Abstract
The alarming rate of global pollinator decline has made habitat restoration for pollinators a conservation priority. At the same time, empirical and theoretical studies on plant-pollinator networks have demonstrated that plant species are not equally important for pollinator community persistence and restoration. However, the scarcity of comprehensive datasets on plant-pollinator networks in tropical ecosystems constrains their practical value for pollinator restoration. As closely-related species often share traits that determine ecological interactions, phylogenetic relationships could inform restoration programs in data-scarce regions. Here, we use quantitative bee-plant networks from Brazilian ecosystems to test if priority plant species for different restoration criteria (bee species richness and visitation rates) can be identified using interaction networks; if phylogenetic relationships alone can guide plant species selection; and how restoration criteria influence restored network properties and function. We found plant species that maximised the benefits of habitat restoration for bees (i.e., generalists and those with distinct flower-visitor species) were clustered in a small number of phylogenetically-diverse plant families, and that prioritising the recovery of bee visitation rates improved both stability and function of restored plant-pollinator networks. Our approach can help guide restoration of pollinator communities, even where information on local ecosystems is limited.
Similar content being viewed by others
Introduction
Growing concern over the negative environmental impacts of human activities has put ecological restoration at the forefront of the fight to reverse biodiversity loss and to recover ecosystem functions and services1. Historically, restoration programs have focused on structural indicators, such as the recovery of plant communities relative to a reference ‘pristine’ state (e.g., pre-European settlement)2, with the assumption that fauna and ecosystem processes (e.g., pollination, seed dispersal, nutrient cycling) recover in tandem3. However, more recent evidence has demonstrated that selecting plant species to restore biotic interactions, rather than habitats, is critical to maintain ecosystem functions and health4,5,6,7.
Animal-mediated pollination is an essential ecological function in most terrestrial ecosystems8. Among pollinator taxa, bees are considered to be the most important group in both natural and agricultural ecosystems9,10. Unfortunately, many species are sensitive to anthropogenic disturbance, particularly the loss and fragmentation of natural habitats, and are globally in decline10,11. Outside of agricultural systems, for which there is considerable evidence on the benefits of habitat restoration for wild bees and crop pollination services (e.g. wildflower strips, hedgerows)12,13,14, bees are rarely considered as an explicit restoration target in natural lands3,15. Given that one of the key aims put forward by the International Union for Conservation of Nature’s (IUCN) Bonn Challenge (www.bonnchallenge.org) is to restore 350 million hectares of degraded land by 2030, a failure to target bees could lead to reproductive failure and even collapse of plant communities in restored habitats15.
To investigate the impacts of habitat restoration on bees, many studies adopt network-based approaches, as they provide quantitative and comprehensive information on how environmental changes spread through ecological communities and influence their stability and function4,16. Studies on plant-bee networks (and more broadly “plant-pollinator networks”) reveal that network generalists, species with many interaction partners, form the core of these networks, and are essential for their dynamic and structural stability17,18. These findings have important implications for ecological restoration, and several network-based measures of species-level generalisation, alongside modularity (i.e. ‘hubs’ and ‘connectors’), and ecological distinctiveness indices (i.e. functional complementarity), have been put forward to select plant species for bees18,19,20,21,22,23. In recent years, conservation scientists have also adopted machine-learning techniques (e.g. genetic algorithms) to inform plant species selection in regional bee restoration programs24. However, together with uncertainty over which approach provides the best outcome for bees and pollination, a major drawback of network-based restoration is the need for accurate and meaningful data on interspecific interactions7. At present, plant-pollinator networks in countries in tropical regions, such as Brazil, remain limited in number, and overrepresented by vertebrate pollinators (e.g. hummingbirds, bats)25,26. This data-deficiency is compounded by accelerating impacts of human activities in the tropics relative to other global regions27,28.
In data-scarce regions, restoration managers could use phylogenetic relationships among plant species as proxies of ecological function29. Phenotypic complementarity between interaction partners (e.g. long-tonged bees visit deep corolla flowers) has been found to be an important predictor of ecological network structure and function (e.g. pollination)30,31,32, although much of the existing data comes from temperate rather than tropical ecosystems. Because closely-related species tend to have similar functional traits, due to shared ancestry33,34,35, they often occupy similar positions in networks, both in terms of number of interaction partners and their partner species’ functional attributes34,36. This trait conservation within evolutionary lineages has great potential for use in bee restoration37. For example, if priority plant species (i.e. those that support high diversity/abundance of bees) occupy defined positions within the phylogeny, land managers could adapt findings from existing networks to habitats with slightly different species compositions38. Such an approach could facilitate the recovery of bee communities in degraded habitats at large spatial scales and across heterogeneous environments.
Here, we use bee-plant networks from across Brazil to test if priority plant species for different restoration criteria (bee species richness and bee visitation rates) can be identified using species-level metrics (for details on chosen network metrics, see Table 1). We compare bee community recovery (i.e. restoration success) under different metrics to outcomes using random plant species selection and machine-learning techniques (genetic algorithms), for quantitative networks sampled in two distinct physiognomies (forest or savannah-like biomes). We then use null models to test if phylogenetic relationships can identify priority plant species in regions for which no information on interaction patterns is available. To provide practical information for bee restoration managers, we additionally investigate whether these plant species are clustered within specific families. Finally, we evaluate if specific restoration criteria influence the structural properties of restored bee-plant networks, and the potential of restored bee species to pollinate wider plant communities using a proxy measure of pollination function (shared flower visitor index).
