Microbiome assembly and maintenance across the lifespan of bumble bee workers

Abstract

How a host's microbiome changes over its lifespan can influence development and ageing. As these temporal patterns have only been described in detail for a handful of hosts, an important next step is to compare microbiome succession more broadly and investigate why it varies. Here we characterize the temporal dynamics and stability of the bumble bee worker gut microbiome. Bumble bees have simple and host‐specific gut microbiomes, and their microbial dynamics may influence health and pollination services. We used 16S rRNA gene sequencing, quantitative PCR and metagenomics to characterize gut microbiomes over the lifespan of Bombus impatiens workers. We also sequenced gut transcriptomes to examine host factors that may control the microbiome. At the community level, microbiome assembly is highly predictable and similar to patterns of primary succession observed in the human gut. However, at the strain level, partitioning of bacterial variants among colonies suggests stochastic colonization events similar to those observed in flies and nematodes. We also find strong differences in temporal dynamics among symbiont species, suggesting ecological differences among microbiome members in colonization and persistence. Finally, we show that both the gut microbiome and host transcriptome—including expression of key immunity genes—stabilize, as opposed to senesce, with age. We suggest that in highly social groups such as bumble bees, maintenance of both microbiomes and immunity contribute to inclusive fitness, and thus remain under selection even in old age. Our findings provide a foundation for exploring the mechanisms and functional outcomes of bee microbiome succession.

1. INTRODUCTION

Understanding how and why microbial communities change over time is a fundamental goal of microbial ecology (Fierer et al., 2010; Nemergut et al., 2013; Shade et al., 2013). For host‐associated microbiomes, the local environment can change dramatically across the host lifespan, influencing their temporal dynamics (Hammer & Moran, 2019; Koenig et al., 2011; Lovat, 1996; Redford & Fierer, 2009; Rera et al., 2012). These dynamics may also have functional consequences, possibly influencing or regulating host development and life history processes (Coon et al., 2017; Hammer & Moran, 2019; Heintz & Mair, 2014; Koropatnick et al., 2004). We have an increasingly clear picture of microbiome succession in humans and in certain models for biomedical and symbiosis research (Costello et al., 2009; Ellegaard & Engel, 2019; Gilbert et al., 2018; Laughton et al., 2014; Obadia et al., 2017; Stephens et al., 2016; Vega & Gore, 2017), for which a range of methods have been used to describe the dynamics of both microbes and host processes in great detail. Patterns can vary substantially across hosts. For example, in primary succession, stochastic colonization dynamics observed in Drosophila melanogaster and Caenorhabditis elegans (Jones et al., 2022; Obadia et al., 2017; Vega & Gore, 2017) contrast with predictable gut microbiome assembly in human infants and honey bees (Bäckhed et al., 2015; Powell et al., 2014; Stewart et al., 2018). On the other hand, convergent patterns are also observed, especially with respect to microbiome maintenance in old age. In humans, gut microbiome composition becomes more variable in the elderly, with losses of core symbiont species (Biagi et al., 2010; Claesson et al., 2011, 2012; Wilmanski et al., 2021). In laboratory models, gut microbiomes also shift (though in various ways) in old age (Erkosar & Leulier, 2014; Smith et al., 2017; Stephens et al., 2016; Thevaranjan et al., 2017); these shifts may constitute a form of senescence, both responding and contributing to deterioration of gut physiology and immunity (Clark et al., 2015; Fransen et al., 2017; Thevaranjan et al., 2017). However, given major biological differences, it is difficult to explain why we see divergent or convergent successional trajectories among these groups.

Marker gene‐based studies of microbiome succession from a greater diversity of hosts suggest a much broader array of temporal patterns in nature. For example, microbiome assembly can differ even between closely related hosts, such as humans and chimps (Reese et al., 2021), and in some hosts, microbiome senescence does not seem to occur (Risely et al., 2021). However, typical marker gene (e.g., 16S rRNA) amplicon sequence data sets lack information on taxa not amplified by the chosen primer set, on absolute abundance, and on activity (e.g., live vs. dead, replicating vs. dormant). Short amplicon data sets of conserved genes also lack phylogenetic resolution below the species (i.e., amplicon sequence variant [ASV]) level, masking subspecies or strain‐level dynamics. Furthermore, many of these studies lack data on host processes that might impact microbes, such as immune responses, especially at spatial and temporal scales relevant to microbial dynamics. For a general understanding of how and why host‐associated microbiomes change over time, it is crucial to develop a broader range of host–microbiome systems, studied with a comprehensive set of host‐ and microbe‐level analyses.

