Study Site

The study sites were distributed in two habitat types: 1) a protected native dry forest in the Chamela-Cuixmala Biosphere Reserve and 2) a disturbed area including secondary vegetation in different successional stages, human settlements, croplands, pastures, and paved and dirt roads.

The conserved sites were located in the conservation area of the Chamela Biological Station (EBCh; 19° 22′–19°39′N, 104°56′–105°10′W; 35 to 120 m.a.s.l.) with an extension of 13,142 hectares. The average annual precipitation is 745 mm but is variable year to year (366–1320 mm) and mainly concentrated during the rainy season from June to October (Bullock, 1986). The dominant vegetation mainly consists of tropical dry forest (TDF) with trees 6 to 12 m tall on average mainly located on slopes with a dense understory. Most plants lose their leaves for 5 to 8 months during the dry season (Lott et al., 1987). Also, along creeks and valleys, tropical semi-deciduous forest (TSDF) is present, with trees taller than 20 m on average but with some individuals up to 25 or 30 m; in this vegetation cover, some trees keep their leaves all year long (Lott et al., 1987; Rzedowski, 2006). The Chamela-Cuixmala dry forest is considered one of the most diverse (≈ 1,100 plant species) in Mexico and possess a high degree of endemism (Banda et al., 2016; Lott et al., 1987).

The disturbed sites were the rural town of San Mateo, 6 km from the EBCh (19°34´32´´ N–105^°05´06´´ W), and the ejido (a communally owned land) La Fortuna, 9 km from the EBCh (19°59´83´´N–105°12´45´´ W), both in La Huerta municipality in Jalisco state. The first site has a population of 647 habitants and mainly consists of houses and businesses, with some bare plots and original vegetation remnants. The second site mainly consists of croplands with watermelon, pepper, and sorghum; papaya, mango, and coconut orchards; and ruderal vegetation with some native trees and plants. Around houses we also observed ornamental vegetation mostly composed of exotic plants.

Plant-Hummingbird Interaction Records

In 2016 (July–December), 2017 (January–December), and 2018 (January–July), we registered the hummingbird-plant interactions in both study areas. We established 78 circular plots (39 in each area) with a 25-m radius at a distance of 200 m from each other (Hutto et al., 1986; Ralph, 1997). At monthly intervals, point counts were performed in each plot for 10 min to record the number and species of hummingbirds feeding on flowering plants. The censuses began around 7:30 a.m. and ended around 1:30 pm. Two individuals started records at different plots and walked in different directions to avoid order effects (Ralph et al., 1995). Binoculars (10 × 40mm) and a field guide (Arizmendi & Berlanga, 2014) were used to identify hummingbirds. Plant specimens were identified at the EBCh herbarium. The sampling completeness of the plant-hummingbird interactions recorded in both study areas was determined using the Chao2 estimator in EstimateS version 9.1 (Colwell, 2013) following Chacoff et al. (2012).

Interaction Networks

We pooled together the monthly hummingbird-plant interaction data records from each habitat (conserved and disturbed) and built two adjacency matrices, where aij = the number of interactions between an individual plant species (i) and an individual hummingbird species (j), with 0 indicating no interaction (Bascompte et al., 2003). From these matrices, we built two plant-hummingbird interaction bipartite networks using the “Bipartite” package in the R software (R Development Core Team, 2019).

To evaluate how the topological properties of the plant-hummingbird mutualistic networks changed in disturbed habitats, we calculated and compared the following metrics: connectance, network specialization, niche overlap, nestedness, and species contribution to nestedness. We chose these metrics because their results can provide an overview of community organization and be compared with previously published studies (e.g., Aizen et al., 2008; Alarcón et al., 2008; Benítez-Malvido et al., 2014; Dáttilo et al., 2013; Tylianakis et al., 2010; Villa-Galaviz et al., 2012). Also, network-level metrics such as nestedness and connectance can be related with emergent properties such as network stability (Tylianakis et al., 2010).