Results
Selecting priority plant species for bee restoration using interaction networks
The number of plant species required to meet the bee species richness and bee visitation rate targets (80% of species/visits recovered) greatly depended on plant selection strategy (i.e. network metrics, genetic algorithms, or random selection), biome type (forested or savannah-like biomes), and network size (Table 2; for weighted model equations, see Table S1, Supplementary Information). As expected, network-based strategies outperformed random selection, requiring at least 50% fewer plant species (Fig. 1). However, only when bee species richness was targeted did a network metric (interaction strength) match the performance of machine-learning methods (genetic algorithms), requiring on average only 11% of plant species per network (Fig. 1a). For the recovery of bee visitation rates, no metric-based strategies performed better than the genetic algorithms, that required on average 20% fewer plant species than the best metric-based strategy – selection by functional complementarity (Fig. 1b). For both restoration criteria, a lower proportion of total plant species per network was required to meet restoration targets in savannah-like biomes compared with networks from forested biomes (Table S1). Results from networks excluding non-native honeybees (Apis mellifera) were qualitatively similar for both restoration criteria (Table S2).
Phylogenetic relationships among priority plant species
Results from two measures of phylogenetic structure (Net Relatedness Index – NRI, and Nearest Taxon Index - NTI) revealed that, across the global species pool (Table 3), but not in individual networks (Table S3, Supplementary Information), plant species selected for restoration criteria were phylogenetically-clustered compared with random samples of equivalent size (null models). Results from full networks and those excluding honeybees showed that phylogenetic clustering as measured by the NTI was more often significant (3 out of 4 cases) than the NRI (1 out 4 cases) (Table 3). This implies that priority plant species occurred in several clusters throughout the phylogeny, rather than in closely-related families (Fig. 2; Fig. S1 – without honeybees), and that for individual networks, a small number of species per cluster would be sufficient for achieving restoration targets. Finally, phylogenetic turnover between biomes did not differ, indicating that compositional differences between forest and savannah-like biomes occurred mainly among lower taxonomic levels (i.e. genus/species) (Fig. S2).
Key plant families
Among common plant families (member species present in networks on ten or more occasions), Malpighiaceae was selected more often than expected by chance to recover bee richness (Fig. 2; full family list, Table S4, Supplementary Information). Nineteen families were selected less often than expected by chance, of which four were considered common in networks: Euphorbiaceae, Polygonaceae, Boraginaceae, and Anacardiaceae (Fig. 2; Table S4). As for bee visitation rates, one family, Lythraceae, was selected more often than expected by chance, and ten families, two being common (Apocynaceae and Asteraceae), were selected significantly less often than expected by chance (Fig. 2; Table S5). Results from networks excluding honeybees were qualitatively similar for the bee richness but differed when restoring bee visitation rates was the priority; only the Malpighiaceae (among common families) was selected significantly more often than expected by chance (Fig. S1, Tables S6 and S7).
Evaluation of recovered network properties and pollination function
The use of different restoration criteria (bee species richness and bee visitation rates) resulted in restored networks differing in both the number of bee species and bee flower visits recovered (bee species - Likelihood Ratio Test (LRT) = 69.77, d.f. = 1, P < 0.001; bee visitation rates - LRT = 3402.80, d.f. = 1, P < 0.001), with restoration simulations that prioritised bee species richness outperforming those that prioritised visitation rates for the recovery of bee species richness, and vice versa for bee visitation rates (Fig. 3). The number of links recovered, interaction diversity (Shannon), and species-level (bees) specialisation indices in restored networks did not differ between criteria (Fig. 3). However, we found marginal and significant differences respectively, in network-level specialisation and the proxy measure of pollination function (potential contribution of restored habitats to wider plant communities via shared flower visitors), between networks constructed under different restoration criteria (H2′: LRT = 3.12, d.f. = 1, P = 0.077; pollination function: LRT = 52.35, d.f. = 1, P < 0.001). Network-level specialisation was reduced (more generalised), and the proxy measure of pollination function elevated in restoration simulations that prioritised the recovery of bee visitation rates versus simulations focused on bee species richness (Fig. 3).
Discussion
Our results show that previous information on plant-bee interactions can be used to optimise restoration programs by the selection of plant species that maximise benefits to bee communities, and that priority species are clustered in a small number of phylogenetically-diverse plant families. We also found that prioritising the recovery of bee visitation rates over bee species richness, may be the most effective means of restoring ecosystem function in degraded Brazilian landscapes. Given the widespread impacts of human activities on tropical biodiversity, our results can help restoration practitioners to recover pollinator communities and ecosystem function in heterogeneous landscapes such as those found in Brazil, even where little is known beyond inventories of local plant communities.
