted the structure of the networks describing interactions between ants and plants with EFNs (n ¼ 12 networks) with networks describing the occurrence of ants on plants without EFNs (n ¼ 12 networks, forming 12 pairs of networks) and described the difference in the number of species (ants and plants) and the total number of interactions between these two types of networks. We also calculated the following network metrics for each of the networks between ants and plants with and without EFNs: connectance (C ), which is the proportion of possible interactions that are actually observed (Jordano 1987), the mean and standard deviation of the number of interactions per plant and ant species, network specialization, modularity, and nestedness. We used these network descriptors because they provide a way to characterize the organization of these networks that allows a direct comparison with the previous work on mutualistic networks. Moreover, both theoretical and empirical evidence associates these metrics with the evolutionary and ecological processes that shape mutualistic networks , Va´zquez et al. 2009a, Donatti et al. 2011, Lange et al. 2013.
We calculated the level of specialization in the networks using the quantitative specialization index (H 0 2 ), which ranges from 0 (extreme generalization) to 1 (extreme specialization). H 0 2 is defined as follows: where i represents one plant species and j one ant species, p ij is the proportion of interactions between i and j, P is the plant species richness, and A is the animal species richness This index is derived from the Shannon entropy and is based on the deviation from the expected probability distribution of chance interactions (Blu¨thgen et al. 2006). H 2 was normalized between the minimum entropy, H 2min , and maximum entropy, H 2max , for associations We estimated modularity using the index M (range 0-1), which estimates the degree to which groups of species (ants and plants) interact more with each other than with the species in the other groups in the network (Newman andGirvan 2004, Olesen et al. 2007). The M index is defined as follows: where N m is the number of modules, L is the total number of interactions in the network, l s is the number of interactions between the species in module s, and d s is the sum of the number of interactions per species of all the species in module s (Olesen et al. 2007). We tested the significance of M for each network through 1000 simulated networks generated by a null model to assess whether the value of M observed in the empirical network was higher than would be expected for networks of equal species richness and similar heterogeneity with regard to interactions among species (null model II; Bascompte et al. 2003). Null model II assumes that the probability that plant species i interacts with ant species j depends on the observed number of interactions of both species, such that P(n ij ) ¼ 0.5[(k i /A) þ (k j /P)], in which k is the number of interactions for the species in the empirical network and n ij is the interaction between ant species i and plant species j in the theoretical network (Bascompte et al. 2003). We constructed each null model network through a routine in MATLAB, and we calculated M using the software Netcarto (Guimeraà nd Amaral 2005). Although the M index was designed for unipartite networks, our null model controls for the potential effects of a bipartite structure on modularity (i.e., interactions only occur between ant and plant species [Pires et al. 2011]). Because the resource secreted by EFN plants is predictable, we expected that these networks would be more modular because we expect that ants will be more likely to forage on the same plant species.
We used the NODF metric (nestedness metric based on overlap and decreasing fill [Almeida-Neto et al. 2008]) to estimate the nestedness value of the networks using the ANINHADO software . In a nested network, if species i and j, both from the same trophic level (plants or ants), differ in their number of interactions (k i , k j ), the set of species interacting with i also interacts with j. As a consequence, in nested networks, (1) species with a large number of interactions from different trophic levels tend to interact with each other, forming a core of highly interactive species, (2) species with few interactions tend to interact with highly interactive species, and (3) interactions between species with few interactions rarely occur . Biologically, nestedness describes the organization of niche overlap within interacting assemblages, and species often show high and asymmetrical niche overlaps in a nested network (Bastolla et al. 2009, Blu¨thgen 2010. The degree of nestedness may be affected by species richness, connectance, and heterogeneity in the number of interactions. To assess whether the observed nestedness value was higher than would be expected based on these other features of the network structure, we tested the degree of nestedness of each network with 1000 networks generated by the same null model II that we used in the modularity analysis. We then used the ensemble of null model networks to compute z scores for each real network, z ¼ (NODF O À NODF T )/r T , in which NODF O is the observed value of nestedness for a given real network, NODF T is the mean value of nestedness for the null model of our theoretical networks, and r T is the standard deviation of nestedness values of the ensemble of null model of our theoretical networks. Therefore, these z scores describe the deviation of the degree of nestedness of the real network from the expected nestedness value for networks that share all the attributes observed in the real networks preserved in the null model. Thus, the analysis of nestedness z scores allowed us to investigate whether the patterns observed in the analysis of raw values of nestedness held after controlling for all the network structural patterns preserved by the null model analysis (Almeida-Neto et al. [2008] and Ulrich et al. [2009] for additional details). We used the paired t test (paired per plot) in R (R Development Core Team 2010) to test the difference of all the metrics described in this subsection for the ecological networks of ants and plants with and without EFNs.
