INTRODUCTION
Mutualistic interactions are important to determine demographic dynamics and co-evolution of the species in communities, as a consequence their study is key to understand ecological processes that generate and sustain biodiversity (Bascompte et al. 2007; Bascompte & Jordano 2008). In particular, identifying factors that determine the importance of species to interaction network is fundamental to developing management strategies that lead to conservation of ecological processes. The structure of interaction networks is derived from the accumulation of interactions between species within communities. In particular, the complex interaction between plants and animals across time and space results in distinctive topological properties that differentiate mutualistic networks from other ecological networks (Vázquez et al. 2009a). This structure arises from spatial and temporal structures and/or evolutionary processes leading to non-random patterns of interactions, and describes the extent what species are organized into networks (Olesen et al. 2007). Despite the removal of keystone species could trigger secondary extinctions and cascading effects (Montoya et al. 2006), the structural properties of mutualistic networks can buffer the effect of primary extinctions (Memmott et al. 2004), conferring robustness to mutualistic networks. Therefore, understanding the architecture of the network and identifying species that are important for its structure are fundamental to conserve and manage ecosystems (Solé & Montoya 2001).
Mutualistic networks are more heterogeneous than expected by chance with many species having scarce interactions and a few species having a very large number of interactions (Bascompte & Jordano 2014; Vázquez et al. 2009a; Vázquez & Aizen 2004). Besides, mutualistic networks in general are highly nested (Jordano et al. 2003; Dupont & Olesen 2009; Bascompte et al. 2003; Montesinos-Navarro et al. 2012), which means that there is a core of generalist species interacting among them and specialists tend to interact with the most generalist species.
Finally, plant-animal mutualistic networks are modular (Olesen et al. 2007) since there are groups of species (modules) that tend to interact more between them than with species from other modules (Krause et al. 2003). This structure increases the stability and coexistence of mutualistic communities (Bastolla et al. 2009; Okuyama & Holland 2008) and makes mutualistic network more robust to loss of keystone species (Solé & Bascompte 2006; Solé et al. 2002). Evidence suggests that species traits (Bascompte et al. 2003; Jordano et al. 2003; Vázquez et al. 2009b; Olesen et al. 2011), evolutionary history (Peralta 2016; Vitória et al. 2018) and abundance (Vázquez & Aizen 2004) of species influence the structure of mutualistic interactions.
Morphological and phenological traits are important for interactions in plant-animal mutualism by constraining interactions between species (Olesen et al. 2011; Stang et al. 2009; Bascompte & Jordano 2007). Morphological constraints generate mismatches between morphology of the species involved in the interaction (Olesen et al. 2011; Stang et al. 2009). On the other hand, the presence of a phylogenetic signal in mutualistic networks, would suggest that inherited ancestral phenotypic traits determine, at least in part, ecological interactions (Thompson 2005; Vázquez et al. 2009a). However, other evidence indicates that neutral processes have an important influence on those observed patterns (Krishna et al. 2008; Canard et al. 2012; Vázquez et al. 2007). The ecological equivalence (neutrality) hypothesis proposes that interactions between individuals are random; consequently, all individuals have the same probability of interaction in ecological networks regardless of species differences (Vázquez et al. 2007). Hence, species abundance and random interactions between individuals are important drivers of mutualistic network structure. Consequently, biological constraints, evolutionary history and neutral processes can determine species role in structuring mutualistic networks (Ives & Godfray 2006; Jordano 2010; Bascompte & Jordano 2014); however, the relative contribution of these factors to network structure is poorly known and requires studying highly diverse biological group with important ecological roles for ecosystem functioning.
Seed dispersal and pollination are mutualistic interactions that contribute to maintenance of the structure and stability of tropical communities (Bascompte & Jordano 2014). Interactions between plants and frugivores, in particular, are crucial in tropical forest since from 50% to 90% of the woody plants are dispersed by animals (Fleming 1987). Bats are the most diverse group of seeds dispersal among mammals, playing a key role for conservation of plant diversity in tropical forests (Muscarella & Fleming 2007). Bats mainly disperse plants with opportunistic strategies that are important for colonization of degraded areas, which are characterized by quick growing and fruits with large number of seeds (Charles-Dominique 1986). Therefore, bats disperse seeds of important plants for regeneration of disturbed forests by intense human activities (Memmott et al. 2007; Hegland et al. 2009; Laliberté & Tylianakis 2010). This process is particularly important for the tropical dry forest, an ecosystems heavily affected by loss of coverage due to agricultural and livestock expansion (Castaño-Salazar 2009). Consequently, understanding the role of mutualistic networks for tropical forests conservation requires assessing factors that determine species’ importance for the structure of interaction networks.
