Spatial segregation and cross-kingdom interactions drive stingless bee hive microbiome assembly

Abstract

Studying host-associated microbiome assembly is key to understanding microbial and host evolution and health. While honey bee microbiome have been a central model for such investigations among pollinators, they overlook the diversity of eusocial dynamics and multi-kingdom interactions. Stingless bees—a diverse group of highly eusocial insects that includes managed species, varies in colony biology, and harbors a symbiotic yeast essential for larval development in at least one species—offer a valuable complementary system to study microbiome assembly under an eco-evolutionary context. Using amplicon sequencing, metagenomics, and microbial experiments, we investigate the drivers of microbiome assembly in stingless bee colonies. We reveal a spatially structured, site-adapted microbiome, where high microbial influx hive components are segregated from the brood, which harbors a stable, multi-kingdom community. We show that the brood microbiome is not only physically protected but also maintained through selective bacterial-fungal interactions and abiotic conditions shaped by bees and their symbionts, such as temperature and pH. Our findings uncover multi-layered mechanisms shaping eusocial superorganism microbiomes, from host biology to cross-kingdom interactions, while providing critical insights into microbiome maintenance of important pollinators.

Here, the authors characterize the microbiome assembly across an entire stingless bee colony, revealing site-adapted within-hive communities, and showing that nest architecture protects brood areas from microbial influx and fosters selective bacterial–fungal interactions essential for colony health and fitness.

Introduction

Microbiomes expand their host’s genetic repertoire, influencing digestion, immunity, behavior, and fitness^1–3. Such benefits often rely on specific and balanced microbial communities, making microbiome assembly and maintenance crucial, and potentially driving host-microbiome codiversification throughout evolution^4–7. Understanding the factors that shape microbiomes thus remains a central question in the field, with implications for uncovering the evolution of microbial and host features, predicting microbiome responses to environmental changes, and informing strategies for microbiome manipulation.

Multiple mechanisms shape microbiomes. For instance, host-to-microbe interactions influence community structure through gut compartmentalization⁸; microbe-to-host interactions secure key symbionts for diet digestion⁹; and microbe-to-microbe interactions support communities via metabolic cross-feeding¹⁰. The influence of mechanisms may vary across parts of an organism. For example, in plant microbiomes microbe-to-microbe interactions play important roles in rhizosphere and bulk soil microbiome assembly, while host-microbe interactions exert extra pressure in root and leaf endospheres^11,12. In social organisms, host behavior may also influence microbial transmission, and body sites with high microbial influx, like oral microbiomes, may further impact microbiome assembly¹³. Expanding studies to diverse systems is thus key to uncovering the mechanisms of microbiome assembly.

Eusocial bees are powerful models for studying microbiome assembly, especially within the corbiculate clade, which includes honey bees (Apini), bumble bees (Bombini), and stingless bees (Meliponini)^4,14. These tribes harbor conserved gut microbiomes shaped by reliable transmission during colony foundation and social interactions, promoting stable associations and the evolution of host-adapted symbionts^4,15. Across adult female workers, core bacterial taxa include Lactobacillus, Bifidobacterium, Bombilactobacillus, Snodgrassella, and Gilliamella^14,16,17. Some taxa show lineage-specific associations—Bartonella and Frischella in honey bees, Schmidhempelia in bumble bees, and Acetobacteraceae in stingless bees, with Gilliamella and Snodgrassella extinct in Melipona¹⁸. These core symbionts are not only adapted to the bee gut but can support host functions. For instance, in honey bees, Gilliamella, Lactobacillus, and Bifidobacterium help in carbohydrate metabolism¹⁹; in bumble bees Gilliamella and Lactobacillus are suggested protect against the common parasite Crithidia^20,21. In addition to bacteria, fungi—primarily Saccharomycetes—are also found in eusocial bee guts, though they are less studied, often considered transient, and their ecological roles remain unclear^22,23.

Beyond worker bees, a defining feature of eusocial insects is that fitness is measured at the colony level²⁴, with all hive components—subcastes, brood, spatial organization—contributing to function and reproduction. This colony-wide perspective makes corbiculate bees particularly informative. Although honey bees, bumble bees, and stingless bees share conserved gut symbionts, they differ in life history traits that likely contribute for microbiome assembly. For example, honey bees form big perennial colonies with workers showing age polyethism. Nurses (young workers taking care of brood and nest) host more Lactobacillus and Bifidobacterium, and foragers (older workers that collect food) are enriched in Gilliamella and Snodgrassella^25,26. Bumble bees, with smaller annual colonies and more flexible task allocation, exhibit homogeneous microbiomes across workers^27,28. Juvenile stages and other hive components are less studied in integrative analyses of entire colonies, yet they play critical roles in colony function and health. Bumble bee larvae develop over pollen masses and harbor gut microbes shared with adults²⁹. In contrast, honey bee larvae receive royal jelly, a highly antimicrobial diet³⁰ that restricts colonization to select taxa such as Bombella apis³¹, a bacterium known to inhibit fungal pathogens affecting honey bee brood³².

Among eusocial corbiculate bees, stingless bees remain the least studied, despite having unique traits that can illuminate host-microbiome ecology and evolution. As the most diverse eusocial bee lineage ( > 500 species³³), they form perennial colonies with complex caste systems—including nurse, forager and, in some species, morphologically distinct soldiers³⁴. Their hives are even more spatially organized than those of honey bees and bumble bees, with food stores completely separated from the brood³⁵, where larvae is mass provisioned. A single batch of pollen/nectar food with high water and sugar content³⁶ is deposited into the brood cell prior to oviposition and sealed—creating a microbe-rich substrate with minimal further bee interaction^37,38. In some species, such as Scaptotrigona depilis, it was already shown that larvae depend on consuming the yeast symbiont Zygosaccharomyces for development^39,40—the only known case in eusocial bees where a clearly non-transient fungus plays an essential nutritional role. Bacterial communities in larvae and provisions still remain poorly characterized, though one study suggests Lactobacillaceae predominance³⁸. Furthermore, like honey bees, stingless bees are managed for honey, propolis and pollination across the tropics, but suffer from population declines which underscore the urgent need for further research⁴¹.

Here, we investigate the ecological drivers of microbiome assembly in the highly eusocial stingless bee Scaptotrigona (S. depilis and S. bipunctata), species with hive organization representative of many stingless bees^33,35. These species harbors the well-documented yeast symbiont^37,40, serve as an established model for stingless bee biology and toxicity assays^42,43, and have developed methods for in vitro larval rearing³⁹, facilitating future studies. Using amplicon sequencing and shotgun metagenomics, we characterize bacterial and fungal communities across hive components—including food stores, brood, and subcastes—and isolate strains to experimentally uncover potentially important microbial interactions. Our results reveal a multi-layered mechanism of microbiome assembly, spanning host biology to cross-kingdom microbial interactions.

