Bees are a diverse group of key pollinators (Michener, 2007; Danforth et al., 2019). As many species are declining (Ollerton et al., 2014; Goulson et al., 2015; LeBuhn and Vargas Luna, 2021), there is an urgent need to understand ecological factors that influence bee health. One such factor is the microbiome, the assemblage of beneficial, neutral, and harmful microbes associated with a host. Though molecular studies of bee microbiomes began over 15 years ago (Jeyaprakash et al., 2003; Mohr and Tebbe, 2006), and culture-based studies much earlier (Batra et al., 1973; Gilliam, 1979), our understanding of bee microbiomes is far from comprehensive. Only a small fraction of the ~20,000 described bee species (Michener, 2007) have been microbially characterized to date. The majority of studies are on the social corbiculate bees and a handful of solitary bee groups (Voulgari-Kokota et al., 2019). Moreover, as bees are holometabolous and nest-dwelling insects, multiple life stages and microhabitats need to be characterized for a complete picture of bee-microbe interactions. The brood cell, which contains pollen provisions and larvae, is much harder to sample than adults, but has a central role in bee biology. In social corbiculate bees, larvae and adults have very different microbial associations (Kwong and Moran, 2016). In solitary bees—which make up the majority of bee diversity (Danforth et al., 2019) — brood cell microbiome composition and function are poorly understood.

16S rRNA profiling-based studies have provided an initial picture of microbiome composition in solitary bee brood cells. Lactobacilli tend to be dominant members of the bacterial community, across distantly related bee species (McFrederick et al., 2012, 2017; Kapheim et al., 2021). However, for large swaths of bee diversity—including entire families—brood cell microbiomes remain uncharacterized. Moreover, potential variation in the absolute abundance and activity of microbes cannot be inferred from 16S rRNA profiles (Hammer et al., 2019). Here, the decades of research on bee nesting biology provide some clues. In many bee taxa, larval provisions are semi-solid or dough-like, and sometimes quite dry (Danforth et al., 2019; Cane and Love, 2021). Microbial growth in these substrates could be limited by low water potential (Bartlett and Roberts, 2000), as well as other factors. However, in the family Colletidae, provisions are liquid or semi-liquid (Rozen, 1984; Michener, 2007; Almeida, 2008; Sarzetti et al., 2013). The brood cell of colletid bees is lined by an impermeable “cellophane” coating, produced by the Dufour’s gland, that both protects the larva during development and prevents the liquid provisions from leaking into the surrounding substrate (Rozen, 1984; Almeida, 2008; Danforth et al., 2019). Unusually for solitary bees, the colletid subfamily Diphaglossinae also have open brood cells, which have been suggested to facilitate gas exchange (Roberts, 1971). Brood cell venting could be particularly important when there are large numbers of metabolically active microbes alongside the developing bee larva.

Within Diphaglossinae, the genus Ptiloglossa has provisions that are particularly watery, and have an obvious odor and taste characteristic of fermentation. This trait was first observed by D. H. Janzen in Veracruz, Mexico in the early 1960s (pers. comm.) and reported by Roberts for a Costa Rican species, P. costaricana, in 1971 (Roberts, 1971). Vigorous microbial growth in healthy brood cells appears to be characteristic for this genus, and not a case of sporadic microbial spoilage (which is not uncommon among bees; Batra et al., 1973). Strong fermentation odors have also been noted in Ptiloglossa brood cells in Brazil and Arizona, United States (Rozen, 1984; de Araujo et al., 2020). Ptiloglossa, and perhaps other colletids (Michener, 1960), may have similarities with insects such as Drosophila, for which microbes make up an important part of the larval diet (Markow and O’Grady, 2008). But colletid bee brood cells have not been studied using molecular methods, limiting our ability to infer the ecological function and evolutionary history of this symbiosis. There are many open questions. Which microbes grow in the larval provisions? Are they unique species domesticated by the bee host? How do they vary between closely related bee hosts and between habitats? Are brood cell microbes vertically transmitted, or acquired from the environment? How do they vary over host development? And what traits might bees be using to “brew” fermenting larval provisions?

