Frugivory and seed dispersal in the Cerrado: Network structure and defaunation effects

Seed dispersal is a fundamental process that is highly threatened by the rapid decline of large‐bodied frugivores worldwide. The Brazilian Cerrado, the largest savanna in the world, represents an ideal site for investigating seed dispersal because of its biodiversity, environmental challenges, and knowledge shortfalls. We performed a systematic literature review to analyze the seed dispersal network in the Cerrado and the potential impacts of the defaunation of large‐bodied frugivores on it. We considered network metrics, calculated the defaunation index of the frugivore assemblage, and compared traits among different fruit‐sized plants and their respective dispersers in the network. We retrieved 1565 interactions involving 193 plant species and 270 animal species. Results show that the Cerrado seed dispersal network is slightly nested and considerably modular, dominated by small‐ to medium‐sized generalist species, such as passerines, marsupials, and mesocarnivores. Nonetheless, large‐bodied frugivores like the lowland tapir have a key role in the network due to their great foraging and network integration capacity. The Cerrado frugivore assemblage is moderately defaunated, with possible effects in its interactions with large‐fruited plants. The Cerrado's defaunation and functional loss of large vertebrates deserve urgent attention to further understand the impacts on seed dispersal mechanisms and ecosystem functioning.


| INTRODUC TI ON
Seed dispersal is a key process for ecosystem functioning, especially in tropical environments, where it has profound ecological and evolutionary repercussions on the individual , population , community (Levine & Murrell, 2003), ecosystem (Rogers et al., 2021), and landscape (Villar et al., 2021) levels. One of the key elements for seed dispersal is the size of the interacting agents. Large plants usually have large seeds (Moles et al., 2005;Moles & Westoby, 2006;Thompson & Rabinowitz, 1989), which are better dispersed by medium-to largebodied frugivores (Corlett, 1998;Kitamura et al., 2002). However, large-bodied vertebrates represent one of the most highly endangered species groups in the world (Barnosky et al., 2011;Dirzo et al., 2014), commonly threatened by hunting, habitat destruction, fragmentation, selective-logging, and human-wildlife conflicts (Markl et al., 2012;Peres & Palacios, 2007;Wright et al., 2007). As a consequence, large-bodied animals worldwide are facing sharp declines in their population size (Barnosky et al., 2011;Ceballos et al., 2017;Young et al., 2016). This fast-paced faunal perishing induced by anthropogenic pressure is termed 'defaunation' (Dirzo & Miranda, 1990), and in a forest context has been referred to as a generalized 'empty forest syndrome' (Redford, 1992). An 'empty forest' refers to not only key species being extinguished, but also to the loss of their ecological interactions with different trophic levels, which may cause cascading effects impacting ecosystem functioning (e.g., Beck et al., 2013;Bello et al., 2015;Galetti & Dirzo, 2013;Rogers et al., 2021;Terborgh et al., 2008).
The Brazilian Cerrado is the most biodiverse savanna in the world (Myers et al., 2000;Oliveira & Marquis, 2002). The biome has highly distinctive levels of species richness and endemism (Klink & Machado, 2005), which are reflected in its diverse ecological interactions. More than half of its woody plant species have animal-mediated seed dispersal, being highly reliant on vertebrates (Gottsberger & Silberbauer-Gottsberger, 1983;Kuhlmann & Ribeiro, 2016a, 2016b. The Cerrado vegetation exhibits a heterogeneous and intricate landscape mosaic that ranges from vast grasslands to dense canopy forests within short distances (Oliveira-Filho & Ratter, 2002;Ratter et al., 1997). These formation assortments are essential to maintain the plant-frugivore mutualism in the biome (Kuhlmann & Ribeiro, 2016b). However, almost half of its original territory has already been lost for conversion to pastures and crops for large-scale agribusiness operations (Strassburg et al., 2017;Machado et al., 2004). This fast land conversion, together with other anthropogenic drivers like habitat fragmentation and hunting, has resulted in defaunation rates of nearly 50% in the Cerrado (Bogoni et al., 2020).
Although Brazil is among the best-studied tropical countries in terms of seed dispersal and frugivory (Donoso et al., 2022), the Cerrado remains largely understudied. Despite the pioneering contributions done by Ribeiro (2016a, 2016b) on the ecological and phylogenetical relationships of Cerrado fruits and frugivores, Darosci et al. (2017) on the seasonality effects on Cerrado plant-frugivore interactions, and Purificação et al. (2020) on the structure of bird-plant interaction networks in savanna-forest mosaics, our understanding of the dispersal dynamics at community-to ecosystem-level is still emergent. There is evidence of ants complementing seed dispersal along with birds in the Cerrado (Christianini & Oliveira, 2009, 2010, which helps us grasp how such dispersers may flourish when larger animals are extirpated or functionally lost (Christianini et al., 2014). Nonetheless, research explicitly addressing Cerrado dispersal dynamics in defaunation or functional-loss contexts deserves urgent scrutiny.
Comprehending how frugivores and fruited-plants interact at the community-level offers powerful tools to tackle conservation issues and propose management actions (McConkey et al., 2012).
For seed dispersal and frugivory research, the interacting network approach is especially useful because it allows analysis of both network structure and properties. It also allows the evaluation of species' functional roles (such as seed dispersal effectiveness- Schupp et al., 2010) when enough information on frugivores' natural history and dispersal activity (pre-and post-dissemination stages) is provided (Heleno et al., 2014;Simmons et al., 2018). Moreover, the detection of structural patterns in mutualistic networks allows for a better understanding of communities' ecological and evolutionary dynamics .
Considering the ecological importance of plant-frugivore interactions and the effects of defaunation on them, we ask the following There is evidence that Cerrado plants are mostly visited by generalist species Francisco & Galetti, 2001Paniago & Silva, 2017) that occur in several vegetation types (Kuhlmann & Ribeiro, 2016b). Some studies also report generalization in frugivory interactions (Darosci et al., 2017), as well as considerable overlaps of frugivore assemblages, regardless of habitat or fruiting season (Maruyama et al., 2019). Thus, we hypothesize the Cerrado has a diverse and plural network, predominantly composed of generalist species. Given the ecological-evolutionary relationship between large animals and large-fruited plants in the Cerrado (Guimarães et al., 2008), we also expect the network may be experiencing impacts from large-dispersers' defaunation, affecting interactions of large-fruited plants.

