Background

Honey bees (Apis mellifera) function as a cooperating group of individuals (colonies) characterized by division of labor [1]. Reproduction is performed by long-lived queen phenotypes while short-lived workers perform a variety of nutrient processing and other tasks that support the reproductive effort [2]. Both longevity phenotypes can result from identical genomes, but queens can live over ten times as long as workers and consume a very different diet [3]. Beginning as newly hatched larvae, queen vs. worker (caste) development is controlled by signaling molecules found in different diets. Pollen exposure halts queen development while royal jelly promotes queen development [4, 5]. Nurse workers gorge on pollen to synthesize royal jelly fed to queens throughout their lives. Royal jelly is functionally analogous to mammalian breast milk comprised of a complete diet and antioxidant, antimicrobial, and immunoregulatory properties [6, 7]. Attributed to caste-specific diets, the phospholipid profile of aging workers becomes increasingly susceptible to oxidative stress, but the queen profile remains stable with age [8]. Consistent with these results, antioxidant gene expression increases in aging workers but not queens [9, 10]. Workers live longer when fed the queen diet (royal jelly) as compared to a pollen diet [11]. Collectively, these results suggest that the drastically different lifespans and diets associated with division of labor in honey bees provide a model for mechanisms of diet, aging, and microbiota [12, 13].

Division of labor in social insects is organized around nutrition and reproduction. In honey bees, this social organization is attributed to the evolutionary repurposing of an egg yolk glyco-lipoprotein (vitellogenin) to serve as nutritional currency throughout the colony [14]. The oldest honey bees leave the hive to forage for nectar, pollen, and water. Collected pollen is converted by young workers into two major forms of nutritional currency, one internal, vitellogenin, expressed mostly by abdominal fat body, and one external, royal jelly, shared as social currency among nestmates. In workers, much of the vitellogenin released into the hemolymph is diverted to worker head glands to produce royal jelly [15]. Royal jelly secretions from young (nurse) bees are fed via oral trophallaxis to growing larvae and the queen. In turn, much of the royal jelly fed to queens is converted internally to vitellogenin, to support massive egg production [14]. Vitellogenin is expressed constitutively throughout the queens’ internal anatomy [9, 16]. Like royal jelly, vitellogenin is a multipurpose superfood that functions in immunity, detoxification, oxidative stress, nutrition, and longevity [9, 16,17,18]. Older foragers no longer produce jelly, but often beg for and receive small doses from younger nurse bees.

Reproductive division of labor underlies changes in microbiota composition both proximately and ultimately [19, 20]. Workers feeding queen vs. worker-destined larvae differ markedly for antimicrobial gene expression associated with royal jelly production in their head glands [21]. Following emergence as winged adults, queen and worker guts are colonized by very different microbiota [20, 22, 23]. Although highly antimicrobial, the queen’s diet of royal jelly enhances the growth in vitro of at least two bacterial species associated with the queen microbiota [7]. Accordingly, the worker phenotype is affected by pollen consumption that occurs concurrent with early adult succession of gut microbiota [24]. Experiments with conventionalized honey bee workers and pollen consumption indicate that bacterial fermentation products produced during the digestion of recalcitrant pollen shells influence host insulin signaling and the production of vitellogenin [25]. Vitellogenin and life expectancy decrease dramatically as workers transition to foraging [8] and the hindgut microbiota shifts with age.

In worker hindguts, fermentation products of gut bacteria are produced according to microbiota structure [25]. A variety of environmental insults can perturb microbiota structure (dysbiosis), altering immune expression, producing oxidative damage and host inflammation [25,26,27,28]. Dysbiotic workers suffer developmental deficiencies and early mortality suggesting that the suppression of oxidative stress via microbiota maintenance is critical for gut health and host longevity [25, 29]. Similar to gut dysbiosis in response to early life insult, age-associated succession of gut microbiota in worker bees shows increased Proteobacteria with relative decreases in core Bifidobacterium and Lactobacillus, the same general results found in many other microbiota models including insects and mammals [12, 30]. Unlike workers, the queen does not show greater antioxidant expression with age suggesting that antioxidant function is performed differently or managed by her diet [9]. Vitellogenin hemolymph concentration, constant royal jelly ingestion, and perhaps the microbiota contribute to antioxidant function in long-lived queens.

While research on the worker microbiota has progressed rapidly, little is known of queens. Based on a small sample size and high-throughput 16S subunit of the ribosomal RNA gene (16S rRNA) gene sequencing from whole guts, the queen and worker microbiota differ in taxonomic membership and community structure [19, 20, 31]. Unlike workers, the early queen gut seems dominated by two distinct species of Acetobacteraceae, Parasaccharibacter apium and an unnamed group referred to as “Alpha 2.1” [12]. Alpha 2.1 is prevalent in guts of older workers, but P. apium occupies a variety of nutrition-rich niches associated with honey bees and thrives in the presence of royal jelly [7, 32, 33]. Capable of gut colonization, P. apium is correlated with disease agents in adult bumblebees [34] and, in honey bees, implicated in poor worker health, increased mortality, worker gut dysbiosis, and strain-dependent effects on larval and pupal survival [12, 29, 33]. Often occurring with P. apium, Lactobacillus kunkeei is prevalent/abundant in queens, also considered a probiotic, and associated with worker disease and dysbiosis [12, 35]. L. kunkeei has been designated as a hive (not a gut) bacterium due to its association with fructose-rich niches like honey and honey-rich pollen storage [32, 36]. Thus, investigations of honey bee microbiota require a careful consideration of social and functional context including host longevity, caste specificity, developmental stage, potential refugia, and transmission from nutrition-related niches [22, 23, 37].

Here we test the hypothesis that lifespan differences in a social insect model are associated with age-based microbial succession. We sample known age queens from different genetic and environmental backgrounds and compare our findings to the extensive preexisting characterization of known age workers. We define the aging queen microbiota by deep sequencing the 16S rRNA gene from four alimentary tract niches that differ in many ways including physiological function, pH, and oxygen exposure. To accompany each amplicon library, we normalize bacterial cell counts with qPCR and rRNA gene copy number. Finally, we quantify protein oxidation in the fat body tissue of each queen to test the hypothesis that biological age differs from chronological age and that the accrual of oxidation products in aging queens is associated with differences in the microbiota.