Independent of which component of bee communities was targeted (species richness, bee visitation rates), the genetic algorithm tool developed by M’Gonigle et al.24 was consistently the most effective means of selecting plant species. Only in the case of bee species richness did a network metric (interaction strength) come close to matching results provided by the algorithms, reinforcing that generalist plant species, which are visited by disproportionately high numbers of bee species relative to other plant species, are essential for restoring bee diversity17,39. Despite the primacy of the algorithms, network indices may provide a better indication of the actual ecological drivers underpinning bee-plant network structure, and bee community recovery. For example, plant species selected by functional complementarity (FC)22, the best-performing metric for bee visitation rates, may reflect important functional differences between plant species (e.g. floral traits, resource type, phenology). As a consequence, this metric probably recovered discrete subsets of closely-interacting species (modules) faster than species-level generalisation metrics (e.g. degree, strength, centrality), that most likely selected plants with a high degree of overlap in their floral traits (e.g. short, open corollae) and visitor communities. However, plant species with high FC scores (i.e. distinct flower-visitor communities) could also include rare species that contributed little to overall bee visitation rates. In contrast, algorithms exclusively-selected plant species based on sum totals of visits recovered, a factor likely to be strongly influenced by an individual plant species’ relative abundance in sampled habitats40. This highlights the importance of standardising species’ abundances (e.g. visits/flowers) to disentangle the relative importance of functional traits and local abundance in structuring plant-pollinator networks, something that many existing plant-pollinator networks (but with some important exceptions4,41,42), do not account for.
An important caveat to our results is that aggregated topological models (>1 yr of repeat surveys) used to simulate the bee community recovery did not account for inherent temporal dynamics of plant-pollinator networks (i.e. temporal overlap in bloom/insect activity periods), nor the plastic behavioural responses of bees to the gain or loss of plant species (i.e. ‘rewiring’)39. Thus, it is unclear to what extent these aggregate networks can be used to predict flower-visitor communities in restored (i.e. partial) habitats. However, our aim was to identify plant species that provide the greatest benefit to bee communities (i.e. maximum number of species/visits), including potential rewiring species (e.g. bee species A only visits plant species B in the absence of plant species C), which can be largely determined using aggregated networks.
Across networks, phylogenetic clustering of priority species was strongest under the nearest taxon index (NTI), indicating that priority species were found in several clusters (i.e. families), rather than one large cluster of closely-related taxa, which would cause the net relatedness index (NRI) to indicate stronger clustering as well43. This demonstrates firstly, that phylogenetic relationships can be used to detect priority plant species for bee restoration, and secondly, by increasing phylogenetic diversity among selected species, we can likely reduce the number of plant species required (and hence costs) in restoration programs. This was supported by the lack of phylogenetic clustering among priority species in individual networks, which given that functional traits are expected to be conserved among closely-related species34,35, would lead to reduced levels of functional complementarity among selected plant species, a key driver of bee diversity in natural and agricultural habitats44. Furthermore, elevated competition among closely-related species may preclude co-occurrence at individual sites45, although this will depend on the relative importance of abiotic (i.e., environmental filtering) and biotic factors (e.g., interspecific interactions) in community assembly processes35. Finally, although biome type was an important factor in determining the relative number of plant species required to meet our restoration targets (higher in forested biomes), we found no evidence of significant phylogenetic turnover between biomes. Therefore, whilst species composition varied between networks in different biomes, overall phylogenetic composition and structure of plant communities did not, allowing our recommendations to be applicable to distinct Brazilian biomes. Nonetheless, many of the evaluated networks were from south-eastern Brazil, and further research is required in underrepresented regions, particularly Amazonia, which was represented by a single network (24 in total), to ensure findings are relevant for the restoration of Brazilian bees.
Considering only native bee taxa, assessments at family level revealed that members of the Malpighiaceae family were clear priorities for bee restoration in studied regions. Malpighiaceae species are well known for their importance to oil-collecting solitary bees46, that together represented 15% of bee species in networks. Collected oil is used as both adult and larval food, and in nest construction46. As well as their importance for oil-collecting taxa, several studies examining pollen stores of native eusocial bees have highlighted the importance of Malpighiaceae pollen, among other families, as larval food47,48,49. Thus, not only do Malpighiaceae flowers form highly specialised interactions with oil-collecting bees, they provide protein-rich pollen for generalist eusocial taxa, underlining their importance for Brazilian bee restoration.
Alongside the Malpighiaceae, several other families were selected more often than expected by chance (e.g. Convolvulaceae, Sapindaceae), and likely to be important priorities for bee restoration in Brazilian ecosystems, although they fell above the significance threshold (α = 0.05). Likewise, families which were selected less often (e.g. Anacardiaceae, Asteraceae), whilst still providing resources for flower-visiting bees, are considered low priorities for bee restoration programs. However, reasons why these families promote (or hinder) the recovery of bee communities, and the extent to which these differences are determined by specific functional traits shared among family members (e.g. elaiophores in Neotropical Malpighiaceae), or multiple traits working in chorus, and the relative importance of different traits across geographic regions, remains unresolved. Thus, future studies should aim to incorporate trait-based approaches to identify functional drivers of patterns observed here and test the validity of ‘trait-matching’ hypotheses in diverse tropical plant-pollinator networks. Despite these drawbacks, our results do provide an initial guide (i.e. key plant families) for Brazilian restoration managers, although, given the limited number of families positively or negative selected, further work is required to translate our findings into actual restoration practices. Eventually, this approach could also be adapted to agricultural settings with the aim of improving crop pollination services through the establishment of ‘bee-friendly’ habitats in croplands (e.g., hedgerows, wildflower plantings)12,50.