Here we explore these questions using microbiome sequencing of field-collected brood cells of three diphaglossine bee species: Ptiloglossa arizonensis and Caupolicana yarrowi in Arizona, United States, and Crawfordapis luctuosa in western Panama. We used 16S rRNA gene sequencing to characterize the composition of bacteria in a range of microhabitats in which they may contribute to bee biology, and among which they may be transmitted: larval provisions, larvae of different developmental stages, pupae, adult guts, and floral nectar. For a subset of P. arizonensis larval provisions, we also sequenced shotgun metagenomes. These data were used to measure the relative abundance of nonbacterial microbes such as fungi, infer ratios of bacterial to plant biomass, and assess strain-level diversity. We also constructed a phylogeny, using assembled 16S rRNA gene sequences, to evaluate host- and habitat-specialization of the dominant lactobacilli.

Here, we summarize new and published natural history observations relevant to interactions between diphaglossine bees and microbes. At our field site in southeastern Arizona, we observed Ptiloglossa arizonensis and Caupolicana yarrowi foraging for pollen from Solanum elaeagnifolium and S. rostratum (Figure 1), matching earlier observations (Linsley, 1962; Rozen et al., 2019). Solanum is a commonly reported pollen source across the range of Ptiloglossa, though non-Solanaceous plants are also used (Janzen, 1968; Roberts, 1971; Sarzetti et al., 2013). We also have evidence that P. arizonensis collects nectar from Agave palmeri flowers. First, bees returning from foraging are often dusted with pollen that is morphologically identical to Agave pollen. This pollen is concentrated on the dorsum of the mesosoma, a location that matches the position of exserted Agave anthers. Second, Agave palmeri nectar has a powerful and unique melon (Cucumis melo)-like odor. We detected the same odor from P. arizonensis brood cells. Caupolicana yarrowi were observed collecting nectar from creosote (Larrea tridentata). Earlier studies from the southwestern US report that P. arizonensis visit Larrea as well (Hurd and Linsley, 1975). We do not have information about food plants of the specific Crawfordapis luctuosa population we studied, but another population in Western Panama was reported to collect pollen from dozens of flowering plant species, mainly belonging to Solanaceae, Melastomataceae, and Begoniaceae (Roubik and Michener, 1984), all of which (except Begoniaceae species) have poricidal anthers and require buzz pollination (Buchmann, 1983).

The timing and duration of foraging influence the availability and quality of floral resources and the risk of parasitism (Wcislo and Tierney, 2009). These factors could alter the types of microbes to which bees are exposed, and could select for different microbially mediated nutritional or defensive strategies. Ptiloglossa are dim-light foraging (crepuscular) bees, with activity concentrated from before, to shortly after dawn (Linsley, 1962; Janzen, 1968; Roberts, 1971; de Araujo et al., 2020). This is the case at our field site in Arizona, where we observed P. arizonensis foraging beginning roughly an hour before dawn and ending just after dawn. Caupolicana has a longer foraging window. In Arizona, we and others (Rozen et al., 2019) have observed Ca. yarrowi foraging later in the morning than P. arizonensis, and as late as mid-afternoon; another Caupolicana species has been collected near mid-day (Michener, 1966). Our Crawfordapis luctuosa study population forages throughout the day, as reported earlier (Roubik and Michener, 1984).

We observed that P. arizonensis larval provisions—particularly those in older cells with large larvae—have a sour taste and a strong odor characteristic of microbial fermentation. According to earlier reports from the neotropics, the odor of fermentation is “unmistakable” (Costa Rica; Roberts, 1971), a “strong sour smell” (Brazil; de Araujo et al., 2020) similar to “fermenting beer or mead” (Venezuela; D. H. Janzen, pers. comm.). Caupolicana yarrowi brood cells did not have a noticeable fermentation odor, in agreement with published descriptions (Rozen, 1984; Rozen et al., 2019). We observed fermentation odors during collection of Crawfordapis luctuosa, although they were not observed in an earlier study (Roubik and Michener, 1984), perhaps due to seasonal or fine-scale temporal differences in fermentation within brood cells. Although these odors are likely to be a useful proxy for overall microbial activity, there are caveats. It is difficult to infer which microbes are abundant, as very different microbes can have overlapping volatile profiles (e.g., Saccharomyces yeast and Lactobacillus bacteria; Hansen and Hansen, 1994), and as different volatiles have different salience to the human nose. Also, unusual odors in bee brood cells are not necessarily a product of microbial metabolism. They can be derived from plant volatiles or, as in the case of the cheesy-smelling brood cells of Anthophora, from glandular substances produced by the bee (Norden et al., 1980).