| Experimental design
We performed a systematic literature review (SLR) using the 'Protocol, Search, Appraisal, Synthesis, Analysis, and Report' (PSALSAR) framework (Mengist et al., 2020). The main objective was to obtain secondary data about plant-frugivore interactions to assemble a Cerrado seed dispersal network. Similar methodologies have already been successfully adopted to model other ecosystems' networks (Oliveira et al., 2015;Almeida & Mikich, 2018;Ong et al., 2022). Table 1 TA B L E 1 Information on the adopted search strings, consulted data bases, and period of search for the systematic literature review.
We covered studies from all Cerrado physiognomies (forest-, savanna-, and grassland-like formations), different locations, and diverse methodologies (Table 3). While the use of different methods with varying sampling efforts can influence the capacity in detecting species interactions (Vitorino et al., 2022), we believe that such limitation was circumvented as most studies used a similar approach.

| Network analyses
We created a binary ( for making comparisons with the literature Sarmento et al., 2014).
The matrix was graphically represented by bipartite graphs, in which species were designated as vertices (points) and their interactions as edges (lines). Plants were grouped according to the fruit-length categories presented by Kuhlmann (2018aKuhlmann ( , 2018b: small-fruited (<1 cm), medium-fruited (1-4 cm), and large-fruited (>4 cm). Birds were classified as small (<100 g) or large (>100 g), whereas mammals were grouped into small (<1 kg), midsized (1-10 kg), or large (>10 kg) (Bernardo & Melo, 2013;Chiarello, 2000;Vidal et al., 2013). Given the importance of considering the ecological context to define the largest fauna of an ecosystem (Hansen & Galetti, 2009), we considered 'large-bodied vertebrates' the largest native species acting as seed dispersers in the Cerrado: midsized to large mammals and large birds. Domestic (livestock) species were not included in our analysis. We used the bipartite R package v2.16 (Dormann & Strauss, 2014) and the Pajek software v5.13 (Batagelj & Mrvar, 1998) to model the network. We adopted several descriptors to characterize the structure of the Cerrado plant-frugivore network, such as "Nestedness", "Modularity", "Connectance", "Link per Species", "Species Strength", and "Shannon Index for Diversity of Interactions" (Carreira et al., 2020;Silva et al., 2015;Tylianakis et al., 2010). Details on the interpretation and methods of each metric can be found in the Appendix S1.
Lastly, we did proportional analyses considering the total amount of frugivores (lato sensu), their biomass (further details about data collection in subsection 2.5), and network interactions. We assessed differences in fruit and seed size among disperser groups with a nonparametric Kruskal-Wallis, followed by a Dunn's post hoc test.