Overall bee community recovery and restoration of pollination function were dependent on the specific criteria being targeted in simulations (i.e., bee species richness or visitation rates). Simulations that prioritised the recovery of bee visitation rates over species richness tended to produce more generalised networks and had higher values for a proxy measure of pollination function (i.e., export of shared flower-visitor species to adjacent habitats). High levels of generalisation imply greater functional redundancy among interacting species, a positive driver of stability and function in mutualistic networks, and a key priority for ecological restoration4. However, given that we defined restoration success as the minimum number of plant species required to meet our a priori restoration criteria (80% of bee species/visits restored), further work is required to make sure that levels of functional redundancy among selected plant species are sufficient to ensure that restored bee communities and pollination services are resilient to future disturbances4.
Improved pollination function in simulations that targeted visitation rates may have been because plant species predominantly targeted hyper-abundant ‘supergeneralists’ (i.e., species that interact with disproportionately high numbers of plant species), such as the eusocial bee species A. mellifera and Trigona spinipes, that together represented 46.3% of total visits. Previous research has highlighted the importance of generalist species for network stability and function17, and more specifically, the contributions of the above mentioned bee species in Brazilian ecosystems51. Assuming recovered bee populations move between restored and adjacent habitats, restoration programs benefiting supergeneralists probably have the greatest potential to quickly recover ecosystem functions across large spatial scales. On the other hand, promoting supergeneralists over the recovery of bee species diversity may exacerbate growing problems of biotic homogenisation in tropical ecosystems52. Furthermore, caution is required in assuming that the Müller (‘shared pollinators’) index is a useful proxy of pollination function in wider plant communities. Firstly, it remains unclear the degree to which bees move between restored and adjacent habitats, a factor likely to be greatly influenced by bee foraging range, behaviour, nesting habit, and surrounding landscape structure1. Secondly, our models do not include information on pollinator effectiveness, or the dependence of plant species on biotic pollination39, and evidence from pollen deposition networks suggests that plant-pollinator interactions may be more specialised than they first appear in flower-visitor networks53. Therefore, just because restored plant species sustain bee species that also visit many other plant species does not necessarily imply that those insects are effective pollinators of all visited plant species. And finally, because bee-plant networks ignore the contributions of other insect flower visitors (e.g. flies, beetles, butterflies/moths), of which many are effective pollinators in tropical plant communities. Thus, enhancing native pollinator species diversity, as well as improving the overall conservation value (A. mellifera is an invasive species in Brazil), is likely to be an essential component in the recovery of pollination function in degraded Brazilian landscapes.
In this study we demonstrate that priority plant species for the restoration of Brazilian bee communities can be selected based on prior information on their interaction patterns. Not only are they largely generalist species that form many interactions with flower-visiting bees but are also non-randomly distributed across the angiosperm phylogenetic tree. This information, along with the identity of key plant families, can provide restoration practitioners with the means to select priority plant species for bee community recovery in degraded lands, even where ecological information on the target ecosystem is limited. Further studies evaluating restoration success in areas where these selection practices are applied in loco will be important to validate our model-based predictions.
Material and Methods
Bee-plant interaction networks
We used 24 quantitative bee-plant networks from Brazil constructed using standardised sampling methods54. All networks were found in published articles or publicly-available theses/dissertations (see Data Availability statement), and representative of communities in several biomes with distinct physiognomies, including Atlantic rainforest, Restinga coastal forest, and savannah-like habitats (Cerrado and Caatinga) (Table S8; Fig. S3 in Supporting Information). Networks from each sample site were considered independent as all sample sites were at least 2 km apart (except one – but conducted in different habitat physiognomies, see Table S8).
Selecting priority plant species using interaction network metrics
Our aim was to find optimal combinations of plant species to maximise two components of bee communities: 1) bee species richness, and 2) visitation rates, respectively (henceforth ‘restoration targets’). We chose these targets as they are basic aims of most pollinator conservation actions, and are positively related to the stability and magnitude of pollination services10,55.
We used four species-level metrics of generalisation and one index of ecological distinctiveness in pollinator assemblages (functional complementarity) to inform plant selection in simulations of bee community recovery (for further details on metrics, see Table 1). All metrics unless otherwise stated were calculated using the bipartite package in R (ver. 2.08, Dormann et al. 2009). Topological models were used to simulate impacts of restored plant species on bee communities in cleared or abandoned habitats (‘from scratch’). During simulations, plant species were added in multiples of two using the functions ‘specaccum’ in R package vegan (v. 2.52)56,57 “and ‘cumsum’ (R base function)”, for bee richness and bee visitation rate models, respectively, until restoration targets were achieved. We chose 80% of bee species/visits from the original (‘pristine’) community as our restoration targets as pollination functions are largely provided by abundant generalist species58, that would likely be supported at such thresholds. Furthermore, additional testing of different thresholds (10 to 90%) revealed that only at such high thresholds do restored networks show robustness to random extinction of higher-level species comparative to that of full networks, and which provide bee flower visitors to the majority of plant species in networks (Fig. S4, Supplementary Information). Strategies that required the fewest number of plant species to meet our targets were considered the most effective strategies for bee restoration.