Microbiomes of diphaglossine bee larval provisions are dominated by lactobacilli (mean relative abundance +/− SEM: Ptiloglossa, 94.3 +/− 3.61%, N = 16; Caupolicana, 99.2 +/− 0.645%, N = 2; Crawfordapis, 97.2 +/− 0.627%, N = 10; Figure 2). Lactobacillus is dominant even in brood cells inferred to have been recently provisioned (i.e., no egg or larva). For two Ptiloglossa brood cells in which we separately sampled the top (more nectar-rich) and bottom (more pollen-rich) stratified layers of the larval provisions, we did not observe a difference in microbial composition (Figure 2). In Ptiloglossa and Caupolicana, the dominant ASVs have 100% sequence identity to various strains of Apilactobacillus, particularly A. micheneri and A. timberlakei. In contrast, Crawfordapis larval provisions are dominated by an ASV that may represent a new bacterial species, with <97% identity to isolates belonging to Lactobacillus, Pediococcus, Nicolia, and related genera. For consistency, we use the default classification provided by the SILVA database (Lactobacillus) in the text and plots describing 16S rRNA amplicon data. Other bacteria, namely Saccharibacter, Fructobacillus, and Acinetobacter, are only sporadically present, but can be abundant in individual brood cells (Figure 2).

Whole-body microbiomes of developing Ptiloglossa larvae are also dominated by Lactobacillus, similarly to their diet (compare larvae to provisions samples with matching letters [a, b, d, e] in Figure 2). Secondary bacterial taxa (most notably, Saccharibacter) that appear in larval provisions (sample PA.LP.6) also appear at a similar relative abundance in the larva from the same brood cell (sample PA.DL.4; Figure 2), suggesting dietary acquisition. (Larvae of most bees, including diphaglossines, only defecate after they finish feeding and immediately before entering the last larval instar, or prepupal stage (Danforth et al., 2019). These larvae cannot contaminate the provisions through defecation). Microbiomes of Ptiloglossa larval provisions and developing larvae are not significantly different in Shannon diversity (post hoc test, adjusted p = 0.98; Figure 3) or in composition (pairwise PERMANOVA, adjusted p > 0.05; Figure 4). In contrast, developing Crawfordapis larvae have high relative abundances of Wolbachia in addition to Lactobacillus (Figure 2), leading to a clear division between larval provisions and developing larval microbiome composition (pairwise PERMANOVA, R² = 0.55, adjusted p = 0.003; Figure 4) and Shannon diversity (posthoc test, adjusted p < 0.01; Figure 3), though not within-group heterogeneity (betadisper, F = 0.067, p = 0.79).

In Ptiloglossa, microbiomes shift, diverging from the diet, as larvae approach the prepupal stage. Mature larvae—those which have consumed all of the provisions, but have not yet defecated—have about twice as high alpha diversity as developing larvae (Figure 3), although this difference is not statistically significant (adjusted p > 0.05). (Note that sample sizes are low: N = 4 per sample type.) Mature and developing larvae also appear to have somewhat distinct microbiome composition (Figures 2, 4), though again this difference is not statistically significant (adjusted p > 0.05). For Crawfordapis, only one out of six mature larvae had enough 16S rRNA reads for analysis. Wolbachia is the only bacterium detectable in this larva (Figure 2).

Lactobacillus, and potentially all bacteria, are cleared from the gut before larvae enter diapause as a prepupa. We sequenced one Ptiloglossa prepupa in which pollen was visibly absent from its gut, confirming that it had already defecated. This sample is nearly devoid of bacteria. 97.8% of the 16S rRNA reads are eukaryotic; based on blastn searches, these are likely to originate from the bee as opposed to plant or other eukaryotic DNA. The remaining bacterial reads do not include any Lactobacillus. Rather, these ASVs are all very rare (<100 total reads) or absent in the other Ptiloglossa and nectar samples, and hence may be transient or spurious. In Crawfordapis, eukaryotic reads are not abundant (<10%) in pupae and pharate (pre-eclosion) adult samples—possibly because of a lower identity to bee rRNA with this primer pair. However, sequencing depth was very low (2–198 reads, mean 47), suggesting low amounts of bacterial DNA in these life stages. Lactobacillus is also absent.

