Neotropical cloud forests (CFs) have the highest biodiversity per surface unit of all terrestrial ecosystems (Dirzo and Raven, 2003), and promote the development of multi-trophic relationships at different levels (Vizentin-Bugoni et al., 2018). However, they are one of the most threatened ecosystems worldwide (Hamilton et al., 1995; Scatena et al., 2010). Specifically, in the neotropics more than 41% of CFs have been converted into secondary forests, agricultural land, or pastures (Martínez et al., 2009; Gibbs et al., 2010; Mendoza-Ponce et al., 2018). These anthropogenic changes negatively affect the composition, structure, and ecological functions of biotic communities (Şekercioğlu et al., 2004; García et al., 2011; Sandor et al., 2022). Despite the incessant increase in agricultural and livestock-related activities in converted CF areas, some crops decrease susceptibility of ecological functions (Harvey et al., 2021). For example, rustic shade coffee plantations maintain the essential ecosystem services provided by biodiversity through multiple interactions between plants and animals (Martínez-Salinas et al., 2022).

Coffee cultivation represents one of the most productive activities in the humid tropics of the Americas, with around 5.2 million hectares cultivated (FAO, 2016), which are often located within forest areas in peasant communities and generally support great biodiversity (Moguel and Toledo, 1999). A large extension of these plantations tends to grow under the canopy of humid mountain forests and have been considered as “friendly” to biodiversity conservation and a great provider of ecosystem services (Williams-Guillen and Perfecto, 2011; Alvarez-Alvarez et al., 2021; Harvey et al., 2021).

Several studies have shown that in these plantations, different bird groups find refuge under various disturbance scenarios and spatial scales (Greenberg et al., 1997; Smith et al., 2015). Other studies have shown that these plantations maintain the structure and composition of the original tree vegetation and provide bird groups with the resources they need to survive (e.g., Tejeda-Cruz and Sutherland, 2004). In addition, it has been demonstrated that some bird groups provide essential services by controlling pests, such as the coffee borer (Hypothenemus hampei), which causes extensive damage to coffee plantations worldwide (Damon, 2000). To a lesser extent, birds can also contribute to the pollination of this crop (Whelan et al., 2015).

In Latin America, many shade coffee plantations have been abandoned due to falling coffee prices (Canet and Soto, 2016; Harvey et al., 2021) or replaced by activities that intensify land use, such as cattle ranching (Alvarez-Alvarez et al., 2022). At abandoned coffee plantations, shrubby vegetation and trees typically undergo regeneration and/or succession processes, which makes the vegetation more complex (Alvarez-Alvarez et al., 2021). Both in highly disturbed areas, such as cattle ranching, or in current or abandoned coffee plantations, changes in landscape composition promote changes in plant and bird species composition, disrupting their interactions (García et al., 2011; Morrison and Mendenhall, 2020). However, studies of bird-plant interactions in sites with different agroforestry management regimes have not been addressed in tropical landscapes. In particular, the interactions of bird groups that perform important ecological functions, such as pollination, have not been sufficiently studied.

Mutualistic interactions between pollinators and plants are among the most important interactions in natural systems, as approximately 70–90% of flowering plants are pollinated by animals (Bascompte and Jordano, 2007; Ollerton et al., 2011). In the Americas, hummingbirds are the most important birds that carry out this ecological function. Buzato et al. (2000) estimated that hummingbirds pollinate 10–15% of angiosperms throughout the neotropics. Moreover, both hummingbirds and plants have diversified due to their interactions and the existence of high levels of reciprocal specialization (Serrano-Serrano et al., 2017).

Pollinator visitation is determined by different traits of both plants and pollinators. In this context, bill size and morphology are one of the most important traits of hummingbirds (Maglianesi et al., 2014; Izquierdo-Palma et al., 2021), while nectar characteristics and the number, size, shape, and color of flowers are important plant-based factors (Fornoff et al., 2017). Meanwhile, other variables that have been recognized as good predictors of bird diversity, such as foliage height diversity (MacArthur and MacArthur, 1961), have rarely been considered in studies of interactions between plants and their floral visitors (e.g., Klecka et al., 2018). The foliage height diversity supposes that greater vertical stratification provides avian communities with resources and microhabitats that serve as feeding, refuge, and reproduction sites (Chmel et al., 2016; Almazán-Núñez et al., 2021). Knowledge of interaction networks and the variables that drive these mutualisms may be relevant to a better understanding of tropical agroforestry ecosystems.