Our goal was to evaluate the influence of cranial morphology, phylogenetic history and abundance of frugivorous bats on the structural properties of a mutualistic network between frugivorous bats and plants in a tropical dry forest. We expected that: (a) interaction network would have be heterogeneous, nested and modular; and (b) closely related species and species with similar cranial morphology would have similar importance for network structure and that abundant species would be the most important for network structure.
MATERIALS AND METHODS
Study area
The study was conducted at a fragment of tropical dry forest (Holdridge 1967) corresponding to “Parque Regional Natural El Vínculo” (PRN El Vínculo), Buga, Valle del Cauca, Colombia. The PRN El Vínculo is surrounded by a matrix of sugar cane crops and grasslands, and is covered by forests in different successional stages (intervened forest, secondary forest and scrub land) with different levels of anthropogenic disturbance (Torres 2012). The park ranges from 977 to 1150 m. of elevation, and the climate is characterized by average annual temperature of 24°C and average annual of precipitation 1379 mm (Torres 2012). The area presents two periods of heavy rainfall (March-June and SeptemberDecember) and two dry periods (January-April and July- August), this seasonality is typical of the tropical dry forest (Marulanda et al. 2003).
Field methods
We captured bats monthly, from July 2017 to February 2018, at the different successional stages of PRN El Vínculo during five nights per month. In each sampling period, five 12m x 3m mist nets were set at different zones, during each day, for a total sampling effort of 40 nights (246 hours) during eight months. We randomly selected sites for net placement, before each sampling day, with consecutive nest separated by at least 100 m. To capture bats during the period of greatest activity, the nets were set in the evening (18:00 h) and pulled at midnight (24:00 h); mist nets were checked every 30 minutes. Identification of the bats captured in field, to lowest possible taxonomic category, was based on specialized taxonomic keys (Gardner 2007; Diaz et al. 2016). We held captured bats individually in cloth bags, for one hour, in order to collect their fecal samples (Thomas 1988), which were deposited individually in properly-labeled paper bags. In addition, we walked tracks monthly with the purpose of collecting fruits potentially consumed by bats in order to create a seed reference collection for the study area. Vegetation samples were identified until the least possible taxonomic level with specialized literature (Ríos et al. 2004; Lobova et al. 2009; Linares & Moreno-Mosquera 2010).
Seeds were sorted out from fecal samples of bats, and identified (whenever possible) using taxonomic keys (Ríos et al. 2004; Lobova et al. 2009; Linares & Moreno-Mosquera 2010) and comparing with the seed reference collection for the study area. Since we were interested in the potential function of bats as seed dispersers, we identified the links between bats and plants by seeds from bat feces.
Cranial morphology
To quantify skull morphology, we measured 20 linear variable traits reflecting shape and size that are associated with bite performance and diet in Phyllostomidae bats (Santana et al. 2012; Dumont et al. 2012): Total skull length (TL), Postorbital Width (PW), Zygomatic Breadth (MZB), Posterior Skull width (PSW), Palatal Width at canines (PW), Palatal Width at M1 (PW1), Total Palate Length (TPL), Anterior Skull Length (ASL), Post-Palatal Length (PPL), Maxillary Toothrow Length (MTL), Dentary Depth under m1 (DD), Coronoid Process Height (CPH), Condyle Height (CH), Condyle to Canine bite point (CC), Condyle to m1 point (Cm1), Condyle to m3 point (Cm3), Total Dentary Length (TDL), Condyle-Coronoid Length (CCL), Coronoid-Angular Length (CAL) and Mandibular toothrow length (MAN). For measurements, we used a digital caliper with 0.01-mm accuracy. To describe shape and size variation in cranium, we performed a principal components analysis of functional units (skull and jaw) based on the morphometric variables we measured. Then, we averaged species scores for the first three principal components and used them to represent cranial morphology in subsequent analyses.
Data analysis
We constructed an interaction matrix with bat species in rows and plants species in columns; in every cell we registered the number of interactions (seeds from bat feces) observed between each bat and plant species.
Sampling effort
We assessed sampling representativeness by comparing observed and expected number of species (Bascompte & Jordano 2014). Thus, we calculated the asymptotic richness of species (bats and plants) and interactions, with the nonparametric estimator of abundance Jackknife 1 (Chacoff et al. 2012; Chiu et al. 2014). Posteriorly, we calculated the percentage of representativeness of sampling using the ratio between observed and expected species richness. For analysis, we used vegan package 2.2-3 (Oksanen 2015) in R 3.4.1 (R Core Team 2013).