Because the ant-plant networks shown here are significantly nested (see Results), we explored the question of whether species abundance affects differently the interactions between highly interactive species (the core of the network) and the interactions between other species in the network (the periphery). We first used a core-periphery analysis (see Diaz-Castelazo et al. [2010] for a similar approach) to identify the core of highly connected species in each network of ants and plants with EFNs. The network core was formed by ant and plant species with (k i À k )/r k . 1, in which k i ¼ the mean number of interactions for a given plant/ant species, k ¼ the mean number of interactions for all plant/ant species in the network, and r k ¼ the standard deviation of the number of interactions for the plant/ant species (Da´ttilo et al. 2013a). Accordingly, the periphery was formed by species for which (k i À k )r/ k 1 If the ant species forming the core of the mutualistic networks were those species with adaptations for the foraging of liquid resources, we should expect only a small subset of ant species to be recorded as core species. We tested this prediction performing a randomization test (n ¼ 1000 randomizations). First, we computed the observed core ant species richness, which is the total number of ant species occurring in all 12 ant-EFN-plant networks. We then retained the same number of core and peripheral species in any given ant-EFN-plant network while randomizing the identity of the ant species between groups (core and periphery) and computed the core ant species richness after randomization. Last, we recorded the probability of a given randomization showing a core ant species richness equal to or smaller than that observed in the empirical 12 ant-EFN-plant networks.
We then computed a probability matrix based on the abundance of interacting species, assuming the random mixing of individuals (Appendix B). We used this matrix to identify the interactions between ants and plants that occurred more frequently than would be expected from the species abundance. We computed the number of interactions showing a frequency that is higher than would be expected based on the species abundance, separately for the core, E c and the periphery of the network, E p . Because the core and the periphery differ in species richness, we standardized both E c and E p by the total number of possible interactions among the core species and among the peripheral species, respectively.
Using a paired t test, we compared the frequency of interactions that occur more often than would be expected based on the species abundance, between the core species and between the peripheral species in the networks of ants and plants with EFNs.

Additive partitioning of diversity
To evaluate whether there is a particular assemblage of ant species that forage on EFN plants, we evaluated the turnover of ant species composition (b diversity) between the networks of ants and plants with and without EFNs and calculated the additive partitioning of diversity (Veech et al. 2002). Based on the total ant richness found in each plot (c diversity), we calculated the a diversity, a mean ¼ (a 1 þ a 2 )/2, where a 1 is the ant richness in the networks of EFN plants found in the plot and a 2 is the ant richness in the networks of non-EFN plants found in the plot. We then calculated the b diversity as b ¼ c À a mean . Additionally, to evaluate whether the presence of EFNs caused an increase in ant species richness, we randomized (n ¼ 1000 randomizations) the plant classification (presence or absence of EFNs) while keeping the c diversity fixed. At the end of each randomization, we counted the ant richness in the networks of EFN plants (a 1 ) and plants without EFNs (a 2 ) to assess whether the ant richness in the networks of EFN plants was higher than would be expected for the ant richness in the networks of plants without EFNs. We conducted the randomization through a routine in MATLAB. In addition, to summarize the composition of the ant community on plants with and without EFNs, we ordered the similarity between the points using nonmetric multidimensional scaling (NMDS) and tested the difference in the ant species composition through a permutation test (10 000 permutations) based on an analysis of similarities (ANOSIM; Appendix C).  1A). For ants, the mean and standard deviation of plant species visited per ant species did not differ between the networks of EFN plants (2.67 6 0.14) and plants without EFNs (2.64 6 0.10, t ¼ 0.258, P ¼ 0.801; Fig.  1B).