Eusocial corbiculate bees (honey bees, bumble bees and stingless bees) are a promising group for comparative, in‐depth studies of microbiome dynamics. First, these bees are key pollinators in natural and agricultural ecosystems, and bacterial symbionts have functional roles in host health (Hammer, Le, Martin, & Moran, 2021; Kwong & Moran, 2016; Menezes et al., 2015). Therefore, the dynamics of these microbes could have important consequences for bees as well as plants. Bees are threatened by a variety of anthropogenic stressors (Goulson et al., 2015), and baseline temporal variability in the microbiome needs to be measured in order to observe perturbations and to study resilience (Faust et al., 2015; Zaneveld et al., 2017). Second, these three bee clades are related, are ecologically similar in many respects and share some conserved symbiont taxa, but they also differ in key life history and ecological traits, as well as in the composition and functional potential of their microbiomes (Cerqueira et al., 2021; Hammer, Le, Martin, & Moran, 2021; Kwong & Moran, 2016). This contrast among close relatives provides an opportunity to study how host traits shape the evolution of microbiome dynamics. Third, social bees have host‐specific and very simple gut microbiomes, dominated by just a few core bacterial lineages (Kwong, Medina, et al., 2017). This simplicity makes it easier to delve below community‐level patterns to study the temporal dynamics of individual species and strains within the microbiome. Distinct microbial taxa may exhibit different life history strategies that shape colonization and persistence (Ho et al., 2017; Livingston et al., 2012; Smith et al., 2018; Yawata et al., 2014). These strategies are poorly understood for microbes within host‐associated communities, but are probably important for shaping community‐level succession (Ho et al., 2017).

Gut microbiome temporal dynamics have been relatively well studied in the Western honey bee (Apis mellifera). The gut microbiome is completely different between larvae and adult workers (Martinson et al., 2012), and it continues to change in composition and abundance as workers age (Cai et al., 2022; Ellegaard & Engel, 2019; Kapheim et al., 2015; Kešnerová et al., 2020; Powell et al., 2014). However, as A. mellifera workers go through a highly stereotyped sequence of tasks over time (age polyethism) (Seeley, 1982), age and task effects are difficult to disentangle. For example, honey bees do not defecate until they become foragers and leave the hive for the first time (Winston, 1991); this could contribute to a decrease in microbial abundance from young nurses to older workers (Kešnerová et al., 2020). Indeed, differences in task performance alone (in‐hive tasks vs. foraging) are associated with microbiome differences in age‐matched workers (Jones et al., 2018). Furthermore, the oldest workers are those that overwinter and enter a phase with distinct metabolic, immunological, thermal and behavioural (e.g., lack of defecation) characteristics (Amdam & Omholt, 2002; Kešnerová et al., 2020; Steinmann et al., 2015). The gut microbiome can be quite stable into old age in these long‐lived workers (Maes et al., 2021). It is unclear to what extent gut microbial succession in A. mellifera will extend to other bees that lack strong age polyethism and unique overwintering phenotypes.

Bumble bees (Bombus spp.) differ from honey bees in many ways that probably relate to microbiome dynamics (Hammer, Le, Martin, & Moran, 2021). Workers exhibit comparatively weak age polyethism (temporal division of labour); tasks are generally carried out by workers of all ages, though some tasks are more likely to be performed at certain ages (Cameron, 1989). Symbionts are transmitted between generations by a single queen, instead of by a large group of workers as in honey bees; this changes the bottleneck size and, potentially, selection on caste‐specific maintenance processes (Hammer, Le, Martin, & Moran, 2021). They also lack certain bacteria characteristic of honey bees and have gained Candidatus Schmidhempelia bombi (hereafter, Schmidhempelia; Kwong, Medina, et al., 2017; Martinson et al., 2014). A unique practical advantage of bumble bees is that full colonies can be reared indoors. This provides an opportunity to study intrinsic ageing processes under optimal conditions, in the absence of environmental variation, and to sample microbiomes of very old bees that would normally be rare due to extrinsic mortality. Moreover, the bumble bee gut microbiome seems uniquely prone to disturbance: field‐collected workers are often found to lack the core symbionts and instead harbour opportunistic environmental bacteria (Hammer, Le, Martin, & Moran, 2021; Koch et al., 2012; Li, Powell, et al., 2015). This phenomenon has been linked to colony age, but old colonies will also tend to have older workers on average (Koch et al., 2012). Whether microbiome disturbance is due to individual senescence has not been fully resolved.