First, connectance was calculated as the proportion of realized links out of all possible ones (Winemiller, 1989). Then, we obtained network specialization (H₂′) based on the deviation of species’s realized number of interactions from their total expected number of interactions. This parameter ranges from 0 (no specialization) to 1 (perfect specialization). This index is robust to the number of interacting species and to changes in sampling intensity (Blüthgen et al., 2006). Niche overlap for plants and hummingbirds was calculated as the mean similarity in interaction pattern between species of each group according to Horn’s index. Values near 0 indicate no common use of niches, 1 indicates perfect niche overlap (Horn, 1966). We assessed the significance of these network metrics comparing our results to the expectations of the r2dtable null model (N = 1,000), which maintains the matrix sum and row/column sums constant (Dormann et al., 2008; R Development Core Team, 2019).

To determine significant differences in the structure of the hummingbird-plant networks at both sites (i.e., connectance, specialization and niche overlap), we first obtained the difference between the observed metric in the conserved and the disturbed site. Then, we randomly generated 1,000 matrices using the Patefield algorithm (Patefield, 1981). After this, we calculated the chosen metric for each of the randomly generated matrices and obtained the difference between the 1,000 pairs of matrices. Finally, we plotted the distribution of these values to determine which proportion of these values was smaller than the real (observed) differences. If the observed differences were larger than at least 95% of the differences from the random matrices, we considered them to be significant.

To determine changes in the use of available resources, we calculated the hummingbirds’ niche width in both networks according to Shannon’s niche breadth index, which measures the diversity of resources used, in the R software using the spaa package. Then, we calculated the percentage variance between these two values by subtracting the benchmark number (the niche width value at the conserved site) and dividing the result by the benchmark number. Additionally, to evaluate changes in species roles in network structure at both sites, we classified plant and hummingbird species as either core or periphery using the following formula: Gc = (k_i–k_mean)/σ_k, where k_i= mean number of links for a given plant/hummingbird species, k_mean= mean number of links for all plant/hummingbird species in the network, and σ_k, = standard deviation from the number of links for a given plant/hummingbird species. Gc > 1 are species with a larger number of interactions in relation to other species of the same trophic level and are therefore considered to be part of the generalist core. Gc < 1 are species with a lower number of interactions in relation to other species of the same trophic level and are therefore considered to be periphery species (Dáttilo et al., 2013).

We also calculated the degree of nestedness of each network according to the NODF parameter (Almeida-Neto et al., 2008) in Aninhado 3.0 (Guimarães & Guimaraes, 2006). Nestedness describes a pattern in which a core of generalist species interacts with each other, and with extreme specialist species interacting only with generalists. NODF values close to 1 indicate a high degree of nestedness (Bascompte et al., 2003; Guimarães et al., 2007). NODF’s significance was estimated using null model 2 (Bascompte et al., 2003), which generates random matrices from the original one. The probability of interaction between any pair of species is calculated proportionally to the total number of interactions (i.e., their degree of interactions). The P value was obtained for the random matrices that had a NODF value equal to or larger than the original matrix.

Finally, we calculated species contribution to nestedness with the nestedcontribution function in the Bipartite package. A value larger than 0 indicates a positive contribution to nestedness and a lower value indicates a negative contribution. With this parameter, we identified the proportion of idiosyncratic species (i.e., species with negative values, thus with interaction patterns tending away from a perfectly nested pattern) (Atmar & Patterson, 1993) in addition to the contribution of exotic plant species to network nestedness.

Species Turnover and Network Parameters

Mutualistic networks are typically characterized by fewer observed than possible interactions (e.g., Chacoff et al., 2012; Gonzalez & Loiselle, 2016), as observed herein. In the present study, most native plant species were lost in the disturbed network (68%), and the conserved and disturbed sites only shared a few plant species (16%). Most native plant species remaining in the disturbed network were trees, which besides being present in the conserved area also grow outside the Chamela reserve along roads, in spare lots, or around town. Such countryside habitats may preserve elements necessary for the conservation of some species or, at best, sustain a moderate fraction of the native biota (Daily et al., 2001) (e.g., Tabebuia rosea, Caesalpinia eryostachys, and Hesperalbizia occidentalis).