RESULTS
We did not find significant differences in specialization between the networks (EFN plants, 0.07 6 0.05; plants without EFNs, 0.07 6 0.04; t ¼ À0.7, P ¼ 0.498). The network connectance for ants and EFN plants was higher (0.140 6 0.03) than the connectance for ants and plants without EFNs (0.109 6 0.02, t ¼ À3.528, P ¼ 0.005; Fig. 1C). The degree of modularity was similar to the degree observed in the null model networks, keeping the heterogeneity of interactions at any sampling point (P . 0.05). The modularity in the networks of ants and EFN plants (0.459 6 0.059) was lower than in the networks of ants and plants without EFNs (0.519 6 0.061, t ¼ 3.552, P ¼ 0.005, Fig. 1D). Nestedness was higher in the networks of ants and EFN plants (21.01 6 4.46) than in the networks of ants and plants without EFNs (15.75 6 3.33, t ¼À3.427, P ¼ 0.006, Fig. 1E). The same pattern held using the z score estimates of nestedness to control for the effects of species richness and heterogeneity in the number of interactions (z scores of nestedness in networks of ants and EFN plants, 3.63 6 1.51, and for networks of ants and plants without EFNs, 2.34 6 1.35; t ¼ À2.270, P ¼ 0.04; Fig. 1F).
The number of interactions of ant species with EFN plants was associated with ant abundance at all sampling points (Spearman's rho ¼ 0.59 6 0.19, all P , 0.05), and the core-periphery analysis showed that one to three (2.41 6 0.66) species of ants were present in the generalist core of each network. In fact, of the 149 ant species recorded, only 10 species were present in the core of highly interactive species in at least one network of ants and EFN plants (expected core ant species richness ¼ 21.09 6 1.77, P , 0.001). These recurrent core species have adaptations to utilize a liquid diet and are members of the genera Azteca, Brachymyrmex, Camponotus, Crematogaster (see Plate 1), and Dolichoderus. In addition, on average, most of the interactions between the highly interactive species of ants and plants that formed the core of the network occurred more frequently than would be expected from their abundance (81.9% 6 19.4% of all possible interactions; Fig. 2). In contrast, the less connected species of the network interacted less than would be expected from their abundance (8.7% 6 1.5% of all possible interactions, t ¼ À13.28, P , 0.0001; Fig. 2).
The b diversity turnover in the ant species composition between the paired networks of ant and plant species with and without EFNs involved approximately one-third of the total diversity and was similar across all plots (Fig. 3). The turnover in ant species composition was reproduced by simulations in which the plant classification (i.e., presence or absence of EFNs) was randomized between the pairs of networks (P . 0.05 for all plots). Furthermore, the presence of EFNs did not trigger an increase in ant a diversity (t ¼ 1.448, P ¼ 0.176). An additional NMDS ordination of the ant community showed a similar composition for the networks formed by ants and EFN plants and for the networks formed by ants and plants without EFNs (ANOSIM; P ¼ 0.999, see Appendix C).

DISCUSSION
Recent studies have shown that relative species abundance is a main driver in structuring mutualistic networks (Va´zquez et al. 2007, 2009b, Krishna et al. 2008, Verdu´and Valiente-Banuet 2011. These studies have found that the nested and asymmetrical interactions among species within mutualistic networks are often reproduced by null models parameterized with information on species abundance. In the present study, by combining null model analyses with a novel approach in which replicated pairs of mutualistic networks and networks generated by sampling the abundance of ants foraging on plants without EFNs, we were able to untangle three aspects of the network organization of ant-plant interactions. First, we corroborated the notion that abundance is a key factor organizing mutualistic networks: more abundant ant species often interacts with more EFN plant species. Similar patterns were observed in other mutualisms, such as plant-frugivore interactions (Jordano 1987), pollination networks (Va´zquez et al. 2007, Blu¨thgen et al. 2008, and ant-plant networks in other environments (Chamberlain et al. 2010). More importantly, even networks describing the occurrence of ants on plants without any type of predictable resource show some degree of nestedness (Da´ttilo et al. 2013b), providing empirical evidence for the theoretical predictions derived from null models based on neutral theory (Krishna et al. 2008) and probabilistic matrices (Fischer and Lindenmayer 2002, Va´zquez et al. 2007, Blu¨thgen et al. 2008, Blu¨thgen 2010) that nestedness is expected solely by a neutral, abundance-based mixture of individuals.
Second, the information on the difference in abundance among interacting species was insufficient to FIG. 2. The 12 interaction matrices between ants and plants with extrafloral nectaries (EFNs). The rows represent the ants, and the columns represent the plants. The numbers above each network represent the plot studied. Cells represent interactions (i) involving core species and occurring in a frequency higher expected by their abundances (black), involving peripheral species and occurring in a frequency higher than expected by their abundances (gray), or occurring in a frequency lower than expected by their abundances (white, including species that did not interact). explain ant-plant network organization. Nestedness in networks of ants and EFN plants was higher than that observed in networks of ants and plants without EFNs. An analysis of z score estimates of nestedness suggested that these differences between networks involving plants with and without EFNs hold after controlling for the effects of species richness, connectance and heterogeneity of interactions. Furthermore, highly connected species interacted more with other species than would be expected from the random mixing of individuals based on their relative abundance. Therefore, the structure of the mutualistic ant-plant interactions found in this study is different from that observed for the natural distribution of ants foraging on plants that do not have EFNs and cannot be simply associated with differences in species abundance.
