The sampling area of the study is located in a fragment of the Caatinga situated on the Campus of Agrarian Sciences of Universidade Federal do Vale do São Francisco (UNIVASF) (9°19’44,2’’S, 40°33’30,1W) in the municipality of Petrolina, state of Pernambuco, Brazil ( Figure 1A ). The Caatinga is one of the phytogeographical domains of Brazil. It has a high level of endemism and is classified as a wooded steppe savanna (Andrade-Lima, 1981). The climate in the region is considered semiarid, dry and warm (BSh according to the Köppen classification), with low rainfall (annual average of 433 mm) and two well-defined seasons: dry from May to October and rainy from November to April. The annual average temperature is 24.8°C (Alvares et al., 2013).

The area is in the Senator Nilo Coelho Irrigation Project (C1), which is one of the main fruit farming centers in Brazil (Bustamante, 2009). Orchards of plants of agricultural interest surround the area, such as mangos (Mangifera indica L.), grapes (Vitis vinifera L.), coconut trees (Cocos nucifera L.) and acerola (West Indian cherry) (Malpighia emarginata DC.) ( Figure 1 ).

The vegetation in the area is classified as low Caatinga (Andrade-Lima, 1981), with a predominance of bush vegetation, such as Cenostigma microphyllum (Mart. ex G. Don) Gagnon and G.P.Lewis and Senegalia tenuiflora (L.) Britton and Rose ( Figure 1B ), as well as an herbaceous stratum composed mainly of species of Lamiaceae, Leguminosidae, Malvaceae and Rubiaceae ( Figure 1C ), with the presence of large trees being rare.

Monthly samples of bees were collected between November 2017 and October 2018 to record plants visited by the species. Collections were performed on two consecutive days from 6:00 to 17:00 h, totaling 22 hours per month, in eight preestablished transects of different distances ( Figure 1D ). One collector carefully inspected each flowering plant for maximum five minutes and both female and male bees were caught randomly with an insect net during visits to the flowers. Whenever possible, the resource collected by the bee was observed. To avoid recapture, bees were marked on the thorax with a nontoxic pen (UniPosca ®) and released. Specimens of both species were sacrificed, mounted and deposited in the Entomological Collection of the UNIVASF Entomology Lab.

To characterize the composition of the pollen load on the females, monthly visits were performed in two perimeters ( Figure 1E ) on two consecutive days between the years 2018 and 2021. In this period, a maximum of five samples were collected per month based on the presence of females, totaling 60 samples (30 for each species). Females with pollen adhered to the scopa were caught with an insect net during flight. The pollen on the scopa was removed using a rinsing process following the method proposed by Silva et al. (2010) and transferred to 15-ml Falcon test tubes containing 1.5 ml of distilled water. The females were then dried with paper towel, marked on the thorax with nontoxic pens (UniPosca ®) of different colors to avoid recapture and released. The pollen removed was stored in 70% alcohol until acetolysis.

To understand the composition of the larval diet, samples of larval provisions were collected from 60 brood cells (30 from each species) from 30 nests of C. xanthomelaena and 27 nests of C. perforator ( Figure 1F ) between the years 2017 and 2021. Due to the low availability of C. perforator nests, no minimum number of samples per month was established for this species. Additionally, more than one brood cell was used of the nests built by two C. perforator females that had more than 12 cells/nest, with cells selected from different stages of construction of the nests. All samples were collected from fresh provisions soon after the conclusion of the nests by the females and stored in 70% alcohol until acetolysis.

The pollen samples were submitted to chemical treatment (acetolysis), as proposed by Erdtman (1960). At the end of the process, slides were mounted for optical microscopic analysis. Four slides were mounted with Kisser glycerol gelatin (Kisser, 1935) for each sample. After mounting, qualitative analysis was performed for the identification of the pollen types on each slide and quantitative analysis was performed, with counts of 1000 grains of pollen per sample (250 per slide) (Montero and Tormo, 1990). Pollen types that appeared at a frequency of less than 0.01% were considered contaminants and excluded from the quantitative analysis.