Hummingbird richness was higher in the disturbed area outside the Chamela reserve, possibly as a result of several factors, including the greater density of floral resources compared to forested areas, and hummingbirds’ capacity to explore new available resources provided mainly by the exotic species (Mendonça & Dos Anjos, 2005). Some ornamental plant species, for instance, can bloom all year long (e.g., Bouganvillea glabra and Hibiscus rosa-sinensis), providing constant, readily available resources that can be used, especially when resources inside the reserve become scarce. Even though standard measures such as species richness and abundance were higher in the disturbed area, these metrics tend to mask important compositional shifts associated with agricultural intensification, as forest-dependent species are often replaced by generalists or disturbance-adapted species (DeClerck et al., 2010).

Overall, the hummingbird community of the Chamela region mostly included generalist species with a wide distribution throughout Mexico. Most of these species are found in dry forests but also in open areas, edges (Arizmendi & Berlanga, 2014), and other human-modified landscapes, such as gardens and urban treescapes. Typically, moderate urban development increases ornamental vegetation, water sources, primary productivity, and the amount of edge between habitats (Mooney & Gulmon, 1983; Whitney & Adams, 1980). Because bird abundance and distribution are associated with resource availability, changes in resources influence the presence of individual hummingbird species and the hummingbird community as a whole. Resident hummingbird species such as Cynanthus latirostris and Amazilia rutila are perfectly adapted to and very common in human settlements (del Olmo, 2007; Des Granges, 1979). These species might be considered “urban exploiters” or species adept at exploiting urban habitats and, consequently, may reach their highest densities in these sites (Blair, 1996).

Other hummingbird species such as Archilochus spp. can use a wide variety of habitats along their migratory routes and are commonly linked with disturbed areas (Arizmendi & Berlanga, 2014; Finch, 1991; Udvardy & Farrand, 1994). These “suburban adaptable” species can exploit resources characteristic of moderate levels of urban development, such as ornamental vegetation (Blair, 1996).

On the other hand, H. constantii had a low abundance in both study sites. This species is highly insectivorous, and its presence is usually linked with abundant plant resources (Arizmendi & Ornelas, 1990; Des Granges, 1979), such as Tabebuia rosea. Meanwhile, A. violiceps is a wanderer and altitudinal migrant (Des Granges, 1979; Lopez-Segoviano, 2018) that is very rare in the area (Arizmendi & Ornelas, 1990); however, it may be present in disturbed areas when exotic tree species with broad floral displays such as Spathodea campanulata or Delonix regia are in bloom. Overall, all hummingbird species responded positively to the higher number of plant species in the disturbed sites, widening their niche. The only exception was Chlorostilbon auriceps, which did not seem to benefit from the extra resources in the disturbed sites, as evidenced by its reduced niche width and more common occurrence inside the conserved forest. This “urban avoider” species can be particularly sensitive to human-induced changes in the landscape and, consequently, reaches its highest densities in the most natural sites (Blair, 1996). In Brazil, hummingbird communities in natural areas were categorized according to their varying preferences for open savannah, forest habitats (Maruyama et al., 2014), and even urban areas (Maruyama et al., 2019). Something similar occurs with C. auriceps: This species seems to prefer forest habitats and is less commonly found in open, human-modified landscapes. Accordingly, it is likely more vulnerable to habitat loss, which can highlight the importance of maintaining natural areas such as the Chamela-Cuixmala Biosphere Reserve and other conserved dry forests.