Methods

Queen sampling

Our sampling design distinguished environmental exposure from chronological age. We sampled four different sets of queens: young queens (first year, n = 31) aged 4–6 months and old queens (second year, n = 32) aged 16–18 months. To control for source variation, we sampled old and young queens from similar and different backgrounds. The primary multivariate analysis of variance (MANOVA) model contains two main effects and an interaction effect, asking whether the variation in queen microbiota depends on age, background, or an interaction of both factors.

We sampled a total of 63 queens. Half (n = 32) of these queens were sourced from a large migratory beekeeping operation based in southern California. Referred to as the “CA” source, these Italian queens (Apis mellifera ligustica) were purchased from the same queen breeder in different years (mid-March of 2015 and 2016). Both sets of queens were sampled in mid July 2016. Thus, “older” CA queens (n = 16) were sampled from colonies that had survived 16.5 months and experienced almond pollination and two seasons of alfalfa pollination in the Imperial valley of southern California. Following almond pollination in 2016, “young” CA queens (n = 16) were introduced via colony splits in March, experienced one season of alfalfa pollination, and were sampled at 4.5 months of age.

The other half of our sampled queens, referred to as “AZ” queens (n = 31) were sourced from two very different environmental and genetic backgrounds. We sampled “young” AZ queens (n = 15) from the Carl Hayden Bee Research Center in Tucson, Arizona. Delivered and installed with 3000 young worker bees (package bees) in early May 2016, these Italian queens (Apis mellifera ligustica) were exposed to varied pollen and nectar sources typical of the Sonoran desert, but not intensive agriculture or bulk transportation events. Young AZ queens were sampled in early October 2016 at 5.7 months of age. In contrast, old AZ queens originated from a Northern migratory beekeeping operation that raises Apis mellifera carnica queens. These queens were introduced in April of 2015 to colony splits in the foothills east of Turlock, California following almond pollination. Colonies then experienced the summer in North Dakota making honey, pollinating oilseed crops, sunflowers, and canola. Colonies then overwintered in a temperature-controlled warehouse in Idaho (November–January) and pollinated almonds in central California (February). The colonies were then delivered to Tucson, Arizona in March of 2016 where they flourished for 7 months before queens were sampled in early October at 18 months of age.

All 63 queens were collected into sterile 2.0-ml tubes and immediately frozen on dry ice and stored at − 20 °C for DNA extraction. Queens were dissected under sterile conditions. Four tissue types were extracted to typify the queen microbiota: mouth parts, midgut, ileum, and rectum. Mouthparts were unfolded out of the head capsule and detached proximal to the labrum with sterile scissors. Individuals were then pinned through the thorax and the digestive tract was accessed by removing the dorsal abdominal sclerites. The entire digestive tract was removed and floated in 70% EtOH to wash and separate the midgut, ileum, and rectum. The abdominal fat body and attached dorsal sclerites were retained as a single unit to quantify biological age.

Queen aging assay

As a proxy for biological age, we quantified molecular by-products that cannot be excreted, but accumulate with age in abdominal fat body tissues. In honey bees, the accumulation of oxidized proteins (carbonyl groups) in the fat body is recognized as a marker of chronological age [38]. Carbonyl content of total fat body protein homogenates was determined using a Protein Carbonyl Content Assay Kit (MAK094; Sigma-Aldrich). Briefly, whole fat bodies were homogenized in 600 μl of 1X TE buffer. The supernatant was treated with a final concentration of 10 mg/ml streptozotocin to precipitate nucleic acids. The supernatant was decanted then reacted with 2,4-dinitrophenylhydrazine (DNPH) to form stable dinitrophenyl hydrozone adducts. Derivatized proteins were precipitated with trichloroacetic acid and were washed three times with acetone. The samples were resuspended in 100 μl of 6 M guanidine (pH 2.3). Protein oxidation, expressed as nanomoles of carbonyl groups per milligram of protein, was calculated by absorbance at 345 nm relative to the millimolar extinction coefficient of aliphatic hydrozones (22 mM−1 cm−1). The protein content of each sample was determined using a bicinchoninic acid (BCA) assay [39].

DNA extraction and qPCR

Dissected tissues were placed immediately into 2-ml bead-beating tubes containing 0.2 g of 0.1-mm silica beads and 300 μl of 1X TE buffer. Samples were bead beaten for a total of 2 min at 30-s intervals. To each sample, 100 μl lysis buffer (20 mM Tris-HCl, 2 mM EDTA, 5% Triton X-100, 80 mg/ml lysozyme, pH 8.0) was added and the samples were incubated at 37 °C for 30 min. DNA was then purified using a Thermo Fisher Scientific GeneJet Genomic DNA Purification Kit according to the manufacturer’s instructions for gram-positive bacteria.

We quantified total bacterial abundance for each of the four tissue types with a real-time PCR (qPCR) assay of 16S rRNA gene copies [40]. This assay provides significantly broader coverage than previously reported universal bacterial quantification assays. A standard curve was generated using a 10-fold serial dilution series of a plasmid standard containing a full-length Escherichia coli 16S rRNA gene. A 466 bp fragment in the V3–V4 region of the bacterial rRNA gene was amplified from total DNA using universal primer pair (5′-CCTACGGGDGGCWGCA-3′ and 5′-GGACTACHVGGGTMTCTAATC-3′). Quantitative PCRs were carried out on a BioRad CFX96 thermocycler in 12 μl reactions containing 9 μl of iTaq Universal SYBR Green Supermix (BioRad), 0.5 μl forward primer, 0.5 μl reverse primer, and 2 μl of DNA template. The cycling conditions were 95 °C for 3 min followed by 40 cycles of 95 °C for 10 s and 60 °C for 60 s. The assay was validated for use on honey bee-associated bacteria by confirming amplification against individual plasmid templates harboring full-length 16S genes corresponding to major gut phylotypes. The qPCR results were expressed as the total number of 16S rRNA gene copies per DNA extraction (200 μl volume elution).