Results

Scaptotrigona hive microbiome is dominated by Lactobacillaceae and Saccharomycetaceae

Current knowledge of Scaptotrigona microbiome is limited to culture-dependent studies or amplicon sequencing of forager bees^37,44,45. To characterize its hive microbiome and determine if a consistent microbial community is associated with the bee genus or species, we performed 16S rRNA (bacteria) and ITS (fungi) amplicon sequencing on two species, S. depilis and S. bipunctata. Two hives per species from three meliponaries were sampled for various components (Fig. 1a), including food stores (pollen and honey), developing bees (larvae and larval diet), and adult bees guts (worker subcastes: nurse, forager, and soldiers). A total of 691 bacterial amplicon sequencing variants (ASVs) were recovered from 65 samples, and 910 fungal ASVs from 50 samples (Supplementary Data 1) that successfully passed all steps of sample processing. ASV richness and diversity varies significantly among hive components (Fig. 1b, Supplementary Data 2). Both bacterial (Kruskal-Wallis: χ² = 31.86, df = 6, p = 1.735 × 10^⁻5) and fungal (Kruskal-Wallis: χ² = 30.58, df = 6, p = 3.05 × 10^⁻5) microbiomes vary in richness. Honey samples and soldier bees, similar in age to foragers but tasked with nest defense, have the richest microbial communities. Diversity also differs significantly when comparing the fungal microbiome between hive components (ANOVA: F(6, 12) = 2.71, p = 0.033), but the differences were even more pronounced in the bacterial community (Kruskal-Wallis: χ² = 28.15, df = 6, p = 8.812 × 10⁻⁵), with adult bees exhibiting the highest diversity. Bacterial evenness is higher in adult bees, larvae, and larval diet (Kruskal-Wallis: χ² = 23.83, df = 6, p = 5.62 × 10^⁻4), despite lower richness and diversity in the latter.

Fig. 1. Multi-kingdom microbiome of Scaptotrigona stingless bee hive.

a Above view of Scaptotrigona hive with separate pots for pollen and honey storage on one side, and horizontal combs with developing brood on the other side, covered with an additional protective cerumen layer. As shown in the “Diet / brood comb”, each individual brood cell contains the complete diet for larval development, deposited by nurse bees before the queen lays the egg, and is immediately capped afterward. Image credit: Irmgard I. W. Caesar. b Alpha-diversity metrics of amplicon sequencing variants (ASVs) for bacterial (top) and fungal (bottom) communities across sample types (color key), including pollen (n = 8, 8), honey (n = 11, 7), larval diet (n = 10, 9), larvae (n = 7, 10), nurse (n = 12, 7), forager (n = 11,5), and soldier bees (n = 6, 4). Box plots show interquartile ranges, center lines indicate medians, and whiskers denote minima and maxima; letters indicate significant differences (BH-adjusted p < 0.05). Microbiome richness varied significantly (Kruskal-Wallis, Bacteria: χ²=31.86, df = 6, p = 1.735 × 10⁻⁵; Fungi: χ² = 30.58, df = 6, p = 3.05 × 10⁻⁵), as did diversity (Kruskal-Wallis, Bacteria: χ² = 28.15, df = 6, p = 8.812 × 10⁻⁵; ANOVA (two-sided), Fungi: F(6, 12) = 2.71, p = 0.033). Differences in evenness were only observed in bacterial communities across sample types (Kruskal-Wallis: χ² = 23.83, df = 6, p = 5.62 × 10⁻⁴). c Beta-diversity analysis shows bacterial community compartmentalization across hive components (PERMANOVA (one-way); Bray-Curtis dissimilarity: Df = 6, R² = 0.428, F = 7.45, p = 0.001), although Lactobacillaceae remain the dominant group across samples. d The fungal community was also different across sample types (PERMANOVA (one-way); Bray-Curtis dissimilarity: Df = 6, R² = 0.257, F = 2.48, p = 0.001), with larval diet and larvae samples particularly colonized by Saccharomycetes (bottom barplot).

When comparing bacterial microbiome composition (Fig. 1c), hive component is the most influential factor (PERMANOVA; Bray-Curtis dissimilarity: Df = 6, R² = 0.428, F = 7.45, p = 0.001), as shown by the distribution of samples in the non-metric multidimensional scaling analysis. Most hive components are dominated by Lactobacillaceae, with less abundant taxa contributing to community differences. For example, food stores also carry Acetobacteriaceae and Acinetobacter, while adult bees host Floricoccus, Bifidobacteriaceae, and Bombilactobacillus. Differences in microbiomes of some hive components, such as soldier bees and honey, are also explained by the meliponary (two-way PERMANOVA; Bray-Curtis dissimilarity: Df = 11, R² = 0.134, F = 1.34, p = 0.026, Supplementary Fig. 1), possibly due to constant environmental exposure. These two hive components also harbor the highest number of bacterial species and unique taxa, based on 97% amplicon identity cutoff (Supplementary Fig. 2, Supplementary Data 3).

The fungal community (Fig. 1d) also shows most significant differences among hive components (PERMANOVA; Bray-Curtis dissimilarity: Df = 6, R² = 0.257, F = 2.48, p = 0.001). Unlike the bacterial community, these differences are mostly explained by larvae and larval diet. While food stores and adult bees harbor a more variable fungal composition, larvae and larval diet microbiomes are predominantly comprised of Saccharomycetales, mostly Saccharomycetaceae. This group may include unclassified Zygosaccharomyces and Starmerella—known associates of Scaptotrigona bees²² and potentially underrepresented in public ITS databases. At species-level (97% identity cutoff), ITS amplicons were classified into 29 species in larval samples (including 14 Saccharomycetales) and 123 species in larval diet samples (including 39 Saccharomycetales) (Supplementary Fig. 2, Supplementary Data 3).

There are no differences for bacterial or fungal ASV composition between bee species, nor across hive components (two-way PERMANOVA; Bray-Curtis dissimilarity, Bacteria: Df = 6, R² = 0.061, F = 1.06, p = 0.4; Fungi: Df = 6, R² = 0.116, F = 1.18, p = 0.09). However, significant differences in bacterial community composition are observed when ASVs are clustered into species-level (PERMANOVA: Df = 13, R² = 0.465, F = 2.41, p = 0.001).

Segregation of bacterial species and limited strain sharing across hive components

To gain deeper insights into the microbial species and strains in a Scaptotrigona hive, we performed shotgun sequencing on sample pools from two hives of each stingless bee species within a single meliponary. The 28 metagenomes yielded 9–28 million paired reads. By mapping reads to SSU barcode regions, we see that the Scaptotrigona microbiome is primarily bacterial, with fungi present in much lower proportions (Supplementary Fig. 3)—thus, no metagenome-assembled genomes (MAGs) of fungi were recovered. We recovered 14 high-quality bacterial MAGs (Supplementary Data 4) after dereplicating at species level; representing prevalent bacteria identified in our amplicon data. Nine MAGs are from bacteria consistently associated with at least one hive component (coverage >1X, breadth >50%, to minimize false positives): Two Acetilactobacillus, two Bombilactobacillus, one Nicoliella (5 Lactobacillaceae); one Acinetobacter pollinis (Moraxellaceae), Bifidobacterium (Bifidobacteriaceae), Raoultella-like (Enterobacteriaceae), and Wolbachia pipientis (Ehrlichiaceae) (Fig. 2a). Except for Wolbachia, these MAGs represent novel species, potentially genus, with average nucleotide identity (ANI) ranging from 67% to 75% to their closest relatives. The Wolbachia MAG had 96% ANI with its closest relative, Wolbachia wLug. Combined with its detection only in guard bee guts (Fig. 2b, Supplementary Data 1), unexpected for a vertically inherited symbiont, this suggests Wolbachia reflects residual contamination from recent interactions with parasites. Inspection of eukaryotic contigs from samples with Wolbachia reads revealed only candidate non-bee DNA from wasps, ants, and moth sequences, although not in higher abundance than in samples without Wolbachia reads (Supplementary Data 5). Among MAGs in the same genera, the two Bombilactobacillus share 87% ANI, while the two Acetilactobacillus share 77.7% ANI. Their closest relatives—B. thymidiniphilus, Apilactobacillus apisalvae and Acetilactobacillus jinshanensis⁴⁶—have all been isolated from stingless bees⁴⁷.

