Over the last few centuries, forest ecosystems have been converted into small patches by habitat loss and fragmentation promoted by human activities (Laurance et al., 2014; Haddad et al., 2015). In this very Anthropocene scenario, land-use intensification can affect ecological processes both at the landscape and local-scale (Tscharntke et al., 2005). The decreasing amount of forest cover at the landscape scale as a result of deforestation and habitat conversion is a powerful predictor of species loss in many tropical forests (Dudley, 2005; Fahrig, 2013). Increasing fragmentation and inter-patch isolation decrease the amount of suitable habitat for wild species, affecting their dispersal abilities and movement capacity throughout the landscape (Newbold et al., 2015). Also, changes in local abiotic conditions (e.g., increasing light incidence and temperature, decreasing humidity) along a gradient from the edge to the forest interior are commonly observed at the local scale in fragmented landscapes (Murcia, 1995; Laurance et al., 2007). Therefore, the synergy between processes operating at both the landscape and the local scales may result in a marked change in species composition in those forest fragments, in which the loss of sensitive species might be compensated by the proliferation of species adapted to anthropogenic habitats (Supp and Ernest, 2014; Morante-Filho et al., 2015). However, how the interplay between landscape and local processes affects the ecological interactions between the persisting animal and plant species remains poorly understood.

Habitat loss and fragmentation strongly determine the trajectory of the local forest structure (Arroyo-Rodríguez et al., 2017), leading to a retrogressive succession process toward the homogenization of the vegetation and the predominance of pioneer plant species (Lôbo et al., 2011; Thier and Wesenberg, 2016). In the long-term, it leads to the functional erosion of plant traits related to reproduction that provide food resources for animals (Marcilio-Silva et al., 2016; Rocha-Santos et al., 2020). The loss of functional diversity of plant community might influence the availability and variability of floral, fruit, and seed resources for the associated fauna, thus increasing their vulnerability to future disturbances (Magnago et al., 2014; Pessoa et al., 2017a,b; Lim et al., 2020; Rocha-Santos et al., 2020). Specifically, in forest fragments embedded in highly deforested landscapes, the vegetation homogenization triggers a decrease in fruit biomass production and fruit quality (Pessoa et al., 2017a,b), possibly disrupting animal-plant interactions fundamental for the provision of ecological services (e.g., pollination and seed dispersal) (Mitchell et al., 2015; Valiente-Banuet et al., 2015).

Frugivorous birds are particularly sensitive to changes in local vegetation structure in forest fragments inserted in highly disturbed landscapes (Galetti et al., 2013; Morante-Filho et al., 2018). Larger frugivorous birds are the first to vanish when the small forest fragments do not support the number of fleshy fruits needed to sustain their populations (Donoso et al., 2017; Bomfim et al., 2018; Emer et al., 2018). Also, some forest-dweller frugivorous bird species may not be able to move among forest fragments (Pizo and dos Santos, 2011) and inhabit exclusively the forest interior away from the influence of edges and open areas (McCollin, 1998). Among those birds are the ones that live in the lower forest strata like the understory (Visco et al., 2015; Bradfer-Lawrence et al., 2018). On the contrary, in the remaining small forest fragments small to medium-sized frugivores/omnivores and edge-specialists bird species dominated. Typically, these birds are able to fly relatively long distances looking for food in multiple forest fragments in order to meet their daily metabolic requirements (Lees and Peres, 2009). Therefore, the structure of the understory forest strata is an important component to sustain understory bird species with direct consequences to the interactions they perform and the ecosystem functioning (García et al., 2018).

Ecological networks have been proven to be an efficient tool to unveil how natural and human-driven environmental changes affect the structure, dynamics, and the stability of ecosystems (Memmott, 2009; Hagen et al., 2012; Guimarães, 2020). For instance, in more conserved landscapes, habitat and/or diet specialist species are able to succeed and interact with each other, forming subgroups of cohesive interacting partners that can be depicted by analyzing network modularity (Schleuning et al., 2014). In contrast, species homogenization in more disturbed landscapes can concentrate interactions within some dominant species in the system, reducing interaction evenness (Rodewald et al., 2014). Similarly, the loss of species or decreasing species abundance in smaller forest fragments can affect the number of links per species by reducing the number of partners available for the interactions to occur (Muñoz et al., 2017; Emer et al., 2020).