16S rRNA gene sequencing

The V6–V8 variable region of the 16S rRNA gene was amplified using PCR primers 799F (acCMGGATTAGATACCCKG + barcode) and bac1193R (CRTCCMCACCTTCCTC). Amplification was performed using the HotStarTaq Plus Master Mix Kit (Qiagen, USA) under the following conditions: 94 °C for 3 min, followed by 28 cycles of 94 °C for 30 s, 53 °C for 40 s, and 72 °C for 1 min, with a final elongation step at 72 °C for 5 min. PCR products were confirmed using a 2% agarose gel. PCR products were used to prepare DNA libraries following Illumina TruSeq DNA library preparation protocol. Sequencing was performed on a MiSeq at the University of Arizona Genetics Core.

16S rRNA gene community analysis

16S rRNA gene sequences were processed using MOTHUR v.1.35.0 [41]. Forward and reverse reads were joined using the make.contigs command. After the reads were joined, the first and last five nucleotides were removed using the SED command in UNIX. Using the screen.seqs command, sequences were screened to remove ambiguous bases. Unique sequences were generated using the unique.seqs command. A count file containing group information was generated using the count.seqs command. Sequences were aligned to Silva SSUREF database (v102) using the align.seqs command. Sequences not overlapping at the same alignment coordinates, and columns not containing data, were removed using the filter.seqs command. Sequences were preclustered using the pre.culster command. Chimeras were removed using UCHIME [42], and any sequences that were not of known bacterial origin were removed using the remove.seqs command. All remaining sequences were classified using the classify.seqs command. All unique sequences with one or two members (single/doubletons) were removed using the AWK command in UNIX. A distance matrix was constructed for the aligned sequences using the dist.seqs command. Sequences were classified at the unique level with the Ribosomal Database Project (RDP) Naive Bayesian Classifier [43] using a manually constructed training set containing sequences sourced from the greengenes 16S rRNA database (version gg_13_5_99 accessed May 2013), the RDP version 9 training set, and all full-length honeybee-associated gut microbiota on NCBI (accessed July 2013). Operational taxonomic units (OTUs) were generated using the cluster command. Representative sequences for each OTU were generated using the get.oturep command. To further confirm taxonomy, resulting representative sequences were subject to a BLAST query using the NCBI nucleotide database. Diversity indices were generated using the rarefaction.single and summary.single (alpha diversity) commands.

Statistical analysis

To examine the effect of community size, we multiplied the proportional abundance of OTUs by group or species-specific 16S rRNA gene copy number and total bacterial 16S rRNA gene copies determined with qPCR for each individual queen and niche. All core bacterial genomes contain four 16S rRNA gene copies except L. kunkeei (5), Bifidobacterium asteroides (2), and P. apium (1). Acetobacteraceae Alpha 2.1 (copy number unknown) was designated a value of one, consistent with the copy number of its closest relative, P. apium [44]. In this case, qPCR-normalized abundance is extrapolated from relative abundance of amplicons, so remains compositional. OTUs at low abundance were summed (Σ OTUs 10–500), corrected for community size and mean 16S rRNA gene copy number (4.2) [45], and used to assess the change in collective low-abundance OTUs with chronological age and biological age. As an exploration of “other” or sporadic low abundance bacteria, we present analyses with and without low abundance OTUs.

We used both parametric and non-parametric tests to analyze different properties of our data. To allow the use of parametric multivariate analyses [46], we converted the qPCR-normalized bacterial abundances to ratios among all OTUs [47] using the software CoDaPack’s centered log ratio (CLR) transformation [48]. These transformations reflect the ratio abundance of all taxa in the data set. Nearly all of these transformed data sets were normally distributed [47]. A few samples deviated slightly from normal following transformation. Because our sample size is large (n = 63 for each IV), these tests are robust to slight deviations from normality. As an additional measure, we used Pillai’s Trace test statistic, also robust to violations of multivariate normality and homogeneity of covariance. The MANOVA and multivariate analysis of covariance (MANCOVA) were performed on CLR-transformed data with OTUs 1–9 as dependent variables. The MANOVA examined age and source as independent variables, and post-hoc pairwise analyses were conducted using Tukey HSD. The MANCOVA explored biological age (carbonyl accumulation) as a covariate, while partitioning variation associated with chronological age and source. We compared qPCR-normalized abundance of each bacterial taxon by age without reference to source variation using the Wilcoxon rank sum test followed by Bonferroni, Benjamini-Hochberg, and FDR corrections to account for multiple comparisons.

We examined the relationship between biological aging (carbonyl accumulation) and the microbiota in various ways: (1) we performed distance-based linear model (DistLM) on the top 37 OTUs to test whether the microbiota from each of four distinct tissues is significantly associated with carbonyl accumulation in queens, (2) we examined carbonyl accumulation as a covariate in three separate MANCOVA models (above), (3) we calculated independent Pearson’s correlations between species-specific CLR scores and log transformed carbonyl data, and (4) we performed principle component analysis (PCA) on CLR scores from OTUs 1–9, plotting the relationship of bacterial community composition and age-associated succession relative to carbonyl accumulation by niche. As a confirmation of the clustering and separation produced by PCA, we performed ANOSIM on the raw amplicon reads from the top 37 OTUs to test for microbiota similarity by age and source. For the extended queen data set, we calculated correlations among the top 200 OTUs using sparse correlations for compositional data algorithm (SparCC; [49]) as implemented in mothur [41]. SparCC is robust for compositional data sets with a low effective number of species [50]. Analyses were conducted in JMP_ v11 (JMP_ 1989–2007) and/or SAS_ v9.4 [51].

To compare the microbiota of aging queens with those of aging workers, we analyze worker age data from a recently published manuscript [19]. As one of three studies sequencing both nurses and foragers with the same primer sets and PCR parameters, the Kwong et al. study is the most robust and provides whole gut microbiota values based on 16S rRNA gene sequences from differently aged workers of Apis mellifera, n = 84 workers, 19 foragers (old) and 65 in-hive bees (younger). From this data set, we designated eight core gut bacteria representing 95% of OTU abundance based on known samples of whole gut communities in the literature. The remaining 5% OTU abundance from [19] was comprised primarily (83%) of Proteobacteria, occurred with sporadic abundance and prevalence, and was combined as an “other” category to represent low-abundance bacteria. As stated above for queens, we CLR-transformed relative abundance measures of workers and performed a one-way MANOVA on age to compare old vs. young gut microbiota, calculating post hoc differences between specific bacterial groups. To visually compare the queen and worker results, we transform our tissue-specific queen data to reflect relative abundance values predicted for the whole gut. Tissue-specific bacterial cell counts were used to normalize the relative occurrence of bacterial species by queen tissue then additively produce a single value that represents the expected result of sequencing whole queen guts. These whole gut values highlight hypothetical differences in abundance and prevalence between workers and queens.