Network structure
To describe network heterogeneity, we separately calculated the accumulated distributions of interactions for plants and animals, and assessed the fit to three models (Jordano et al. 2003): exponential, power law and truncated power law; using the degreedistr function of bipartite Rpackage (Dorman et al. 2017). To assess for network nestedness, we measured the Nestedness metric based on Overlap and Decreasing Fill index (NODF) that varies from 0 (no nestedness) to 100 (perfect nestedness). In a non-nested network, interactions tend to be symmetric, generating a vulnerable structure to the loss of interactions (Bascompte et al. 2003). The significance of NODF was estimated with a resampling procedure in which we randomly generated 1000 bootstrap samples from the original matrix with Patefield’s (1981) algorithm, as implemented in function r2dtable in R statistical software version 3.4.1 (R Core Team 2013); this algorithm fix marginal totals to distribute the interactions and produce networks in which all species are randomly associated. We tested the estimates of nestedness by calculating 95% confidence intervals for NODF values generated from the random matrixes, and we considered a network as antinested (less nested than expected) or nested (more nested than expected) if the observed NODF was lower or higher than values in confidence interval, respectively.
We measured modularity with the QuaBiMo (Q) algorithm (Dormann & Strauss 2014); which computes modules on pondered bipartite networks, based on a hierarchical representation of the heights of the species’ links and the optimal assignation to the modules (Dormann & Strauss 2014). Q index ranges from 0 (random network without modules) to 1 (maximum modularity) (Dormann & Strauss 2014). To test for significance of modularity, we generate 1000 random matrixes from the original matrix with a bootstrap procedure using Patefield’s (1981) algorithm as before. We tested the estimates of modularity by calculating 95% confidence intervals for Q values generated from the random matrixes, and we considered a network as modular if the observed Q was higher than values in the confidence interval.
We assessed robustness of the network to species extinction with index R (Burgos 2007), which corresponds to the area under the curve of extinctions: relationship between the amount of species that survive in the network against the accumulated number of extinct species of the other group. This index measures how fast one group of the network collapses with the extinction of species from the other group. Values of R close to 1 indicate a curve that decrease slowly with the elimination of species, while values close to 0 indicate a network that collapses quickly when the first species are removed. We generated 1000 simulations of extinctions for each one of the tropic levels of the network (bats and plants) with a bootstrap approach, in which species were removed from the network following three strategies: random elimination, elimination from specialists to generalists and elimination from generalists to specialists. Analyses were performed with the bipartite package (Dorman et al. 2017), implemented in program R 3.4.1 (R Core Team 2013).
Contribution of bats to network structure
We calculated several species-level indexes that estimate the importance of bats to different aspects of network structure: (1) z measure the importance of species to connectivity inside modules (Vidal et al. 2014), (2) c measure the importance of species to connectivity among modules (Vidal et al. 2014), (3) species strength measures the extent to which an assemblage depends on a specific species of the other group (Bascompte et al. 2006), (4) partner diversity is a measure of species’ generality of interactions (Dormann 2011), (5) closeness centrality measures the proximity of a species to all other species in the network (Martín-González et al. 2010; Mello et al. 2015) and, (6) betweenness centrality quantifies species’ importance as a network connector (Martín-González et al. 2010; Mello et al. 2015). These indexes can be correlated since species may have similar importance for distinct aspects of network structure, so we performed a principal component analysis (PCA) on the correlation matrix, and considered the first principal component (PC1) as a measure of species contribution to the structure of the interaction network (Vidal et al. 2014; Sazima et al. 2010). For index estimation, we used bipartite R-package (Dorman et al. 2017) in R 3.4.1 (R Core Team 2013).
We evaluate the importance of abundance and morphology of the bats to the network structure by performing Pearson’s correlations between the index of species importance for the network and: a) bat abundance, and b) the first three principal components of cranial morphology. Finally, to evaluate the contribution of phylogenetic closeness to network structure, we assessed the phylogenetic signal of the index of species contribution to network structure using the lambda method (λ) (Pagel 1999) with Phylosig function of Phytools R-package (Revell 2012). For this, we used the phylogenic tree for Noctilionoidea of Rojas et al. (2016); from which we extracted a tree with the bat species captured during the fieldwork. For analysis, we used software R 3.4.1 (R Core Team 2013).
