Temporal Changes in Gut Microbiota Composition and Pollen Diet Associated with Colony Weakness of a Stingless Bee

Compared to honeybees and bumblebees, the effect of diet on the gut microbiome of Neotropical corbiculate bees such as Melipona spp. is largely unknown. These bees have been managed for centuries, but recently an annual disease is affecting M. quadrifasciata, an endangered species kept exclusively by management in Southern Brazil. Here we report the results of a longitudinal metabarcoding study involving the period of M. quadrifasciata colony weakness, designed to monitor the gut microbiota and diet changes preceding an outbreak. We found increasing amounts of bacteria associated to the gut of forager bees 2 months before the first symptoms have been recorded. Simultaneously, forager bees showed decreasing body weight. The accelerated growth of gut-associated bacteria was uneven among taxa, with Bifidobacteriaceae dominating, and Lactobacillaceae decreasing in relative abundance within the bacterial community. Dominant fungi such as Candida and Starmerella also decreased in numbers, and the stingless bee obligate symbiont Zygosaccharomyces showed the lowest relative abundance during the outbreak period. Such changes were associated with pronounced diet shifts, i.e., the rise of Eucalyptus spp. pollen amount in forager bees’ guts. Furthermore, there was a negative correlation between the amount of Eucalyptus pollen in diets and the abundance of some bacterial taxa in the gut-associated microbiota. We conclude that diet and subsequent interactions with the gut microbiome are key environmental components of the annual disease and propose the use of diet supplementation as means to sustain the activity of stingless bee keeping as well as native bee pollination services.


Introduction
There is overwhelming evidence that climate change, pollutants, and pathogens as well as poor nutrition are leading to worldwide bee population declines [1][2][3][4]. When bees are exposed to more than one of these stressors, their resulting effects on bee mortality are often synergistic, but the mechanisms of stressor interactions are not fully understood yet [5]. For social bees, there are feedbacks between nutrition and disease, i.e., poor nutrition increasing the susceptibility to pathogens [6], and infections also affecting colony health through behavioral perturbation that changes colony-level nutrition [7]. Furthermore, these feedback loops intersect a third dimension of bee health, the gut microbiome, which is altered by host diet on one hand, and on the other hand is capable of modulating parasitism [8].
A large part of the metagenome of the honeybee-associated microbiota is dedicated to carbohydrate metabolism, in which symbionts seem to exhibit labor division [9]. Gut bacteria also stimulate the honeybee innate immune system, which might be a host mechanism to regulate the microbiota [10]. Symbiont-mediated stimulation of the innate immune system in honeybees results in the production of host antimicrobial peptides that can differentially modulate microbial constituents in favor of particular core members and protect against pathogens [11].
The gut of Apis mellifera adult workers is dominated by nine bacterial phylotypes, five of which are ubiquitous, while four others sometimes absent [12]. There are large strain-level differences in the core gut microbiota of closely related honeybee species [13] and deeper symbiont differences between distantly related corbiculate bees (family Apidae). For example, Bartonella and Frishella, which are common symbionts of the Apini (honeybees), are not found in the Bombini (bumblebees) and Meliponini (stingless bees) [14]. Moreover, stingless bees from the genus Melipona, which radiated in South America during the Miocene [15], lack members of the core microbiome of other Apidae, i.e., Snodgrassella and Gilliamella [16].
To some extent, bee microbiomes seem to be shaped by the host dietary habits. Necrophagous bees show a higher richness of acidophilic bacteria than their pollinivorous relatives [17]. Furthermore, the microbiome of a single pollinivorous species may vary in composition according to the composition of the flowering plant community within its habitat [18], and the core bacteria of the honeybee gut microbiome may shift in abundance throughout a season, probably influenced by dietary changes [19]. Thus, understanding the factors shaping gut microbiomes of Neotropical stingless bees, and the influence of gut microbiomes on their health, is crucial to sustain pollination services of Neotropical ecosystems.
In Brazil, there are at least 244 valid species of native stingless bees-and about 89 undescribed forms-placed in 29 genera [20]. Stingless beekeeping in Brazil, known as meliponiculture, is a traditional activity with significant potential both to assure plant biodiversity in many natural ecosystems and fulfill the growing agricultural demand for pollination, but remains as an informal activity needing technical knowledge [21]. Considering that there are 23 Brazilian stingless bees reported in official lists of endangered species [22], most of them being managed, the optimization of meliponiculture is urgent.
One of the most serious issues affecting meliponiculture is disease. Melipona quadrifasciata, the second most important stingless bee kept through meliponiculture [21], and listed as endangered in Southern Brazil [22], suffers from an annual syndrome that outbreaks in March, causing high mortality rates of adult bees and eventually colony losses [23]. Our previous studies suggest that the syndrome results from increased susceptibility to pathogens such as viruses, probably caused by poor nutrition and other sub-lethal environmental factors [24,25]. The goal of the present study was to characterize the changes in diet and the associated microbiota of forager bees throughout a season, in order to answer the following questions. (1) Is there a core microbiome of Melipona quadrifasciata worker bees? (2) How is it affected by diet? (3) Are there feedback loops between the microbiome and diet influencing the health of these stingless bees? By identifying some key microbiota and diet healthrelated factors, we aim to improve M. quadrifasciata colony management.