Fig. 2. Bacterial species and strains in Scaptotrigona hives.

a Maximum likelihood phylogenetic analysis of high-quality metagenome-assembled genomes (MAGs), substitution model Q.yeast+I + G4, based on the alignment of 72 single-copy orthologs present in >80% of 501 reference genomes. MAGs from Scaptotrigona hives are highlighted in pink, with the nine species consistently associated with hives (in >1 sample, >1× coverage, and >50% breadth) shown in bold and included in subsequent analyses. Dots indicate nodes with >95% bootstrap support. b Proportions of bacterial species in hive components for each species, shown in order: two S. bipunctata and two S. depilis samples. To avoid misleading interpretations, for larval microbiomes—which are only minimally colonized—coverage proportions refer to the proportion to all MAGs mapped (Supplementary Fig. 4). c Differences in the percentage of variable sites per bacterial species (Kruskal-Wallis χ² = 22.27, df = 8, p = 0.004), shown as boxplots with points representing individual samples. Box plots show interquartile ranges, center lines indicate medians, and whiskers denote minima and maxima; letters indicate significant differences (BH-adjusted p < 0.05). Hive components (colors) also show differences in strain diversity (Kruskal-Wallis χ² = 20.75, df = 5, p = 9.03 × 10⁻⁴). Acetilactobacillus sp. 1, n = 13; Acetilactobacillus sp. 2, n = 14; A. pollinis, n = 8; Bifidobacterium, n = 6; Bombilactobacillus sp. 1, = 3; Bombilactobacillus sp. 2, n = 4; Nicoliella sp., n = 7; Raoultella-like p., n = 4; W. pipientis, n = 2. d Bacterial species with strains shared at least once between sample pairs (e.g., sample A and sample B carry the same strain), with bars indicating the number of sharing events. e Acetilactobacillus sp. 1 strain-sharing network, highlighting hive components that host the bacteria (dots, triangles) and samples carrying the same strain (black line).

Bacterial species exhibit clear segregation within the hive (Fig. 2b), also corroborating findings from amplicon sequencing. Food stores and larval diet primarily host Lactobacillaceae, predominantly the phylogenetically derived strains Acetilactobacillus sp. 1 and Acetilactobacillus sp. 2 (each predominant in one bee species), as well as Acetobacteraceae and Acinetobacter pollinis. Larvae are minimally colonized by bacteria, with no MAG coverage above 1X depth or 50% breadth, and non-persistent hive strains likely reflect transient microbes (Supplementary Fig. 4). Following development, nurse bees, young workers tasked with feeding and caring for the brood, host Acetilactobacillus—especially species also abundant in larval diet—along with Bombilactobacillus and Bifidobacterium. Forager bees, older workers responsible for collecting nectar, pollen, water, and propolis, no longer carry Acetilactobacillus but maintain Bombilactobacillus and Bifidobacterium. Nest entrance-defending soldier bees are mostly colonized by Bifidobacterium. However, some also carry Enterobacteriaceae, such as Raoultella sp., related to opportunistic pathogens⁴⁸ and, as Wolbachia, may reflect transient environmental contamination acquired through interactions with hive intruders.

Among hive components, honey, pollen, and forager bees serve as the primary entry points for bacterial strains, showing the highest strain diversity, as indicated by the percentage of variable sites in MAGs (Fig. 2c; Kruskal-Wallis χ² = 22.27, df = 8, p = 0.004). Nicoliella has the higher strain diversity, followed by Acetilactobacillus sp. 1, both predominantly associated with food stores (Kruskal-Wallis χ² = 20.75, df = 5, p = 9.03 × 10^⁻4). The larval diet, a mixture of pollen and nectar/honey prepared by nurse bees, shows lower strain diversity compared to honey and some pollen samples. Despite overlapping ASVs and bacterial species, strain sharing is rare, mostly seen in food store samples (Fig. 2d, e, Supplementary Data 4), where, for example, the same Acetilactobacillus sp. 1 strain is detected across pollen, honey, and larval diet. This strain is also present in both Scaptotrigona species, suggesting priority effects may contribute to the dominance of bacterial lineages in each host.

Bacteria carry genes to handle synthetic compounds and degrade plant polysaccharides

To further understand microbiome assembly dynamics, we functionally annotated all bacterial contigs from each sample, alongside the eight MAGs consistently associated and representing novel bacterial species. The bacterial contig annotations confirm that larvae are minimally colonized by bacteria, with the total length of bacterial contigs in larvae samples ranging from 0.01 to 0.05 Mb, while adult bees range from 2 to 13 Mb and food stores from 14 to 33 Mb. Fewer than three proteins from larvae samples were classified into functional groups, leading us to exclude those samples from further analysis (Supplementary Data 6). Hive components differ significantly in the number of predicted proteins from bacterial contigs (ANOVA: F(5,18)=28.91, p = 5.28×10⁻⁸). The predicted proteins were classified into KEGG Orthology (KO) groups and organized into BRITE functional hierarchies. On average, 45% ( ± 6%) of proteins were assigned to KOs, with over 99.5% of these KOs linked to BRITE categories. The overall number of bacterial proteins positively correlates with expanded functional capacity, indicated by a greater number of unique KOs (Pearson correlation: r = 0.56, p = 0.01; Fig. 3a). However, this correlation is not significant when we include diet samples (Pearson correlation: r = 0.33, p = 0.12), which have high protein counts but low KO classification (around 40%; Supplementary Data 6)—suggesting bacteria, potentially the abundant Acetilactobacillus and related species, carrying novel functions in this environment.

Fig. 3. Functional repertoire of bacteria across Scaptotrigona hive components.

a Number of proteins and non-redundant KEGG Orthology (KO) terms predicted from bacterial contigs in each sample per hive component (color) and species (shape). A positive Pearson correlation between protein count and number of unique KOs was observed only when larval diet samples (highlighted in grey) were excluded due to their lower KO assignment rates (Supplementary Data 6). b Non-metric multidimensional scaling (NMDS) analysis of samples by species (shape) and type (color) based on normalized coverage of BRITE categories using Bray-Curtis dissimilarity (PERMANOVA (one-way): Df = 5, R² = 0.205, F = 7.98, p = 0.001) and presence/absence using Sørensen distances (PERMANOVA (one-way): Df = 5, R² = 0.082, F = 3.56, p = 0.001). c Normalized coverage of BRITE categories per sample, displaying the top 15 most abundant categories across all samples (Supplementary Data 7) and the top 10 categories enriched in guard and honey samples (Wilcoxon rank-sum test (two-sided), * = BH-adjusted p < 0.05; Supplementary Data 2). Coverage values are row-scaled for better visualization. d Number of genes involved in different carbon metabolism pathways across the six most prevalent bacterial species in Scaptotrigona hives, with columns clustered based on Euclidean distance. Colored circles at the bottom indicate the hive components where these species are most prevalent—following color key in (a, b).