Results

16S rRNA gene sequencing and qPCR

Next-generation sequencing returned 7.2 million quality trimmed reads (400 bp assembled) across the 252 libraries (Additional file 1: Table S1). Only four mouthpart libraries were excluded from the analysis, and read coverage in the remaining 248 libraries was sufficient for all downstream characterization and statistics (Additional file 2: Table S2). The queen rectum was represented by 2.4 million reads averaging 38 K per library, the ileum by 1.9 million reads averaging 30 K per library, the midgut by 1.6 million reads averaging 25 K per library, and the mouth by 1.3 million reads averaging 21 K per library. Like the gut microbiota of workers, that of queens is taxonomically simple and represented by only 7–12 conspicuous OTU’s dependent on niche. The remaining OTUs were highly sparse and typically at less than 10% prevalence (Additional file 1: Table S1). The nine most common 97% OTUs accounted for 98.8% of all reads across the combined niches. Given the low effective number of OTUs, unique OTUs were manually assessed to verify 97% species clusters. Subtracting the rare biosphere (1.2%), these nine OTUs are the focus of most analyses and are what we present in figures. Summed across the four niches, the nine most abundant OTUs according to raw amplicon read totals were Lactobacillus firm5 species cluster (51.3%), Parasaccharibacter apium (27.1%), Lactobacillus kunkeei (7.6%), Lactobacillus firm4 species cluster (6.8%), Acetobacteraceae Alpha 2.1 species cluster (2.0%), Bifidobacterium asteroides (1.5%), Snodgrassella alvi (1.8%), Gilliamella apicola (0.3%), and Delftia spp. (0.2%). In honey bees, Delftia is genus of Burkholderiales not recognized in workers.

Similar to the abundance pattern in worker guts, the queen rectum harbors an average of 121.2 M 16S rRNA gene copies, a magnitude more than the ileum (17.9 M) or midgut (14.2 M). The mouth (1.4 M) contains the least bacteria. Total amplicon reads returned for the mouth, midgut, and ileum were significantly correlated with bacterial abundance as determined by qPCR (Additional file 2: Table S2). OTU dominance in the queen increased with community size in the mouth, midgut, and ileum (Additional file 2: Table S2). Extrapolating qPCR results to estimate community size, P. apium, and L. kunkeei decrease in abundance approaching the rectum, while L. firm5, L. firm4, B. asteroides, and Alpha 2.1 increase (Additional file 3: Table S3). S. alvi and G. apicola occur sporadically at low (< 1%) average relative abundance throughout all queen niches.

MANOVA of queen microbiota by chronological age and source

The two-way MANOVA performed for each of the four queen niches revealed significant variation due to chronological age, source, and interaction (Table 1, Additional file 4: Table S4). In the mouth, P. apium and L. firm5 increased with age, while Alpha 2.1 and Delftia were more abundant in young queens. The midgut and ileum aged similarly; in both niches, B. asteroides and L. kunkeei were more abundant in old queens while Alpha 2.1, Delftia, and “other” all decreased with age (Fig. 1). Most abundant in the ileum relative to other queen tissues, S. alvi bloomed in 4 of 63 individuals and increased with age, while P. apium and G. apicola decreased (Additional file 3: Table S3). In the rectum, where L. firm5 represents the majority of total gut bacteria, B. asteroides abundance increased with age while L. firm5 and both core Acetobacteraceae (P. apium and Alpha 2.1) decreased (Fig. 2). Wilcoxon rank sum tests revealed significant differences by chronological age, many of which agree with age-specific differences detected in the two-way MANOVA (Table 1, Additional file 4: Table S4). Wilcoxon tests provide context comparing chronological age without regard to environmental source (Additional file 5: Table S5).

Table 1 Queen microbiota: results by niche, chronological age, source and biological age
Full size table
Fig. 1
figure 1

Age-associated bacterial succession of distinct longevity phenotypes. Honey bee host differences (pink panels) reflect aging physiology. In the context of life history theory, workers are literally the “disposable soma,” while queens represent reproduction [10]. Vertical arrows indicate the direction of change with increasing age. Lactobacillus and Bifidobacterium are listed in the blue panels and Proteobacteria in the yellow panels. All listed bacterial groups differ significantly in ratio abundance. The microbiota of the short-lived worker phenotype represents a meta-analysis of Apis mellifera gut libraries from Kwong et al. [19]. Queens were analyzed in the present study (see Table 1)

Full size image
Fig. 2
figure 2

The honey bee queen microbiota by tissue. Color-coded bars represent relative abundance corrected by species-specific 16S rRNA gene copy number (see Additional file 3: Table S3 for normalized abundance). The 4 × 4 panel displays the top nine most abundant OTUs by niche, age, and source. Black represents the summation of OTUs 10–500. Old queens in the upper two rows are 16–18 months of age and young queens in the bottom two rows are aged 4.5–5.7 months (Fig. 4)

Full size image

MANOVA of worker microbiota by age

Analyzing data published in Kwong et al. (2017), we investigated age-associated changes in the gut microbiota according to the behavioral role of the worker. The one-way MANOVA of microbiota by worker age (task) revealed major proportional shifts among core gut bacteria (Fig. 1). Bifidobacterium, L. firm5 and L. firm4 decreased significantly with age while “other” bacteria, Acetobacteraceae Alpha 2.1, and Bartonella apis increased (Additional file 6: Table S6). Of note, S. alvi decreased in relative abundance with age but was borderline insignificant (p = 0.06). With reference to these results, and other worker data sets in the literature [19], we use a cutoff value of 50% prevalence at 0.5% average relative abundance to define bacterial taxa specific to each longevity phenotype. Applying this value, we identify four worker-specific gut groups, all Proteobacteria, two queen-specific groups, and four groups shared by longevity phenotypes (Fig. 3). In general, the four bacterial groups with fidelity for the rectum are shared by longevity phenotypes while the ileum-specific groups differ by longevity phenotype.