Experimental Setting
Temporal changes in the M. quadrifasciata-associated microbiota were assessed during 4 months (January, February, March, and April) of 2019. March is characterized not only by high mortality of M. quadrifasciata in Southern Brazil, but also by a reduction in colony temperatures and an increase in colony humidity, possibly reflecting changes in M. quadrifasciata behavior [25]. Three pairs of motherdaughter (MD) colonies (1, 2, and 3) were obtained by division during the summer of 2018 and kept in a small agricultural property in the municipality of Bom Principio (BP; 29°31′2.30′′S/51°17′29.00′′W) until next spring. Daughter colonies were then transferred to the vicinity of a secondary forest located in Porto Alegre (PA; 30°2′4.7292''S/ 51°13′3.5724''W). This experimental setting was also used to monitor changes in forager bee expression, whose data have already been published elsewhere [25]. Starting from 1 3 January 2019, five forager bees were sampled with an entomological aspirator from each of the six colonies every month (totalizing 120 bees), weighed, and stored at − 80 °C until DNA extraction.

DNA Extraction and Metabarcoding
Total DNA was purified with the DNeasy Blood & Tissue Kit (Qiagen) from individual abdomens, which were separated from bee bodies in aseptic conditions using sterile scalpels. Prior to the standard DNA purification protocol, a lysozyme treatment of abdomen homogenates was performed to break the cell walls of gram-positive bacteria. DNA concentrations were determined on a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific). PCR was performed using modified versions of the original 16S and ITS2 rRNA primers, in which an individual tag barcode of 8 nt was added to the 5' end. Primers for the 16S target were 341F (V3 region; CCT ACG GGNGGC WGC AG) and 805R (V4 region; GAC TAC HVGGG TAT CTA ATC C), generating an amplicon of about 550 bp [26], and for ITS2, ITS3_KYO2 (GAT GAA GAA CGY AGY RAA ), and ITS4 (TCC TCC GCT TAT TGA TAT GC) generating 300-400 bp amplicons [27].  Table 1). PCR included an initial denaturation at 95 °C for 5 min, followed by 20 cycles of 94 °C denaturation for 30 s, 55 °C annealing for 30 s (with a touchdown of 0.5 °C every cycle), and 72 °C extension for 20 s, and another 15 cycles with an annealing temperature of 45 °C, with a final extension step at 72 °C for 10 min. Purified amplicons from all samples were pooled for building 16S and ITS2 libraries with the TrueSeq Nano Library Preparation kit (Illumina), and paired-end sequenced (2 × 300 nt) on separate Illumina MiSeq flowcells with the MiSeq Reagent kit v3 (Illumina). Sequencing data were deposited as a BioProject at NCBI under the accession PRJNA751106.