Adult Ptiloglossa arizonensis collect Agave nectar, regurgitating it from the crop into the larval provisions and likely consuming it themselves (see Natural History). Bacterial communities in Agave nectar, Ptiloglossa adult guts, and Ptiloglossa larval provisions are distinct. Shannon diversity is much higher in Agave nectar than in larval provisions (adjusted p = 0.011; Figure 3). Microbiome composition also differs between each of these habitats (pairwise PERMANOVAs, R² = 0.17–0.45, adjusted p = 0.01–0.02; Figure 4). As evident in the ordination (Figure 4), within-group microbiome heterogeneity varies among sample types (betadisper, F = 2.78, p = 0.043), with higher heterogeneity for adult guts and Agave nectar. Lactobacillus is present in adult guts as well as Agave nectar, but at much lower and more variable relative abundances than in larval provisions (Figure 2). In addition to Lactobacillus, adult gut microbiomes contain a variety of bacterial genera common among bees and other pollinating insects, such as Entomomonas and Fructobacillus (McFrederick et al., 2017; Hammer et al., 2020; Wang et al., 2020; Handy et al., 2023). In a smaller set of samples, collected in 2018 from the same site in Arizona, adult guts of Ptiloglossa, Caupolicana, and Protoxaea gloriosa (Andrenidae) were all dominated by Lactobacillus (Figure 2). Alkanindiges, a gammaproteobacterium detected in other Hymenoptera (Suenami et al., 2019; Koto et al., 2020), was also abundant in one Caupolicana adult. Lactobacillus and Fructobacillus are again dominant bacterial genera in Crawfordapis adult microbiomes (whole abdomen samples), alongside Wolbachia (Figure 2).

There are multiple Lactobacillus amplicon sequence variants (ASVs) in our three datasets, representing distinct species or strains. We analyzed the distribution of these ASVs to further investigate transmission and potential host-symbiont specificity. In the Ptiloglossa dataset, Lactobacillus populations across provisions, larvae, and adult guts predominantly belong to a single ASV (Figure 5). However, many of the samples contain one or more additional Lactobacillus ASVs. General patterns are similar in the Caupolicana and Crawfordapis datasets, with 1–2 dominant Lactobacillus ASVs (alongside a few rarer ASVs) in provisions, larvae, and adult guts (Figure 5). Within the 2018 Caupolicana-focused dataset, we also see evidence of Lactobacillus ASV sharing among co-occurring bee species. The dominant ASV in Ptiloglossa is present in Caupolicana, as well as Protoxaea gloriosa (Figure 5), a distantly related bee in the subfamily Oxaeinae, which also produces liquid provisions (Sarzetti et al., 2014).

Shotgun metagenomic data from larval provisions of two Ptiloglossa brood cells corroborate the 16S rRNA gene amplicon data. Based on taxonomic classification of all SSU rRNA genes identified in the metagenomes, microbial community structure is dominated by Lactobacillus (Figure 6). Bacterial sequences not classified as Lactobacillus mostly (78.7%) belong to the order Lactobacillales; these either belong to related genera or lacked a genus-level classification. We also used these data to quantify the relative abundances of non-bacterial microbes such as fungi. The latter are rare (Figure 6). Fungi occur at 0.502 and 0.120% in the two larval provisions samples. 87.7% of the unidentified eukaryotic sequences—those not classified as host (Metazoa), plant, or fungi—lacked any classification below the Domain level and may be artifactual. Acanthamoeba and Plasmodium were present, but at very low levels (≤ 32 sequences total). As a control, a mock community (Zymo) was also sequenced and processed alongside the larval provisions. The proportion of Lactobacillus in this sample is fairly accurate (13.1% versus expected 12%) and fungi are somewhat overrepresented (13.5% versus expected 4%).

As amplicon sequencing and metagenomics are compositional methods, they do not directly quantify the absolute abundance (biomass) of microbes in a sample. However, variation in the ratio of bacterial to plant SSU rRNA sequences may be used as a proxy for variation in bacterial biomass (relative to plant biomass). We find that the bacteria:plant ratio in Ptiloglossa larval provisions is substantially higher than nine non-colletid bee species assayed with the same methodology (Figure 6). This suggests there is likely more bacterial biomass per brood cell in Ptiloglossa as compared with many other bee species, in line with inferences of microbial activity based on the presence or absence of fermentation odors. Note it is possible that Ptiloglossa females provision brood cells with less pollen than other bees, which would also increase the bacteria:plant ratio.