Previous work has outlined the early stages of gut microbiome succession in bumble bees (Meeus et al., 2013; Su et al., 2021; Wang et al., 2019), but there is no information on what happens to the microbiome in old age. We also lack information on the temporal dynamics of endogenous processes (e.g., immunity) in the bee gut that control gut microbes, and may be controlled by them (Evans & Lopez, 2004; Kwong, Mancenido, & Moran, 2017). Gut physiology and immunity senesce in many animals (Clark et al., 2015; Peters et al., 2019; Rera et al., 2012; Thevaranjan et al., 2017), but these processes—and senescence generally—may operate quite differently between different castes of eusocial insects (Heinze & Giehr, 2021; Keller & Genoud, 1997). In bumble bees, a solitary queen founds the colony and produces cohorts of (mostly) nonreproductive workers; reproductive offspring (queens and males) are produced toward the end of the colony cycle (Goulson, 2003). Given that: (i) age‐specific survival probabilities are similar over much of the bumble bee worker lifespan (Goldblatt & Fell, 1987; Müller & Schmid‐Hempel, 1992), (ii) even old workers contribute to colony reproduction (Cameron, 1989) and (iii) there is a need to transmit the core gut symbionts—and not pathogens or parasites—to new queens (Hammer, Le, Martin, & Moran, 2021), one may expect only minimal senescence of worker microbiomes, gut physiology and immunity. Indeed, some aspects of systemic immunity in bumble bees remain stable or increase with age (Moret & Schmid‐Hempel, 2009).

Our main research questions are: how is the bumble bee gut microbiome assembled and maintained through the lifespan—are patterns predictable, and are they convergent with other host systems? Do species within the gut microbiome vary in their temporal dynamics, and can this give us clues into ecological differences? How does the host's gut transcriptional landscape change in concert with the microbiome? And, is the microbiome disturbance that is widely observed in wild bumble bee populations due to individual senescence? To address these questions, we conducted a cross‐sectional microbiome and transcriptomic survey of Bombus impatiens, focusing on dynamics during the adult stage of workers. We used high‐temporal‐resolution sampling and a variety of molecular methods (16S rRNA amplicon sequencing, metagenomics, quantitative PCR [qPCR], and RNAseq) to provide a detailed characterization of microbiome succession and gut processes over the lifespan. Our findings develop bumble bees as a case study with which to compare dynamics with other social insects and hosts generally, and have implications for microbiome disturbance and bumble bee health.

2. MATERIALS AND METHODS

2.1. Bumble bee rearing

For the main study, three commercially reared bumble bee (Bombus impatiens) colonies were obtained from Koppert Biological Systems and reared in the laboratory. Upon arrival, all of the cocoons (containing worker pupae) present in each colony were moved to separate containers in a 35°C incubator. We monitored the cocoons daily, and marked all newly emerged adult worker bees with numbered tags, affixed with wood glue to the thorax. Tagged bees were then returned to their colony of origin. Three newly emerged bees per colony were sampled (see below) instead of returned to the colony. To maintain colonies, we provided nonsterile pollen dough (ground pollen mixed with syrup) every 3–4 days. Nonsterile sucrose syrup (50% w/v) was provided ad libitum through an enclosed foraging area connected to the main nest.

We used cross‐sectional sampling to measure changes in gut microbiomes and transcriptomes over the worker lifespan (Figure S1). For the first week of adult life, we sampled one bee per colony per day, in order to have higher temporal resolution for the colonization phase, which we expected to be dynamic. Thereafter, for colonies B and Y—which had more tagged bees available, because more pupae were present—we sampled one bee every other day in age (e.g., 9, 11, 13 days old). For colony W, sampling occurred every fourth day in age. Sampling entailed anaesthetizing bees on ice and removing the gut with 70% ethanol‐sterilized forceps. The midgut and hindgut were separated at the pylorus and each stored in 0.1 ml DNA/RNA Shield (Zymo) at −80°C until nucleic acid extractions.

Sampling continued until all of the originally tagged bees had either died or been collected—up to 59 days old (colonies Y and W) or 75 days old (colony B) (Figure S1). These maximum ages are similar to, or greater than, the average lifespan for indoor‐reared workers of B. impatiens (Hagbery & Nieh, 2012; Kelemen et al., 2019) and other Bombus species (Blacher et al., 2017; Smeets & Duchateau, 2003). They greatly exceed the average lifespan of free‐foraging bumble bee workers (Brian, 1952; Cartar, 1992; Rodd et al., 1980).

A smaller set of samples were collected from colonies reared from field‐collected queens of B. impatiens (three colonies) and B. ternarius (one colony). Queens were collected from New Hampshire, USA (B. impatiens: 44.221788, −71.735138; B. ternarius: 44.221034, −71.774747). They were then reared in small Ziploc containers in the closet of a private residence at ~60% relative humidity and at 28°C. The colonies were fed pollen and nectar as described above. Newly eclosed (emerged) bees were tagged and returned to the colony, and combined midgut and hindgut samples were collected from younger (4–14 days old) and older (37–47 days old) workers and males. Samples were stored in 95% ethanol at −20°C. Finally, we also sampled 11 larvae from two additional commercial B. impatiens colonies. Whole larvae were stored in 95% ethanol at −20°C.