Studies of heavily modified landscapes have found that some hummingbird species are resistant to disturbance (Sonne et al., 2016; Bustamante-Castillo et al., 2018; Maruyama et al., 2019), and may establish new interactions with plants, particularly invasive ones (Aizen et al., 2008), resulting in less functionally diverse communities and more generalist networks (Maruyama et al., 2019). Studies on interaction networks in modified primary forests have revealed the flexibility and vulnerability of mutualisms in the face of species loss (Bascompte et al., 2003, 2006; Morrison and Mendenhall, 2020). Thus, ecological networks serve as a valuable tool within applied ecology because they can be used to monitor the impact of anthropogenic disturbances on ecosystems and evaluate the efficiency of restoration programs (Kaiser-Bunbury and Blüthgen, 2015).

In this study, we describe the composition of hummingbird-plant interaction networks in three types of land use: (1) forests in advanced succession (hereafter, late forest), (2) shade coffee plantations (hereafter, coffee plantations), and (3) cattle pastures, that had been derived from CF in southern Mexico. We also test whether there is an association between plant structural traits and the number of hummingbird visits to flowers in relation to types of land use. Based on these aims, we propose the following predictions:

• 1.

  Hummingbird-plant interactions will be more intense and connected in late forests and coffee plantations compared to cattle pastures, due to their diverse floristic composition and complex vegetation structure.

• 2.

  Cattle pasture networks will have greater nestedness and generalization than late forests because habitat specialists tend to be affected by disturbed environments (Beal-Neves et al., 2020).

• 3.

  Cattle pastures will display a more robust network, favored by the presence of invasive plant species which are more likely to create new connections with generalist hummingbirds (Aizen et al., 2008; Díaz-Infante et al., 2020).

• 4.

  Interactions along the disturbance gradient will be influenced by the foliage height diversity because stratification enables the coexistence of bird species that exploit similar resources (Chmel et al., 2016).

The study findings have important implications for the conservation of mutualistic pollination interactions in CFs ecosystems, which are one of the most threatened ecosystems worldwide.

Our results showed how the composition of hummingbird and plant species is affected as land use change intensifies in a predominantly agroforestry landscape and the direct effect of such change on the structure and configuration of interaction networks. According to our first prediction, the coffee plantations and late forests had more plant and hummingbird species and interactions that the cattle pastures. Given that shade coffee plantations generally maintain tree cover (Valencia et al., 2014), which is a stratum that does not usually differ from that found in late forests (Alvarez-Alvarez et al., 2021), these agroforestry systems allow some pollinators and canopy species to increase their presence in these plantations (Vergara and Badano, 2009). Additionally, coffee plantations are usually combined with banana plantations (M. paradisiaca), which are frequently visited by some hummingbird species (Itino et al., 1991). Conversely, in cattle pastures, sparse vegetation cover is the main factor in the decreased food supplies of some pollinators; consequently, the number of floral visitors and their interactions are significantly affected, as has been reported in other studies (e.g., Bustamante-Castillo et al., 2018; Maruyama et al., 2019).

The networks of the three types of land use had a heterogeneous structure, which was adjusted to a power law distribution. This indicate that most species had a few links, while few species had many more interactions than expected based on random networks with similar properties (Jordano et al., 2003; Gilarranz et al., 2011). This could explain why our networks did not have high nestedness values, ​​compared to other mutualistic networks in disturbed tropical environments (e.g., Bustamante-Castillo et al., 2018; Maruyama et al., 2019), giving it a compartmentalized structure.

In contrast to the quantitative matrices, the network based on presence or absence of coffee plantations followed a nested pattern in which the interaction was organized by a core group of species: S. beryllina, E. poliocerca (two Emerald species notably generalists, Rodríguez-Flores et al., 2019) and C. fragans and M. paradisiaca (specialist and generalist nectar resource, respectively; Bascompte et al., 2003). Once the number of interactions was included, the nested pattern disappeared disclosing the modular organization and denoting the role of these nodes in maintaining the cohesive pattern in this network.