Comparisons of functional categories show that larval diet, nurse bees, and pollen have some overlap, while honey, forager, and soldier samples exhibit more distinct functional profiles (Fig. 3b, Supplementary Data 7). Indeed, honey and soldiers have more unique KOs and are the only hive components with enriched BRITE categories (Supplementary Data 2), but category abundance contribute more strongly to hive component microbiome differentiation (PERMANOVA; Bray-Curtis dissimilarity: Df = 5, R² = 0.205, F = 7.98, p = 0.001) than presence/absence patterns alone (PERMANOVA; Sørensen dissimilarity: Df = 5, R² = 0.082, F = 3.56, p = 0.001). Soldiers, for example, show enrichment in bacterial infectious disease proteins and ansamycin biosynthesis (Wilcoxon rank-sum test, adjusted-p = 0.04, fold-change = 1.75; adjusted-p = 0.04, fold-change = 7.77) (Fig. 3c), which can be from pathogens or being used against them. Honey microbiome is enriched in proteins related to antimicrobial compound biosynthesis, such as phenazines and monobactams (Wilcoxon rank-sum test, adjusted-p = 0.03, fold-change = 4.22; adjusted-p = 0.01, fold-change = 4.29), with potential roles in competition and spoilage prevention. Honey is also enriched for proteins involved in the degradation of synthetic compounds, such as nitrotoluene, fluorobenzoate, and polycyclic aromatic hydrocarbons (Wilcoxon rank-sum test, adjusted-p = 0.02, fold-change = 6.51; adjusted-p = 0.01, fold-change = 5.35; adjusted-p = 0.01, fold-change = 6.32). The hive microbiome overall shows relative high abundance of genes for carbohydrate and amino acid metabolism, as well as cofactors and vitamins metabolism (Fig. 3c). For bacterial species prevalent in each hive component, we additionally predicted carbon metabolism (Supplementary Data 8). Acetilactobacillus and Nicoliella, lactic acid bacteria (e.g., D-lactate dehydrogenase), are abundant in food stores and may contribute for spoilage control through fermentation. In contrast, Bifidobacterium and Bombilactobacillus, which are abundant in adult bees, carry more genes for plant polysaccharide digestion (e.g., glycoside hydrolases, polysaccharide lyases targeting pectin), potentially contributing to bee nutrition (Fig. 3d).

Cross-kingdom interactions help shape the characteristic brood microbiome

The functional analyses revealed bacterial adaptations to hive components, including potential roles in adult bee metabolism. In non-bee hive components, bacteria likely persist due to their ability to thrive in high-sugar and fermented environments (Fig. 3d). However, we described a consistent bacterial, but also fungal community, especially in larval diet (Fig. 4a)—where microbial interactions may play a significant role on assembly once there is no bee innate immune system or extra host selective pressures. We thus wondered how bacteria interact with fungi in this hive component, especially since S. depilis depends on the yeast Zygosaccharomyces sp. for its development⁴⁰. The dominant bacteria in the larval diet are related to taxa with known antifungal roles^49,50—suggesting that bacteria in stingless bee larval diet may not inhibit fungi or may exhibit selective interactions. To investigate such interactions, we used multiple media to isolate microbes directly from the larval diet and conducted in vivo and in vitro experiments. Along the isolations process, we successfully isolated three bacterial species for which single representative strains were tested (Supplementary Data 1): two Lactobacillaceae (Apilactobacillus sp. and Weissella sp.) and one Acetobacteraceae (Bombella sp.). Both Lactobacillaceae had ASVs present in the larval diet, with Apilactobacillus closest to the dominant taxa (Supplementary fig. 5). Bombella is rare but belongs to a genus with known antifungal symbionts in honey bee larvae³². Our competition assay shows that all strains, in high or low abundance, inhibit the honey bee pathogen Aspergillus (Fig. 4b)—confirmed to be potentially pathogenic in our honey bee larval assays (Supplementary Fig. 6). Apilactobacillus sp. exhibits the strongest inhibition, completely preventing both mycelia growth and sporulation. This left us with a conundrum: if these bacteria share the larval diet with symbiotic fungi, do they also inhibit their growth in vitro?

Fig. 4. Competition assays between bacterial strains isolated from stingless bee brood and fungi, including fungi isolates from honey bees.

a Growth of Zygosaccharomyces in Scaptotrigona depilis brood combs brought to the laboratory and after egg removal, allowing yeast proliferation normally suppressed by larval consumption post-hatching. Image credit: Irmgard I. W. Caesar. b In vitro competition assays between S. depilis larval diet bacterial isolates and the honey bee pathogen Aspergillus (n = 3). c Competition assays between the same bacterial isolates and fungal species isolated from the same S. depilis brood samples (n = 3 each). d Competition assays between the same bacterial isolates and fungal species isolated from Apis mellifera larvae (n = 3 each). b–d The control condition (grey) shows fungal growth in the absence of a bacterial lawn (15 μL MRS + 100 spores or 5 μL yeast at OD600 = 0.15). Other conditions (green, orange, purple) show fungal growth in the presence of a bacterial lawn (15 μL of 10⁸ CFU/mL). For filamentous fungi, bars show mycelial diameter and lines indicate spore counts; for yeast, bars show OD600 of PBS-diluted culture—all measured at the experiment’s end, when controls had sporulated or grown. All comparisons showed significant differences (Kruskal–Wallis, p < 0.05). e Light microscopy images of Zygosaccharomyces sp. isolates from stingless bees and honey bees growing on an Apilactobacillus sp. lawn—only the stingless bee brood symbiont forms pseudomycelia, consistently observed across replicates (n = 10). Cyan scale bar = 0.25μm.

We next tested whether bacteria from the S. depilis larval diet could inhibit fungi isolated from the same environment—including core taxa and potentially transient fungi. For this, we isolated and used single-strain representatives of the core Saccharomycetaceae Zygosaccharomyces sp. (Scaptotrigona symbiont; Supplementary Fig. 7) and Starmerella sp., and filamentous fungi Talaromyces sp. and Penicillium sp. (Supplementary Data 9)—only the latter showed potential pathogenicity in larval assays (Supplementary Fig. 5). Our results show that while both Lactobacillaceae strains strongly inhibited mycelial growth and all three bacterial strains inhibited sporulation across all fungi tested, only Bombella specifically inhibited Zygosaccharomyces growth (Fig. 4c). Interestingly, Apilactobacillus sp. does not inhibit Zygosaccharomyces sp. but promotes its growth, also accelerating pseudomycelium formation (Fig. 4e) compared to controls without a bacterial lawn—structure suggested to support its floating in larval diet (Fig. 4a)⁴⁰. Apilactobacillus sp. also promotes Zygosaccharomyces sp. growth even in the presence of Bombella sp. (Supplementary Fig. 8). Overall, similar results were also observed in the experiments using lower bacterial CFUs (Supplementary Fig. 9).

Apilactobacillus inhibits honey bee-associated fungi

To assess the spectrum of this selective antifungal role, particularly that of Lactobacillaceae, we used an eco-evolutionary framework and tested how these bacteria interact with fungi we isolated from Apis mellifera larvae, including a Zygosaccharomyces sp., a Starmerella sp., and the filamentous fungi Penicillium sp. and Mucor sp. (Supplementary Data 9). These fungi are not known to be essential for honey bee health and are generally considered transient colonizers²³. The Zygosaccharomyces sp. shares 96.7% 18S identity with the stingless bee brood symbiont, is more similar to Zygosacharomyces species found in other parts of stingless bee hives (Supplementary Fig. 7) and does not form pseudomycelia (Fig. 4e). In vivo assays further indicate that Penicillium is potentially pathogenic to larvae (Supplementary Fig. 6). Apilactobacillus sp. and Weissella sp. from Scaptotrigona inhibit the non-native filamentous fungi but not the Starmerella sp. Interestingly, although both Lactobacillaceae exhibited similar inhibition patterns in competition assays, only Apilactobacillus sp.—which promotes Zygosaccharomyces from S. depilis—inhibited the honey bee-associated Zygosaccharomyces (Fig. 4b), suggesting this selective interaction is not a general trait of Lactobacillaceae. As assays with Zygosaccharomyces sp. from stingless bees were conducted in 30 G media (the only medium supporting its growth), we repeated the assay with honey bee Zygosaccharomyces sp. in 30 G media and observed the same result (Supplementary Fig. 10). Following the same framework, we further performed inhibition assays with A. kunkeei and B. apis, bacterial symbionts of honey bee larvae (Supplementary Data 9). Both bacteria inhibited all filamentous fungi (Supplementary Fig. 11), confirming their potential defensive role in honey bees, and also inhibited most yeasts, including the honey bee-associated Zygosaccharomyces sp. Interestingly, A. kunkeei showed a selective pattern similar to the Apilactobacillus sp. from S. depilis, promoting the stingless bee-associated Zygosaccharomyces sp. Altogether, these results suggest that inhibition by Bombella and especially Apilactobacillus may represent ancestral traits, and the ability of Apilactobacillus to inhibit one Zygosaccharomyces species while promoting another implies a possible counter-strategy by the yeast symbiont to evade bacterial inhibition.