Fig. 3
figure 3

Average abundance and prevalence of gut bacteria in queens (n = 63) and workers (n = 83). Workers are whole gut samples from Kwong et al. [19]. Queen data was normalized by tissue-specific community size to reflect relative abundance values expected from sampling whole guts. The red bars represent average abundance, black bars are prevalence defined at ≥ 0.5% relative abundance, and the bar apex is prevalence defined as two or more reads per gut library. We did not detect F. perrara* in any of the four sampled queen alimentary tract niches

Full size image

Biological age, source, and the queen microbiota

We measured carbonyl accumulation in the queen fat body as a proxy for queen biological age. While some variation in carbonyl accumulation is due to genetics and environment, difficult to excrete waste products accumulate in a clock-like fashion with age. We found that chronological age was strongly associated with carbonyl content in the fat body of the queen (Fig. 4). Carbonyl accumulation differed by both chronological age and source (Additional file 7: Table S7). Examining all pairwise combinations, only first year queens (CA1 and AZ1) did not differ in average carbonyl accumulation. In both sets of young and old queens, chronological age did not agree with biological age. In both age classes, queens from the Imperial Valley of California (source CA) were chronologically younger, but biologically older with greater carbonyl accumulation (Fig. 4). The finding that biological age differed significantly from chronological age suggests that “source” (environmental exposure and genetic background) is a major contributor to biological age and is a confounding factor for the analysis of chronological age. For this reason, we emphasize the statistical tests and interpretations that control for source variation.

Fig. 4
figure 4

Carbonyl accumulation (protein oxidation) in queen fat body differs by chronological age (F3, 59 = 48.3; P < 0.0001***) and source (t = 2.2; df = 30, P = 0.03*)

Full size image

To further explore the relationship of biological age and source variation with queen microbiota, we ran a set of related analyses that partition variation by different strategies. DistLM revealed a significant association between microbiota composition and carbonyl accumulation in each of the four tested communities (Mouth, pseudo-F61 = 4.1: P = 0.01; midgut, pseudo-F61 = 3.7: P = 0.006; ileum, pseudo-F61 = 4.1: P = 0.004; rectum, pseudo-F61 = 3.9: P = 0.005). Although all four communities were significantly associated with carbonyl accumulation, little variation was explained by the collective community (mean R-sq = 0.06) due to opposing species variation within communities. The separation of background (source), chronological age, and carbonyl accumulation via MANCOVA analyses detailed OTU-specific changes in the microbiota (Additional file 8: Table S8). Pearson’s correlations examining species-specific CLR log-transformed OTU abundance and log-transformed carbonyl values agree with the main MANCOVA results examining source as the dependent variable with carbonyl accumulation as the covariate without reference to chronological time (Additional file 9: Table S9). Most notably, throughout the gut, B. asteroides was correlated significantly with the accumulation of carbonyl in abdominal fat body tissue (Fig. 5). Although at similar abundance in chronologically old and young queens, L. firm5 abundance was also correlated strongly and positively with carbonyl accumulation. Rare throughout the queen gut, an undescribed Burkholderiales, Delftia spp. showed the strongest negative relationship with carbonyl content, decreasing dramatically with age and varying by source (Table 1).

Fig. 5
figure 5

Principle components analysis by niche based on the top nine most abundant OTUs and carbonyl accumulation. The colored symbols illustrate differences among the chronological sample cohorts; pink and orange are young and blue and green are old. The green vector illustrates carbonyl accumulation relative to community structure, shows strong affinity with increased Lactobacillus and B. asteroides in the gut, and is largely allied with the biologically oldest queen cohort (CA2). Orange symbols are biologically the youngest and consistently allied with P. apium in the hindgut and Acetobacteraceae Alpha 2.1 and Delftia throughout the system. Biplot constructed with normalized bacterial cell abundance data, transformed to centered log ratios (CLR) that represent the change in taxon abundance (covariance) relative to all other taxa in the data set. The species vectors are proportional to the standard deviation of the ratio of each taxon to all other taxa. In general, clustered groups of points contain similar groupings of taxa with similar ratio abundances, and longer OTU vectors result from greater variation in CLR scores. The parentheses below each niche label contain the percent variation explained by the first and second principle component respectively (Additional file 10: Table S10)

Full size image

To visualize variation associated with biological age in the queen microbiota, we performed PCA analysis using centered log ratios from the top nine OTUs and associated carbonyl values from the fat body of each queen (Fig. 5, Additional file 10: Table S10). Across each niche, the first two principle components explained approximately 50% of the variation in log ratio abundance scores. Because the queen microbiota has shallow, deep, and noisy structure, the third and fourth principle components for each niche explained an average of 14 and 8%, respectively (Additional file 10: Table S10). Although only 50% of the variation is presented in the two dimensional PCAs, a strong and consistent separation of two queen cohorts is realized in every niche, young Arizona (AZ1) and old California (CA2). In each niche, the carbonyl vector indicates CA2 as the oldest, and AZ1 as the youngest cohort, consistent with determinations of biological age. Also consistent with results for biological age, ANOSIM results support the PCA clustering and separation (Additional file 11: Table S11).

We examined microbiota correlations using SparCC, an approach that incorporates the structure of the data matrix to identify potential species interactions and generates null expectations based on permutation of OTU columns in the transformed data matrix. Based on SparCC, the mouth and midgut reveal a number of significant positive relationships between core bacteria within niche (Additional file 12: Table S12). We note that SparCC results are unreliable when OTU sparsity exceeds 70% zero values but robust to communities with a low effective number of species [49] (see Table 1). The ileum reveals a marked decrease in positive relationships and the first occurrence of significant negative relationships. As the relevant dynamic, the two major Acetobacteraceae (Alpha 2.1 and P. apium) associate positively in the ileum, but both associate negatively with L. firm5 and Bifidobacterium. The strongest negative correlation occurs between L. firm5 and P. apium, the two most abundant ileum species (Additional file 12: Table S12). With more detailed investigation, age-specific Pearson’s correlations on log-transformed normalized abundance shows that as queens age, the relationship of L. firm5/P. apium cell number shifts from mildly negative (Pearson’s r = − 0.27, p < 0.07) to strongly positive (Pearson’s r = 0.49, p < 0.005), concurrent with the loss of Acetobacteraceae (Alpha 2.1 and P. apium).