2.2. Nucleic acid extractions and qPCR

All samples were homogenized with a sterile pestle prior to extractions. For hindguts sampled from the commercial colonies, we extracted both DNA and RNA using the Zymo Quick‐DNA/RNA kit, following the manufacturer's protocol. For all other samples we extracted only DNA using the ZymoBIOMICS DNA kit. Six extraction blanks and three cross‐contamination controls (0.l ml of a OD₆₀₀ 10.0 suspension of Sodalis praecaptivus cells in phosphate‐buffered saline) were included alongside the gut samples.

For hindguts and midguts of the commercial bees, bacterial titres were measured by SYBR Green‐based qPCR targeting the 16S rRNA gene (with universal 27F/355R primers), as described in Powell et al. (2014). Absolute copy numbers were calculated using standard curves generated from serially diluted plasmid DNA carrying the target gene. Estimates of copy numbers per gut sample were calculated by multiplying values from qPCRs (containing 1 μl template) by the volume of genomic DNA (gDNA) eluted from each extracted sample.

2.3. Library preparation and sequencing

For 16S rRNA gene sequencing, gDNA (excepting larval samples) was first PCR‐amplified using universal primers targeting the V4 region (515F/806R) and conditions as detailed in Powell et al. (2021). Addition of dual‐indexed barcodes, magnetic bead purifications and additional library preparation steps also followed the protocols in Powell et al. (2021). Libraries (including the extraction blanks and three PCR no‐template controls) were pooled and sequenced on an Illumina iSeq with 2 × 150 chemistry. Samples were split among three separate sequencing runs, as listed in the Appendix S1. For larvae, library preparation and 16S rRNA gene (V4 region) sequencing (Illumina NovaSeq 2 × 250) were conducted separately by Novogene.

A total of 57 hindguts from the commercial colonies, spanning the range of ages in our sample set (including newly emerged bees), were selected for RNAseq. Library preparation for host mRNA sequencing was conducted by Novogene using the NEB Next Ultra II RNA library preparation kit. Libraries were sequenced on an Illumina NovaSeq with 2 × 150 chemistry, resulting in an average of 23.7 million raw paired‐end reads per sample. The same set of hindguts used for RNAseq were initially selected for shotgun metagenomics, except the newly emerged bees, which had very low amounts of bacterial DNA. Five gDNA samples did not pass QC, and three of these were replaced by other hindgut samples from bees “adjacent” in age, for a total of 46 samples. Library preparation was conducted by Novogene, using the NEB Next Ultra II DNA Library preparation kit. Libraries were sequenced on an Illumina NextSeq with 2 × 150 chemistry, with an average of 22.4 million raw paired‐end reads per sample.

2.4. Bioinformatic analyses

16S rRNA gene amplicons from the three iSeq runs were combined for data processing. Adapters and primers were removed using cutadapt (Martin, 2011). Sequences were then quality‐filtered, trimmed and denoised to generate ASVs by dada2 (Callahan et al., 2016). Only the forward reads were used, as the reverse reads were of poor quality. Taxonomy was assigned to ASVs using the SILVA version 138.1 database (Quast et al., 2013). Data processing and analysis was conducted in R version 4.1.1. Additional detail is provided in the Appendix S1.

The workflow for processing the RNAseq data, from raw reads to gene‐level counts, is described in the Appendix S1. Analysis of count data in R followed the general approach of Law et al. (2018), using the limma (Ritchie et al., 2015) and edgeR (Robinson et al., 2010) packages. Genes were filtered using the filterByExpr function, with normalization factors calculated by the TMM method. To use pairwise differential expression analyses, we grouped bees into four age classes: new: 0–1 days, N = 12; young: 3–19 days, N = 16; middle: 23–43 days, N = 15; old: 47–75 days, N = 14. Age classes were delineated such that they would have roughly similar sample sizes, and were chosen before statistical analysis of the data. We calculated the number of differentially expressed genes (DEGs) between age classes using linear models of log counts‐per‐million (log‐CPM) values in limma. The design matrix (~0 + age + colony) and contrasts were designed for pairwise comparisons of sequential age classes (e.g., young vs. middle‐aged). DEGs were defined as genes with a p value <.05 after false discovery rate (FDR) adjustment for multiple comparisons. To analyse expression patterns of genes that might be linked to microbiome dynamics, we focused first on antimicrobial peptides (AMPs) and dual oxidase, which generates reactive oxygen species (ROS). AMPs and ROS are major effectors in the insect gut epithelial immune response, and are known to regulate gut microbes in bumble bees and other insects (Deshwal & Mallon, 2014; Engel & Moran, 2013; Lemaitre & Hoffmann, 2007). To investigate how host immune regulation may change with age, we also analysed key genes in the Imd and Toll pathways, which control expression of these effectors (Buchon et al., 2014; Lemaitre & Hoffmann, 2007). The specific genes we included are listed in Table S1.