Results from network-based strategies were compared to mean values under random selection (1000 simulations for each multiple of two plant species), and those determined by a freely-available ‘genetic algorithm’ tool designed to select priority species from networks based on a priori restoration criteria (i.e., maximise bee richness)24. Results from random selection and genetic algorithms were respectively, considered sub-optimal and near-optimal solutions for each restoration target, and therefore means to evaluate bee community recovery under different network-based strategies.
Statistical analyses
Recovery of bee communities under different plant selection strategies was evaluated using generalised linear mixed-effects models (GLMMs) with binomial errors using ‘glmer’ function in lme4 package59. Our response variables were the number of ‘successes’ (plant species selected) and ‘failures’ (plant species not selected) in each network required to meet targets for bee richness and visitation rates (i.e., 80% of species/visits recovered from the original network). Thus, for each network response values closer to 0 require fewer plant species than those with values closer to 1. Fixed effects included plant selection strategy (i.e. network metrics, genetic algorithms, and random species selection), network size (number of unique pairwise interactions), biome type (‘savannah’ or ‘forest’ biomes), and all two-way interactions between predictor variables. Biome type was included in models because ordination analyses of plant communities (Non-Metric Multi-Dimensional Scaling and Analysis of Similarity) revealed that networks in forest biomes (Atlantic rainforest, Restinga) differed in their species composition from those in savannah-like biomes (Cerrado and Caatinga) (Fig. S5, Supplementary Information). Network was included as a random effect to account for non-independence of data from simulation models.
All candidate models were checked for collinearity using the function ‘max.r’ (https://github.com/rojaff/dredge_mc). Best-fitting models were selected from candidate model list using ‘dredge’ function in R package MuMIn60, where all models < 2 ΔAICc (corrected for small sample sizes) from top model were considered statistically equivalent61. Model residuals were visually checked for homogeneity of variance. Importance of fixed effects in explaining variation in response variables was assessed using relative importance values (sum of Akaike model weights across all models in which fixed effect occurs) and regression coefficients derived from average (weighted) models (<2 ΔAICc). All previous steps and those presented in sections below were also performed on networks excluding flower visits and plant species exclusively-visited by non-native honeybees (A. mellifera) (present in all but one of 24 networks).
Phylogenetic relationships among priority plant species
To investigate the influence of phylogenetic relationships among priority plant species, we used different measures of relatedness for a given sample of species/individuals relative to the overall population mean, known as phylogenetic structure43,45. We tested phylogenetic structure of plant species combinations selected by the optimal plant selection strategy (fewest plant species required) for each of our two restoration targets. Phylogenetic structure analyses were performed on individual networks and all networks together (global species pool) to determine the scale at which phylogenetic relationships can inform pollinator restoration programs (for details on phylogenetic tree construction see Appendix I, Supplementary Information).
Evaluating phylogenetic structure
To test whether priority plant species were randomly or non-randomly distributed across plant phylogenies, we used null models to generate 10 thousand random subsets of plant species of equivalent size to the optimum selection in individual networks and across all networks. Phylogenetic distances of observed (selected species) and null plant communities were used to compute: a) the negative Net Relatedness Index (NRI), calculated from the mean phylogenetic distance between all species pairs in a sample, and b) the negative Nearest Taxon Index (NTI), considering only the phylogenetic distance to the nearest species43. Results from NRI and NTI may or may not be correlated, depending on whether species clusters are ‘clumped’ together in closely-related families (correlation likely), or more widely-distributed across the phylogenetic tree (correlation unlikely)43.
Null models were constructed using unweighted (all species have equal probability of selection) and weighted procedures. Weighted models were constrained by plant ‘abundance’ (individual networks - number of bee visits; global species pool – sum number of species occurrences) using the picante package in R62. Results from both null model approaches were qualitatively similar, and so only those from weighted models are presented. One-sample z-tests were used to test differences between observed and expected levels of phylogenetic structure. As plant species composition differed between forest and savannah biomes, we also tested degree of phylogenetic turnover between biomes by comparing mean phylogenetic distances between network pairs using the R package ‘picante’ and function comdist62. We then used ANOSIM to test if mean phylogenetic distances within biome groups were smaller than between groups.
Identification of key plant families
To determine key plant families for bee restoration, we first estimated the expected number of species to be selected as restoration priorities from each family based on their relative frequency in individual networks. For example, if 25% of plant species in a network belong to Family A, and 12 plants are selected for restoration, we expect that Family A will be represented by three species. We repeated this process for all 121 families and all networks and then used one-sample z-tests to compare expected (null) values, i.e. in the absence of ecological factors, with observed values (actual number of species selected). Where confidence intervals did not overlap 0 indicated a plant family that was selected more/less often than expected by chance. We included the additional caveat that families must be represented at least 10 times in networks to be considered relevant for restoration.