Third, we showed that only a small subset of component species form the highly interactive core of ant-plant networks, and we hypothesize that this result documents the role of traits in shaping the organization of mutualistic networks. In mutualisms, a dependence on the resource provided by the mutualistic partner (Krishna et al. 2008), trait matching (Stang et al. 2006) and animal size (Chamberlain et al. 2010) also affect the network organization, generating departures from the expected results based on models parameterized with information on species abundance. We proposed two, non-mutually exclusive hypotheses to explain why antplant networks are characterized by a core of highly interactive ant species that represent a small subset of ant species richness.
If EFN plants within the core are the optimal resource, optimal foraging theory predicts that competitively superior ant species would monopolize these plants, whereas less competitive ants would be forced to use suboptimal resources. In fact, competition has been identified as an important factor structuring ant communities (Lo´pez et al. 1994, Chesson 2000, Parr et al. 2005. Moreover, the quantity and quality of nectar vary among plant species and individuals (Schupp andFeener 1991, Heil et al. 2000). Thus, we hypothesize that the core of highly connected species might be formed by competitively superior ant species that can exclude others, as is expected in ant mosaics in forest canopies (Leston 1978). This hypothesis led to two predictions that future studies should investigate: (1) core plant species have better resources and (2) competitively superior ant species (i.e., by massive recruitment and/ or aggressive behavior) monopolize these more-valuable resources McKey 2003, Blu¨thgen andFiedler 2004a, b). In fact, the workers of the core ant species in this study are from Azteca, Brachymyrmex, Camponotus, Crematogaster, and Dolichoderus, which are known to typically exhibit massive recruitment, allowing these ants to dominate the resources and to outnumber other ants when using the resources.
An alternative and not mutually exclusive hypothesis is that this core of highly connected species might be the result of adaptations of a few species related to the exploitation of extrafloral nectar as a resource. The few component species of the core of ant-plant networks have adaptations for liquid feeding, and all the core ant species are from genera that have adaptations to utilize a liquid diet. The workers of these genera can spend more time feeding on nectar because they can perform oral trophallaxis and ingest a large amount of nectar due to proventricular adaptations and gastric expansion (Davidson et al. 2003(Davidson et al. , 2004. In addition, the species from these genera are known to have physiological adaptations for the digestion of liquid resources secreted by EFNs (Fowler et al. 1991, Davidson et al. 2003. It is important to note that, although core ant species often show adaptations for feeding on liquid resources and may be competitively superior to other ant species, we found a low turnover of ant species foraging on plants with and without EFNs (as in Schoereder et al. 2010). In fact, some ant genera that lack adaptations for feeding on liquid resources can carry liquids externally and actively use extrafloral nectar (e.g., Acanthoponera, Ectatomma, Heteroponera, Pachycondyla, Paraponera, and Pseudomyrmex [Oliveira and Brandão 1991, Davidson et al. 2003). We hypothesize that competition and/or adaptations affect the identity of the species forming the core and the periphery of the networks of ants and plants with EFNs, leading to less specialized ants and weak competitors to participate in these networks as mostly peripheral species.
One theoretical consequence of a core that is formed by species with adaptations to use EFNs, as studied here, is that these adaptations increase the potential for the coevolutionary convergence of traits among species from both sides of the interaction within the community (Bascompte et al. 2003, Thompson 2005, Guimarães et al. 2007. The low variability we observed in the ant core composition suggests a stable network core, as originally proposed by Bascompte et al. (2003). The stability of a highly connected core is predicted by coevolutionary theory (Thompson 2005), is observed in long-term studies in other mutualisms (Olesen et al. 2011, Fang andHuang 2012) and is corroborated by the niche conservatism found across ecological networks (Stouffer et al. 2012), specifically for highly connected species within ant-plant networks (Sugiura 2010, Da´ttilo et al. 2013a, Lange et al. 2013. Future empirical studies evaluating the consequences of spatial and temporal variations in ant-plant interactions and other types of mutualistic networks must give particular attention to the network core composition due to its implications for the ecological and evolutionary relationships of communities. Moreover, modeling approaches could be useful to investigate the implications of variations in the species composition of network cores for coevolutionary dynamics.