The workflow for processing the shotgun metagenomic data is described in the Appendix S1. phyloflash (Gruber‐Vodicka et al., 2020) was used to analyse the nonbacterial taxonomic composition of small subunit (SSU) rRNA genes. To assemble the data, we used megahit (Li, Liu, et al., 2015; default parameters) for single‐sample assemblies (Meyer et al., 2022). After binning (see Appendix S1), we used drep (Olm et al., 2017) to obtain a dereplicated set of 15 high‐quality, approximately subspecies‐level (Olm et al., 2021) bacterial metagenome‐assembled genomes (MAGs) with an average nucleotide identity (ANI) threshold of 98% (Table S2). MAGs were classified using gtdb‐tk (Chaumeil et al., 2020). For further analyses, we mapped each sample's reads against the concatenated set of MAGs using bowtie2. The relative abundance of each MAG in each sample, normalized by sequencing depth, was measured as the number of reads per kilobase per million mapped reads (RPKM). instrain (Olm et al., 2021) was used to resolve strain‐level diversity. Specifically, we characterized strain‐level clusters (generated by the instrain compare function) belonging to each of the MAGs, and visualized their distribution across bee gut samples using cytoscape (Shannon, 2003). The default Prefuse Force Directed Layout was used to visualize the bee‐strain network. We also conducted a nonclustering‐based and MAG‐specific analysis of strain sharing with 99.99% population ANI as a cutoff for differentiating strains. Finally, we used irep (Brown et al., 2016) to estimate the instantaneous population‐average replication rates for MAGs of Schmidhempelia and Gilliamella, the two taxa that varied in abundance with age. These data can provide insight into the relative contributions of cell replication and mortality to bacterial population dynamics (Brown et al., 2016). Additional detail for these analyses is included in the Appendix S1.

2.5. Statistical analyses

To model changes in bacterial titre with age, we fitted logistic curves to the data using the SSlogis and nls functions in R. To test whether changes in community composition were associated with age and colony, we used distance‐based redundancy analysis (db‐RDA) with the Bray–Curtis dissimilarity metric, as implemented in the vegan package (Oksanen et al., 2019). To identify bacteria whose relative abundance changed with age after the colonization phase, we focused on eight dominant genera, leaving aside very low‐abundance taxa (<1% mean relative abundance across gut samples) that are less likely to influence host function or overall microbiome dynamics. Then we conducted Spearman's correlations and adjusted p values using FDR. For the three taxa with an FDR‐corrected p value of <.05, we used linear mixed effects models to further test whether age predicted changes in relative abundance, including colony as a random effect. The latter approach was also used to test whether the relative abundance of Schmidhempelia and Gilliamella MAGs varied with age. Strain partitioning by colony vs. age (the four discrete age classes described above) was analysed by the following method: for all MAGs, chi‐squared tests were conducted to test whether bees belonging to the same colony or age class tended to have a higher number of shared strains; p values from these tests were corrected for multiple comparisons by FDR. To model replication indices of Schmidhempelia and Gilliamella as a function of age and colony, we first conducted linear regressions including the interaction term; these did not provide a significantly better fit to the data than models lacking an interaction, so the results we report are from the latter. We used the glht function in the multcomp package for post hoc tests of Schmidhempelia replication indices among the three colonies.