Abiotic factors, including bacteria-driven, contribute to a yeast rich larval diet

As part of the bacteria-fungi competition assays, we also accounted for brood temperature differences between bee species. The two S. depilis yeast species tested grew better at ~30 °C⁴²—typical of Scaptotrigona brood—compared to ~34 °C⁵¹ for A. mellifera brood (yeast-only controls; Fig. 4, Supplementary Fig. 11). While other abiotic factors may also influence fungal growth, our assay design is particularly prone to bacterial effects on nutrient availability and pH through media consumption and fermentation—we tested these variables. To mimic reduced nutrient availability, we cultured fungi in 50% diluted medium. This condition did not inhibit filamentous fungi and in some cases increased spore production; yeasts, including both Zygosaccharomyces, were only mildly affected (Supplementary Fig. 12). Also, all three Lactobacillaceae strains (stingless bees: Apilactobacillus sp. and Weissella sp.; honey bees: A. kunkeei) acidified the media to pH<5 (Supplementary Fig. 13). To test the effect of acidification, we cultured fungi in media adjusted to pH 4.2, matching the lowest pH observed. Acidic conditions inhibited all filamentous fungi, while yeast growth was largely unaffected; Zygosaccharomyces isolates from both stingless bees and honey bees were even slightly promoted (Supplementary Fig. 12). These results suggest that bacterial acidification of the larval diet likely contributes to filamentous fungi inhibition and, alongside brood temperature, promotes yeast proliferation. However, despite both Zygosaccharomyces species growing slightly better under acidic conditions, their contrasting responses to Apilactobacillus—one promoted, the other inhibited—indicate that additional factors beyond abiotic conditions contribute for those interactions.

Discussion

Microbiome assembly and maintenance are essential for hosts that rely on microbial partners. In eusocial insects, however, direct fitness is measured at the colony level²⁴, so understanding microbiome dynamics across the entire hive is critical. The corbiculate clade of eusocial bees, particularly honey bees and bumble bees, have been serving as important systems for such investigations, although stingless bees—the most species-rich and biologically diverse tribe of the group⁵²—remain understudied. Here, we present the first comprehensive microbiome characterization across multiple components of a highly eusocial colony, focusing on stingless bees (Scaptotrigona spp.). Using amplicon sequencing, metagenomics, and experiments with representative microbial isolates, we show that: (I) Scaptotrigona hives hosts a consistent microbiome, dominated by Lactobacillaceae and Saccharomycetaceae; (II) the microbiome is spatially structured, with novel taxa showing signatures of local adaptation and high-microbial influx components segregated from the protected brood; and (III) in the brood, the larval diet—a mass provision unique among eusocial corbiculates—is colonized by a Saccharomycetaceae community maintained by brood comb features (e.g. temperature and bacteria-driven acidification) and potential selective interactions with locally abundant bacteria. Together, these findings reveal multi-layered mechanisms—spanning host biology and microbial cross-kingdom interactions—that structure colony-level microbiome assembly and maintenance (Fig. 5).

Fig. 5. Multi-level mechanisms in stingless bee microbiome assembly.

Many mechanisms likely shape colony microbiomes, and based on our results, we discuss at least two key ones in stingless bee hives: host biology and cross-kingdom microbial interactions—both potentially coevolving in a way that promotes and protects one of the most important components of a social insect colony, the developing brood and associated microbiome. Our findings of microbial segregation within spatially distributed hive components (e.g., food stores and subcastes) suggest greater opportunities for local adaptation as well as better control of microbial influx. Hive components apart from the brood combs show high bacterial strain diversity (see Food Stores: Honey, pink gradient) and carry many transient bacteria and fungi (see Subcastes: Soldiers), potentially pathogenic. Below all the physical layers protecting the brood, our results also reveal a consistent community dominated by Lactobacillaceae and Saccharomycetaceae, which is not only assembled as a consequence of the hive biology (e.g. temperature), but also by bacteria-driven acidification of the diet and potentially by selective bacteria-fungi interactions. Our in vitro assays with single representative strains from the larval diet suggest that bacteria can potentially promote the growth of the beneficial yeast symbiont while inhibiting other fungi, including potential pathogens. Illustration(s) by Julie Johnson (Life Science Studios).

Social interactions can homogenize microbiomes through reinoculation, as seen in group-living spiders⁵³ or wild lemurs⁵⁴. In eusocial insects, hive connectedness driven by nest architecture, worker movement and care behaviors may create similar patterns. While bumble bee workers generally exhibit homogeneous gut microbiomes^27,28, honey bee workers show greater differentiation, largely attributed to age polyethism²⁶. In Scaptotrigona spp., our amplicon sequencing and metagenome-assembled genome analyses also shows subcaste-specific microbiomes: nurse bees gut hosts a core community of Acetilactobacillus, Bombilactobacillus and Bifidobacterium, while foragers and soldiers exhibit more diverse profiles and a marked decrease in Acetilactobacillus. Because foragers and soldiers are possibly workers of same age³⁴, those differences we observe—such as a reduction in all Lactobacillaceae in soldiers—are more likely a consequence of task behavior rather than age, a potential confounding factor in many studies of honey bee microbiomes. Characterizing nest components, we observe more examples of microbiome segregation. Unlike the connected frames and wax pots of honey and bumble bees—where food stores and open brood are closely associated—stingless bees build highly compartmentalized hives, with individual pots for pollen and honey that are physically separated from the well protect brood (Fig. 1a)³⁵. Consistently, while honey bees and bumble bees tend to share similar microbial communities between food stores and brood^29,55, Scaptotrigona shows more marked compartmentalization: Acetilactobacillus and other Lactobacillaceae dominate honey and pollen, with pollen further enriched in Acetobacteraceae and Acinetobacter, while larvae are minimally colonized. In contrast, the larval diet—mass provisioned—harbors a less rich Lactobacillaceae community alongside Saccharomycetaceae, forming a consistent multi-kingdom assemblage not yet observed in other eusocial corbiculate bees.

Our results show that bacterial genomes we recovered exhibit signatures of local adaptation—for example, bacteria in the larval diet encode genes for tolerance to high-sugar and acidic substrates, while those in adult bees carry genes involved in the digestion of plant-derived diets. Similarly, yeasts that are particularly predominant in the larval diet grow better at the Scaptotrigona brood temperature and low pH typical of larval diet^36,42. All these results go in line with similar observations on genes carried by related bacterial taxa in honey bees and bumble bees—particularly supporting sugar and pollen digestion¹⁹—or bacteria that can withstand royal jelly³¹. But, beyond these genomic and physiological adaptations, spatial compartmentalization also appears to play a key role in shaping colony microbiome assembly. Compartmentalization has been proposed to contribute to microbial control by rewarding cooperative symbionts, preventing undesirable invaders, and reducing direct conflict among microbial taxa⁸. Our amplicon and strain-level analyses identify, for example, honey stores and soldiers—both well appart from central hive parts, such as the brood—as components with high microbial influx. The repeated evolution of a soldier subcaste in stingless bees, likely to counter hive invaders³⁴, also aligns with our finding of higher bacterial richness in this subcaste. Environmental microbes, including the potential pathogen Raoultella-like sp.⁴⁸, were mostly found on soldiers. Additionally, stingless bee brood combs are enclosed by a waxy involucrum, and brood cells are sealed after egg-laying over the larval diet—which likely represent an important evolutionary adaptation linked to their mass provisioning strategy³³, unlike the progressive provisioning of honey bees and bumble bees. Similar patterns of spatial structuring have evolved across diverse systems. In both bees and humans, gut tissues have diversified in morphology, physiology, and immunity to create compartments that filter and shape microbiomes^56–58. Analogous structures, like legume root nodules⁵⁹ and insect bacteriocytes⁶⁰, house symbionts in protected, host-regulated niches. Ants offer another interesting example among eusocial insects hives. For instance, Azteca ants nests show strong spatial microbial organization, with distinct communities in worker and brood chambers, such as ant brood chambers having lower levels of pathogenic fungi and lower levels of bacterial richness overall⁶¹.