| Network ecological role
Species were classified according to the ecological roles they performed in the network. When a species exhibited a disproportionate number of interactions, especially within its module, it was classified as a "hub". On the other hand, if the species was essential to connect different modules in the network, it was considered to be a "connector". Some species were classified as both when they had many recorded interactions and showed a high capacity to bind modules (Olesen et al., 2007).

| Seed dispersal potential
Aiming to appraise the ecological contribution of the main frugivores registered in the network, we adapted Kuhlmann's (2018aKuhlmann's ( , 2018b "Seed Dispersal Potential" classification system, which is based on the combined evaluation of morphological and behavioral attributes exhibited by fauna attracted to fruited-plants. Kuhlmann's system considers body size, feeding capacity, frugivory degree, and fruit processing to rank species accordingly to their dispersal capacityviz. 1 (low potential), 2 (medium potential), or 3 (high potential). We converted his ranking categories to decimal scale and normalized it to range from 0 (non-existent dispersal potential) to 1 (maximum dispersal potential) after accounting for the network frugivore assemblage. Species that did not originally feature in Kuhlmann's frugivore ranking could not be considered for the seed dispersal potential

Selection criteria Decision
The study addresses the topic of seed dispersal, frugivory, or their related conservation science (e.g., ecological restoration, rewilding, deforestation, defaunation)

Inclusion
The study presents the search string terms, at least, in their title, abstract, or listed keywords

Inclusion
The study provides clear information on interactions of fruited-plant species and frugivores (i.e., which species interact with each other)

Inclusion
The study addresses the subject precisely in the Cerrado biome Inclusion Studies that are duplicated within the search documents Exclusion TA B L E 2 Selection criteria used to include or exclude studies in the systematic literature review.
(SDP) estimation as there was no available score for them. Hence, we appraised a reduced sample size of frugivores (n = 125; Table S2) to calculate the SDP for the Cerrado network.

| Defaunation and frugivore-loss impact
We calculated a defaunation index (DI) (sensu Emer et al., 2020) based on the Bray-Curtis dissimilarity index (Legendre & Legendre, 2012): where "B f " and "B r " represent the biomass of species k in the focal 'f" and the reference "r" assemblages, respectively, in a universe of "s" total species. DI ranges from 0, when species in both assemblages are completely similar (non-defaunated), to 1, when all species are absent in the focal assemblage (complete defaunation) (Carreira et al., 2020;Giacomini & Galetti, 2013).
In this study, "Focal Assemblage" referred to the species surveyed in the SLR, and "Reference Assemblage", to the expected frugivores for the Cerrado biome. As a proxy for the biome species pool, we used the list of frugivorous species reported by Kuhlmann (2018aKuhlmann ( , 2018b for the Cerrado (Table S2 -

| Structure of the cerrado seed dispersal network
We

| Defaunation and frugivore-loss impact
Comparing the assemblage of frugivores in the network with the expected baseline (Table S2), we found a defaunation index of 0.43.
The total disperser body mass did not significantly differ among small-, medium-, and large-fruited plants (χ 2 = 0.54; df = 2; p = .76; Figure S4). We found a low similarity between the species composition of the large-bodied frugivore group and the actual dispersers of large-fruited plants in the network (Sørensen index = 0.198).

