The Similar Usage of a Common Key Resource Does Not Determine Similar Responses by Species in A Community of Oil-collecting Bees

The comprehension of the factors that affect variations in abundance and species richness in natural systems is one of the most explored questions in ecology (Morin, 1999; Agrawal et al., 2007). In many situations, such variations seem to be determined by an interaction between both biotic (such as competition or predation) and abiotic (such as temperature and rainy fluctuations) factors (Morin & Lawler, 1995; Ritchie, 2000; Dyer & Letourneau, 2003). However, there are groups of species that present similar specialized traits related to foraging behavior, predator avoidance strategies or even adaptations to specific conditions. In these cases, it may be that the existence of a similar adaptation makes the species bearing it more dependent on the variation of one or a Abstract Variations in abundance and species richness among communities are often determined by interactions between biotic and abiotic factors. However, for communities composed of species that share a common specialization (such as similar foraging adaptations) it may be a key ecological factor involved in the common specialization that affects community variations. To evaluate this possibility, we characterized the guild of oil-collecting bees of a Neotropical savanna in Brazil and tested whether differences in Byrsonima abundance and availability of floral oil explain differences in species richness and abundance of oil-collecting bees of different tribes. Both the number of species and total abundance of Centridini species increased with the abundance of Byrsonima. One plausible explanation for the stronger adjustment between the abundance of Centridini and Byrsonima is that the abundance of these plants affects not only the availability of floral oil, but also of pollen. These findings indicate that the existence of a common specialization among different species does not homogenize their response to variations in a common explored resource. Sociobiology An international journal on social insects


Introduction
The comprehension of the factors that affect variations in abundance and species richness in natural systems is one of the most explored questions in ecology (Morin, 1999; few key ecological factors.Consequently, variations in such factors should trigger similar responses in the species that share similar adaptations. One species group that may respond to variations in a key ecological factor is oil-collecting bees that explore the resources produced by oil-flower species of different families.These plant-pollinator systems are highly specialized, with interactions based one resource of high energetic value (oil).Few plant families provide oil on specialized flower structures (elaiophores), which are accessible only to flower visitors that are able to "scratch" epithelial elaiophores with specialized structures (combs) on their legs, or use specialized setae on their legs to obtain oil from trichomatic elaiophores (Neff & Simpson, 1981;Vogel & Machado, 1991).As consequence of these adaptations, oil-collecting bees and oil-producing plants have interactions of high intimacy: floral oil is essential to feed the larvae and build nests (Vinson & Frankie, 2000;Aguiar & Garófalo, 2004;Michener, 2007), while the pollination services provided by these bees are needed for the plant reproduction (Vogel & Machado, 1991;Sigrist & Sazima, 2004;Costa et al., 2006;Sazan et al., 2014).
Despite the high dependence between oil-collecting bees and oil-producing flowers, the effects of the abundance of floral resources on the communities of oil-collecting bees had been poorly studied.The few existing studies report that both the abundance and richness of Centridini species (a tribe composed of oil-collecting bee species) may be determined by the abundance Malpighiaceae species, mainly Byrsonima sericea DC.This relationship between bee abundance or richness and Malpighiaceae species seems to hold both in local and regional scales (Ramalho & Silva, 2002;Rosa & Ramalho, 2011).However, it is important to note that other bee tribes, such as Tetrapediini and Tapinotaspidini possess similar specializations and dependency to explore oil-flowers as Centridini bees.Consequently, if the foraging specialization observed in these bees is responsible for the strong relationship between Centridini and Malpighiaceae species, this pattern should also be detected in the other two bee tribes.In addition, since oil is the main food source for oil-collecting bees, it is possible that the abundance and richness of these insects are more dependent on oil availability than plant abundance.
To evaluate the hypothesis that the specialization in foraging behavior is responsible for the tight relationship between the abundance and richness of oil-collecting bees and oil producing flowers, we made focal samplings of the flower visitors of two Byrsonima species.We tested whether differences in the amount of floral oil available and in the local abundance of two Byrsonima species explain differences in the richness and abundance of oil-collecting bees of different tribes at local scale.