RESULTS
We captured 11 species of frugivorous bats (111 individuals) during the sampling period; which belong to the Phyllostomidae family and to the genera Artibeus, Carollia, Chiroderma, Dermanura, Sturnira and Uroderma (Table 1. We obtained 45 out of 75 samples of bat feces with seeds; the remaining came from individuals that consumed fleshy fruits with large seeds such as Mango (Mangifera indica, which was one of the most abundant trees around the sampling site. Uroderma species were not considered for analyses since we did not collect seeds in their feces (Table 1). We found 38 plant species (32 genera and 23 families) potentially consumed by frugivorous bats during the monthly censuses in the study area. The most abundant families in those censuses were Moraceae, Piperaceae, Rubiaceae and Solanaceae. However, we only found seeds of nine of those plants in bat feces (Table 2). We identified 55 interactions between nine species of bats and nine species of plants for the network of PRN El Vínculo (Fig. 1). The species of bats considered generalists, those that consumed the most variety of fruits, were A. lituratus (seven interactions) and, C. perspicillata (three interactions). For plants, Ficus sp. (five) and Piper holtonii (four) presented the most interactions (Fig. 1).
Sampling effort
The sampling was representative for bats and plants, since we found that the richness corresponded to 75% and to 81% of the expected richness, respectively. On the other hand, we recorded 65% of the expected pairwise interactions between bats and plants.
Network structure
The general structure of the mutualistic network was heterogeneous since distribution of interactions fitted to the truncated power-law model (truncated power-law AIC=-16. 50; Exponential AIC=-14.37; power law AIC=-4.75); which indicated that most species have a few interactions and few species have a very large number if interactions (Fig. 2). Furthermore, connectance was low since we only registered the 26% of total possible interactions, whereas density indicated that in average each species interacts with 2.5 species from the other group. On the other hand, the network was significantly less nested than expected (antinested) since the observed value of the NODF (0.39) was lower than confidence interval values of the null model (95% Confidence interval=0.41-0.76). Additionally, the mutualistic network presented a modular structure since the observed value (Q=0.48) was not included in the confidence interval of the null model (95% CI=0.19-0.29). This indicated that the network is organized in subgroups of bats and plants (modules) that are more connected among them than with the other species (Fig. 3). Finally, the simulated extinction of specialist to generalist presented high values of robustness for plants and for bats (R=0.90 and 0.83, respectively), with the random elimination of species presented intermediated values (R=0.67 plants and 0.66 bats), and with extinctions from generalists to specialists presented low values (R=0.43 and 0.41); which indicated that the network of interactions is most sensitive to the loss of generalist species. In addition, our results showed that bat assemblage is equally resilient to loss of generalist species as plants, but is less resilient to the loss of specialist plants than bats.
Contribution of bats to network structure
The first principal component (index of importance) had similar and positive factor loads (z=0.42, c=0.36, species strength=0.39, partner diversity=0.45, closeness centrality=0.36, betweenness centrality=0.44), which indicated that species with high values (PC1 scores) contribute the most to network structure. Consequently, A. lituratus was the species thatmost contributed to the structure of the interaction network (species scores: A. lituratus=5.34; A.planirostris=0.32; C. perspicillata=0.26; Dermanura sp.=0.12; C. brevicauda=0.03; S. lilium=-0.67; A. jamaicensis=-1.62; C. salvini=-1.66; D. phaeotis=-2.11). Furthermore, the most abundant species were themost important for network structure since we found a positive correlation between bat abundance and importance index (r=0.95, p<=0.01).
On the other hand, the first three PCs account for 98.05% of the variation in cranium and jaw morphology; with PC1 scores representing skull size, PC2 scores representing relationship between skull width (Posterior Skull Width) and rostrum length (Total Palate Length), and PC3 scores representing relationship between cranial width (Zygomatic and Palatal Amplitude at M1) and length (Post Palatal Length) (Table 3). Consequently, these components represented changes in skull size and shape of bats. The first two morphological components were not associated with the index of importance (PC1 r=0.39, p=0.34; PC2 r=0.05, p=0.91), whereas association with the third morphological principal component was between 1.6 and 12.8 times higher than for first two PCs (r=0.64, p=0.08). This indicated a trend for bat species with a short and wide skull to be the most important for network structure. Finally, the contribution of the species to network structure was not associated with phylogenetic relationships since the index of importance did not have a strong phylogenetic signal (λ=6.61e-05, p=1.00).