Here, we combine network and landscape ecology to evaluate the interplay between landscape and local scale processes on understory bird seed-dispersal networks on a hyperdiverse yet highly disturbed tropical forest. We studied 19 forest sites embedded in two regions with contrasting patterns of land use in the still poorly studied northeast of the Brazilian Atlantic Forest. Thus, for each site, we evaluated: (1) the total number of interactions between frugivorous birds and plant species; (2) the mean number of links per species; (3) interaction evenness; and (4) modularity. Given that our forest fragments present simplified understory frugivorous bird assemblages (Morante-Filho et al., 2015, 2018) and low functional diversity, quality, and biomass of fruits (Pessoa et al., 2017a,b), we expect to record bird seed-dispersal networks of exceedingly small size and reduced number of interactions, especially in deforested landscapes, in which the emergent structural properties may be compromised by the loss of both interacting partners and the interactions they perform. Additionally, as our study assessed a subset of the frugivorous bird assembly, i.e., mainly the understory species, and used a specific method to quantify the interactions, we believe that these particularities could contribute to limited-sized networks. At the landscape scale, we predicted a decrease in the number of interactions and on the mean number of links per species as forest cover decreases, and edge densities and isolation increase. In this scenario, modularity is expected to decrease due to the loss of specialist bird-plant interactions and the persistence of a few generalist bird species (i.e., species adapted to use different food resources and occupy environments with distinct disturbance intensities) able to establish links with many species and hence connecting the whole network (Blake and Loiselle, 2002; Emer et al., 2018). Thus, the interactions might be concentrated within a few dominant bird species in the system, consequently decreasing network evenness in more deforested landscapes. Likewise, at the local scale we expect that the lower fruit availability associated with changes in forest structure will reduce the number of interactions, the mean number of links per species, interaction evenness and modularity due to the predominance of generalist bird species able to use the wherever resource available.

We captured 491 birds along 19 forest sites (mean = 25.84 birds ± SD = 15.91 per site), but we obtained samples of fruit or seed only from 185 birds (mean = 9.74 birds ± SD = 9.10 per site) belonging to 14 families and 28 species (Supplementary Appendices 7, 8), forming the interplay of interactions used to build the local network. Even after six days sampling and 600 net-hours we did not reach an asymptote in the number of interactions in several forest sites (see Supplementary Appendix 4), especially in those inserted in more forested landscapes. We failed to record a single interaction in the forest site with the lowest amount of forest cover (15%) because no birds were captured therein. In fact, the low number of captured birds with food content contributed to limited sampling sufficiency of interactions, given that no interaction was recorded for 62% of the birds caught in the mist nets (Supplementary Appendix 8). In contrast, sites located in LFC region showed reasonable sample completeness (Supplementary Appendix 4).

In terms of bird diversity, the families Pipridae (five species, n = 145 individuals), Dendrocolapitidae (four species, n = 6) and Thraupidae (four species, n = 4) were the most representative (Supplementary Appendix 7), showing the greater diversity of bird species reported consuming fruits. The most abundant bird species were Ceratopipra rubrocapilla (n = 52) and Dixiphia pipra (n = 51), followed by Machaeropterus regulus (n = 26), Manacus manacus (n = 15), and Schiffornis turdina (n = 6). Confirming our expectations, there was a significant difference in bird richness between regions (t = 5.0, df = 13.4, p < 0.001), in which we recorded 22 (mean = 5.2 ± SD = 1.69 per site) bird species in the HFC and 12 bird species (mean = 2.22 ± SD = 0.83 per site) in the LFC region. Among those, 16 bird species were recorded exclusively in the HFC region, six were exclusive from the LHC and the other six were common to both regions (Supplementary Appendix 7).

In turn, plant diversity was characterized by 30.837 seeds/fruits in the birds’ stomach content and feces, distributed in 68 morphospecies, of which 10 met the species level identification and another 13 the genus level. We observed a significant difference in the number of seed morphospecies between regions (t = 3.7, df = 11.2, p = 0.003), with 56 morphospecies in the HFC region (mean = 9.3 ± SD = 4.9 per site), and only 17 morphospecies in the LFC region (mean = 3.2 ± SD = 1.6 per site). Miconia hypoleuca was the only plant species found in all studied sites, being the most consumed by the majority of bird species (n = 22 species, 78.6% of the bird community) with a total of 28.734 seeds recorded, showing to be an important resource for the understory bird assemblages.