3. RESULTS

We focused our study on changes with ageing during the adult stage, when the characteristic gut microbiome is known to be present (Hammer, Le, Martin, & Moran, 2021). As microbiome colonization in adult bees could be influenced by larval symbionts that persist through metamorphosis (Hammer & Moran, 2019), we also characterized microbiomes in larvae. Larval microbiomes are dominated by Lactobacillus and Apilactobacillus (Figure S2), which are also present in the gut of adult worker bees. Despite this overlap, microbiomes are largely restructured across metamorphosis, with other adult‐associated bacteria very rare in larvae (mean relative abundances in 16S rRNA amplicon libraries: Schmidhempelia, 9.92 × 10⁻⁴; Gilliamella, 4.12 × 10⁻⁵, Snodgrassella, 3.97 × 10⁻³). Newly emerged adults (<24 h post‐emergence) have very few bacteria in either the midgut or hindgut (Figure 1a). 16S amplicon profiles (Figure S3) show large proportions of reagent contaminants, such as Burkholderia, the most abundant taxon in our extraction blanks (see Appendix S1), further indicating a scarcity of bacteria in these bees' guts (Salter et al., 2014). These <24‐h‐old bees are not included in further 16S‐based analyses. As bees mature, the gut bacterial community exhibits logistic growth, stabilizing after ~4 days, with much higher abundances in the hindgut than in the midgut (Figure 1a). Therefore, we focus on the adult worker hindgut in the following analyses, which involve commercially reared colonies unless otherwise noted. Alpha diversity also increases quickly in young bees, from a monodominance of Schmidhempelia to a stable community of about eight bacterial groups (Figures 1b and 2a). There was no evidence of a change in absolute abundance or alpha diversity in old bees (Figure 1a,b). These patterns are highly consistent among the three replicate colonies (Figures 1 and 2). Community composition also changes with age (db‐RDA, F = 17.9, p < .001; Figure 1c), and only weakly differs between the three replicate colonies (db‐RDA, F = 2.15, p = .036).

FIGURE 1.

Changes in gut microbiome abundance, diversity and composition over the worker lifespan. A total of 103 bees were sampled, consisting of 44, 23 and 36 from colonies B, W and Y, respectively. (a) qPCR‐based measurements of bacterial titre as a function of age, showing patterns for each replicate colony and gut region. Solid lines are logistic curves fitted to the data. (b) Alpha diversity of bacterial communities in hindguts of ≥1‐day‐old bees only, characterized by 16S rRNA gene sequencing. (c) Beta diversity of the same hindgut samples visualized as an ordination (nonmetric multidimensional scaling) of Bray–Curtis dissimilarities.

FIGURE 2.

Dynamics of dominant hindgut microbiome taxa (≥1% mean relative abundance across samples) over the lifespan. (a) 16S‐based relative abundances of the top genera. A total of 94 bees are shown, consisting of 41, 20 and 33 from colonies B, W and Y, respectively. One taxon belonging to the Bifidobacteriaceae was not classified to the genus level using the SILVA database. Also note that the sampling interval varied among the three colonies (see Section 2 and Figure S1). (b) 16S‐based relative abundances (in the same hindgut samples) for Schmidhempelia and Gilliamella, the only two taxa that varied significantly with age. Lines are linear models fitted to the data, with 95% confidence intervals in grey. (c) Relative abundances of Schmidhempelia and Gilliamella in whole guts of seven workers from three Bombus impatiens colonies reared from wild queens.

Despite exposure to microbes present in the diet and rearing environment, gut microbiomes of workers from both commercial Bombus impatiens and wild‐queen‐derived B. impatiens and B. ternarius colonies are almost entirely dominated by the core, host‐specialized bacterial taxa known to be prevalent in bumble bees (Hammer, Le, Martin, & Moran, 2021; Figure 2a; Figure S3). Bacteria previously observed in microbiome‐disrupted bumble bees, such as Enterobacteriaceae (Li, Powell, et al., 2015; Meeus et al., 2015; Parmentier et al., 2016) and Fructobacillus (Koch et al., 2012; Krams et al., 2022), are virtually absent from the 16S rRNA gene amplicon data sets, including commercial bee midguts (Figure S4) and hindguts (Figure 2a) and wild‐queen‐derived colonies (Figure S5). The single exception is a male bee from one of the latter colonies, which has a large proportion of Klebsiella (Enterobacteriaceae) and fungal sequences (Figure S5).

After the colonization phase, hindgut microbiome composition is generally stable throughout the adult stage (Figure 2a), with only three taxa changing in relative abundance: Schmidhempelia steadily decreases (t = −4.47, p < .001) (Figure 2b) while Gilliamella (t = 7.02, p < .001; Figure 2b) and Bombiscardovia (t = 3.05, p = .003) increase. These are relative abundances, derived from compositional 16S rRNA amplicon profiles. Changes in relative abundances of taxa can be misleading when the absolute abundance of the entire community changes (Rao et al., 2021). Indeed, using taxon‐specific population sizes estimated by correcting relative abundances with qPCR data, Bombiscardovia does not significantly increase with age after the colonization phase (t = 1.31, p = .19; Figure S6). Otherwise, similar patterns are found: Schmidhempelia decreases, Gilliamella increases and other dominant bacterial taxa generally remain stable (Figure S6). Schmidhempelia and Gilliamella show the same pattern in B. impatiens colonies reared from wild queens (Figure 2c).