DISCUSSION
The mutualistic network of interactions between frugivorous bats and plants was heterogeneous, antinested, modular, and was sensitive to the loss of generalist species. The importance of species to network structure was determined by cranial morphology and abundance; with Artibeus lituratus being the bat species contributing the most to network structure. However, phylogenetically related species were not equally important. Thus, our results suggest that bats are not functionally equivalent for network structure.
Network structure
Contrary to species inventories, there is a considerable fraction of non-possible interactions, due to biological constraints or non-co-occurrence in space or time that cannot be sampled. Besides, using seeds from feces would underestimate the network between large bats and plants with large fruits since they can disperse large seeds without swallow them. In plant-seed disperser networks, around 15% of all potential links can be forbidden because of morphological constraints (Olesen et al. 2011) and 46-61% due to phenological uncoupling, especially in highly seasonal habitats (Jordano 2016) such as dry forest. In addition, the truncated nature of the distribution suggests that there is a limit to the number of interactions of the most connected species in our study area (Bascompte & Jordano 2014). Consequently, the fact that we observed more than 65% of the interactions seems to be appropriate for confidence in results based on mutualistic networks analyses (Jordano 2016).
High robustness is caused by some structural patterns such as heterogeneity, nestedness and modularity (Krause et al. 2003; Memmott et al. 2004; Teng & Mccann 2004; Burgos 2007; Olesen et al. 2007; Bascompte & Jordano 2014). As expected for mutualistic interactions between frugivorous bats and plants (Mello et al. 2011), the network was heterogeneous and modular, but was less nested than expected based on null models (NODF observed=0.39, 95% Confidence interval of NODF expected=0.410.76). The network was heterogeneous since some species displayed a large number of interactions (generalist), which correspond to the bats A. lituratus and C. perspicillata and to the plants Ficus sp. and P. holtonii, whereas other displayed few interactions (specialists). However, previous works have reported heterogeneous bat fruit networks fitting to an exponential function (Zapata-Mesa et al. 2017) not a truncated power-law, as we found. The truncated nature of the distribution suggests that there is a limit to the number of interactions of the most connected species (Bascompte & Jordano 2014). Heterogeneity implies that each plant in the network is connected with a high number of frugivores, and this redundancy increases network robustness to bats extinction (Memmott et al. 2004).
Contrary to our expectations, the interaction network studied did not exhibit a nested linkage pattern; in which specialist interact with subsets of the partners of generalists. Nine bat-fruit datasets were analyzed (Mello et al. 2011) and most net works showed higher nestedness (NODF x=0.566, minimum=0.41, maximum=0.75) than our dataset (NODF=0.39). We found an antinested structure since the missing observations of specialist bats feeding on generalists’ preferred plant species could result in a lower nestedness than in the null models; which can result from competition- induced host utilization and compartmentalization (Dorman et al. 2017). This pattern can result if the most generalist bats (A. lituratus and C. perspicillata) can act as dominant species that monopolizes specific plant species (Ficus sp. and Piper holtonii, respectively) forcing specialist bats to feed onto other plant species (Dorman et al. 2017). Consequently, whether species substantially decrease plant overlap in bat assemblages need to be tested.
Furthermore, the modular structure indicates that the network consists of subgroups of plants and bats that interact more strongly among them than with the other groups. This modular structure has been reported previously and corresponds with the associations that have been reported between particular genera of bats and plants (Mello et al. 2011; Zapata Mesa et al. 2017). Bat species of Stenodermatinae subfamily are consumers of Ficus plants (Giannini & Kalko 2004), with exception of Sturnira bats that prefer plants of Solanum (Saldaña-Vázquez et al. 2013; Montoya-Bustamante et al. 2016); while species of subfamily Carollinae are associated with plants of genus Piper (Marinho-Filho 1991; Thies & Kalko 2004; Saldaña-Vázquez et al. 2013; Montoya Bustamante et al. 2016). We observed these associations in the modules corresponding to A. lituratus, A. jamaicensis, C. salvini and C. brevicauda with Ficussp.; and C. perspicillata and C. brevicauda with P. adumcun and C. rhomboideum. Consequently, our results suggest that bat species from the same genus share morphological traits that allow them to exploit similar plant resources (Olesen et al. 2007).