All this protective layers potentially facilitated the establishment of Zygosaccharomyces symbiosis in S. depilis³⁹. Our results, further confirmed with culture-independent data the enrichment of larval diet by a community of Zygosaccharomyces and other Saccharomycetaceae, but also a conserved bacterial community—already suggesting cross-kingdom interactions through shared substrates. Lactobacillaceae such as Acetilactobacillus, Nicoliella, and Apilactobacillus are among the most abundant core taxa (Supplementary Fig. 5). Isolating these microbes can be challenging—the Zygosaccharomyces symbiont grows only on sugar-rich selective media and requires over 15 days to form colonies, whereas other yeasts grow within 1 day. Additionally, key bacterial taxa such as Acetilactobacillus and Nicoliella have not been yet cultured³⁸. We successfully isolated from larval diet Apilactobacillus, a strain closely related to the most abundant bacteria and showed that it selectively inhibits non-native and non-core fungi while only promoting the growth of the symbiont Zygosaccharomyces. While we tested only single strains of representative species, further studies are needed to determine the directionality of these interactions, assess their consistency across additional strains and species, and clarify their ecological relevance during larval diet maturation and development. That Weissella, another hive-associated Lactobacillaceae, does not promote the symbiotic yeast suggests this interaction is not a general trait of the family and is unlikely driven by media acidification, as both symbiotic and non-symbiotic Zygosaccharomyces grow in low pH. Targeted studies on Acetilactobacillus and Nicoliella—closely related to Apilactobacillus—may help clarify whether these selective interactions are shared within the clade. Apilactobacillus also only weakly inhibited Starmerella, a yeast that promotes Zygosaccharomyces growth⁶², suggesting additional ecological interactions may be involved. Moreover, we only characterized the microbiome of fresh larval diet; sampling across larval development could provide a more comprehensive view of microbial community dynamics.

Selective cross-kingdom microbial interactions have also evolved in other hosts, such as attine ants, where Pseudonocardia—originally a gut symbiont⁶³—became a cuticular mutualist that protects the fungal cultivar by inhibiting fungal pathogens⁶⁴. These microbes are vertically transmitted, promoting codiversification between ants and their symbionts⁶⁵. In stingless bees, the swarming behavior—unique among eusocial corbiculates and involving the transfer of nest material by workers—may similarly enable vertical transmission and contribute for the establishment of hive-level symbionts. Although previous studies found no clear 18S phylogenetic clustering of Zygosaccharomyces by bee species³⁷, our bacterial data suggest that microbiome codiversification may be occurring. For example, Scaptotrigona species harbor distinct bacterial species profiles, and each hosts a different dominant Acetilactobacillus. Future studies using high-resolution genomic approaches will be important to test this hypothesis and to identify potential genomic signatures in the Zygosaccharomyces symbiont that support positive interactions with antifungal bacteria.

Our findings support the view that microbiome traits must be understood by considering microbial function, transmission, and evolution in the context of host biology and behavior⁵⁷—all of which appear to shape hive-level microbiome assembly in stingless bees. This work also holds applied significance since stingless bees are key pollinators in tropical ecosystems, yet face increasing threats from habitat loss, novel pathogens, agrochemicals, and climate change⁴¹. A deeper understanding of their symbionts can help guide conservation and management efforts to support pollination and ecosystem health.

Methods

Sampling from managed stingless bee hives

Samples from two hives of S. depilis (SD) and two of S. bipunctata (SB) were collected from three locations in Rio Grande do Sul, the southernmost state of Brazil: BP (29.4926° S, 51.3550° W), BG (29.1667° S, 51.5170° W), and SC (29.5911° S, 51.3761° W). Sampling was carried out in April 2021, under SISGEN collection permission no. AE1AFA6. The hives were maintained in meliponaries alongside other stingless bee species, without interventions, such as the use of antibiotics, diet supplementation, or heaters. Two hives per species, which have Zygosacharomyces growing in the brood cells, were sampled for honey (from 3 pots), pollen (from 3 pots), larval diet (from 1 egg comb), 3rd instar larvae, nurses, foragers and soldiers. All samples were transported to the laboratory on ice and processed on the same day for microbial isolation or stored at −80°C for subsequent DNA extraction.

Amplicon and metagenome sequencing

DNA was extracted from each sample using the PureLink™ Genomic DNA Mini Kit (Thermo Fisher, K182001), following the manufacturer’s protocol with an additional initial lysis step using Lysozyme (Thermo Fisher Scientific, 89833). Honey (3 mL) and larval diet (300 µL) samples were first diluted with 1 volume of 1X PBS, vortexed, then centrifuged at 4500 × g for 10 min, and the pellet used for DNA extraction. For adult bees, individuals were surface-sterilized with bleach and water, then the guts of 7 individuals were dissected using sterile forceps and pooled. Similarly, 7 entire 3rd instar larvae surface-sterilized were pooled for DNA extraction. DNA quality was assessed using the Genomic DNA (gDNA) ScreenTape assay on an Agilent 2200 TapeStation (DIN > 7 cutoff). DNA concentration was measured with the Qubit dsDNA HS kit (Invitrogen) on a ThermoFisher Qubit4 fluorometer. To control for possible contaminants, we used blank extractions and sterile water as negative controls during the PCR.

PCR amplification was performed using 16S rRNA and ITS primers from the Earth Microbiome Project⁶⁶, with the Illumina adapter, a primer pad, and a linker added to the 5’ end, and a barcode added to one primer direction. The V4 region of the 16S SSU rRNA was targeted using primers 515 F (GTGYCAGCMGCCGCGGTAA) and 806 R (GGACTACNVGGGTWTCTAAT), generating an amplicon of ~390 bp^67,68. For the ITS region, primers ITS1 (CTTGGTCATTTAGAGGAAGTAA) and ITS2 (GCTGCGTTCTTCATCGATGC) were used, generating an amplicon of ~260 bp^69,70. PCR reactions were performed in triplicate, with each 25 µL reaction containing 2 µL of template DNA (100 ng total), 12.5 µL of 2X Phusion® High-Fidelity PCR Master Mix with HF Buffer (NEB, M0531L), and 0.75 µL of each primer at 10 µM. For 16S rRNA, the PCR conditions included an initial denaturation at 98 °C for 3 min, followed by 30 cycles of denaturation at 98 °C for 20 s, annealing at 71 °C for 30 s, and extension at 72 °C for 45 s, with a final extension at 72 °C for 10 min. For ITS, the annealing was performed at 54 °C for 35 s, and the extension at 72 °C for 7 s. The triplicates were pooled, and the amplicons were first checked on a 1.5% agarose gel, then quantified using the Quant-iT™ PicoGreen™ dsDNA Assay Kit (Invitrogen, P7589) before pooling. The 16S and ITS libraries were prepared using the TrueSeq Nano Library Preparation Kit (Illumina) and paired-end sequenced (2 × 250 nt) on an Illumina MiSeq flowcell with the MiSeq Nano Kit v2 (500 cycles).