Discussion

Our results support the hypothesis that gut microbiotas of queen and worker longevity phenotypes differ in structure and membership and show age-associated changes concurrent with host physiology, diet, and behavior (Fig. 1). Because long- and short-lived phenotypes are produced from the same genotype, microbiota establishment and age-associated changes likely reflect host regulatory anatomy and environmental exposure, primarily diet. Long-lived (queen) phenotypes are fed royal jelly throughout their lives to replenish internal levels of vitellogenin. In their youth, short-lived (worker) phenotypes consume pollen to produce a discrete pulse of vitellogenin that fuels royal jelly synthesis in their head glands. In old age, workers forage for pollen and nectar consuming honey to support flight metabolism. This fundamental difference in diet and behavioral task reflects a suite of age-associated host gene expression, highlighted by differences in vitellogenin production, immunity, insulin signaling, and antioxidant levels [9, 14, 16, 25, 52, 53]. These core changes in host physiology are consistent with the distinct microbiota compositions and age-based succession of honey bee longevity phenotypes (Fig. 1). In general, aging worker guts show decreased Lactobacillus and Bifidobacterium and increased Proteobacteria adding to the list of insect and mammal systems where this pattern has been documented [30]. In stark contrast, the gut microbiota of aging queens is depleted of core and other Proteobacteria and accumulates core hindgut bacteria typically considered probiotic like Lactobacillus and Bifidobacterium.

We tracked chronological queen age, but also quantified carbonyl accumulation in the queen fat body as a marker of biological aging (Fig. 4). We hypothesize that the portion of microbiota variation best explained by biological age may be associated with environmental variables including diet quality. Vitellogenin is the major fat body lipoprotein and plays an essential role in oxidative stress resistance and honey bee lifespan [14, 18]. The rate of vitellogenin synthesis in laying queens is 20 times that of workers, and much of the oxidative damage in queens is absorbed by vitellogenin [14, 18]. Changes in diet quality or other environmental variables that influence colony homeostasis (i.e., overheating and intensive water collection) may affect the production and quality of royal jelly produced by workers and fed to the queen. In turn, the royal jelly diet provides defense against oxidative damage in the queen gut and fuels the production of vitellogenin throughout the queen hemolymph. Similarly, workers deprived of pollen are extremely low in vitellogenin and do not develop a normal microbiota suggesting that pollen serves as a prebiotic in the worker gut [24] and/or that the regulatory anatomy associated with the conversion of ingested pollen to vitellogenin is intimately linked to the development of a normal gut microbiota [25].

Longevity phenotypes differ in core membership

Six of the gut bacterial groups (or species) show strong specificity by longevity phenotype (Fig. 3). The four Proteobacteria we classified as worker-specific show distinct patterns of rarity in queen guts. In an evolutionary context, the two most recent additions to the worker gut microbiota are B. apis and Frischella perrara [19]. Perhaps a result of this novelty, these bacteria exhibit a relatively narrow niche breadth. F. perrara is specific to the worker pylorus and results in host melanization response, while B. apis appears in the hindguts of older foragers [28, 54,55,56]. In queens, B. apis was extremely rare and F. perrara not detected, not even on the mouth, suggesting that these particular Proteobacteria are not tolerated by the queen or excluded via some mechanism perhaps the antimicrobial mechanisms of royal jelly. In contrast, worker-specific G. apicola and S. alvi are tolerated at low levels in queen guts, and S. alvi showed sporadic abundance in 4 of 63 seemingly healthy queen ileums (Fig. 2). In workers, this species pair is omnipresent, accounts for 20–60% of the ileum microbiota, and represents a core syntrophic relationship critical to gut oxygen balance [25, 26, 57, 58]. G. apicola produces acetate under low oxygen, and S. alvi is an obligate aerobe that consumes acetate to fuel its oxygen metabolism [25]. Although at low abundance in queens, this species pair is highly correlated throughout all sampled queen niches occurring with ≤ 10% prevalence defined at 0.5% average abundance (Fig. 3). Queen-specific bacterial species are Parasaccharibacter apium and Lactobacillus kunkeei, both showing strong fidelity for queen mouth, midgut, and ileum. These two species may be associated with the royal jelly diet of the queen, but occur with sporadic abundance in worker ileums under conditions of putative dysbiosis and oxidative stress [12, 23, 29, 34].

The four bacterial groups shared by queens and workers differ in abundance and prevalence showing strong niche fidelity (Fig. 3). Of these four, only Lactobacillus firm5 is core to both the ileum and rectum of both longevity phenotypes. Considering whole guts independent of age, queens and workers average 75 and 24% relative abundance of Lactobacillus firm5, respectively. Three of the four shared core groups (L. firm5, L. firm4, and B. asteroides) populate 100% of workers by 3 days of age [22]. Of these three, B. asteroides is significantly more abundant and prevalent in workers than queens (particularly young workers), perhaps associated with pollen consumption [24, 59]. However, B. asteroides increases significantly in the hindguts of aging queens independent of pollen consumption. The fourth shared group, Acetobacteraceae Alpha 2.1, is abundant in young queens but not typically detected in young workers. It decreases with queen age but becomes prevalent and abundant in older workers [19, 31, 60,61,62,63,64].