As expected the network can be considered to be robust to species extinction, particularly when such extinctions occur at random or affect the least linked species in the network first (Mello et al. 2011; Bascompte & Jordano 2014). Network robustness to the random loss of interacting species (R=0-66 bats and 0.67 plants) tended to be higher for bats and intermediate for plants with respect to previous studies (bats=0.41-0.69 and plants 0.58-0.84, Mello et al. 2011); suggesting that the network is resilient to the changes in species and interactions, and to the arrival of other species (Díaz-Castelazo et al. 2010). When specialist species were extinct first, bats presented a lower robustness in comparison to the plants, suggesting that bats have higher dependence for specialist plants.
Contribution of bats to network structure
According to our expectations, cranial morphology was important for determining bats’ contribution to network. Cranial morphology has been associated with bite’s force (Santana et al. 2012; Dumont et al. 2012) and strict frugivorous diet in Stenodermatinae bats (Dumont et al. 2012); which, in turn, has been a key factor in the diversification of phyllostomids (Dumont et al. 2012). Morphological traits associated with the importance of the species for the interaction network we studied (zygomatic amplitude, palatal amplitude and the post palatal longitude) has been identified as important for bite force, which is related with the consumption of fruits in Phyllostomidae (Dumont et al. 2012). The importance of A. lituratus for network structure was derived from having a short and wide skull, which is associated with a strong bite (Dumont et al. 2012), to feed on fruit of different characteristics; which make this bat a module connector. Results suggest that cranial morphology of A. lituratus is translated in an advantage for the use of diverse fruit resources in the study area, since these characteristics enable A. lituratus to have a stronger bite than the other abundant species (i.e. C. perspicillata) and to consume not only hard (i.e. Ficus plants) but also soft fruits (i.e. Piper plants).
As we expected, abundance of bats was associated with their contributions to network structure. This has been found in mutualistic networks, where variations in animal abundance can potentially affect the structure and robustness of the mutualistic networks (Rooney et al. 2006; Bascompte & Jordano 2014; Ramos-Robles et al. 2016; Laurindo et al. 2017). Although species from Artibeus genus reported in the study area (such as A. lituratus and A. planirostris) are morphologically similar, they did not have the same level of importance since A. lituratus had a greater abundance than A. planirostris. However, the relative importance of A. lituratus with respect to C. perspicillata was greater than expected based on differences in abundance. Furthermore, in spite of the differences in abundance between C. perspicillata and A. planirostris, these species presented a similar level of importance for the interactions network. Collectively, these results, suggest that morphology actually plays a determining role in the contribution of bats to network structure. Several studies have suggested that species abundance is the leading factor in the structuration of the mutualistic networks (Jordano 1987; Olesen et al. 2002; Vázquez & Aizen 2004; Gonzalez & Loiselle 2016), which has relevant ecological or evolutionary implications for interactions network (Bascompte & Jordano 2007). This agrees with the neutral theory of diversity that proposes that species are functionally similar (Hubbell 2001), that is biological characteristics of individuals do not matter but the factors like their abundance determine interactions in ecological networks (Krishna et al. 2008; Bascompte & Jordano 2014; Ramos-Robles et al. 2016; Laurindo et al. 2017). On the contrary, our results suggest that frugivorous bats are not functionally equivalent since differences in morphology grants them differential capacities to exploit feeding resources, indicating that the importance A. lituratus is not exclusively derived of its abundance; as the neutral hypotheses would predict.
CONCLUSIONS
Assessing the differential contribution of species to the structure of interaction networks is essential to develop conservation and management strategies that lead to the safeguarding of the ecological processes that generate and maintain diversity. The removal of keystone species could trigger secondary extinctions and cascading effects affecting network stability (Montoya et al. 2006). Therefore, identifying those species is important to conserve and manage ecosystems (Solé & Montoya 2001). In our case, A. lituratus was the species that contributed the most to network structuration, so this species is fundamental for the regeneration process of PRN El Vínculo by the virtue of its seed dispersal capability. This species is generalist and common, representing almost 30% of captures in tropical forest (Muylaert 2017); but their functional relevance for ecological processes in tropical forest has been overlooked. Generalist species connect peripheral species together into modules, but also connect modules keeping the cohesiveness of the network (Guimarães et al. 2007). Therefore, they are considered very important for the conservation of mutualistic interaction networks. The importance of A. lituratus was derived from both a cranial morphology that allows access to several types of fruits and a high abundance. We conclude that cranial morphology is important for interactions between frugivores and plants, and that the contribution of morphologically similar species to network structure may depend on their relative abundances. Taken together, our results suggest that there is not functional equivalence among bats of the studied assemblage; on the contrary, species’ contribution to the network was derived from a combination of biological restrictions and neutral processes but not evolutionary history.