Sequencing Data Processing
Illumina reads were filtered by quality scores (Q30) and size (> 100 bp) with fastp 0.20.0 [28] and demultiplexed using Mothur v. 1.45.1 [29]. Forward and reverse pairedend reads were merged and assigned to their respective sample according to the tag barcodes (Supplementary Table 1). Sequences shorter than 250 bp, containing ambiguous bases, with homopolymer stretches longer than 15 bases or having mismatches in primer sequences, were discarded. For the 16S dataset, non-bacterial sequences were removed by performing a preliminary classification using the SILVA v138 nr database [30]. A similar procedure was performed for ITS2 using the UNITE v8.3 database with all eukaryotes [31]. Two separate datasets were built for ITS2, i.e., one including plants and another containing the fungi. In spite of having been designed to amplify the fungal ITS2 region, we confirmed in silico that primers ITS3_KYO2 and ITS4 [27] are adequate for barcoding pollen from the plant families previously identified in our colonies using morphological data [25]. This was done by mapping representative plant rRNA sequences against both primers using the "map to reference" tool in Geneious Prime v. 2021.1.1 (Biomatters). Chimeric sequences were removed with Mothur's implementation of UCHIME [32]. The final 16S and ITS2 datasets were obtained by discarding singleton sequences. Non-singleton sequences from the same sample were pooled. Finally, Operational Taxonomic Units (OTUs) were identified at 97% (16S) or 95% (ITS2) sequence similarity using the nearest neighbor algorithm (16S) or AGC clustering (ITS2) and classified with a confidence threshold of 80% with the SILVA (16S) or UNITE (ITS2) databases.

Relative Quantification of Bacteria
To compare bacterial load between samples, quantification of 16S rRNA genes was carried out relative to the amount of host bee DNA. Quantitative real-time PCR (qPCR) was performed using universal 16S rRNA gene primers 16S-F (AGG ATT AGA TAC CCT GGT AGTCC) and 16S-R (YCG TAC TCC CCA GGCGG) [33]  Reactions generating Ct values < 30 were determined to be positive results and melting curve analysis was used to confirm amplification specificity. Primer amplification efficiency was calculated from the slope of a five-point 1:10 serial dilution of pooled M. quadrifasciata DNA samples. The fold changes in bacterial abundance were calculated relative to the endogenous control (PO), taking the PCR efficiency for each target into account.

Statistical Analyses
After assigning taxonomy to OTUs, alpha-diversity estimators such as Shannon Diversity Index (SDI) and Pielou's Evenness (EVEN) were obtained with Mothur for each individual host on subsamples of 10,000; 1000; and 800 sequences from bacteria, fungi, and plants, respectively. Subsample sizes were stipulated in order to maximize the number of samples retained, while still capturing a high proportion of the total diversity in each sample. Rarefaction curves were generated with Mothur using 1000 randomizations. The number of OTUs retained for analyses (common OTUs) was determined using a cutoff of at least 10,000; 1000; and 800 sequences per OTU from bacteria, fungi, and plants, respectively, and pooling the less representative OTUs within a class of "rare OTUs." All statistical tests and plots were performed in R using RStudio v. 1.4.1106. Our codes and datasets are available at https://github.com/ klhaag/metabarcoding-melipona. Hypothesis testing was non-parametric, i.e., by Kruskal-Wallis tests, correcting for multiple comparisons with the Benjamini-Hochberg (BH) method, followed by Wilcoxon rank-sum tests. We used linear (Pearson's) models to evaluate whether microbiota alpha diversity, relative amounts of bacteria, and forager bee weight are correlated to each other.
Beta-diversity (Bray-Curtis distance) was used to identify whether microbiota and pollen diet species composition was associated with colony, MD colonies, locality, month, or colony health. Interactions between factor locality, month, and colony health on Bray-Curtis distances were also evaluated. Significance was calculated through PER-MANOVA with 10,000 permutations, using the "adonis" function from the vegan package (Oksanen et al. 2020). Shifts in beta diversity were visualized by non-metric multidimensional scaling (NMDS). To identify differentially abundant taxa, we calculated their relative abundances and tested for significant differences with the Kruskal-Wallis test correcting for multiple comparisons with the BH method. We used non-linear (Spearman) correlations to evaluate whether a particular microbiome taxon was associated with a particular pollen taxon in foragers' diets.