Spatial differences in the number and identity of species forming interaction networks have been previously reported. For example, according to Hagen and Kraemer (2010), the species richness and abundance of plant and bee communities and their mutualistic flower-visitor networks strongly differed between structurally diverse farmland habitats and the neighboring forest understory in Western Kenya. Specifically, the largest networks, diversity, and abundance of bees and plants were found at the forest edge and in farmlands (i.e., disturbed sites). According to Villa-Galaviz et al. (2012), the plant-herbivore network in the Chamela area in Mexico showed high resilience: Differences were only found in recently abandoned fields, as most network descriptors of secondary and mature forest did not differ. Also, no significant nestedness or modularity in the network structure was found. Plant-herbivore network properties appear to quickly reestablish after disturbance despite changes to species richness and composition. Dáttilo et al. (2013) showed that, although the ant and plant composition of networks changed spatially following disturbance, the central core of generalist species and network structure remained unaltered over a geographic distance up to 5,099 m in the southern Brazilian Amazonia. Meanwhile, Benítez-Malvido et al. (2014) assessed the topological structure of individual Heliconia–invertebrate networks in Chiapas, Mexico, and found that H. collinsiana richness was greater in riparian vegetation and that no differences existed in the diversity of the invertebrates associated with particular Heliconia species and habitats. Invertebrate abundance was, however, greater on H. latispatha in gaps and on H. collinsiana in riparian vegetation, indicating species turnover in human-disturbed habitats. Therefore, Heliconia-invertebrate network properties appear to be maintained in human-disturbed habitats despite differences in species richness, abundance, and composition and host number and quality. Finally, Tinoco et al. (2017) analyzed the similarity of hummingbird species in three Andean valleys in Ecuador and found, on average, a hummingbird species turnover of 33% between networks.

Integration of Exotic Plant Species Into the Network

In the disturbed network, around 40% of plant species were exotic (22 spp.), a smaller percentage in comparison to other hummingbird studies in human-modified landscapes. For example, Mendonça and Dos Anjos (2005) reported that 60% of species were exotic (22 spp.) in the University of Parana, Brazil, including some species in common with the present study such as Hibiscus rosa-sinensis and Spathodea campanulata. Marcon (2016) also reported 61% of plant species as exotic in another disturbed habitat in the south of Brazil. However, most studies on hummingbird-plant interactions in disturbed landscapes do not specify whether the plant species visited by hummingbirds are natives or exotics. In one review, Maruyama et al. (2016) found that only around 8% of plants in hummingbird networks were exotic, but this review mostly included studies in conserved areas where exotic plants are not common. This highlights the need to continue to study hummingbird-plant networks in disturbed areas and to explicitly identify whether plants are native or exotic. A global data set study that explored the integration of exotic plants into plant-insect pollination networks showed that these networks are characterized by greater total, plant and pollinator richness, as well as higher values of relative nestedness (Stouffer et al., 2014).

Some important changes in the core plant species were also noted. The number of core plant species in the disturbed site was half of the conserved network even though there were more species at the disturbed site. The plant species lost from the generalist core seemed to be forest-dependent, and they were core generalist because of their varied interactions with hummingbirds but not necessarily because of their ability to grow everywhere. Also, when some exotic species become increasingly abundant and/or increase their interaction strength, they increase their chance of interacting with a large number of partners sequestering interaction frequency and links from the original network (Aizen et al., 2008). Thus, this augmented strength of invasive plants may result not only from its high density, but also from the tendency of these species to produce exuberant flower displays and offer superabundant flower resources (Chittka & Schürkens, 2001). Therefore, these exotic plants become network hubs (super generalist species) occupying a central role in the community (Bartomeus et al., 2008), and making it more difficult for other plant species to become core species.

Also, whereas inside the forest only two of the core species were trees (20%), in the disturbed area, five out of six core species (83%) were trees. This agrees with prior hummingbird studies in natural areas where pollinators are rarely associated with trees and more commonly associated with herbs and shrubs (Arizmendi & Ornelas, 1990; Buzato et al., 2000; Stiles, 1978). Even though trees were less important in the conserved forest, the only core plant shared between the sites was the tree Tabebuia rosea. Therefore, this tree species contributes toward conserving pollinators in both locations and should be considered in reforestation programs, as part of live fences/corridors in agricultural areas, or as an ornamental/shade tree in rural or urban zones.