We further used the shotgun metagenomes to explore sub-ASV-level diversity of lactobacilli in the two Ptiloglossa brood cell samples. Single-sample assemblies resulted in one medium-quality and one low-quality metagenome-assembled genome (MAG) per sample (see The Genome Standards Consortium et al., 2017, for quality definitions; Table 1). The medium-quality MAGs are classified as Apilactobacillus, matching results from blastn searches of dominant ASVs in the 16S rRNA amplicon dataset. One sample contains an A. micheneri MAG, while the other contains an A. timberlakei MAG, with high average nucleotide identity (98.5–99%) to previously sequenced genomes (Table 1). As measured using single-copy core genes, bins have high levels of strain heterogeneity (Table 1). This metric evaluates the degree to which reported contamination comes from conspecific strains, versus heterospecifics (Parks et al., 2015). High values suggest that multiple closely related strains were co-assembled into the same MAG. Thus there is likely intrageneric Apilactobacillus diversity within each of the two brood cells, including sample PA.LP.9, which contains only a single ASV classified as Lactobacillus in the 16S rRNA amplicon data (Figure 5).

To investigate possible host-specificity between Ptiloglossa and Apilactobacillus, we constructed a phylogeny using full-length 16S rRNA gene sequences assembled from the metagenomes. Only a single Apilactobacillus sequence was reconstructed from each of the two brood cells using phyloFlash. This does not necessarily conflict with the evidence of within-sample Apilactobacillus diversity (Figure 5; Table 1), as the assembler tends to collapse strain-level variability into a single, approximately species-level, consensus sequence (Gruber-Vodicka et al., 2020). Closely matching sequences and outgroups were collected from GenBank. The phylogeny (Figure 7) generally agrees with prior phylogenomic analysis of Apilactobacillus, with two sister clades corresponding to A. micheneri and A. timberlakei (Vuong and McFrederick, 2019). Matching the assembly data (Table 1), one Ptiloglossa brood cell (PA.LP.1) has a consensus 16S sequence belonging to A. micheneri, while the other (PA.LP.9) has a consensus sequence belonging to A. timberlakei. Thus, Ptiloglossa exhibit some degree of species-level flexibility in their symbiosis with Apilactobacillus (note that these two species are very closely related; Vuong and McFrederick, 2019; Wittouck et al., 2019). Conversely, Apilactobacillus are not specialized to particular bee hosts. The Ptiloglossa-associated A. micheneri has 99.9–100% sequence identity to A. micheneri strains isolated from non-colletid bees and from nectar (McFrederick et al., 2018), and to a clone from the gut of a Caupolicana yarrowi adult collected in 2006 from southeastern Arizona (Martinson et al., 2011; Figure 7). Likewise, the Ptiloglossa-associated A. timberlakei has 99.9–100% sequence identity to additional isolates from non-colletid bees and nectar, and to a clone from the gut of a Diadasia opuntiae (Apidae) adult bee collected in 2012 from Sonora, Mexico (Martinson et al., 2011; Figure 7). Diadasia and Ptiloglossa co-occur in Arizona but use temporally non-overlapping floral resources. Diadasia opuntiae forage nearly exclusively on cactus (Opuntia and Carnegiea) flowers for nectar and pollen (Ordway, 1984). Therefore, sharing of lactobacilli between these bees is likely mediated by other flower visitors.

Across all three diphaglossine bee species, in both the Sonoran desert (Arizona, United Sates) and premontane tropical rainforest (Western Panama), microbial communities of larval provisions are near-monocultures of lactobacilli. Larval provisions of Ptiloglossa and Caupolicana specifically harbor Apilactobacillus while those of Crawfordapis harbor unique and potentially novel Lactobacillus-related taxa. Thus, diphaglossine bees fit an emerging pattern of symbiotic interactions between bees and lactobacilli, spanning solitary species, social species, the adult stage, and brood cells (Kwong and Moran, 2016; McFrederick et al., 2017; Voulgari-Kokota et al., 2019; Tang et al., 2021; Handy et al., 2023).

Consistency of the diphaglossine bee brood cell microbiome is unlikely to come from vertical transmission. Lactobacilli disappear from larvae before metamorphosis—a common phenomenon in gut symbionts of holometabolous insects (Hammer and Moran, 2019). Furthermore, phylogenetic evidence from Ptiloglossa brood cells indicates that Apilactobacillus are not host-specific; host specificity (or host restriction) is a common feature of vertically transmitted symbionts (Moran et al., 2019). Apilactobacillus appear to be frequently exchanged with other bees via flowers, as is the case in other solitary bee-lactobacilli associations (McFrederick et al., 2014, 2017). Brood cell microbiome consistency is likely mediated instead by horizontal transmission coupled with strong ecological filtering (or partner choice; Sachs et al., 2004). Agave palmeri nectar (visited by Ptiloglossa) does contain detectable levels of lactobacilli, but amid a diverse and heterogeneous background of bacteria. Similarly, a culture-based study in southern Arizona found that Apilactobacillus is present in Agave palmeri nectar, but at a low abundance relative to other bacteria and yeasts (von Arx et al., 2019). The traits that allow Apilactobacillus to colonize brood cell microbiomes are not known, but may include tolerance of low pH, osmotic stress, and oxidative stress, and the ability to metabolize nectar carbohydrates (particularly fructose) and attach to host tissues (Vuong and McFrederick, 2019). Where and when filtering takes place are also unknown, but given that Apilactobacillus dominates even freshly provisioned brood cells, one possibility is that it begins in the female bee’s crop. Microbes have been observed at higher densities in the bee crop as compared with nectar (Batra et al., 1973), and the crop of Xylocopa sonorina bees is strongly enriched in Apilactobacillus (Handy et al., 2023). We cannot directly address this hypothesis because our adult bee samples include the midgut and hindgut in addition to the crop, and these regions may have highly distinct microbiomes (Kwong and Moran, 2016).