Phylogenetic Analyses
Representative sequences from the most common OTUs within our bacterial 16S and fungal ITS2 datasets were aligned with other sequences reported for the M. quadrifasciata microbiome in previous publications [16,23]. Multiple alignments were done with MUSCLE and then used for inferring phylogenetic relationships with FastTree, both implemented in Geneious Prime v. 2021.1.1 (Biomatters), resampling the alignment sites by bootstrapping, with 1000 iterations.

Results
We sampled 120 foragers from 3 pairs of mother-daughter colonies of M. quadrifasciata colonies maintained in two separated locations (BP and PA), from January until April of 2019. In March, two pairs of colonies manifested symptoms of the annual disease in both locations (PA2 and BP2, PA3 and BP3), with adult bees appearing dead in front of the hives, and BP colonies suffering heavier mortality than their daughter colonies from PA.
High throughput amplicon sequencing of 16S and ITS2 targets from these samples resulted in 4,747,544 and 4,509,510 reads, respectively (Table 1), which ended in 28,946 (16S) and 12,894 (ITS2) sequences per sample, on average. Table 1 shows the total number of sequences, number of unique sequences, and number of OTUs identified in our datasets. Rarefaction analyses show that the sequencing depth employed in our study is sufficient to represent the richness of bacterial, fungal, and pollen representative OTUs within samples ( Supplementary Fig. 1). The sample metadata and alpha diversity estimators are listed on Supplementary Table 1 Alpha Diversity, Bacterial Quantification, and Forager Bee Weight We found an overall increase in the amount of bacteria in the abdomen of forager bees between January and February (p = 0.0019; Fig. 1A). Between February and March, our forager bees became lighter (p = 0.0000; Fig. 1B) with a concomitant increase in average gut bacterial diversity (SDI; Fig. 1C; p = 0.0412) and evenness (EVEN; Fig. 1E; p = 0.0275). Bees from PA were lighter, on average, than those from BP (p = 0.0486). Curiously, there was a weak but statistically significant positive correlation between forager weight and bacterial SDI (r 2 = 0.2141; p = 0.0352). On the other hand, there was a negative correlation between SDI and bacterial quantities (r 2 = − 0.2703; p = 0.0167), suggesting that larger bacterial quantities per bee were not evenly distributed among all bacterial taxa. We did not observe statistically significant differences neither in bacterial alpha diversities between locations nor in fungal and pollen diversities between months or locations. Nevertheless, the average fungal diversity exhibited an opposite trend compared to bacteria, with a slight increase between March and April ( Fig. 1D and  E). Furthermore, the richness of the fungal community associated to forager bees was outstanding. We found 23 times more OTUs of fungi than for bacteria, many of them unclassified.