Based on the quantitative analysis, the frequency of occurrence was calculated for the different types of pollen in each sample and mean frequencies were calculated for each type among the samples (Villanueva-Gutiérrez and Roubik, 2004; Silva, 2009). The percentage of each pollen type was determined and categorized using the classification proposed by Jones and Bryant (1996): very frequent (> 50% of all pollen on the slide), frequent (20 to 50%), infrequent (10 to 20%) and rare (< 10%).

In the period from 2018 to 2019, monthly monitoring was performed of plants in bloom in the study area (UNIVASF Campus of Agrarian Sciences), with a total of 153 species sampled. A reference slide collection was created of the pollen obtained from the flowers of the species, following the method proposed by Erdtman (1960). The pollen collection contributed to a more precise identification of the pollen types in the samples from the bees. Vouchers of the plants were identified by specialists and deposited in the Vale do São Francisco Herbarium.

The Shannon-Weaver diversity index (H’) (Shannon, 1948) was used to assess the diversity of plants that composed the diet, considering the richness and abundance of pollen types. The uniformity of the floral resources was tested by calculating Pielou’s evenness index (J’) (Pielou, 1977). The dominance (D) of pollen types was determined using the Berger–Parker index (Magurran, 2004). The number of the most important pollen types in the diet of the species was calculated from the apparent S (apparent S = e H ' ). The analysis of trophic niche overlap of the two species was performed using the Jaccard similarity index (Sj) and the proportion of plant species in the diets was determined using the percentage similarity index (PS). (The formulas are available in the Supplementary Data 1 ).

The bipartite package of the R program 3.4.1 (R Core Team, 2017) was used for the representation of interaction networks in Eulerian graphs. To visualize the relationship of the plants on the genus level and the bee species, a weighted network was created through the analysis of quantitative matrices using the Pajek program version 5.13 (De Nooy et al., 2018) and the ‘Kamada-Kawai free’ method, making the subgroups and centrality of each vertex more evident.

A total of 433 individuals of the two species were collected: 209 specimens of Centris perforator and 224 specimens of Centris xanthomelaena ( Table 1 ). Together both species visited 32 plant species (C. perforator = 22 species and C. xanthomelaena 19 species). Leguminosae was the botanical family with the largest number of interactions (n = 274), the majority of which involved the subfamily Caesalpinioideae (n = 226). Lamiaceae was the family with the second largest number of interactions (n = 104), despite the small number of species recorded in the area.

Centris perforator females interacted with the flowers of 19 plant species and males visited 14 species. The largest part of the pollen sources used for larval feeding was nearly exclusively from plants of the family Leguminosae, with Chamaecrista and Senna as the genera with the largest number of visits (n = 73 and 24, respectively). Cenostigma microphyllum (Mart. ex G.Don) Gagnon and G.P.Lewis was the main source of nectar visited by both male and females of C. perforator. However, Rhaphiodon echinus (Nees and Mart.) Schauer and Libidibia ferrea (Mart. ex Tul.) L.P.Queiroz also received a number of visits by males of the species. The only source of floral oil visited by the females was Malpighia emarginata, which was also used sporadically as a source of pollen. During the flowering period of Tabebuia aurea (Silva Manso) Benth. and Hook.f. ex S.Moore (September-November), nectar collecting by both males and females seems to concentrate on this plant.

Females of the species C. xanthomelaena interacted with the same number of plants as those of C. perforator (n = 19). However, C. xanthomelaena males restricted their visits to smaller number of plants (n = 9). Females restricted pollen collecting to four species of Chamaecrista (L.) Moench (n = 69), followed by Zornia brasiliensis Vogel (n = 6). Both males and females were regularly seen visiting R. echinus, which was the main species visited by females to collect nectar. During pollen collection, females also interacted with R. echinus. Malpighia emarginata and Angelonia cornigera Hook.f. were the two oil sources visited by the females. Table 1 displays the complete list of plants on which males and females of both species were observed or collected.