Although brood cell microbiomes are homogenous at the level of bacterial genera—Apilactobacillus in Ptiloglossa and Caupolicana, unclassified Lactobacillus in Crawfordapis—there is diversity at finer phylogenetic scales. Most individual brood cells of all three bee species harbor lactobacilli comprising multiple amplicon sequence variants (ASVs). In Ptiloglossa, there is evidence of further diversity not detectable at the ASV level. Across brood cells, Ptiloglossa cultures at least two closely related species of Apilactobacillus—micheneri and timberlakei. Within a brood cell, there is also strain-level diversity (Table 1). Thus, ecological filtering in Ptiloglossa brood cells is somewhat porous, permissive of multiple species and strains of Apilactobacillus. Whether sub-generic diversity of Apilactobacillus has functional consequences for bee development is not known. A. micheneri and A. timberlakei are sister species, and gene sets involved in carbohydrate metabolism generally overlap between their genomes (Vuong and McFrederick, 2019). But there is also evidence for divergence, particularly in genes mediating tolerance of environmental stressors (Vuong and McFrederick, 2019). Partnering with multiple symbiont species and strains can be advantageous to hosts (Batstone et al., 2018). For example, the ability to culture multiple microbial partners with distinct niches could buffer the brood cell symbiosis against environmental variation and against a heterogeneous distribution of Apilactobacillus species in flowers. At the same time, bees may not be able to discriminate among closely related lactobacilli. For example, legumes are often unable to selectively exclude certain strains of rhizobia from colonizing root nodules, even those that are ineffective mutualists (Hahn and Studer, 1986; Simms and Taylor, 2002).

Microbiomes of developing Ptiloglossa larvae are statistically indistinguishable from those of their diet. Further, microbial variation among provisions is reflected in the corresponding larvae. A similar pattern occurs in caterpillars, where ingested bacterial communities pass transiently through the gut (Whitaker et al., 2016; Hammer et al., 2017). This result suggests that: (i) there are no substantial populations of unique symbionts inhabiting the gut of Ptiloglossa larvae; (ii) there is no additional ecological filtering of the provisions once ingested. Larvae of several other solitary bees, as well as honey bee larvae, appear to be similar in these respects (Martinson et al., 2012; McFrederick et al., 2014, 2017; Kapheim et al., 2021). A strategy of suppressing bacterial colonization makes sense in light of the fact that most developing solitary bee larvae do not defecate until after they have completed feeding (Danforth et al., 2019); it may be too risky to allow microbial proliferation without a way of expelling excess cells. In contrast, we do observe a strong difference between provisions and whole-body larval microbiomes in Crawfordapis, mediated by the presence of Wolbachia in larvae. Wolbachia is also abundant in adult Crawfordapis. Wolbachia is fairly common in temperate bees (Gerth et al., 2015), and in at least some tropical bees such as Megalopta centralis (McFrederick et al., 2014).

Although dominated by lactobacilli, diphaglossine bee brood cells do harbor a variety of rarer bacteria. Saccharibacter is sporadically abundant in Ptiloglossa brood cells and adults, and in Agave nectar. Saccharibacter and other aerobic, acid-tolerant, and osmophilic Acetobacteraceae are commonly found in insects with sugar-rich diets (Crotti et al., 2010), on pollen (Jojima et al., 2004), and in the crop (Handy et al., 2023) and pollen provisions of bees (McFrederick et al., 2012). Fungal sequences are also present, but at very low relative abundances, in shotgun metagenomes from Ptiloglossa larval provisions. The lack of a substantial fungal community contrasts with earlier reports of yeasts in neotropical Ptiloglossa brood cells (Roberts, 1971; Batra et al., 1973). One possible explanation is that the presence of yeasts differs between Ptiloglossa in temperate deserts (studied here) versus tropical forests. Another explanation, which cannot be fully excluded, is a technical bias against fungi. However, our mock community metagenome harbors more yeast sequences than expected, and the DNA extraction protocol we used is similar to that used for fungal sequencing in sourdough starters, soil and other habitats (Rousk et al., 2010; Landis et al., 2021).