The DNA samples from BP meliponary were also used for shotgun metagenome sequencing. Libraries were prepared using the Illumina DNA Prep kit and IDT 10 bp UDI indices and sequenced on an Illumina NextSeq 2000, producing 2 × 151 bp reads. Demultiplexing, quality control, and adapter trimming were performed with bcl-convert (v3.9.3) from Illumina. Raw shotgun sequencing data can be found within the NCBI BioSample accessions: SAMN46439779, SAMN46439780.

Amplicon data and statistical analyses

Illumina fastq files were demultiplexed, barcodes removed and concatenated into a separated barcode files directly after sequencing. The following analyses were conducted with DADA2⁷¹. As part of the read quality control step, 16S rRNA reads were filtered other than the default using truncLen=c(200,225) and minLen=175, and ITS reads with truncLen=c(190,140), minLen=100. After reads were dereplicated, inferred ASVs, merged, and chimeric ASVs were removed. The database SILVA v.138.1 (silva_nr99_v138.1_train_set.fa and silva_species_assignment_v138.1.fa) and UNITE v.9 (sh_general_release_dynamic_25.07.2023) were used for the taxonomic assignment of the ASVs, following DADA2 default. Contaminants in the samples found in the negative controls were removed using decontam R package. We additionally removed any ASVs identified as mitochondria or chloroplasts. The ASVs and related data can be found in Supplementary Data 1. Additionally, ASVs were clustered at species level (97% identity cutoff) using CD-HIT⁷² with flags cd-hit-est -c 0.97 -n 10 -d 0, and counts of clustered ASVs were summed to describe species present ( ≥ 5 reads) in each hive component.

All statistical analyses, which primarily focus on ASVs, and figure plots were performed in R version 4.3.1. To describe alpha diversity in the samples we use the functions estimate, diversity and specnumber from the package Vegan⁷³. To test for differences between samples and groups, the estimates with normal distribution according shapiro.test and variances are equal according bartlett.test were compared using t test or ANOVA. In case of no normal distribution and/or variances significantly different, we used Wilcoxon rank sum test or Kruskal-Wallis rank sum test. Following statistically significant ANOVA or Wilcoxon rank sum test we used TukeyHSD or pairwise.wilcox.test as post hoc tests. To investigate beta diversity, we first calculated Bray-Curtis and Sørensen dissimilarity dissimilarities and a non-metric multidimensional scaling (NMDS) using vegdist and metaMDS function from Vegan. We also used these dissimilarity matrices to carry out PERMANOVA (999 permutations) with function adonis2. All the results were visualized and plotted with ggplot2 package⁷⁴.

Metagenome assembly and binning

Paired-end raw sequencing reads from the 28 samples from BP meliponary were first trimmed by length and quality using Trimmomatic v.0.36⁷⁵, with options LEADING:28 TRAILING:28 SLIDINGWINDOW:6:25 MINLEN:75, then assembled into contigs with metaSPAdes v.3.15.5 using default settings. Only contigs >1500 bp were used for the following steps. Contig coverage was recovered by mapping trimmed individual sample reads back to all contigs with Bowtie2 v.2.4.2⁷⁶, using the options --no-discordant --no-mixed. Next, we used Metabat v.2.15⁷⁷ for binning, with flag -m 1500, followed by CheckM v.1.2.2⁷⁸ to recover bin quality. Bins with >80% completeness and <5% contamination were dereplicated with dRep⁷⁹ dereplicate, --P_ani 0.95 --S_ani 0.95 --cov_thresh 0.85. No fungal bins were recovered, with 99% of bins having <5% completeness according to BUSCO⁸⁰, using fungi_odb10 (2024-01-08, number of genomes: 549, number of BUSCOs: 758) and none of those being confirmed as fungi according to best BLAST hits. Moving on with the bacterial bins, to recover their coverage per sample we used again Bowtie2 v.2.4.2⁷⁶ to map trimmed individual sample reads back to the bins in a single database. Samtools v.1.17⁸¹ was used for file transformation, the depth command with option -a and idxstats command were used to recover the number of reads covering each base and the entire contig. Bin average coverage was calculated as the number of mapped reads*150/bin length, and proportions (%) plotted. To estimate the approximate ratio of bacteria to fungi, trimmed reads per sample were mapped with Bowtie2 v.2.4.2⁷⁶ reads against SILVA SSU database v.138.1, which contains 2,224,740 SSU rRNA sequences. Samtools v.1.17⁸¹ was used for file transformation, and idxstats command to recover the number of reads per sequence contig of different taxa. Total number of reads mapping to bacteria, fungi and insecta was plotted.

Strain diversity metagenome-assembled genomes

To estimate strain richness of MAGs in each sample we used the command profile from inStrain v1.4.0⁸², with --database_mode flag, on the bam mapping files generated previously (See section “Metagenome assembly and binning”). As part of inStrain protocol, only MAGs with >1x and 50% breadth were considered, and only sites covered at least 5x and the allele frequency of at least 0.05 were counted as SNVs. The output with genome information was used to calculate the percentage ratio of the number of single nucleotide variants (SNVs) to the total length of the MAG covered (percentage of variant sites = number of SNVs/(MAG length*breadth)). To characterize the distribution of strains and sharing events we used compare command from inStrain, which conducts paired comparisons of popANI metric (i.e., average nucleotide identity among locations along the genome where no alleles are shared between either sample). A popANI value ≥ 99.99% when MAG in both samples being compared had >5x and >50% breadth was considered evidence for strain sharing.

Taxonomic assignment of MAGs and isolated strains

Contigs from each MAGs were used as queries in a BLASTx⁸³ against the nr database from NCBI (version November 11/10/2023). Resulted best hits guided the choice of species reference genomes to retrieve from NCBI and include in the phylogenetic analysis, in addition to outgroups. Single-copy orthologs were recovered with OrthoFinder v.2.5.4⁸⁴, aligned with MAFFT v.7.520⁸⁵, trimmed with trimAl option -automated1, and then concatenated for IQ-TREE v.2.2.0.3⁸⁶ maximum likelihood analysis, options -m TEST -bb 1000 (where -m TEST invokes the built-in ModelFinder to select the best-fit substitution model according to the Bayesian Information Criterion (BIC)). Phylogeny summary statistics are available in the figure captions and detailed in Supplementary Data 10. The website iTol was used for tree visualization. Similarly, phylogenies from Sanger sequenced amplicons of isolated strains (see below, “Bacteria and fungi isolation and taxonomy”) only skip the single-copy orthologs recovery step. As an additional step for confirming MAG taxonomy, GTDB-Tk v2.1.1⁸⁷ was used with default parameters.

Microbiome and MAG encoded functions

Contigs from each sample had coding sequences (CDS) predicted with Prodigal through Prokka v.1.14.6⁸⁸, using default options. Diamond v2.1.7.16⁸⁹ was then used to blast the CDS against the nr database from NCBI (version November 11/10/2023), using --max-target-seqs 1 flag, and the corresponding best hit taxa was recovered with NCBI Entrez API. The broad consensus taxa of each contig (i.e., bacteria, fungi, insecta or other) was determined as the majority of contigs’ CDS taxa classification. Having a list of all bacterial contigs, we extracted their CDS sequences to run eggnog-mapper, using emapper v.2.1.1 and eggNOG DB v.5.0.2⁹⁰, -m diamond. Predicted KEGG ortholog IDs (ko:K#####) for each protein were used to recover BRITE classification. To test for functional enrichment in each hive component, coordinates of each CDS were recovered from Prokka gff ouput file, and used to recover the mean coverage from contig depth file (see “Metagenome assembly and binning”) which was then normalized for reads per million (RPM=mean coverage/total mean coverage)*1000000). The CDS from MAGs were also predicted and imported to KBase to run DRAM and recover additional information about carbon metabolism.