Metagenomic data provide further support for a Schmidhempelia/Gilliamella transition with age. Using read mapping to MAGs as another compositional measure of bacterial abundance, we find the same switch (Schmidhempelia: t = −3.27, p = .002; Gilliamella: t = 4.62, p < .001; Figure 3). Metagenomes also show that gut microbiomes are dominated by core bacteria. All of the MAGs belong to bee‐specific bacterial taxa (Table S2); SSU rRNA genes from fungi (homologous to bacterial 16S rRNA genes) are generally rare relative to those from bacteria, though with elevated proportions in a few of the youngest and oldest bees in our sample set (Figure S7). SSU rRNA genes from other nonbacterial microbes are practically nonexistent. We also detect diet‐derived plant sequences. Except for the youngest bees, proportions of plant sequences are generally low and do not show any clear trends with age (Figure S7).

FIGURE 3.

Coverage‐based abundance estimates of all metagenome‐assembled genomes (MAGs) in 46 worker hindgut samples from the three commercial colonies. Abundance for a given sample is normalized to sequencing depth and MAG size, by measuring reads per kilobase per million mapped reads (RPKM). Lines are linear models fitted to the data, with 95% confidence intervals. Some genera contain multiple MAGs with <98% average nucleotide identity; in these cases, congeneric MAGs are shown in different colours. MAGs are listed and described in Table S2.

Analysis of ASVs, the finest level of resolution available with our 16S sequencing approach, shows that the major core taxa comprise only a single ASV generally ubiquitous across samples (Figure S8). We used metagenomic data to reveal further layers of diversity beyond ASVs. Some (but not all) of the major bacterial groups comprise multiple MAGs with <98% ANI—“subspecies,” following Olm et al. (2021) (Figure 3; Table S2). Using instrain, which compares single‐nucleotide variants between samples' reads aligned to a common reference (Olm et al., 2021), we find that MAGs contain additional strain‐level diversity. For most MAGs, this diversity is clearly partitioned by colony, but not by age (Figure 4; Figure S9). All of the MAGs from Gram‐negative bacterial taxa (Snodgrassella, Schmidhempelia, Gilliamella), but only some of those from Gram‐positive taxa, are more likely to be shared within than between colonies (FDR‐adjusted p < .05; Figure S9). We also used metagenomic data to examine in situ population‐average replication rates, focusing on the two taxa that shift with age. Schmidhempelia has much lower replication indices in colony B (post hoc pairwise contrasts: W vs. B, t = 10.69, p < .001; Y vs. B, t = 14.29, p < .001; W vs. Y, t = 1.35, p = .38; Figure 5). There is also a weak negative effect of age on Schmidhempelia replication (t = −2.65, p = .013). Gilliamella replication indices do not differ significantly between colonies (t = −1.32, p = .22) or due to age (t = 0.197, p = .89; Figure 5), although sample sizes are also smaller due to lower coverage.

FIGURE 4.

Networks of bacterial strain composition in the 46 worker hindgut metagenomes, showing bee gut sample (large circle) grouping by colony vs. age class. Strain clusters (small diamonds) from all MAGs are shown; strain sharing within and between colonies is shown for each MAG individually in Figure S9. Clusters are derived from hierarchical clustering of pairwise comparisons of population ANI, a metric calculated by instrain (see Section 2).

FIGURE 5.

Instantaneous population‐average replication rates, estimated for Schmidhempelia and Gilliamella, the two taxa that vary in abundance with age. A replication index value of 1.5 corresponds to half of the cells making one copy of their genome; a value of 2 corresponds to all cells making one copy (see Brown et al., 2016). However, note that these are population averages, and bacteria can make multiple copies of their genome simultaneously. Some data points are missing due to low coverage of the MAG in a given sample.

Host gene expression profiles in the hindgut change as bees mature and reach “middle age” (~3–6 weeks old; Figure 6a). Between newly emerged and young bees, and young and middle‐aged bees, there are 2696 and 6136 DEGs, respectively. Thereafter, gut gene expression profiles do not change with age in a consistent way (Figure 6a): there are zero DEGs comparing middle‐aged and old bees. Of the immunity effectors we analysed, most show low levels of gene expression in newly emerged bees, with upregulation in older age cohorts (Figure 6b). Dual oxidase—which generates ROS (Engel & Moran, 2013; Lemaitre & Hoffmann, 2007)—and three of the four antimicrobial peptides, increase in expression as bees mature. Catalase, which degrades ROS to maintain redox balance (Ha et al., 2005), is highly expressed in newly emerged bees—possibly to prevent self‐harm in the absence of abundant microbial cells (Kwong, Mancenido, & Moran, 2017)—and is subsequently downregulated (Figure 6b). Signalling genes in the Imd and Toll pathways show variable patterns. Imd and relish decrease in expression with age, while cactus and dorsal expression do not significantly differ between any age classes (Figure S10).