Another explanation for the low nestedness found in land use networks is related to the small size of each network, which has been observed in other studies (i.e., Rivera-Hutinel et al., 2012; Vizentin-Bugoni et al., 2016). Recently, Arizmendi et al. (2021) analyzing the hummingbird-plant interaction network in the same study area, recorded interacting species that were not observed in our study (e.g., Selasphorus rufus, S. platycercus, and S. calliope hummingbird species and Vernonia cordata, Guarea glabra, Ixora coccinea and Dombeya wallichi plant species). They also recorded interactions between network partners (e.g., E. poliocerca and D. wallichi) that we did not observe in our sampling (i.e., missing links; Olesen et al., 2011). Meanwhile, we recorded new interactions in the study area (e.g., between L. brachylophus and C. fragans and between P. mexicanus and Hibiscus uncinellus); only 14 interaction pairs were consistent in both studies. The discrepancies in these interaction records could be attributed to the 10-year time difference between the study periods, plant phenology, and the migratory movements of the species involved. Even when in an interannual analysis, Chávez-González et al. (2020) did not find differences in the network’s connection pattern, they find differences in species composition (i.e., high interaction turnover), suggesting that the presence of hummingbirds and flowers remained highly dynamic over time. In addition, land use changes that occurred between our study and that of Arizmendi et al. (2021) could also determine the interaction dynamics (Dalsgaard, 2020).

Contrary to our prediction, the cattle pasture network was not nested. Similar studies observed higher nestedness in the networks of disturbed areas, compared to primary forests (e.g., Bustamante-Castillo et al., 2018; Maruyama et al., 2019; Díaz-Infante et al., 2020). Generally, the most disturbed sites negatively affect functionally specialized hummingbirds, especially those with long bills, such as hermits and fandangueros (Maglianesi et al., 2015). At such sites, hummingbird species are functionally less diverse, and interaction networks are more nested. Nevertheless, some recent studies have shown that some hummingbird-plant networks are heterogeneous and have low nestedness (Vizentin-Bugoni et al., 2014, 2016; Chávez-González et al., 2020).

In the present study, the modularity values ​​between the types of land use indicated that the late forest and coffee plantation networks were more modular than the cattle pasture network. These results are in line with that obtained by Maruyama et al. (2019) in an urbanization gradient, which is explained by the functional trait diversity present in a biotic community. Studies have shown that, in disturbed environments, the number of hummingbirds with functionally specialized traits decreased (Hardley et al., 2018), resulting in a more generalized interaction networks (Maglianesi et al., 2015; Maruyama et al., 2018). In the present study, in the late forest and shade coffee plantation networks, species interacted more strongly with each other than with other species in some modules. For example, some modules were made up of large, long-billed hummingbirds (i.e., C. hemileucurus, H. constantii, H. longirostris, and P. mexicanus) that frequently visited flowers with long corollas (i.e., C. pictus, H. coronarium, M. paradisiaca, and Heliconia spp.), which were rarely visited by short-billed hummingbird species (i.e., L. brachylophus and T. dupontii). Species that present a morphological mismatch have limitations or prohibited links due to the inability of some hummingbirds to reach the nectar of the flowers (Temeles et al., 2002; Vizentin-Bugoni et al., 2014).

The information on the roles of the species showed that S. beryllina and E. poliocerca were nodes that maintained network cohesion across the different types of land use; as they were highly connected species, they formed generalist nodes. Both species are usually abundant in humid environments throughout their geographic range (Arizmendi and Berlanga, 2014) and tend to present generalist behavior (Rodríguez-Flores and Arizmendi, 2016). In our study, both species were observed actively defending their floral patches or feeding sites and engaging in aggressive behavior, potentially limiting the interaction frequencies of other hummingbird species (e.g., L. brachylophus, T. dupontii) by scaring them away. This behavior pattern has been commonly observed in both hummingbird species, especially in S. beryllina (Rodríguez-Flores and Arizmendi, 2016; López-Segoviano et al., 2018).