Beta Diversity
Bacterial and fungal microbiome compositions differed significantly between colonies, locations, and months during  Table 2; see also Supplementary Fig. 2). For the Fungi, we found significant differences between healthy and diseased, as well as between MD colony pairs, though only about 3% and 4% of the variation in fungal beta diversity are explained by the colony health status, or colony genetic relatedness ( Table 2). For bacteria, both the 2-dimensional and 3-dimensional NMDS plots (stress = 0.2330 and 0.1438, respectively) revealed a temporal trend in community composition change, as well as a slight differentiation between locations ( Fig. 2A and B; Supplementary Fig. 2). These results are further supported by PERMANOVA, where the factors "month" and "location" result in significant effects (Table 2). February represents a transitional period, since the pairwise comparisons between months showed no statistically significant differences between January vs. February, and February vs. March (Table 2). Moreover, combining two factors in a single PERMANOVA revealed an interaction between "month" and "disease," as well as between "location" and "disease" ( Table 2).
For the fungi, both the 2-dimensional and 3-dimensional NMDS plots (stress = 0.2208 and 0.1396, respectively) and PERMANOVA revealed significant effects at all levels ( Fig. 2C and D; Supplementary Fig. 2; Table 2), which may reflect the outstanding high richness of the fungal microbiota composition of M. quadrifasciata forager bees. For the pollen, the 2-dimensional NMDS (stress = 0.1434) shows that the diet of forager bees shifted between January and March  2E) and differs between locations (Fig. 2F). "Month" and "location" were significant factors shaping the betadiversity of pollen diets, but there was no significant difference between diseased and healthy colonies ( Table 2).

Relative Abundance of Taxa
Lactobacillaceae, represented by nine OTUs with variable abundances (Supplementary Fig. 3, Supplementary Table 2), and Bifidobacteriaceae, represented by a single OTU, were the most representative bacterial families in our bees, with median relative abundances of 0.384 and 0.271, respectively. Their amounts showed significant temporal variation (p = 0.0018), but with opposite trends (Fig. 3A). The relative abundance of Bifidobacteriaceae was significantly higher in PA (p = 0.0391), and of Acetobacteraceae was higher in BP (p = 0.0003).
Candida and Starmerella were the most representative fungal genera, with median relative abundances of 0.263 and 0.079, respectively. Their abundances varied significantly over time, with Candida showing a peak in January (p = 0.0264) and Starmerella in February (p = 0.0000), and both being more abundant in PA than in BP ( Fig. 3B; p = 0.0044 and p = 0.0448, respectively). The most abundant fungal OTUs were OTU1, identified as Candida apicola, and OTU2, identified as Starmerella cellae ( Supplementary  Fig. 3). Two fungal genera showed significant association with the colony health status, i.e., Zygosaccharomyces and Debaryomyces, the first being less abundant (p = 0.0292) and the latter more abundant (p = 0.0096) in diseased colonies. Both were more abundant in BP than in PA (p = 0.0030 and p = 0.0001, respectively), and Zygosaccharomyces showed the lowest abundance during the outbreak period (Fig. 3B).
The pollen of Eucalyptus and Eugenia was the most frequently found in the diet of our forager bees (median = 0.0725 and 0.0706, respectively), but while Eucalyptus was the most consumed pollen in March (p = 0.0008), Eugenia was less consumed during the outbreak than in the other three summer months (Fig. 3C). Eucalyptus pollen was also more consumed in BP (p = 0.0000), which is consistent with the fact that these trees are extensively cultivated in BP surrounding areas.

Correlations Between Pollen Diet and the Gut Microbiome
The abundance of Eucalyptus pollen in forager diets was negatively correlated with the presence of several bacterial taxa. The strongest correlation was found with OTU8 (ρ = − 0.5448; p = 0.0307), followed by OTU4 (ρ = − 0.3728; p > 0.05), OTU7 (ρ = − 0.3188; p > 0.05), and OTU6 (σ = − 0.3102; p > 0.05). Figure 4 shows all possible pairwise correlations between the representative pollen and bacterial taxa within our study. OTU8 and OTU4 are still unclassified Lactobacillaceae that were named as Firmicutes Group Y in our previous study on the M. quadrifasciata microbiota [23] (Fig. 5). OTU7 belongs to the genus Bombella (Acetobacteraceae; Supplementary Fig. 4) and OTU6 is a Lactobacillus named as Firmicutes Group Z in our previous study, which is also known as the A. mellifera Firm-5 phylotype (Fig. 5). Another significant negative correlation was found between the abundance of Eugenia pollen and bacterial OTU1 (ρ = − 0.5823; p = 0.0075), a Lactobacillaceae named as Firmicutes Group U in our previous study (Figs. 4 and 5).