Evaluation of recovered network properties and pollination function
We compared the effectiveness of bee community recovery (i.e., pollination of the wider plant community) and the stability of networks constructed under different restoration targets. As networks size differed between restoration targets (see results), we standardised the number of plants in individual networks to that required to meet the bee richness target (least number of plant species). For all recovered networks, we compared the following metrics: total bee richness, total visitation, interaction richness (number of links), interaction diversity (exponent of Shannon entropy H2), specialisation level of bee species (d’poll), and network-level specialisation (H2′). Total number of species, visits and interactions in restored networks were converted into proportions based on values from full networks (e.g. 10 of 20 bee species restored = 0.5). Differences between networks recovered under different restoration criteria (bee species richness or visitation rates) were assessed using linear mixed effects models (H2) and binomial GLMMs, with network included as a random effect. Significance of fixed effects were assessed using likelihood ratio tests.
To calculate a proxy measure of the potential contribution of restored bee communities to pollination function (including pollination services) in both restored habitats and wider plant communities, we estimated the average contribution of each restored plant species to the pollination of other plant species via shared flower visitors. To do that, we use an index developed by Müller et al.63 to assess apparent competition between resource species via shared consumers. Specifically, the index quantifies how much an individual plant species (‘acting species’) contributes to the diet of the consumers (pollinators) of another plant species (‘target species’)64. Values close to 1 signify that the acting species contributes greatly to the diet of the target species’ pollinators (sum of dependencies), and scores given to individual plant species represent the sum value across all species pairs in which that plant is the acting species. Müller index (MI) scores were calculated using the function ‘PAC’ in the R package bipartite. We summed scores for all plant species selected for different restoration targets and divided these by scores from full networks to provide a proxy measure of how much ‘pollination function’ was recovered under different scenarios. Differences in MI scores were assessed using a binomial GLMM with network included as a random effect. Significance of fixed effects were assessed using likelihood ratio tests.
Data availability
All datasets used in this publication are freely available at the following website: Scientific information system about neotropical bees. http://abelha.cria.org.br/index
References
Montoya, D., Rogers, L. & Memmott, J. Emerging perspectives in the restoration of biodiversity-based ecosystem services. Trends Ecol. Evol. 27, 666–672 (2012).
Thorpe, A. S. & Stanley, A. G. Determining appropriate goals for restoration of imperilled communities and species. J. Appl. Ecol. 48, 275–279 (2011).
Tonietto, R. K. & Larkin, D. J. Habitat restoration benefits wild bees: A meta-analysis. J. Appl. Ecol. 55, 582–590 (2018).
Kaiser-Bunbury, C. N. et al. Ecosystem restoration strengthens pollination network resilience and function. Nature 542, 223–227 (2017).
Sobral, M. et al. Author Correction: Mammal diversity influences the carbon cycle through trophic interactions in the Amazon. Nat. Ecol. Evol. 1, 1782 (2017).
McCann, K. Protecting biostructure. Nature 446, 29 (2007).
Harvey, E., Gounand, I., Ward, C. L. & Altermatt, F. Bridging ecology and conservation: from ecological networks to ecosystem function. Journal of Applied Ecology 54, 371–379 (2017).
Ollerton, J., Winfree, R. & Tarrant, S. How many flowering plants are pollinated by animals? Oikos 120, 321–326 (2011).
Ballantyne, G., Baldock, K. C. R., Rendell, L. & Willmer, P. G. Pollinator importance networks illustrate the crucial value of bees in a highly speciose plant community. Sci. Rep. 7, 8389 (2017).
Ollerton, J. Pollinator Diversity: Distribution, Ecological Function, and Conservation. Annu. Rev. Ecol. Evol. Syst. (2016).
Potts, S. G. et al. Global pollinator declines: Trends, impacts and drivers. Trends Ecol. Evol. 25, 345–353 (2010).
Morandin, L. A. & Kremen, C. Hedgerow restoration promotes pollinator populations and exports native bees to adjacent fields. Ecol. Appl. 23, 829–839 (2013).
Scheper, J. et al. Environmental factors driving the effectiveness of European agri-environmental measures in mitigating pollinator loss–a meta-analysis. Ecol Lett 16, (2013).
Wood, T. J., Holland, J. M., Hughes, W. O. H. & Goulson, D. Targeted agri-environment schemes significantly improve the population size of common farmland bumblebee species. Mol. Ecol. 24, 1668–1680 (2015).
Menz, M. H. M. et al. Reconnecting plants and pollinators: challenges in the restoration of pollination mutualisms. Trends Plant Sci. 16, 4–12 (2011).
Kaiser-Bunbury, C. N. & Blüthgen, N. Integrating network ecology with applied conservation: A synthesis and guide to implementation. AoB Plants 7, (2015).
Memmott, J., Waser, N. M. & Price, M. V. Tolerance of pollination networks to species extinctions. Proc. R. Soc. London B Biol. Sci. 271, 2605–2611 (2004).
Martín González, A. M., Dalsgaard, B. & Olesen, J. M. Centrality measures and the importance of generalist species in pollination networks. Ecol. Complex. 7, 36–43 (2010).
Russo, L., Debarros, N., Yang, S., Shea, K. & Mortensen, D. Supporting crop pollinators with floral resources: Network-based phenological matching. Ecol. Evol. 3, 3125–3140 (2013).