Study area
We carried out the study in the surroundings of the Chapada Diamantina National Park (12°25' S; 41°29' W), Palmeiras municipality, state of Bahia, northeastern Brazil.The local climate is hot, with humid summer, and four to five dry months.Rains occur from December to April (Jesus et al., 1983;Nimer, 1989) and the annual rainfall varies from 600 to 1,000 mm (CEI, 1994).The region of Chapada Diamantina has diverse vegetation, including campos rupestres (sandstone outcrop vegetation), campos gerais (grasslands), cerrados (savannas), caatingas (deciduous thorny woodlands), and montane forests, which form a vegetation mosaic (Conceição et al., 2005).We collected samples in a cerrado area, with a predominance of herbaceous plants and shrubs, and sparse trees with up to 5 m.

Sampling
We carried out field work during the flowering period of Byrsonima sericea and Byrsonima cydoniifolia A. Juss (Malpighiaceae), the most common oil-producing plants in this area.The total abundance of Byrsonima was variable among the sampled sites (23 to 233 individuals per transect), but in general there were more than 55 individuals per transect.Details on the morphology and floral biology of these plant species can be found in Teixeira and Machado (2000), and Sazan et al. (2014).The flowers of B. sericea are attractive to bees for about a day and a half (Teixeira & Machado, 2000), and the flowers of B. cydoniifollia for approximately two days (Sazan et al., 2014).In this region, B. sericea flowering season lasts from December to March, while B. cydoniifolia blooms in December and January (personal observation).
We selected four sites, with distances varying from 1.1 to 3.8 km from one another.In each site, we established three transects, 1,000 m in length.In each month, we randomly selected four transects (one per site) for sampling on consecutive days.Each whole transect was sampled from 0900 h to 1600 h, by two collectors.The collection of bees on B. sericea and B. cydoniifolia flowers lasted eight months, from January to March 2011, from October to December 2012, and from January to February 2013, a total of 175 h of sampling effort.The specimens were deposited in the Johann Becker Entomological Collection (Zoology Museum of the State University of Feira de Santana, MZFS) and in the Entomological Collection of the University of Brasília.

Quantification of the number of flowers
To estimate the number of flowers produced by B. sericea and B. cydoniifolia, we randomly picked ten individuals of each species, during the peak flowering period.For each plant individual, we randomly placed on the canopy a "square box" measuring 25 x 25 cm x the height of the "square box" (range 1.40 m -1.70 m), which varied according to plant height.Then, we counted all inflorescences that fitted inside the "square box".After counting, we removed ten inflorescences of each individual plant to count the total number of open flowers.We also measured the radius and height of the tree canopy of each plant and used these measures to estimate the volume of flowers per canopy on each plant individual (10 individuals per each Byrsonima species).Hence, based on the volume of the "square box" and the volume of the canopy flowers, we estimated the amount of inflorescences per Byrsonima individual (number of inflorescences per canopy on blossom= number of inflorescences x (volume of the canopy on blossom/ volume of the "square box").To estimate the total number of flowers produced per individual of B. sericea and B.cydoniifolia, we multiplied the total number of flowers per inflorescence by the average number of flowers per inflorescence.Finally, we counted all individuals of B. sericea and B. cydoniifolia observed in each transect, to estimate the abundance of Byrsonima.