| What is the structure of the Cerrado seed dispersal network?
Our study demonstrates the heterogeneous structure of the   (Dugger et al., 2019). Low nestedness also indicates more interactions between generalist species in the network .
The considerable modular pattern in the Cerrado seed dispersal network encompassed nine subgroups that are more intimately connected with each other than with species from other modules (Olesen et al., 2007). In mutualistic networks, modular topologies are generally associated with complex communities .
Additionally, modular networks ensure robustness and allow greater stability against disturbances because impacts can be retained by a single module, dampening their propagation to the other ones (Thébault & Fontaine, 2010). If cascade effects arise from frugivore loss, a modular network will prevent further extinctions from occurring (Correa et al., 2016;Ramos-Robles et al., 2018).
We must also consider the repercussion of forbidden links (i.e., unobservable connections due to linkage constraints) while determining modules in the seed dispersal network. Restrictions caused by phenological uncoupling (e.g., seasonal availability of fruit and frugivores; interactions of migratory species) or biogeographical mismatches (e.g., species restricted to one of the biome's biodiversity supercentres) prevent our observation of potential links among species (and consequently modules), regardless of sample efforts (Olesen et al., 2011).
The low connectance found in the Cerrado network seems to reflect a pattern reported for tropical systems. Due to their high species richness, tropical environments have a relatively low number of established interactions in their networks, thus displaying particularly low connectivity (Jordano, 1987). Consequently, if a species were randomly selected, it would have very few interactions with other species, making it more sensitive to the secondary effects of other species' extinctions (Silva et al., 2007). Although the Cerrado network has low connectance, the network may have sufficient robustness to tolerate disturbances, according to its observed modularity pattern.
Nonetheless, further evidence is needed to decipher whether the de- The predominance of small bird interactions is explained by, first, this group's abundance in the Cerrado (Pinto et al., 2008;Tubelis & Cavalcanti, 2001), and, second, small-and medium-sized birds are highly effective seed dispersers (Godínez-Alvarez et al., 2020), interacting with multiple plants. We also highlight that most seed dispersal research focuses on particular frugivore taxa, often birds (Vidal et al., 2013).
F I G U R E 2 Bipartite representation of the Cerrado seed dispersal network. The graph illustrates the pairwise interactions (lines) between frugivore dispersers (multi-colored boxes, left-hand side, n = 274 spp.) and fruiting plant species (green boxes, right-hand side, n = 193 spp.) in the biome. Larger boxes mean species with a greater amount of interactions performed (in the case of dispersers) or received (in the case of plants). Major disperser groups had their boxes and lines color-coded: ants (brown), reptiles (black), small birds (<100 g; blue), large birds (>100 g; turquoise), bats (purple), small mammals (<1 kg, non-volant; light pink), midsized mammals (1-10 kg; orange), large mammals (>10 kg; ruby). The network was modeled by compiling secondary data on plant-frugivore interactions already observed in the biome. All Cerrado vegetation physiognomies were considered (savannah, forest, and grassland formations). Interactions are weighted equally since the network's matrix was binary. The full list of species can be found in Tables S3 and S4 for  dispersers  All these mentioned generalists had distinct network roles and high species-strength metrics, suggesting their relevance as likely keystone species.
Identifying species' network role is key for conservation. If a connector becomes extinct, the network can fragment into several isolated modules-although their inner structure may be preserved (Donatti et al., 2011). This would be the probable scenario if tapirs, an already vulnerable large-sized mammal (Varela et al., 2019), go extinct: the Cerrado seed dispersal network would become less resilient and more susceptible to subsequent extinctions (Correa et al., 2016;Ramos-Robles et al., 2018). On the other hand, if a hub species disappears, the module it belonged to might fragment, bringing cascade effects on neighboring modules (Donatti et al., 2011).
For example, the extinction of the stricto sensu frugivore A. galeata could disintegrate the core structure of the modules it mostly interacts (Silva & Melo, 2011). Whatever its network role, when a species is lost, disperser richness becomes increasingly important as disperser abundance declines-meaning the identity of this lost species could greatly impact the number of plants dispersed (Rumeu et al., 2017). It must also be considered indirect or unforeseen impacts caused by disperser extinction on plant communities (Rogers et al., 2021).
In terms of seed dispersal potential, the frugivore assembly in the Cerrado network has a moderate ecological capacity (SDP = 0.48), probably due to the diversity of species acting in the network. Animals interact differently with fruits and seeds: while some frugivores swallow fruits entirely, leaving seeds intact, others destroy seeds when feeding (Levey, 1987;Ragusa-Netto, 2014). In heterogeneous landscapes like the Cerrado, multiple species providing the dispersal service are important to ensure complementary network efficiency (Rother et al., 2016). In fact, a generalized system (i.e., multiple dispersers of contrasting efficiency) can be more advantageous for plants rather than a specialized one (i.e., few but highly efficient dispersers). This is because several dispersal vectors diversify interactions with plants, increase the system's overall seed dispersal efficiency, and ensure plants' maintenance even if their most efficient agents go extinct (Schupp et al., 2010).
In the Cerrado, the synergistic effects of birds and ants help to stabilize temporal fluctuations in the number of seeds dispersed from Cerrado sensu stricto trees (Campagnoli & Christianini, 2021).
Likewise, secondary dispersal promoted by different agents is pivotal for the Cerrado vegetation. In Arecaceae species, seeds are dispersed by a rich assemblage of ground-and tree-dwelling frugivores, which assist their short-to medium-distanced dispersal (Blanco et al., 2019).