Discussion
We observed striking changes in the associated microbiota composition concomitant to the loss of weight of forager bees, in accordance to findings obtained for other bees [34]. Furthermore, there was an uneven increase in the quantity of gut-associated bacteria 2 months before the outbreak. Together, these findings corroborate our previous proposal that the events culminating with the annual syndrome initiate earlier in the season [25].

A Putative Core Gut Microbiome of M. quadrifasciata Forager Bees
Five OTUs belonging to three bacterial families were represented in all samples within our study, suggesting that they represent the "core" bacterial microbiome of M. quadrifasciata. Together, OTUs 1, 3, and 5 (Lactobacillaceae); OTU2 (Bifidobacteriaceae); and OTU10 (Acetobacteraceae) comprised 74% of the total number of sequences (Supplementary Table 2). All nine Lactobacillaceae OTUs observed in the study are closely related to other 16S sequences previously found in the gut of M. quadrifasciata foragers, which were separated in four phylogenetic clusters named Firmicutes groups U, X, Y, and Z by Díaz et al. [23]. The core Lactobacillaceae belong to Groups U and X (known as Firm-5). Group X, represented by OTUs 3 and 5 in the present study, is widespread among the Apidae. Firm-5 phylotypes are known for their high degree of host specificity, but there is no evidence of microbiome-host codiversification [14]. Contrastingly, Group U, represented in our study by OTU1, has never been reported for other bees [23]. Our phylogenetic analysis reveals that this Melipona-specific clade is highly diverse, with several lineages displaying long branch lengths. Thus, we speculate that some of these lineages constitute separate species and may represent host-adapted phylotypes. We found a single highly prevalent Bifidobacteriaceae phylotype (OTU2), closely related to the bumblebee symbiont, Bifidobacterium communae. While Bifidobacterium is known to be core in all corbibulate bees, the Acetobacteraceae appear to be particularly prevalent in the Meliponini [14]. Among the five Acetobacteraceae OTUs identified in our study, OTU10 (Commensalibacter) was the most abundant and omnipresent, followed by OTU9 (unidentified) and OTU7 (Bombella). Contrasting with bacteria, for which a total of 364 OTUs have been inferred in our study, fungal ITS2 sequences were clustered into 8557 OTUs. There are difficulties inherent to fungal metabarcoding using ITS as marker sequence [35], due to its high amount of polymorphism. We used a lower threshold of 95% similarity for clustering fungal OTUs and opted to focus on the 16 more abundant OTUs representing about 72% of the fungal ITS2 sequences within our study, in order to discard nectar-associated fungi only transiently present in forager bee guts. Some of the highly abundant fungi identified in our samples are well known stingless bee symbionts, such as Candida apicola (synonym of Starmerella apicola), Starmerella spp., and Zygosaccharomyces [36][37][38]. In the Scaptotrigona depilis stingless bee, Zygosaccharomyces provides sterol precursors required for host metamorphosis [38], and its growth is controlled by volatile compounds produced by C. apicola [39]. Furthermore, we found significant differences in the fungi associated to forager bees between locations, suggesting a strong influence of the environment on fungi acquisition, similar to what has been found for the generalist small carpenter bee Ceratina australensis in Australia [18].