Although similar to related diphaglossines and to other solitary bee groups in terms of composition and host specificity (Figure 2; McFrederick et al., 2013, 2017; Kapheim et al., 2021), the abundance and activity of Ptiloglossa brood cell microbiomes appear to be unusual. First, larval provisions of Ptiloglossa have a much higher ratio of bacterial to plant DNA as compared with several other ground-nesting solitary bee species. Second, they are consistently observed to exhibit strong fermentation odors, which are only rarely reported from healthy brood cells of solitary bees. These two lines of evidence suggest that there has been a quantitative, but not qualitative, microbiome shift within diphaglossine bees. As more abundant beneficial microbes will generally confer stronger benefits to hosts (up to a point; Hammer et al., 2019), this shift may constitute a functionally novel form of symbiosis in Ptiloglossa.

Novel host-microbe symbioses often evolve through a process analogous to domestication, with capture and vertical transmission of host-restricted symbionts (Moran et al., 2019; Ganesan et al., 2022). Humans also domesticate microbes, propagating starter cultures used for fermentation of certain foods and beverages (Gibbons and Rinker, 2015; Steensels et al., 2019). In contrast, Ptiloglossa appear to culture undomesticated bacteria in their larval provisions. If confirmed, this result would demonstrate that quantitative microbiome shifts can evolve without changes in host specialization. A caveat is that symbiont domestication can occur rapidly (Stallforth et al., 2013; Bodinaku et al., 2019), without concomitant changes in 16S rRNA gene sequences. Another caveat is that our results are limited to Ptiloglossa arizonensis. The common ancestor of Ptiloglossa was most likely a tropical species, given that most extant diversity is in the tropics and subtropics (Michener, 2007), and the sister genus Crawfordapis (following, Velez-Ruiz, 2015) is also tropical. Hence it is not yet clear whether the undomesticated nature of P. arizonensis brood cell symbionts is ancestral or derived (e.g., related to adaptation to desert environments).

Our results suggest that the unusually fermentative brood cell microbiomes of Ptiloglossa evolved not through domestication, but simply by modification of the culturing environment. This path to symbiotic novelty has parallels with animals such as Riptortus bugs, which use modified gut structures to enrich non-host-specialized Burkholderia from the diet (Kikuchi et al., 2007; Ohbayashi et al., 2015). Some other animals behaviorally modify their environment to promote the growth of undomesticated “crops” (Zhu et al., 2016; Selden and Putz, 2022). Analogously, spontaneous (or natural) fermentation of certain foods and beverages relies on wild, undomesticated microbes. Lactobacilli often participate in this process; for example, Apilactobacillus micheneri (strain 11D in Figure 7) is a dominant member of the bacterial community in kôso, a fermented vegetable drink (Chiou et al., 2018). What traits enable Ptiloglossa to culture Apilactobacillus at high densities in their brood cells? Facilitated by waterproof brood cell linings, colletid bees in general, and Ptiloglossa in particular, tend to have more liquid larval provisions than other solitary bees (Roberts, 1971; Rozen, 1984; Almeida, 2008; Cane and Love, 2021). We hypothesize that this relatively high water content may facilitate microbial growth. How exactly highly liquid provisions are achieved is not known, but one possibility is that Ptiloglossa collect a large volume of nectar relative to pollen. Ptiloglossa often forage from flowers (like Agave) that produce large quantities of nectar (see Natural History). Another possibility is that by foraging in the early morning, Ptiloglossa collect particularly dilute nectar. Nectar is generally more dilute when first produced (Cane and Love, 2021).