For both sample types (pollen load on females and larval provisions), 47 pollen types were found, corresponding to at least 16 families and 26 genera of plants ( Table 2 ).

Centris xanthomelaena use a greater diversity of pollen types (H’ _nest = 2.08; H’ _body = 1.78) compared to C. perforator (H’ _nest = 1,42; H’ _body = 1,81). The number of species considered important in the diet of the immature forms (apparent S) varied with the type of sample, ranging from four to eight ( Table 2 ). Three to nine pollen types were found in each subsample for C. perforator ( Χ ¯ = 5.4 ± 1.44; n = 60), whereas three to 13 pollen types were found in each subsample for C. xanthomelaena ( Χ ¯ = 7 ± 2.35; n = 60).

The two species used plants of the genus Chamaecrista as the main source of pollen ( Figure 4 ), diverging on the species level. While Ch. repens was used by C. perforator as the main source of pollen, C. xanthomelaena collected the large portion of pollen from Ch. calycioides. Moreover, Senna martiana composed a considerable fraction of the larval diet for C. perforator. A difference was also found with regards to the sources of nectar, with Ce. microphyllum as the main source for C. perforator and R. echinus as the most important source for C. xanthomelaena. The Jaccard index (Sj) revealed 36.8% similarity in the pollen samples from the body of all females and 43.3% in the larval provision samples. The proportions of plant species in the diet (PS) were similar, with 43.52% overlap in the samples from the body of the females and 47.89% in the samples from the nests.

The analyses of the pollen load on the females and the nests of Centris perforator and Centris xanthomelaena demonstrated a strong association with plants with poricidal anthers, such as the genus Chamaecrista, which corresponded to the largest part of the pollen types and was present in all samples analyzed. Pollen types from at least four species of Chamaecrista were identified in the samples, together corresponding to more than 45% of the pollen types found on the scopas and in the nests. Certainly, Chamaecrista is the main source of pollen for both species of Centris. The results of the present study lend strength to the hypothesis put forth by Aguiar et al. (2003) that species of Centris endemic to the Caatinga domain, such as Centris hyptidis Ducke and C. xanthomelaena, restrict pollen collecting to fewer sources in comparison to more widely distributed species of the genus. The description of pollen sources found in the nests of Centris analis (Fabricius), Centris tarsata Smith and Centris trigonoides Lepeletier studied in an area of the Caatinga demonstrated a strong association with species of Leguminosae, especially the genera Chamaecrista and Senna (Aguiar et al., 2003; Dórea et al., 2010a, Dórea et al., 2010b; Dórea et al., 2013; Cruz et al., 2015). Vivallo and Zanella (2021) in a collection of plants visited by C. xanthomelaena also pointed to Chamaecrista as the only source of pollen recorded to date. Moreover, Sabino et al. (2019) found that Chamaecrista was a constant or dominant source of pollen for Centris burgdorfi even in different types of vegetation.

The two studied bee species differ regarding to the species Chamaecrista exploited for pollen. Ch. calycioides was the predominant pollen type in the samples of C. xanthomelaena while Ch. repens was the dominant pollen types in the samples of C. perforator. This difference may be related to the size of the flowers, as Centris xanthomelaena, which is smaller, exhibited a preference for smaller Chamaecrista flowers ( Figure 5 ), unlike the dominant species in the samples of C. perforator, thus diminishing competition among the females (see the details of measurements in Supplementary Table 1 , Supplementary Figure 1 ).