Isaacs, R., Tuell, J., Fiedler, A., Gardiner, M. & Landis, D. Maximizing arthropod-mediated ecosystem services in agricultural landscapes: The role of native plants. Front. Ecol. Environ. 7, 196–203 (2009).
Forup, M. L., Henson, K. S. E., Craze, P. G. & Memmott, J. The restoration of ecological interactions: plant–pollinator networks on ancient and restored heathlands. J. Appl. Ecol. 45, 742–752 (2008).
Devoto, M., Bailey, S., Craze, P. & Memmott, J. Understanding and planning ecological restoration of plant-pollinator networks. Ecol. Lett. 15, 319–328 (2012).
Olesen, J. M., Bascompte, J., Dupont, Y. L. & Jordano, P. The modularity of pollination networks. Proc. Natl. Acad. Sci. USA 104, 19891–19896 (2007).
M’Gonigle, L. K., Williams, N. M., Lonsdorf, E. & Kremen, C. A Tool for Selecting Plants When Restoring Habitat for Pollinators. Conservation Letters 10, 105–111 (2017).
Archer, C. R., Pirk, C. W. W., Carvalheiro, L. G. & Nicolson, S. W. Economic and ecological implications of geographic bias in pollinator ecology in the light of pollinator declines. Oikos 123, 401–407 (2014).
Vizentin-Bugoni, J. et al. Plant-Pollinator Networks in the Tropics: A Review. In Ecological Networks in the Tropics (eds Dátillo, W., & Rico-Gray, V.) 73–92 (Springer, 2018).
Sloan, S. & Sayer, J. A. Forest Resources Assessment of 2015 shows positive global trends but forest loss and degradation persist in poor tropical countries. For. Ecol. Manage. 352, 134–145 (2015).
Barlow, J. et al. The future of hyperdiverse tropical ecosystems. Nature 559, 517–526 (2018).
Flynn, D. F. B., Mirotchnick, N., Jain, M., Palmer, M. I. & Naeem, S. Functional and phylogenetic diversity as predictors of biodiversity–ecosystem-function relationships. Ecology 92, 1573–1581 (2011).
Garibaldi, L. A. et al. Trait matching of flower visitors and crops predicts fruit set better than trait diversity. J. Appl. Ecol. 52, 1436–1444 (2015).
Junker, R. R. et al. Specialization on traits as basis for the niche-breadth of flower visitors and as structuring mechanism of ecological networks. Funct. Ecol. 27, 329–341 (2013).
Rezende, E. L., Jordano, P. & Bascompte, J. Effects of phenotypic complementarity and phylogeny on the nested structure of mutualistic networks. Oikos 116, 1919–1929 (2007).
Aizen, M. A. et al. The phylogenetic structure of plant-pollinator networks increases with habitat size and isolation. Ecol. Lett. 19, 29–36 (2016).
Rezende, E., Lavabre, J., R Guimarães, P., Jordano, P. & Bascompte, J. Non-random extinctions in phylogenetically structured mutualistic networks. Nature 448, (2007).
Gastauer, M. et al. Mine land rehabilitation: Modern ecological approaches for more sustainable mining. J. Clean. Prod. 172, 1409–1422 (2018).
Chamberlain, S. A. et al. Traits and phylogenetic history contribute to network structure across Canadian plant–pollinator communities. Oecologia 176, 545–556 (2014).
Verdú, M., Gómez-Aparicio, L. & Valiente-Banuet, A. Phylogenetic relatedness as a tool in restoration ecology: a meta-analysis. Proc. R. Soc. B Biol. Sci. 279, 1761 LP–1767 (2012).
Raimundo, R. L. G., Guimarães, P. R. & Evans, D. M. Adaptive Networks for Restoration Ecology. Trends in Ecology and Evolution https://doi.org/10.1016/j.tree.2018.06.002 (2018).
Kaiser-Bunbury, C. N., Muff, S., Memmott, J., Müller, C. B. & Caflisch, A. The robustness of pollination networks to the loss of species and interactions: a quantitative approach incorporating pollinator behaviour. Ecol. Lett. 13, 442–452 (2010).
Thompson, J. D. How do visitation patterns vary among pollinators in relation to floral display and floral design in a generalist pollination system? Oecologia 126, 386–394 (2001).
Chacoff, N. P. et al. Evaluating sampling completeness in a desert plant–pollinator network. J. Anim. Ecol. 81, 190–200 (2012).
Castro-Urgal, R., Tur, C., Albrecht, M. & Traveset, A. How different link weights affect the structure of quantitative flower-visitation networks. Basic Appl. Ecol. 13, 500–508 (2012).
Vamosi, S. M., Heard, S. B., Vamosi, J. C. & Webb, C. O. Emerging patterns in the comparative analysis of phylogenetic community structure. Mol. Ecol. 18, 572–592 (2009).
Blüthgen, N. & Klein, A.-M. Functional complementarity and specialisation: The role of biodiversity in plant–pollinator interactions. Basic Appl. Ecol. 12, 282–291 (2011).
Cavender-Bares, J., Kozak, K. H., Fine, P. V. A. & Kembel, S. W. The merging of community ecology and phylogenetic biology. Ecol. Lett. 12, 693–715 (2009).