F I G U R E 3
Performance of frugivores in the Cerrado seed dispersal network -evaluated in the number of interactions (y-axis) and species strength metric (bubble size) -in relation to their body mass. Each point on the graph represents a frugivore species, color-coded according to its main taxonomic group: ants (orange), birds (blue), mammals (pink), and reptiles (green). Point size (bubbles) varies according to species' strength metric (i.e., the larger the bubble, the greater the species' strength metric). Points plotted higher on the chart represent species with the highest number of interactions in the network. Frugivore species with a high number of interactions and a high score of species strength were labeled in the chart (except for Atta and Wasmannia auropunctata, which were placed to identify the outlier points in the body mass scale). A complete list, detailing species strength metrics for each frugivore of the network, can be found in Table S5.
High degrees of generalism-as seen in the Cerrado networksuggest ecological redundancies to seed dispersal, so that a single fruited-plant species can be dispersed by various, distinct agents.
Redundancy might even uncover "rescue effects" to the biome's system, in which plants missing their main disperser are unlikely to get extinct due to the "dispersal rescue" exerted by alternative, generalist agents.
Our SDP score represents an initial attempt to numerically evaluate the ecological potential of Cerrado frugivores for seed dispersal.
Its value is not fixed, however, as it might change if spatially and temporally analyzed. Plant-frugivore interactions usually vary by region and season, susceptible to fruit availability and frugivore abundance (Fleming & Kress, 2013;Howe, 2014). Given Cerrado's extension and seasonality, these factors might also be pondered. SDP is not an estimation of Seed Dispersal Effectiveness (SDE). SDE would require accounting for quantitative and qualitative aspects of seed dispersal beyond our data availability-for example, amount of seeds removed per species, seed viability after interactions, seed deposition site, and seedlings' fate (Schupp et al., 2010).