In the Caatinga domain, Leguminosae is the best represented botanical family and is an important resource for various species of bees (Queiroz, 2002). The high availability and proximity of plants in this family to nesting sites optimizes the provisioning phase and, consequently, diminishes the energy expenditure of females in search of resources. Despite being an abundant resource, pollen collection from various species of Leguminoseae, especially those of the subtribe Cassiinea, is limited to bees with the capacity to perform collection by vibration, which is known as “buzz pollination”, such as those of the genera Bombus, Centris, Epicharis, Melipona and Xylocopa (Nunes-Silva et al., 2010). With buzz pollination, pollen is released from the poricidal anthers through the vibration executed by the thoracic muscles of the bees (Buchmann, 1983). The high protein content of the pollen produced by plants with poricidal anthers may result in the preference on the part of females due to the quality of the resource (Roulston et al., 2000; Vaudo et al., 2020).

Richness in the pollen samples ranged from 18 to 27 species. The diversity indices in the samples of C. perforator varied mainly due to the presence of contaminant species, with percentages lower than 1% (12 species) on the body of females. We believe that various of these species were visited for nectar consumption (even the undetermined ones) or only represent contamination during foraging of the females. The number of species and Shannon’s diversity index (H’) were lower in the nests of C. perforator (S = 18; H’=1.42) and the brood cells are evidently supplied with pollen from Leguminosae – Caesalpineaceae (> 75% Chamaecrista + 16% Senna), thus explaining a higher dominance index in comparison to pollen from the scopa. The richness of pollen types used by C. xanthomelaena was slightly higher in the samples from the scopa than those from the nests. In contrast, dominance was greater in the samples from the brood cells of the two species.

The evenness indices (J’) were similar in all samples (difference of less than 10%) and slightly higher in the sample from the nests of the two species of Centris. The Jaccard indices (Sj) indicated 36.8% similarity in the samples of pollen from the body of the females of the two species and 43.3% for the samples from nests. The proportions of plants species in the diet (PS) varied, with 43.52% overlap in the samples from the scopas of the females and 47.89% in the samples from the nests of the two species.

Despite that close similarity in the sources of pollen exploited by the two species of Centris, a difference was found in the plants visited to obtain nectar. Rhaphiodon echinus was the main source of nectar recorded in all samples of C. xanthomelaena, with M. tomentosa and W. albicans as secondary sources. Apparently, R. echinus is an annual source of nectar composing the diet of females and immature forms, even in periods of resource scarcity, and accounted for approximately 10% of the pollen types in the samples. Martins et al. (2020b) found that nectar is one of the constituents of the larval provisions of C. xanthomelaena and added together with the oil soon after pollen collection trips. Rhaphiodon echinus was also reported as a source of nectar for other species of Centris in the Caatinga domain, such as C. analis (Dórea et al., 2010a), C. tarsata (Dórea et al., 2010b) and C. trigonoides (Dórea et al., 2013). Besides being endemic to the Caatinga, R. echinus is one of the main plants of the herbaceous stratum in the study area (Harley et al., 2015). Floral visitation data confirm that this herbaceous plant is a key species for the permanence of C. xanthomelaena in the area.

The sources of nectar for C. perforator were more diverse. The pollen from Cenostigma microphylum appears with greater frequency in the samples, but the main source of nectar alternated with the period of the year. In samples from September to November, Tabebuia aurea was the main source of nectar visited by the females. This species has mass flowering in the dry season (Barros, 2001), when resource scarcity naturally occurs.

In the same way as the females, the males of both species diverged in their trophic niche with regards to the main sources of nectar exploited. Males of C. xanthomelaena visited flowers of R. echinus more intensely, whereas the two main sources of nectar visited by males of C. perforator were Ce. microphylum and T. aurea. Moreover, males of this species demonstrated territorial behavior at these plants, underscoring their importance for the trophic niche. To maximize reproductive success, patrolling for virgin females on flowers is one of the tactics employed by males of different species (Paxton, 2005), which would explain the preference of males of C. xanthomelaena and C. perforator for the main sources of nectar used by the females.