The queen microbiota improves with age

Our results are consistent with the body of work detailing biological aging and oxidative stress in queens, workers, and social insects in general [8,9,10, 65]. Results from the queen carbonyl assay demonstrate that queens accrue oxidative damage with age (Fig. 4) and that chronological age can differ significantly from biological age possibly due to environmental differences including climate, nutrition, toxins, and other landscape variables. Despite similar signs of biological aging in both queens and workers, the aging gut microbiota of older queens seems to reflect a refined structure with greater efficiency. It is unlikely that queens ever develop a senescence physiology and associated microbiota as seen in workers. Under natural conditions, queens accrue biological damage associated with aging, but are not allowed to grow “old” because fecundity is critical to colony survival, and workers routinely replace substandard queens [66]. With increased oxidative damage, gram-positive bacteria decrease in workers [19] but increase in queens (Fig. 1). Of note, core Lactobacillus and B. asteroides in queens show greater correspondence with biological than chronological age suggesting that these species may track and/or signal host physiology (Additional file 8: Table S8). Consistent with decreased antioxidant expression and less generation of reactive oxygen species in queens [9, 65], the bacteria that increase with queen age do not rely on oxygen, but generate continuous fermentative metabolism in the queen hindgut (Fig. 2). In turn, these fermentation products (i.e., butyrate) are considered fundamental to host physiology and homeostasis [25, 59].

Results from conventionalized bee experiments suggest that butyrate produced by the honey bee worker hindgut microbiota plays a key role in host metabolism [25, 59]. In human colons, positive butyrogenic effects are considered a result of cross feeding by butyrate-producing Firmicutes and Bifidobacterium [67]. Feeding worker honey bees’ relevant amounts of sodium butyrate results in gene expression considered beneficial to general health, broadly affecting immunity and detoxification [68]. We found that bacterial communities implicated in butyrate production were diminished in aging workers but seemingly enhanced in aging queens. Better explained by biological than chronological age (Fig. 5), B. asteroides increased significantly with age in the queen midgut, ileum, and rectum (Additional file 8: Table S8 and Additional file 9: Table S9). Moreover, L. kunkeei increases in the midgut and ileum, while community changes in the ileum favor the persistence of Lactobacillus firm5 (Table 1), and suggests a more efficient relationship that emerges with queen age. Lactobacillus firm5 is the most plentiful bacteria in the queen hindgut, and combined with increased B. asteroides, may add to the butyrogenic effect in queens concurrent with increased biological age. B. asteroides itself was recently identified as a major bacterium associated with host-derived signaling molecules in worker honey bees [59]. Interestingly, B. asteroides abundance in both queens and workers is often low and/or highly variable so may be affected by diet or strain variability [24].

We compared the gut microbiota of young in-hive bees to older foragers within and among studies. Foraging is the last functional role workers serve before death. But as a behavioral cohort, both in-hive bees and foragers can range greatly in chronological age and environmental exposure [69]. Also, comparing across next-generation sequencing studies can be misleading due to differences in methodology like primer choice or analysis pipeline [70]. Despite these and other sources of potential error, we found that the worker gut microbiota ages in a highly predictable fashion, becoming depleted of core hindgut Lactobacillus firm4, firm5, and B. asteroides (Fig. 1). Of seven available studies sequencing the 16S gene from forager guts, three used the same primers and methods to sequence both foragers and nurses [19, 31, 63], and we used these studies as a point of reference for examining worker aging. We analyzed the largest and most variable of these three data sets [19] defining six significant differences in microbiota between young and old workers (Fig. 1). The collective results from six of seven studies are largely in agreement and suggest that age-associated shifts in worker microbiota are strongly predictable despite study particulars including the use of different primer sets [19, 31, 60,61,62,63,64]. The changes we report in Fig. 1 [19] represent a functional change from a fermentative to proteolytic hindgut environment, involving significant shifts in core bacterial structure. Acetobacteraceae Alpha 2.1 increases in all seven studies, B. apis, and “other” bacteria in 6 of 7. One to three major core hindgut bacteria are depleted significantly in six of seven studies, while studies were more variable concerning shifts of S. alvi and G. apicola, the species pair that dominates the worker ileum.

Early gut succession

Similar to workers [37], the rectum of the mature queen contains 84% of the total bacteria found in the queen gut (Additional file 2: Table S2). On average, a whole gut sample from a mature laying queen would be highly biased toward rectum species, dominated by Lactobacillus firm5 (Fig. 3). In contrast, whole gut samples of queens during the mating process show a dominant Acetobacteraceae (P. apium and Alpha 2.1) profile [20]. This finding is consistent with our detection of significantly more P. apium and Alpha 2.1 in the guts of younger queens (Fig. 2, Table 1) and suggests that the bacterial succession leading to a Lactobacillus firm5 dominant hindgut in queens may require many weeks, perhaps months. Given that worker gut succession occurs throughout the life of the worker [12, 22], we speculate that the early queen microbiota [20] represents a pioneer community that primes the gut environment or host immune system and/or potentially aids disease prevention during the days-long mating process that involves queen flight metabolism and mating with > 20 males. A successfully mated queen is fed massive amounts of royal jelly as she begins to lay eggs. The decrease and stabilization of cell replacement rate in early queen midguts [71] suggests a more stable gut environment emerges around 40 days of age, perhaps influencing bacterial succession.

Queen niche breadth

In queens, the occurrence patterns and numerical dominance of P. apium in the mouth and midgut and Lactobacillus firm5 in the ileum and rectum suggests that the extended gut structure is important for host function (Fig. 3). The antimicrobial properties of royal jelly may be more manifested in the mouth and midgut. The taxonomic shift at the pylorus demarcates a steep physiological gradient in the adult bee gut. Recently characterized in workers, this change occurs just upstream of the ileum where Malpighian tubules feed host waste products back into the gut, and microoxygenic and pH gradients affect bacterial establishment and persistence [25, 29, 54]. Host excretions provide a different niche for bacterial co-evolution including an influx of nitrogenous waste compounds, a decrease in oxygen availability and lower pH [25]. While the effect of pollen consumption on host signaling has been investigated in workers [24, 25], the effect of the queen’s diet (royal jelly) on gut physiology and host signaling remains unknown. The reliable and predigested nature of the queen diet may generate very different collection of waste products, supporting hindgut bacterial strains distinct from those found in workers.