Temporal Changes in Diet and Gut-Associated Microbiota
We found a pronounced increase in the relative abundance of Bifidobacteriaceae in our samples between January and February, concomitant to a decrease in Candida. One month later, the relative abundance of Lactobacillaceae and Starmerella decreased. These changes in the gut microbiota occurred simultaneously with a switch in pollen diet. Between January and February, forager bees replaced native Myrtaceae such as Myrciaria and Eugenia in their diets by the pollen of Eucalyptus, an exotic tree in Brazil. Pollens differ in their fatty acid content, and Eucalyptus species are particularly poor in lipid concentration, with the majority below 2% [40], and the typical range in A. mellifera harvested pollen around 4.7% of dry weight [41].
In our study, the relative amount of Eucalyptus pollen in forager bee diets was negatively correlated with putatively core Lactobacillaceae OTUs, Bombella, and with C. apicola. It is known from studies with A. mellifera that monofloral diets with Eucalyptus pollen lead to impaired bee development due to lipid deficiency [42] and the loss of core Lactobacillaceae [43]. Furthermore, reduced abdominal lipid stores of worker bees induce the transition from nursing to foraging in honeybees [44]. Thus, the effects of diet on the associated gut microbiota might be direct, by altering the substrates available for gut bacteria, and indirect, by changing host physiology and development [45,46].
Noteworthy, the pollen stored within our bee colonies showed a similar abrupt switch between January and February, as reported in our previous publication [25]. However, the pollen switch in colony storages was from Myrtaceae to   Fabaceae. In the present study, Fabaceae pollen was identified in bee diets later in March. One explanation for this discrepancy between the pollen stored and the one effectively consumed by worker bees is that adult stingless bees may consume the stored pollen only after it undergoes significant microbially driven nutrient changes, as previously suggested [47]. If confirmed, such a behavior would contrast with A. mellifera, which prefers freshly stored pollen [48].

Diet and Microbiome Effects on Stingless Bee Health
Nutritional stressors affect the health of bees by intersecting at the gut microbiome [11,49]. Although we did not find significant differences in the composition bacterial associated microbiota of foragers from diseased colonies, PERMANOVA interactions suggest that the gut microbiota temporal changes differ between healthy and diseased colonies. Thus, by altering the gut microbiota composition, diet switches may have affected the health of our bees. Remarkably, our previous study showed that the transcriptomes of M. quadrifasciata diseased bees are depleted in transcripts related to integral membrane components [25]. Moreover, colonies that remained healthy during the outbreak period had a significantly higher expression of apolipophorin, involved in fatty acid transport, 3 months before the outbreak [25]. We do not rule out involvement of a bee genetic component in disease susceptibility, since MD colony pairs became diseased in both locations. However, we think that nutritional deficits resulting from monofloral diet on lipidpoor Eucalyptus pollen probably result in an altered lipid metabolism that has multiple physiologic consequences and increased susceptibility to disease, as previously observed for A. mellifera [43,50,51]. In addition, the underrepresentation of key symbionts as a consequence of monofloral Eucalyptus diet may render bees defenseless against pathogenic microorganisms. Besides modulating host immune responses, some gut bacteria such as Lactobacillus sp. [52], and Bombella apis [53], for example, show direct antimicrobial activity against bee pathogens.

Concluding Remarks
We showed that the pollen diet of M. quadrifasciata foragers changes dramatically during one season, and that such changes are correlated with the loss of putative core members of the stingless bee gut microbiota. It is unknown whether such changes are seasonal, or if they persist in the long-term, but it was shown for other stingless bees that some of the dietdriven microbiota changes have strong legacy effects [54]. Together with our previous studies, the present data suggest that the events culminating with annual M. quadrifasciata disease outbreaks begin earlier in the season and include pronounced diet shifts. Our findings raise concerns regarding the effects of Eucalyptus reforestation on Neotropical stingless bee populations, which we think could be attenuated by enforcing polyfloral diets and introducing supplementary diets in stingless bee management. Diets aiming for a supplementation of fatty acids, as well as those focused on probiotics enriched with Lactobacillus, would have a great potential in meliponiculture.