Candidate non-bee eukaryotic sequences present in samples with Wolbachia

Taxonomic classifications were recovered from the top Diamond v2.1.7.16⁸⁹ blastp hit for each predicted protein, using the same dataset generated for functional annotation. Contigs were classified as potentially originating from hive invaders, parasites, or pathogens if the majority of their proteins were assigned to Metazoa (excluding Anthophila, i.e., bees) and if at least one protein had a top hit with >95% nucleotide identity to a non-Anthophila metazoan. To estimate the abundance of these contigs, we calculated reads per million (RPM) based on mapping data described in the “Metagenome assembly and binning” section, comparing samples with and without Wolbachia.

Bacteria and fungi isolation and taxonomy

Bacteria and fungi were isolated from larvae and larval diet samples using a variety of media recipes, with the final type used for each strain stated (Supplementary Data 9). Briefly, raw, macerated and PBS 1X diluted samples were plated in Petri dishes, then incubated at derived-bee species brood temperature (i.e., 30 °C for stingless bee isolates and 34 °C for honey bee isolates). Plates were checked every day for microbial growth, then colonies, spores or mycelia pieces replated until isolation. Stocks of cells for bacteria and yeast or spores for filamentous fungi were stored in cryotubes at −70◦C with 20% glycerol. Strain isolation and taxonomy was confirmed by sequencing 16S rRNA gene for bacteria (primers: 27 F 5′-AGAGTTTGATCCTGGCTCAG-3′ and 1492 R 5′-TACGGYTACCTTGTTACGACTT-3′⁹¹) and 18S rRNA gene (primers: NS1 5′-GTAGTCATATGCTTGTCTC-3′ and NS4 5′-CTTCCGTCAATTCCTTTAAG-3′) or ITS region (primers: ITS5 5’ GGAAGTAAAAGTCGTAACAAGG 3’ and ITS4 5’ TCCTCCGCTTATTGATATGC 3’⁹²) for fungi, observing single clean peaks in chromatograms. A maximum likelihood phylogeny was inferred for Apilactobacillus, Weissella, Zygosaccharomyces and Starmerella strains as described in “Taxonomic assignment of MAGs and isolated strains” section.

Culturing bacteria and fungi for in vivo and in vitro experiments

Three media types were used for strain cultivation and experiments: 30 G (30 g glucose, 3 g malt extract, 3 g yeast extract, and 100 mL deionized water³⁷), PDA (Difco™ Potato Dextrose Agar, 213400), and MRS (BD Difco™ Dehydrated Culture Media: Lactobacilli MRS Agar, 288210). Single bacteria and yeast colonies from each strain were transferred to 5 mL broth and grown for the experiments at their respective isolation temperature for 24 h, shaking at 230 rpm. After 24 h grow all bacterial strains reached 10⁸ cells/mL. Yeasts were centrifuged for 8 min, at 4,500 × g, and the cell pellet resuspended in 500 µL of sterile water for OD600 measurement and normalization. Filamentous fungi were grown in PDA agar under respective temperature until spores were observed. Spores were collected with sterile water and gently use of cell scraper, then counted with the help of a hemocytometer for normalization. Acidification of the media due to bacterial metabolism was measured after 24 h of growth in 30 G and MRS broth using a pH meter. Additionally, acidification via fermentation was qualitatively assessed by two methods: (I) spotting 2 µL of each bacterial culture onto MRS agar supplemented with bromophenol blue (yellow <pH5, blue >pH5) to final concentration of 0.002%, and (II) adding the same concentration of bromophenol blue to MRS broth and documenting the color change after 24 h of incubation. MRS was chosen for these assays due to its lighter base color compared to 30 G, which facilitated visual detection of pH shifts.

In vivo larvae infection assays

Honey bee larvae were exposed to high loads of fungi cells or spores to test for potential pathogenicity to bees. Schmehl protocol was followed for larvae rearing, which currently is the only protocol using artificial larval diet⁹³. This ensured that no diet-derived microbes were present in the experiment, with only microbes inherent to the larvae or intentionally offered. Briefly, at day 0, first-instar larvae were grafted and deposited into plastic queen cups in 48-well plates, which already contained 10 µL of diet A. Larval diet was prepared no more than 24 h in advance of each feeding, always UV-sterilized for 20 min. On day 3, which is the first day of diet C, we added to each cup an additional 2 µL of 10⁸ fungal cells or spores/mL diluted in sterile water. For the negative control 2 µL of sterile water was added. At day 7, when larvae had finished to consume the last diet offer, they were transferred to a defecation 48-well plate with filter papers instead of queen cups. Development/survival was then followed until pupae reached brown-eyes stage, then weighed.

Bacterial vs fungi competition assay

The experiment was conducted in 6-well agar plates, including two experimental conditions, as well as bacterial and fungal growth controls, each with three replicates. One experimental condition included a higher bacterial load (15 µL lawn of 10⁸ CFU/µL), while the other contained a lower bacterial load (15 µL lawn of 10⁶ CFU/µL). After the bacterial lawn had dried, a 5 µL drop containing yeast at OD600 = 0.15 or 100 spores was placed at the center of the well. The bacterial control received 5 µL of sterile water at the center, while the fungal control contained a lawn of MRS broth, the medium used for bacterial growth and dilutions. The experiment was incubated at 30 °C for stingless bee bacterial strains and at 34 °C for honey bee strains. Plates were monitored until the fungal positive controls showed complete yeast growth or spores from filamentous fungi; pictures were taken for qualitative assessment of fungal growth. Additionally, for yeasts, cells from each replicate were harvested in 500 µL PBS, and OD600 was measured from a 100 µL aliquot. For filamentous fungi, mycelial diameter was measured, and spores were collected by scraping the agar after adding 3 mL PBS; spore counts were performed using a hemocytometer. Zygosaccharomyces from stingless bees were grown on 30 G agar, their required medium, while Zygosaccharomyces from honey bees were tested on both MRS (used in the overall experiment) and 30 G.

Fungal growth in acidic and low-nutrient media

Fungi were grown on 6-well agar plates under two experimental conditions: acidic medium (pH 4.2) and low-nutrient medium (50% diluted), alongside controls of regular MRS or 30 G media, each with three replicates. A 5 µL drop of yeast at OD600 = 0.15 or 100 spores was placed at the center of each well. Plates were incubated at 30 °C and monitored until controls reached complete yeast growth or spores from filamentous fungi. As described above, fungal growth was qualitatively assessed by imaging and quantitatively evaluated by yeast OD600, filamentous mycelial diameter, and spore counts. Zygosaccharomyces from stingless bees were grown on 30 G agar, their required medium, while Zygosaccharomyces from honey bees were tested on both 30 G and MRS for comparison.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization – L.C., I.N.; Methodology – L.C., IN; Experiments and Data Collection – L.C., C.B.; Data Analysis – L.C.; Writing Original Draft – L.C.; Review and Editing – C.B., V.H.V., I.N.; Resources – V.H.V., I.N.; Funding acquisition  – L.C., I.N.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

Raw metagenomic data and MAGs have been deposited to the NCBI Sequence Read Archive (SRA) under the Project ID PRJNA1216660. MAGs are also stored on GitHub https://github.com/liliancaesarbio/scaptotrigona_microbiome.

Code availability

All custom scripts used in this study are available on GitHub (https://github.com/liliancaesarbio/scaptotrigona_microbiome) and archived on Zenodo (10.5281/zenodo.17488340)⁹⁴.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-66678-9.