Rasmussen, C., Rasmussen, C., Olesen, J. M. & Olesen, J. M. Oil flowers and oil-collecting bees. Scand. Assoc. Pollinat. Ecol. honours Knut Faegri (2000).
Ramalho, M., Kleinert-Giovannini, A. & Imperatriz-Fonseca, V. L. Important bee plants for stingless bees (Melipona and Trigonini) and Africanized honeybees (Apis mellifera) in neotropical habitats: a review. Apidologie 21, 469–488 (1990).
Rech, A. R. & Lúcia Absy, M. Pollen sources used by species of Meliponini (Hymenoptera: Apidae) along the Rio Negro channel in Amazonas, Brazil. Grana 50, 150–161 (2011).
Aleixo, K. P., Menezes, C., Imperatriz Fonseca, V. L. & da Silva, C. I. Seasonal availability of floral resources and ambient temperature shape stingless bee foraging behavior (Scaptotrigona aff. depilis). Apidologie 48, 117–127 (2016).
Garibaldi, L. A. et al. Mutually beneficial pollinator diversity and crop yield outcomes in small and large farms. Science 351, 388–391 (2016).
Giannini, T. C. et al. Native and Non-Native Supergeneralist Bee Species Have Different Effects on Plant-Bee Networks. PLoS One 10, e0137198 (2015).
Solar, R. R. & de, C. et al. How pervasive is biotic homogenization in human-modified tropical forest landscapes? Ecol. Lett. 18, 1108–1118 (2015).
Ballantyne, G., Baldock, K. C. R. & Willmer, P. G. Constructing more informative plant–pollinator networks: visitation and pollen deposition networks in a heathland plant community. Proc. R. Soc. B Biol. Sci. 282, (2015).
Sakagami, S. F., Laroca, S. & Moure, J. S. Wild Bee Biocoenotics in Sao Jose dos Pinhais (PR), South Brazil. J. Fac. Sci. Hokkaido Univ. (1967).
Vázquez, D. P., Morris, W. F. & Jordano, P. Interaction frequency as a surrogate for the total effect of animal mutualists on plants. Ecol. Lett. 8, 1088–1094 (2005).
Dormann, C. F., Frund, J., Bluthgen, N. & Gruber, B. Indices, Graphs and Null Models: Analyzing Bipartite Ecological. Networks. Open Ecol. J. 2, 7–24 (2009).
Oksanen, J. et al. Vegan: Community Ecology Package. R Project 1, (2008).
Winfree, R., Fox, W., Williams, J., Reilly, N. M. & Cariveau, J. R. D. P. Abundance of common species, not species richness, drives delivery of a real-world ecosystem service. Ecol. Lett. 18, 626–635 (2015).
Bates, D., Mächler, M., Bolker, B. M. & Walker, S. C. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015).
Barton, K. MuMIn: Multi-model inference. R package version 1.9.13. Version 1, 18 (2015).
Anderson, D. R. & Burnham, K. P. Avoiding Pitfalls When Using Information-Theoretic Methods. J. Wildl. Manage. 66, 912–918 (2002).
Kembel, S. W. et al. Picante: R tools for integrating phylogenies and ecology. Bioinformatics 26, 1463–1464 (2010).
Muller, C. B., Adriaanse, I. C. T., Belshaw, R. & Godfray, H. C. J. The Structure of an Aphid-Parasitoid Community. J. Anim. Ecol. 68, 346–370 (1999).
Carvalheiro, L. G. et al. The potential for indirect effects between co-flowering plants via shared pollinators depends on resource abundance, accessibility and relatedness. Ecol. Lett. 17, 1389–1399 (2014).
Acknowledgements
The authors thank the National Council for Scientific and Technological Development (CNPq 443381/2015–1 and 300037/2018–0 for AJC grant; 421668/2018-0 and 421668/2018-0 for funding LGC). We also thank the São Paulo Research Foundation (FAPESP project number 2004/15801–0) and project coordinators, Vera Lucia Imperatriz-Fonseca and Astrid M. P. Kleinert, for data access. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Author information
Authors and Affiliations
Contributions
T.C.G. devised study and sourced datasets, L.G.C. devised study and advised on statistical analyses, A.J.C. ran network analyses and wrote an initial draft of manuscript, M.G. constructed phylogenies and ran phylogenetic structure analyses, M.A.N. advised on statistical analyses and study design. All authors contributed substantially to the discussion of results, revision process and approved the final draft of the submitted manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Campbell, A.J., Gigante Carvalheiro, L., Gastauer, M. et al. Pollinator restoration in Brazilian ecosystems relies on a small but phylogenetically-diverse set of plant families. Sci Rep 9, 17383 (2019). https://doi.org/10.1038/s41598-019-53829-4
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-019-53829-4
This article is cited by
-
Network-based restoration strategies maximize ecosystem recovery
Communications Biology (2023)
-
Assessing entomological collection data to build pollen interaction networks in the tropical Amazon forest
Arthropod-Plant Interactions (2023)
-
Integrating network theory and biodiversity conservation: do different species selection approaches result in different recommendations?
Biodiversity and Conservation (2023)
-
Agricultural landscape influences on the solitary bees and wasps that nest in ecological restoration sites
Biodiversity and Conservation (2023)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.