Such interplay of bird and plant diversity resulted in 268 quantitative interactions between frugivorous birds and fleshy-fruited plants, with an average of 14.11 ± 12.98 per forest site (Supplementary Appendix 8). The higher number of birds with food content, the higher number of interactions and therefore the larger-sized networks were observed in the more forested landscapes, mainly in the HFC region (Figure 2). Indeed, the number of interactions was the only network descriptor that responded to changes in both landscape and local predictors and that was mainly driven by region (Supplementary Appendix 9). Seven models were considered equally plausible to explain the relationship between local and landscape-scale predictors and the number of interactions (Table 1 and Supplementary Appendix 9). However, only two models, which consider the region and forest cover separately, were significant (Table 2 and Figure 3). Specifically, the number of interactions was only positively related to forest cover at the landscape scale (Figure 3A), with a higher number of interactions recorded in the HFC region (Figure 3B). Also, all models used to explain the variation in the number of interactions at forest sites showed a good fit (Supplementary Appendix 10) and did not show spatial autocorrelation (p > 0.05) (Supplementary Appendix 11). In contrast, the number of links per species, interaction evenness, and modularity were not affected by any landscape or local predictors evaluated in this study (Supplementary Appendix 9 and Supplementary Tables 6–8). Also, when network size was controlled for, the number of links per species, interaction evenness, and modularity did not differ from the null models (Supplementary Appendix 12), indicating the lack of significant structural properties in these simplified bird-plant networks.

We showed that deforestation reduces the number of interactions between understory frugivorous birds and fleshy-fruit plant species at the landscape level but that does not change the structure of small-sized seed-dispersal networks in highly depauperate tropical forests. The observed decrease in the number of interactions reflects the decline or even the loss of bird population of specific species, as already detected in previous assessments (Morante-Filho et al., 2015, 2018). Despite such numeric decline in interactions, no significant network structural change was observed in both HFC and LFC regions, possibly because both regions encompass highly impoverished assemblages of understory bird species that have already lost part of its functionality due to defaunation (Canale et al., 2012). Importantly, our study provides further evidence about the pervasive effects of landscape-scale forest loss, not only for understory bird richness, but also for animal-plant interactions and probably the ecological services they provide.

Forest cover at the landscape scale was the best and only predictor explaining the decrease in species interactions. Previous studies have also reported the detrimental effects of forest loss on plant diversity, especially late-successional, shade-tolerant and zoochoric tree species (Benchimol et al., 2017; Rocha-Santos et al., 2017). Similarly, forest loss has been cited as the main factor driving frugivores loss in tropical forests (Coster et al., 2015; Morante-Filho et al., 2018). Indeed, forest loss can act as an important selective force, filtering frugivorous birds with specific phylogenetic and functional features, and thus interfering in the seed dispersal networks (Peña et al., 2020). In particular, recent studies reported significant changes in network structure (e.g., higher nestedness and lower modularity) after forest loss in Atlantic forest sites (Emer et al., 2020; Marjakangas et al., 2020) or changes in bird functions in deforested landscapes (Coster et al., 2015). Here, we found highly simplified seed-dispersal networks with only a few pairs of interacting species, whose structural properties have been depleted due to the loss of forest cover in these human-modified landscapes.

Our results also showed that the number of interactions depends on the regional context in which forest fragments are inserted, with a higher number of interactions recorded in the region with higher amounts of forest cover. In our study, HFC region is composed by more forested landscapes, including native remnants immersed in a biodiversity-friendly matrix such as the cocoa agroforestry systems. Altogether, these features allow a greater diversity of birds and plants in forest fragments (Faria et al., 2006; Sambuichi et al., 2012) and hence a higher number of interactions. In contrast, the LFC region is composed of small forest fragments that present declines in their regional diversity (Morante-Filho et al., 2016). Additionally, we also observed a marked change in bird and plant species composition between LFC and HFC regions (Supplementary Appendix 13). From the total interactions observed (n = 268), only 16% were recorded in the LFC region, occurring among a limited number of bird and plant species (Supplementary Table 1). In particular, many captured bird species in the LFC region are occasional frugivores, which may help to explain the extremely low number of interactions recorded. Conversely, in the HFC region we observed a more diversified bird assembly, including large frugivores such as Ramphastos vitellinius (Supplementary Appendix 7). This frugivorous bird can facilitate seed dispersal, including large-sized seed species, among forest fragments due to their high vagility. Accordingly, the absence of certain bird species can limit seed dispersal and the recruitment rates of plant species, compromising the vegetation regeneration of small and severely disturbed forest remnants (Gomes et al., 2011; Donoso et al., 2017; García et al., 2018).