Oil quantification in Byrsonima flowers
To estimate the amount of oil produced by B. sericea and B. cydoniifolia in each transect, we multiplied the average amount of oil produced by one flower by the estimated number of flowers per plant.We used the value obtained as an estimate of the amount of oil produced by an individual plant.Then, to estimate the total amount of oil available in each of the 12 transects, we multiplied the amount of oil produced by an individual plant by the number of individual B. sericea and B. cydoniifolia with elaiophores recorded in each transect.
To quantify oil production per flower, we collected ten inflorescences from ten individuals of B. sericea and B. cydoniifolia separately, with pre-anthesis flowers that had opened on the same day.We dried the flowers in an oven for 72 h at 60 ºC, and then removed the petals and reproductive structures of each flower, leaving only the floral receptacle and the elaiophores.Then, for each of the two Byrsonima species, we separated the dried material into eight samples of 1 g each.We extracted floral oil using the Soxhlet system, with hexane solvent.We placed each sample (n = 8) on a cellulose disk (filter paper) and maintained it in the extractor system for 24 h.We subjected the resulting mixture (solvent plus floral oil) to a low pressure system in a rotaevaporator, to remove the excess of solvent and recover the oil.We weighted the oil obtained per sample in a high precision analytical balance.The oil content (OC) per sample was calculated as a percentage, using the equation: Where: Bu = biomass of flowers used (g); V% = humidity extracted of the flowers (%) To calculate the amount of oil produced per flower, we divided the amount of oil extracted from 1 g of flowers by the total number of flowers used for the extraction.

Data analysis
To estimate the species richness of bees collected on Byrsonima flowers, we used the Chao-2 and Jackknife-1estimators.These estimators are based on the incidence of unicates and duplicates and on the number of singletons, respectively, using 150 randomizations (Colwell, 2013) assess whether the sampling effort was suitable, we built a rarefaction curve based on the species observed (Mao Tau).These analyses were made in the program EstimateS 8.20.
We used generalized linear mixed-effects models to assess whether the number of bee species and individuals of different tribes varied with oil availability and plant abundance.We built models containing bee richness or abundance as the response variables, and tribe, oil abundance, and plant abundance as fixed factors.However, as oil availability and plant abundance were correlated (r = 0.53, p < 0.001), we built different models containing only oil availability or plant abundance as explanatory variables.To assess which model better explained the variations in abundance and richness of bees, we used the Akaike information criterion corrected for small samples (AICc - Burnham and Anderson 2002).The AICc represents a heuristic approach used to select the most parsimonious model that best approximates the true process under investigation among a set of candidate models.Consequently, for each response variable, the model with the smaller AICc value will represent the best one.In all models, we fitted each sampled area nested per field trip as a random factor.Whenever necessary, we performed planned contrasts to test for differences in response variables among Centridini, Tapinotaspidini, and Tetrapediini tribes.We made all analyses using the packages lme4 (Bates et al., 2014), bbmle (Bolker & R Development Core Team, 2012), and multcomp (Hothorn et al., 2008) in R software (R Development Core Team, 2014).

Results
The guild of oil-collecting bees sampled on flowers of B. sericea and B. cydoniifolia was composed of 22 species (Table 1).The species rarefaction curve showed signs of stabilization, with the sampled species richness corresponding to 89% and 97% of the estimates provided by Jackknife-1 and Chao-2, respectively.The Centridini tribe showed higher richness (14 species; 64%) and abundance (n=374; 59% of all oil-collecting bees sampled) than the other tribes.Centris aenea (Lepeletier) and Tetrapedia amplitarsis Friese were the most abundant species (Table 1 3).Despite the high abundance of Byrsonima in transects T3 and T5, they had relatively low oil availability (T3: 3,005.8g and T5: 4,021.4g).This low oil availability resulted from the predominance of B. cydoniifolia in these transects.
The abundance of oil-bees was more strongly related to the abundance of Byrsonima than to floral oil availability (Table 4).The most parsimonious model indicated that the relationship between total oil-bee abundance and Byrsonima abundance differed among tribes (χ 2 = 334.18,df = 5, p < 0.001).In fact, the removal of the interaction term between tribe and plant abundance from the model indicated that this interaction was important to explain variations in total bee abundance (χ 2 = 248.41,df = 2, p < 0.001).The total abundance of Centridini increased with Byrsonima abundance (Fig 1A), whereas the effect was the opposite for Tetrapediini species (Fig 1B).The total abundance of Tapinotaspidini species was unrelated to plant abundance (Fig 1C , Table 5).
The number of oil-collecting bee species was also more strongly related to Byrsonima abundance than to oil availability (Table 4).The most parsimonious model indicated that the relationship between oil-collecting bee richness and Byrsonima abundance differed among tribes (χ 2 = 74.42,df = 5, p < 0.001).The removal of the interaction term between tribe and plant abundance from the model indicated that this interaction was important to explain variations in oil-collecting bee richness (χ 2 = 19.04,df = 2, p < 0.001).The number of Centridini species increased with Byrsonima abundance (Fig 2A ), whereas the number of Tetrapediini and Tapinotaspidini species was unrelated to Byrsonima abundance (Fig 2B and  2C, Table 5)