Our networks presented intermediate specialization values, a result that contrasts with that obtained by Morrison and Mendenhall (2020), who identified more specialized interaction networks in primary forests than in coffee plantations. However, the plantations they studied seemed to have a significant reduction in tree cover, while the shade coffee plantations in our study maintained an important tree cover, similar to that observed in mature forest (Alvarez-Alvarez et al., 2021). Moreover, the late forests had the highest number of specialist hummingbird species, such as Cynanthus latirostris, P. mexicanus, and L. brachylophus. The latter two species, which are endemic to a small portion of western Mexico (Arizmendi and Berlanga, 2014), played a specialist role in the coffee plantations. This finding highlights the relevant role that these agroforestry systems play in providing resources for the conservation of species with a restricted range and sensitivity to disturbance (Sonne et al., 2016). On the other hand, in the cattle pastures, we mostly observed generalist hummingbirds, such as A. rutila and Archilochus colubris, which are usually tolerant to disturbance (Maruyama et al., 2019). Regarding plants, J. adenothyrsa, C. pictus, and C. fragans were specialists in their relationships with the hummingbirds in the late forests. These plant species were present also in coffee plantations, which had remnant species of the original forest as well as cultivated species (Valencia et al., 2014; Alvarez-Alvarez et al., 2021).

Our networks seem to be fragile in the face of random species loss and the removal of generalist species (Memmott et al., 2004). However, when plants and hummingbirds were eliminated based on threat degree, the networks were more robust. The robustness index measures changes in network structure following a specific extinction scenario but does not consider the rewiring process between nodes; this process improves robustness against node removal (Vizentin-Bugoni et al., 2019). Despite this, our simulations showed that, when the most vulnerable species in the system based on its risk category and endemism level (i.e., L. brachylophus) was eliminated, the network remained robust. This finding could be explained by the low degree of this hummingbird species and their low contribution to network cohesion. However, it does not seem to be consistent with other microendemic hummingbirds, such as Eriocnemis mirabilis, in CFs in Colombia (Ramírez-Burbano et al., 2017), where this species was found to be the most connected in the network.

The cattle pasture network had greater robustness than the late forest and coffee plantation networks because, in cattle pastures, there were more generalist species that are less selective in their ecological links and are more resistant to disturbance (Santos de Araújo, 2018). Thus, our findings suggest that the transformation of primary CFs into intensively managed land uses leads to communities that are functionally less diverse and more generalized interaction networks, which may have contrasting effects on the conservation and robustness of plant-pollinator interactions.

Our results also confirmed the hypothesis that foliage height diversity is the best predictor of the number of hummingbird visits to plants, even greater than the number of flowers, which is consistent with other studies (e.g., Vizentin-Bugoni et al., 2014). In the study area, the late forests and coffee plantations had the most foliage height diversity, hummingbird richness and frequency of tree visits. The coffee plantations had the greatest combination of mature forest trees (i.e., I. vera, S. grandis), shrubs (i.e., Syzigium jambos), and introduced species of commercial interest (C. arabica, M. paradisiaca, and Citrus limonia) that are frequently visited by some hummingbird species (e.g., P. mexicanus, C. hemileucurus). Plant heterogeneity in coffee plantations promotes greater diversity of foliar strata, which not only increases the number of flowers, but also provides perching, shelter, and nesting for different bird species (Chmel et al., 2016), including hummingbirds. Although the role played by foliar strata in plant-pollinator interactions has been insufficiently studied, Klecka et al. (2018), as well, found this variable to be a good indicator of the frequency of visits by different pollinators.

In Amazonian forests, where vertical stratification is remarkable, the hummingbird community is spatially separated; hummingbirds of the Phaethornithinae subfamily are most abundant in the understory, while those in the Trochilinae subfamily dominate canopy interactions (Cotton, 2008). In the present study, there were more flowers in the cattle pastures than in the late forests due to the abundance of pioneer shrubs and herbs with a high density of flowers. However, the number of flowers at these sites did not determine the number of hummingbirds visits to plants, possibly due to the low food rewards offered by these plants. Attributes such as nectar quality may play an important role in hummingbirds’ plant species preferences (Kim et al., 2011).

Our data suggested that, although hummingbird-plant assemblages can persist in managed environments such as shade coffee plantations, the complete conversion of natural habitats to heavily anthropized landscapes can cause drastic changes in biotic interaction patterns. This undoubtedly have implications for the conservation of mutualistic pollination interactions. Shade coffee plantations and late succession forests, however, present hummingbirds with the highest specialization in the use of floral resources, but little tolerance to disturbance. This same specialization results in the networks of both land uses being more modular compared to cattle pastures. This denotes the relevance of certain agroforestry systems such as shade coffee plantations that, by maintaining floristic elements of the original vegetation, promote the permanence of important functional groups in ecosystem dynamics, as was shown with some hummingbirds in this study.

The authors declare no conflict of interest