Changes in composition and structure of the local vegetation, as well as fruit availability, are known to shape the diversity of frugivores (García et al., 2011; Plein et al., 2013). Previous studies in the same landscapes studied here have found that deforestation leads to both a simplification in the forest structure and a decrease in fruit biomass (Rocha-Santos et al., 2016, 2017; Pessoa et al., 2017b). Nevertheless, we found no evidence that local changes in fruit availability or vegetation complexity affected bird-plant networks, possibly because the most representative bird family recorded (Pipridae) was abundant in all sites. Despite the decrease in abundance or even the loss of bird species such as C. rubrocapilla, D. pipra, and M. regulus, the abundance of other species such as M. manacus increased in more disturbed sites. Indeed, M. manacus was the only member of this family recorded in forest sites located in landscapes with less than 20% forest cover (Supplementary Appendix 7). All the above-mentioned bird species share similar ecological features, i.e., they are forest-dweller frugivorous species that forage in the understory stratum (Supplementary Appendix 7). Thus, we suggest that functional redundancy might explain the results observed here.

Interestingly, none of the evaluated network descriptors was affected by any local or landscape variables assessed here. The lack of significant differences in the number of links per species across landscapes might be explained by the fact that the most captured birds (∼78%) belonged to the same family—Pipridae—which were recorded consuming only fruits of a few pioneer plant species in our fragments. Typically, the predominance of these species is well reported given that they are the dominant frugivorous birds in the forest understory and commonly captured in mist nets (Loiselle et al., 2007). Similarly, we also did not observe changes in interaction evenness, which were mostly high thus indicating the predominance of homogeneity in the distribution of interactions between birds and plants. This result, associated with the small size of the networks, could reflect the low fruit productivity in these forest fragments (Pessoa et al., 2017a). In this situation, the already depauperated bird assemblages (Morante-Filho et al., 2015) would heavily rely on the very few diversity of resources available, which are mainly provided by the pioneer species (e.g., Miconia hypoleuca, Miconia mirabilis, and Leandra rufescens) commonly observed in these bird diet. Indeed, these three plant species corresponded to 97% of the seeds observed in our interaction networks and were recorded in almost all forest sites, regardless of environmental differences in local or landscape-scale here evaluated. This is an alarming result because the reduced richness of frugivorous birds in the forest understory can lead to plant species homogenization via decrease in seed dispersal. In this case, seed dispersal is occurring only for a limited number of plant species, and other plant species that have lost their interaction partners can be locally extinct in the long term (Loiselle et al., 2007).

The lack of significant changes on network modularity at both the regional and local landscape scale indicates that the impoverished understory assemblage of plants and frugivorous birds do not form distinguished subgroups of interacting partners according to any specific trait, ecological niche and/or spatial segregation in these human-modified tropical forests. In our study, the seed-dispersal network is shaped by small-bodied species (e.g., the five dominant species in our study present an average body mass of 13 g) consuming and, most likely, dispersing small-seeded plant species (Silva et al., 2018). Therefore, the relatively low values of modularity found in our study can be associated with the presence of a few remaining forest birds, ecologically plastic and generalists in their diet, that feed on the available small-sized fruits. For instance, about 50% of the recorded birds present a diet consisting of different food items such as invertebrates, seeds and plants (Supplementary Appendix 7). In addition, high modularity in ecological networks is related to megadiverse communities, in which the diversity of animals that interact with particular sets of plant species could make them tightly linked within modules (Donatti et al., 2011). However, in our study we observed highly simplified networks composed by low richness of bird and plant, mainly in the more disturbed region.

Our results revealed a detrimental decrease in the size of bird-plant networks, triggered by landscape-scale deforestation. Depleted communities often exhibit gaps in ecosystem functionality through the loss of functional groups and decrease in the number of interaction partners (Laliberté et al., 2010; Muñoz et al., 2017). However, contrary to our expectations, the observed impoverishment of bird-plant interactions following deforestation was not mirrored by changes in network structure. Most of the interactions exclusively reported in sites immersed in more forested landscapes are shaped by a few small frugivorous birds and pioneer plants that are abundant and widely distributed in these sites.

The maintenance of network structure despite the variation in species richness and composition among sites can suggest a high functional redundancy among the remaining understory bird assemblages, thus assuring the dispersal of a few widespread pioneer species. Yet, networks are limited to a few plant and small bird species interacting among themselves, and also similar among sites regardless of the landscape or local levels of disturbance. Thus, even in our most preserved forest sites, characterized by a larger number of interacting species, network structure is functionally similar to those more disturbed forest sites.

All relevant information, including dataset, are available in the Supplementary Material.

The animal study was reviewed and approved by Comissão de Ética no Uso dos Animais (CEUA) of Universidade Estadual de Santa Cruz.

ÍM, EC, and JM-F: conceived and designed the experiments. ÍM: performed the experiments. ÍM, CE, and JM-F: data analysis and interpretation. ÍM and JM-F: writing-original draft. ÍM, CE, EC, and JM-F: writing the manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer, MP, declared a past co-authorship with one of the authors CE, to the handling editor.