Discussion
The guild of oil-collecting bees of the studied savanna was diverse, with many rare and a few dominant species.The richness of oil-collecting bees of the studied savanna in Chapada Diamantina was in part a result from concentrating the sampling effort on Byrsonima species, with which oilbees have a close relationship (Ramalho & Silva, 2002;Rosa & Ramalho, 2011).In others studies on bee communities in the same region (Martins, 1994;Silva-Pereira & Santos, 2006), sampling effort was equally distributed among all melittophilous flora.The savannas in Chapada Diamantina seem to harbor much richer bee faunas than sandstone outcrop vegetation, which are contiguous to the savannas in many sites of this region.
Centridini, in particular the genus Centris, was the group with highest richness in the studied guild, similarly as recorded in other Brazilian savannas (Martins, 1994;Silveira & Campos, 1995;Andena et al., 2005;2012) and other Neotropical biomes (Araújo et al., 2006;Santos et al., 2013;Silva et al., 2015).Mello et al. (2013), who used a network approach to assess the interactions between oil-flowers and oil-bees at the biome level, highlighted the importance of Centris and Epicharis to the guild of oil-collecting bees.The authors reported that these two taxa occupy highly central roles in each network, because they showed a very large number of interactions (hubs) or connected different modules in the networks (connectors).
Although presenting similar foraging needs, distinct taxonomic groups that compose the local guild of oil-bees responded differently to the abundance of Byrsonima.The abundance of Centridini increased with the abundance of Byrsonima, whereas the abundance of Tetrapediini (Tetrapedia spp) decreased and the abundance of Tapinotaspidini showed no correlation.A similar trend was observed in species richness: the number of Centridini species increased with the abundance of Byrsonima, whereas the number of Tetrapediini and Tapinotaspidini species showed no correlation.These findings indicate that Centris and Epicharis species are more dependent on the resources provided by Byrsonima than Tetrapedia and Tapinotaspidini species.Perhaps species of Tetrapedia and Tapinotaspidini use other plants more intensively as food sources, as suggested by studies on the diet of the larvae of Tetrapedia diversipes Klug (Menezes et al., 2012;Neves et al., 2014) and Tetrapedia curvitarsis Friese (Campos, unpubl.data).In addition, females of Centris and Epicharis species are often larger than females of Tetrapedia and Tapinotaspidini species.Because oilcollecting bees typically "hug" the flower while foraging, it may be that the greater body size of Centridini species prevent flower exploitation by other species.The abundance of Centris and Epicharis species associated with this possible exclusion behavior may result in exploitation dominance of Byrsonima by Centridini.
The total abundance of oil-collecting bees was more strongly related to the abundance of Byrsonima than to the amount of oil offered by these flowers.One possible explanation for the best adjustment between abundance of Centridini and Byrsonima plants (in detriment of oil availability) is that the abundance of these plants affects not only floral oil, but also pollen availability.Pollen represents one of the main aspects of habitat quality for bee populations.Many Centridini species use Byrsonima as a pollen source (Aguiar & Gaglianone, 2003;Dórea et al., 2010;Rabelo et al., 2012), and, therefore, we expect that fluctuations in Byrsonima abundance directly influence bee reproduction.
The local availability of resources to supply food and material to build nests (e.g., floral oil used to coat brood cells and nest plugs) should strongly influence the movement of females in the landscape.This movement in search of resources is expected to affect the local abundance of these bees in parts of the habitat.We can assume that Centridini females that decide to nest in areas with a higher abundance of Byrsonima, and, therefore, with high availability of oil and pollen, face lower costs to produce offspring.Hence, this decision would contribute to an increase in fitness.As a result, we would expect an increase in the local abundance of these bees in habitats with high availability of resources.
The species richness and abundance of Centridini increased with the abundance of Byrsonima, which corroborates Rosa and Ramalho (2011).Since we developed our study in a different vegetation type than previous studies, it is probable that the positive relationship between the abundance of Centridini and Byrsonima does not depend on phytophysiognomy.However, we showed that plant abundance per se was more important to determine the abundance and richness of Centridini than the amount of oil available.This finding suggests that other resources, together with oil, are important for the development and maintenance of these bees.However, such relationship does not seem to be the result of the foraging specialization observed in oil-collecting bees, since the abundance and richness of species of Tetrapediini and Tapinotaspidini were unrelated to the abundance of Byrsonima.Consequently, it seems that even for species that share a common dependency and similar adaptations to exploit a specific resource (e.g.floral oil), other traits may be equally important in affecting community dynamics.