| What are the effects of the defaunation of large-bodied frugivores in the Cerrado seed dispersal network?
We found a DI of 0.43, which means a moderate defaunation intensity (DI = 0.  (Almeida & Galetti, 2007;Vidal et al., 2013). Investigating, thus, their role as seed dispersers is relatively challenging (Vidal et al., 2013).
Some of our findings suggest the effects of large-frugivore defaunation on the Cerrado seed dispersal network, although more evidence and variables should be considered to verify its comprehensiveness. Nonetheless, by the precautionary principle, we call for urgency to the possibility.
In terms of metrics, the considerably modular pattern of the Cerrado network could be a proximal indication of defaunation.
Medium-to large-bodied frugivores disperse a high number of plant species, usually connecting modules. When they are lost, the network could fracture, emphasizing its modular pattern (Donatti et al., 2011). Another sign could be various small-bodied species acting as hubs in the Cerrado network, compensating for the loss of large-bodied species in network topology (Vidal et al., 2013).
Regarding species traits, the lack of correlation between fruit size and disperser body mass is intriguing. Wider fruits are usually associated with larger frugivores (Herrel et al., 2004;Kalko et al., 1996;Lim et al., 2020), and several Cerrado's large-fruited plants exhibit the so-called 'megafaunal syndrome', meaning their main seed dispersers should be large-bodied frugivores (Guimarães et al., 2008). We expected thus these variables would be positively related. Nevertheless, such non-correlation could have other causes rather than defaunation. For example, it could have been influenced by the foraging behavior of tapir, a species capable of ingesting diverse-typed and size-varied fruits (O'Farril et al., 2013). Likewise, Chen and Moles (2015) found a negative relationship between seed size and vertebrates' body size largely due to large ungulates ingesting smaller seeds. Another reason is that we could not isolate the effects of fruit size from other covariates (e.g., morphological traits, chemical components, phylogeny), masking correlation.
Disperser body mass for large-fruited species did not differ from other plant categories. We awaited it would be significantly heavier concerning small and medium-fruited species. A downsizing in the frugivore assemblage, induced by defaunation (Bogoni et al., 2020;Dirzo et al., 2014;Estes et al., 2011), could be an explanation, although this result could stem from limited data availability.
We also saw a low similarity between the network's largebodied species and the actual dispersers interacting with largefruited plants. Consequently, Cerrado large-fruited plants could be relying on a greater frugivore diversity for their dispersal as a result of defaunation. We draw a parallel from the recent geological past when Cerrado megafaunal plants (e.g., Caryocar, Hymenaea, Annona) used to be dispersed by massive frugivores-for example, giant ground-sloths, gomphotheres (Guimarães et al., 2008) -and, nowadays, their disperser pool is composed by relatively smaller frugivores (Guimarães et al., 2008). Similarly, in Atlantic Forest areas where large dispersers are declining or functionally extinct, some large-fruited plants have already experienced rapid microevolutionary changes due to selective pressure, exhibiting smaller seeds capable of accommodating a wider-and smaller-range of frugivores (Carvalho et al., 2016;. Perhaps this could be happening in the Cerrado, yet more data would help test this hypothesis. Information on the number of fruits consumed per species, visitation rate, and population density could help estimate the importance of large-bodied frugivores for Cerrado large-fruited plants' dispersal.

| CON CLUS ION
The Cerrado seed dispersal network is structured in a slightly nested and considerably modular pattern, conferring some resilience to the plant-frugivore interactions of the biome. Most of the interactions are performed by low body-mass dispersers with opportunistic frugivorous behavior, although large frugivores like the tapir also play a critical role in network maintenance and functionality.
Although we offer an initial description of the Cerrado seed dispersal network, with the type of data available in this study (i.e., incidence records on species interactions), we have considerable restraints to describe a species' functional role or its seed dispersal effectiveness. Future research should gather alike information, which is vital for developing effective conservation policies in the Cerrado.
Our findings on defaunation hint at possible impacts on the Cerrado dispersion network. However, they must be evaluated conservatively, recognizing there are significant methodological limitations (e.g., sampling effort, observation difficulties, number of studies), which limit data about the role of Cerrado large-bodied species in seed dispersal. Further research is urgently needed provided the knowledge shortfall on the subject.
Much remains to be answered regarding the plant-frugivore dynamics in South America's largest savannah, especially at the community and ecosystem levels. Nonetheless, this study represents an initial step toward disentangling such a fundamental topic for Cerrado ecology and conservation.

ACK N OWLED G M ENTS
We would like to thank fellow researchers Daiane Carreira, Guilherme Ferreira, Liana Rossi, Marcelo Kuhlmann, Mauro Galetti, Marco A. Pizo, and Chloe Strëvens, whose advice, suggestions, and sharing experiences were of great value to the fulfillment of this project. We appreciate the constructive comments and insightful recommendations from editors and reviewers, which greatly improved the final version of this paper. We also would like to thank the University of Oxford, School of Geography and the Environment, Oxford Ecosystems Lab, and Oxford St. Peter's College for providing academic, technical, and financial support. This work is a product of R.B.C.'s MSc dissertation in 'Biodiversity, Conservation and Management' (University of Oxford, UK).

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare no affiliations, memberships, funding, or interests that could affect the objectivity of this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in the Dryad Digital Repository: https://doi.org/10.5061/ dryad.47d7w m3k4 (Béllo Carvalho et al., 2023).