It is mostly unknown why queens can resist many worker diseases and vice-versa. Early queen death or rejection by the colony has become more common [66, 72], and defining disease states in queens will rely in part on the structure and function of native gut bacteria [12]. Although rare throughout the gut, the occurrence pattern of Delftia (Burkholderiales) suggests it could be detrimental. Not typically detected in workers, Delftia is negatively correlated with Lactobacillus firm5 and B. asteroides in the queen hindgut, shows the greatest negative correlation with carbonyl accumulation, and decreases significantly with biological age (Table 1). Congruently, Delftia is negatively correlated with putatively beneficial bacteria on the queen mouth and midgut (Additional file 9: Table S9 and Additional file 10: Table S10). These two niches are dominated by distinct sequovars of P. apium, a bacterium co-evolved to thrive on royal jelly [7]. Over 95% of the mouth/midgut bacteria classify as P. apium and L. kunkeei, both associated with decreased abundance of honey bee-specific disease caused by bacteria and microsporidia [73, 74]. One primary function of microbes in the queen mouth and midgut may be the exclusion of opportunistic and disease causing microbes. Mouth communities not dominated by P. apium are much smaller in size and contain significantly more Delftia, OTU diversity, and “other” bacteria (Additional file 3: Table S3) suggesting that P. apium dominance in the queen mouth and midgut limits the occurrence of detrimental bacteria in the hindgut. Older queens have significantly more P. apium on their mouths and L. kunkeei in their midguts that may accrue with age and/or improve queen hygiene promoting queen survival (Fig. 2). Pollen exposure and consumption may render workers more vulnerable than queens to frequent pathogen invasion. The queen and her constant diet of royal jelly may discourage novel microbial acquisition and provide a strong selective environment for the evolution of niche specialists. The constant consumption of royal jelly likely represents a form of purifying selection, perhaps even an arms race at the front end of the queen, producing fierce competition among P. apium strains for this constant and complete nutrient source.

Evolution of “queen-specific” gut bacteria

P. apium, L. kunkeei, and close ancestors occur throughout solitary and social bees and may even predate the evolution of corbiculate bees [19, 75,76,77,78,79]. Both P. apium and L. kunkeei grow at extreme sugar concentrations and royal jelly enhances the in vitro growth of some strains [7, 33]. The evolution of host behavior to mechanically concentrate nectar sugars via evaporation was likely a key innovation producing strong selection for these two osmotolerant symbionts. Bacteria adapted to survive in concentrated nectar of solitary bee provisions were well positioned to develop greater fidelity with the host gut. The mature worker ileum is dominated by core bacteria S. alvi and G. apicola that co-occur in a biofilm with lesser amounts of Lactobacillus firm5 [12]. In contrast, the mature queen ileum is dominated by Lactobacillus firm5 that co-occurs with lesser amounts of core gut bacteria P. apium and L. kunkeei (Fig. 2). Reciprocally, worker ileum bacteria S. alvi and G. apicola are found at similarly low levels in the queen ileum and show sporadic abundance in the queen. These symmetrical occurrence patterns suggest antagonistic co-evolution of caste-specific gut bacteria, a hypothesis consistent with host age phenotype and development-specific pathogen strategies.

Over 16 L. kunkeei genomes have been compared, revealing core functionality and a large variety of accessory protein clusters that characterize different strains [80]. Isolated from the gut of A. mellifera, strains MP2 and EFB6 of L. kunkeei, were most related and differ from other L. kunkeei in possessing genes implicated in gut colonization including cell adhesion, biofilm formation, and horizontal transfer [35, 80]. These may represent strains that colonize the queen midgut and ileum. They may also colonize gut environments of workers and larva. That many of the L. kunkeei genomes lack gut-specific genes suggests they may lead more opportunistic life cycles within the hive and pollination environment. Similarly, the genome of Parasaccharibacter apium also reveals multiple functional traits for biofilm life in the insect gut, including survival in low oxygen environments and adhesion to host epithelium [44]. Like S. alvi, and many other acetic acid bacteria, P. apium can grow aerobically and assimilate major fermentation by-products generated by neighboring bacteria. Collectively, this suggests that P. apium metabolism in the queen ileum may be somewhat analogous to S. alvi function in the worker ileum [57].

Patterns of species co-occurrence suggest selection pressure for honey bee gut bacteria to co-exist with other bacteria in a biofilm encouraging competition and co-evolution (Fig. 4, Additional file 12: Table S12). This hypothesis is supported by the complex of highly correlated bacteria on the queen mouth and strongly affiliated species pairs occurring regardless of niche. Not strongly associated with age, niche, or background, at least three pairs of co-occurring species emerge as potential syntrophic relationships throughout the queen microbiota, and may rely on co-evolved traits to ensure niche occupation. This strategy would prove more effective in the queen gut, which provides a more stable long-term environment where partnerships have more generational time to evolve. Many bacterial pairs have evolved strict affiliations with one another and multiple hive niches including P. apium/L. kunkeei, G apicola/S. alvi, and Lactobacillus firm4/B. asteroides (Figs. 3 and 4). Perhaps through their reliance on one another, core bacteria better survive within and outside their preferred niche.

Conclusions

The honey bee is a metabolic model for the effects of aging and diet on microbiota. Sampling the honey bee microbiota with respect to chronological age, biological age, and environmental exposure facilitates an informative partitioning of variation associated with longevity phenotypes. Consistent with established differences in host diet, development and aging, the queen microbiota shifts toward the fermentative metabolism of well-known gram-positive species, while the more rapidly aging worker is progressively depleted of these same species. Given the spectrum of influence of gut microbiota on worker physiology, we suggest that the queen microbiota serves a similarly critical role in host signaling and protection. Separate evolutionary trajectories for caste-specific gut bacteria reflect overt differences in diet and longevity between workers and queens. This trajectory appears to have tracked division of labor evolution, perhaps involving key innovations like nectar concentration to produce honey, and the production of royal jelly in worker hypopharyngeal glands. Once considered bacteria associated with worker gut dysbiosis and larval nutrition, L. kunkeei and P. apium must now be understood as core gut bacteria of Apis mellifera queens. Our results suggest that these two species occupy a functional niche in the queen mouth, midgut, and ileum. The co-occurrence and correlational abundance of multiple core species throughout the honey bee system suggest syntrophic relationships are commonplace. More generally, our study highlights the importance of controlled temporal and tissue-specific data to understand the total diversity and function of the honey bee microbiome.