( 0 .
024%) (t = 13.5;df = 13; p<0.001).The average amount of oil produced per flower was 0.0005 g in B. sericea and 0.0002 in B. cydoniifolia.The amount of oil produced per plant varied from 2.89 to 181.88 g (µ = 75.41± 34.98 sd, n=10) in B. sericea, and from 0.52 to 48.77 g (µ = 11.75 ± 210.53 sd, n = 10) in B. cydoniifolia.The abundance of the two Byrsonima species varied largely among the 12 transects (from 23 to 233 individuals/ transect), which resulted in differences in floral oil availability among transects.The estimates of floral oil produced per transect varied from 1,320.4 to 16,397.2g (Table

Fig 1 .
Fig 1.Total number of individuals of Centridini (A), Tapinotaspidini (B) and Tetrapediini (C) in relation to the number of Byrsonima plants in a cerrado area in Chapada Diamantina, northeastern Brazil.Each dot represents one sampled transect per month (there are superimposed dots).

Fig 2 .
Fig 2. Number of bee species belonging to Centridini (a),Tapinotaspidini (b) and Tetrapediini (c) in relation to the number of Byrsonima plants in a cerrado area in Chapada Diamantina, northeastern Brazil.

Table 2 .
Number of flowers and amount of oil in ten individuals of Byrsonima cydoniifolia and B. sericea (Malpighiaceae) in a cerrado area in the Chapada Diamantina, northeastern Brazil.

Table 3 .
). Abundance of the two species of Byrsonima (BS: B. sericea, BC: B. cydoniifolia), the amount of floral oil estimated at each transect, abundance (A) and richness (R) of each tribe of oil-collecting bees per transect, sampling effort (hours) to capture bees in each transect in a cerrado area in the Chapada Diamantina, northeastern Brazil.

Table 4 .
. Summary of the models describing variation in total bee abundance and richness in relation to the tribe (Centridini, Tapinotaspidini, and Tetrapediini), plant abundance, and oil availability, in two species of Byrsonima (Malpighiaceae) in a cerrado area in Chapada Diamantina, northeastern Brazil.AICc represents the value of the Akaike Information Criterion corrected for small samples; Δi is the difference between the most parsimonious model i, wi is the Akaike weight of model i.Wi is the probability that the model is the best one among all candidate models.

Table 5 .
Planned contrasts for slope differences between tribes.Each slope was calculated from the relationship between bee abundance or bee richness (response variables) and plant abundance (explanatory variable).Estimated values indicate the difference in slope values between the first and the second tribe mentioned in each contrast.