In the conserved sites, most exotic plant species were trees (11 spp.). Of these, Spathodea campanulata, Delonix regia, and Cordia sebestena were also among the five most visited plant species in the disturbed sites, garnering almost 30% of hummingbird visits and the last one being the most visited of all plants. In Brazil, trees were found to be more important than other growth forms for the maintenance of hummingbirds in urban settings because they can provide high amounts of nectar (Maruyama et al., 2019). However, besides attracting hummingbirds (de Andrade et al., 2007; Martínez, 2008; Mendonça & Dos Anjos, 2005; Percival, 1974; Rangaiah et al., 2004) it has been reported in other countries that these tree species attract several other flower visitors such as non hovering birds (Banks, 1997; Dalsgaard et al., 2016; Du Puy et al., 1995; Faegri & Van Der Pijl, 2013; Gentry, 1974; Maruyama et al., 2019), bees (Fohouo et al., 2011), butterflies (Cruden & Jensen, 1979; Percival, 1974), and even mammals such as bats (Ayensu, 1974) and lemurs (Sussman & Raven, 1978) in the case of S. campanulata. Thus, Old World plants with adaptations to bird pollination, such as S. campanulata and D. regia, can more easily integrate into plant-hummingbird interaction networks in the Americas, or at least more so than alien plant species without bird pollination (Maruyama et al., 2016). Thus, some shared traits associated with hummingbird pollination such as odorless flowers with bright colors, copious and diluted amounts of nectar, and exerted anthers, are favored (Banks, 1997; Gentry, 1974) (e.g. Caribbean native tree C. sebestena (Askins et al., 1987; Johnston, 1949).

In addition to the exotic species Spathodea campanulata, Delonix regia, and Cordia sebestena representing one-half of the core species in the disturbed network, overall, exotic plant species in this network received about one-third of all hummingbird visits, making them important resources. Maruyama (2016) stated than an average alien plant is more important for hummingbirds than an average native plant in terms of relative interaction frequency. There is also a tendency for alien plants to have more partners and for some hummingbird species to interact more exclusively with these plants than natives. In the present study, exotic plants played a key role in the disturbed area and strongly integrated into the plant-hummingbird network. Nestedness was only significant outside the reserve, where the exotic plants’ contribution to nestedness was more than double. According to Maruyama et al. (2016), exotic plants have more partners, and hummingbirds show greater dependency on them than on average native plant species (e.g., A. rutila with C. sebestena and D. regia).Thus, exotic plants are important and act as core generalists in disturbed networks (Aizen et al., 2008; Bartomeus et al., 2008; Stouffer et al., 2014; Vila et al., 2009).

Maruyama et al. (2016) also suggest that other plant traits, such as a different flowering phenology with respect to native plants or higher nectar secretion rates, could be important factors that explain the integration of exotic species in these networks (see Chittka & Schürkens, 2001; Godoy et al., 2009). By acting as core generalist species in these networks, exotic plants may impact the entire plant-pollinator network (Traveset et al., 2013) and even modify its eco-evolutionary dynamics (Guimarães et al., 2011). Although exotic plants represent important food resources for hummingbirds in disturbed areas, they could also have negative effects. They may usurp pollinators and reduce visitation to nearby native plants (Bartomeus et al., 2008; Cunningham-Minnick et al., 2020). Thus, a different experimental approach will be helpful to elucidate if hummingbirds (or other pollinators) do prefer exotics over native plant species. Likewise, disturbed habitats may act as ecological traps for hummingbirds if, for example, they prefer to nest at sites where egg and/or nestling survivorship shrinks due to higher exposure to predators (Battin, 2004). Alternatively, floral displays of exotic plants may facilitate or increase the pollination success in adjacent plants by attracting more pollinators to the area (Bartomeus et al., 2008; Cunningham-Minnick et al., 2020). Given this, patches of disturbed habitat may enhance landscape heterogeneity, providing complementary resources to the native remnants (Fonturbel et al., 2017) or provide complementary resources when they are scarce at the conserved areas (Bartomeus et al., 2008).

As a consequence of land-use changes, an increasing number of native species are being “forced” to inhabit human-modified landscapes composed of a mosaic of natural and anthropic land covers (Galán-Acedo et al., 2019). Unfortunately, for many of these species, studies have mostly focused on their ecology within their primary habitat, especially in protected areas, thus limiting our understanding of their use of and tolerance to human-modified landscapes (Galán-Acedo et al., 2019). Therefore, the continued study of these species and their interactions in both natural and modified landscapes is important.