As hypothesized by Roberts (1971), high bacterial densities in larval food likely benefit Ptiloglossa development. The elevated ratio of bacterial to plant DNA suggests that bacterial biomass is a major nutrient source for larvae, alongside pollen and nectar. Although almost all bee species are herbivorous (Michener, 2007), Ptiloglossa may have commonalities with insects that derive nutrition from microbes, such as Drosophila and dung beetles (Markow and O’Grady, 2008; Holter, 2016). Pollen and nectar contain all of the nutrients required for bee development (Roulston and Cane, 2000), but a “brewing” strategy could be economical for Ptiloglossa, a bee that almost exclusively forages from just before, to just after dawn. If bacteria upgrade the nutritional quality of the provisions, Ptiloglossa females may be able to rear more offspring despite a highly constrained foraging window. Brood cell bacteria could also play a role in defense. Ground-nesting, immature Hymenoptera (and their food) are vulnerable to attack by soil-borne microbes. Hence, both endogenous and bacterially based defenses are common (Kaltenpoth et al., 2005; Fernández-Marín et al., 2009; Strohm et al., 2019). For Ptiloglossa, the organic acids (and potentially ethanol) resulting from Apilactobacillus fermentation may suppress microbial invaders (McFrederick et al., 2018; Vuong and McFrederick, 2019) analogously to the anti-spoilage properties of lacto-fermented food. Indeed, despite the common presence of fungi in nectar and soil, fungal growth in Ptiloglossa brood cells appears to be kept to a minimum. Enrichment for lactobacilli occurs even before eggs are laid, potentially acting as a prophylactic antimicrobial defense for offspring, as occurs in some animals (Flórez et al., 2017; Kerwin et al., 2019).

To understand whether the diphaglossine bee brood cell symbiosis constitutes a mutualism, effects on bacterial fitness also need to be determined. Many insects have mechanisms to transmit symbionts vertically (Buchner, 1965), which more closely align fitness interests between hosts and symbionts (Ewald, 1987; Sachs et al., 2004). Many food fermentation practices also involve the reuse of starters or culture vessels, allowing domestication to occur. But, while the diphaglossine bee brood cell is clearly a highly favorable environment for local and short-term growth of lactobacilli, the bacteria may ultimately be digested by the larva or otherwise fail to escape alive. Indeed, lactobacilli consumed by larvae do not persist internally through the prepupal and pupal stages; we hypothesize that adults acquire them anew each generation from flowers. Other mechanisms by which bees could propagate lactobacilli need to be tested but seem unlikely. In theory, adults emerging from brood cells could acquire lactobacilli externally. But first, lactobacilli would need to survive for potentially several months ex vivo, as Ptiloglossa arizonensis, like many bees, overwinter as post-defecating prepupae (Rozen, 1984; Michener, 2007). Opportunities for emerging adults to contact residual lactobacilli in the brood cell are also limited. Postdefecating larvae of most diphaglossines pierce the cell lining such that feces drains into the soil; once the cocoon is spun, there is little direct exposure to remaining fecal material or the cell lining (Rozen, 1984).

If the dead-end hypothesis is correct, it implies a strong contrast with horizontally transmitted mutualisms in which symbionts benefit from their associations with hosts. For example, in the legume-rhizobia and bobtail squid-Vibrio mutualisms, hosts release the symbionts they culture back into the environment in large numbers (Lee and Ruby, 1994; Simms and Taylor, 2002). Indeed, adult diphaglossine bees harbor lactobacilli in their gut, and are likely important for their dispersal and persistence in the bee-flower niche. But the brood cell association may be more exploitative, with little to no long-term benefit to the lactobacilli.

In sum, the nesting biology of these bees appears to create favorable conditions for spontaneous fermentation or “brewing” of generalist lactobacilli: underground, temperature-stable brood cells; maintenance of high water content in provisions; suitable sugars (especially fructose); protection from contamination by the cellophane-like cell lining; open cells, possibly to allow venting; and, potentially, pre-enrichment of lactobacilli in the adult crop. Given the consistency of brood cell fermentation, particularly in Ptiloglossa, the bees likely benefit from culturing lactobacilli. On the other hand, lactobacilli may not benefit from being cultured, as they seem unable to escape the brood cell. While many details remain speculative, our findings provide an initial picture of the microbiology and ecology of a remarkable feat of fermentation.

The data presented in the study are deposited in the NCBI SRA repository, accession number PRJNA925568.

TH, JK, QM, SB, and BD designed the study. JK, QM, WW, and BD helped acquire grant funding. TH, JK, SB, and BD conducted fieldwork. TH, JK, MA-G, and LG conducted labwork. TH and MA-G performed bioinformatic and statistical analyses. TH created the figures and drafted the manuscript, which was then read and edited by all authors. All authors contributed to discussion and interpretation of results.

This work was supported by NSF DEB 1929572 Simons Foundation 429440 USDA-AFRI grant 1018839.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.