Bees of the genus Centris have a strong association with plants of the family Malpighiaceae (Anderson, 1990) and this interaction is reflected in the source of floral oil exploited by C. perforator and C. xanthomelaena. The analysis of the samples revealed that M. emarginata is the main source of oil used by both species. Despite appearing to be the exclusive source of oil for C. perforator, the C. xanthomelaena samples indicate that Agelonia cornigera and Krameria sp. are alternative sources of oil used by the females. Besides floral oil, some studies indicate that Malphigiaceae also furnishes pollen as a reward, such as the genus Epicharis, in which pollen loads can contain more than 90% of the pollen from Byrsonima (Gaglianone, 2005). However, only two samples of C. perforator had M. emarginata as the dominant pollen type.

The presence of pollen from Angelonia cornigera in the samples occurred simultaneously to the rainy period (February to May), in which peak flowering of this plant also occurred. The flowers were found in sites with sandy soil very close to the nesting sites. In the samples containing pollen from A. cornigera, the frequency of grains of pollen from M. emarginata was very low or absent.

Although most species of Centris are adapted to collect oil from epithelial elaiophores (Vogel, 1974; Neff and Simpson, 1981), some species have morphological modifications that enable collecting oil from flowers with trichomal elaiophores (Vogel, 1974; Simpson et al., 1990; Cocucci, 1991; MaChado et al., 2002). This modified “four-legged” pattern (Neff and Simpson, 1981) is found in C. xanthomelaena, C. burgdorfi, C. tricolor and all species of the subgenus Hemisiella (Taniguchi, 2010), enabling these species to expand their sources of oil (Neff and Simpson, 1981). Sabino et al. (2019) studied three populations of C. burgdorfi in different types of vegetation and found plasticity with regards to the sources of oil exploited in flowers with epithelial and trichomal elaiophores of the genera Krameria and Angelonia, respectively. Although pollen from Krameria was found in a low frequency in the samples of C. xanthomelaena, no plant belonging to this genus was found within a 500-meter radius of the nesting sites of the species. There are reports of C. xanthomelaena serving as an effective pollinator of Krameria grandiflora, which also received sporadic visits from C. perforator (Carneiro et al., 2019).

The females of Centris perforator and Centris xanthomelaena demonstrated high dependence on the vegetation mainly in the herbaceous and shrub strata. These strata are fundamental to the survival and permanence of pollinating agents in agricultural areas (Wolowski et al., 2019), but are constantly eliminated within these areas. Acerola (West Indian cherry) requires pollinators for cross fertilization (Freitas et al., 1999; Sazan et al., 2014), which makes good management practices all the more necessary, ensuring not only productivity, but also the quality of the fruit (Magalhães and Freitas, 2013). As the visits of females of C. perforator and C. xanthomelaena are directed to the collection of the floral oil in acerola, the possibility of the management of females in orchards depends on the presence of the surrounding herbaceous and shrub strata, ensuring the availability of sources of pollen and nectar essential to the survival and permanence of the populations of the species in the area.

The availability of floral resources is one of the main factors responsible for the regulation and abundance of native bee populations (Roulston and Goodell, 2011), especially those resources that are indispensable to species survival, such as the case of floral oil for Centris (Silva et al., 2017). Silva et al. (2019) call attention to the change in the regularity of floral resources in response to climate change events, such as long periods of drought, especially in dry forests like the Caatinga. Consequently, the reduction in these floral resources for bees can cause the decline in these pollinators, thus affecting the biome as a whole (Kearns et al., 1998; Kevan and Imperatriz-Fonseca, 2002). The herbaceous stratum constitutes one of the main groups of plants that can suffer these effects more intensely, especially plants endemic to the Caatinga (Silva et al., 2019). In this scenario, populations of C. perforator and C. xanthomelaena could be affected, as both species are highly dependent on plants of this stratum, such as the endemic species Cenostigma microphyllum and Raphiodun echinus.