FIGURE 6.

Dynamics and stability of host hindgut gene expression over the lifespan. Ages and sample sizes of age classes: new: 0–1 days, N = 12; young: 3–19 days, N = 16; middle: 23–43 days, N = 15; old: 47–75 days, N = 14. (a) Principal coordinates analysis showing similarity in gene expression profiles between bees of different age classes. Similarity is quantified as leading log₂‐fold changes, which are defined as the quadratic mean of the largest log₂‐fold changes between a pair of samples. The number of differentially expressed genes (DEGs) is shown for each pair of sequential age classes. (b) Expression levels of key immunity genes (Table S1) normalized to library size (log₂ counts per million) over bee age. Dashed lines show significant differences in expression between sequential age classes (FDR‐adjusted p < .05).

4. DISCUSSION

Our study provides an initial picture of how the bumble bee worker microbiome changes throughout adult life, and how these changes correlate with the expression of key host genes. Overall, the gut microbiome and host transcriptome are highly dynamic during the initial assembly phase. These changes continue over the longer‐term maintenance phase, but their magnitude and direction vary among symbiont species and host genes. Both the microbiome and transcriptome appear to stabilize, as opposed to senesce, in old age. We discuss each of these phases in turn.

4.1. Assembly

In adult bumble bees, microbiome assembly appears to be an example of primary succession. Amounts of bacterial DNA in newly emerged adult guts are very low (Figure 1a), and previous work finding these guts generally devoid of culturable bacteria (Hammer, Le, & Moran, 2021; Sauers & Sadd, 2019) suggests that at least some of this DNA derives from nonviable cells. Larvae harbour Lactobacillus and Apilactobacillus (Figure S2), taxa also present in adult guts (Figure 2a). Although transmission through metamorphosis is theoretically possible (Hammer & Moran, 2019), these bacteria may instead be cleared during pupation and reacquired from the nest environment by newly emerged adults. Other dominant bacteria in adult guts such as Schmidhempelia, Snodgrassella and Gilliamella were either very rare or absent from larvae, indicating de novo colonization of adults. Developmental restructuring of the microbiome has been found in other Bombus species (Parmentier et al., 2018) and in Apis mellifera (Martinson et al., 2012), but why it occurs is not fully clear. Larvae and adults interact through trophallaxis, and both consume pollen and honey from communal stores in the nest (Goulson, 2003; Pereboom, 2000; Smeets & Duchateau, 2003). Thus, nutritional and microbial inputs into the gut are probably similar. Potentially, aspects of larval gut morphology, physiology or immunity create less hospitable conditions for colonization by the core symbionts of adults. The trypanosomatid Crithidia bombi, a common gut parasite in adult bumble bees, is also unable to infect larvae (Folly et al., 2017). High osmotic potential in the larval gut has been suggested to inhibit Crithidia infection (Folly et al., 2017), and may be a factor inhibiting bacterial colonization as well.

Coupled with a low expression of immunity effectors (Figure 6b), the low abundance of pathogen‐protective (Koch & Schmid‐Hempel, 2011; Raymann et al., 2017) core gut bacteria suggests that newly emerged adults are particularly vulnerable to microbiome disruption. Similarly, human infants are prone to infections while their immune system, microbiome and gut microenvironment mature (Sanidad & Zeng, 2020). The microbiome disruption phenomenon widely observed in field‐collected bumble bee workers as well as queens (Hammer, Le, Martin, & Moran, 2021) may begin during the assembly phase. In Bombus griseocollis, workers do not leave the nest for the first couple of days after emergence; most activities, including foraging, begin by the fourth or fifth day (Cameron, 1989). By this point, the core gut microbiome is established and expression of immune effectors has increased (Figures 1, 2 and 6b). The timing of the onset of foraging therefore limits direct exposure to stressors during this vulnerable period. However, environmental microbes and chemicals are present in food stores and other substrates, presenting an opportunity for microbiome perturbations even in bees restricted to the nest.

Microbiome assembly dynamics in bumble bees are both predictable and convergent with other hosts. Temporal patterns of microbiome abundance, diversity and composition (Figure 1) are highly similar among replicate colonies. Moreover, these patterns are evident despite our cross‐sectional study design, suggesting that temporal variation in microbiomes outweighs interindividual variation. Early successional patterns showed similarities to those observed in honey bees (Powell et al., 2014) and human infants (Bäckhed et al., 2015; Bokulich et al., 2016; Koenig et al., 2011; Rao et al., 2021) and, more generally, to heterotrophic microbial communities supplied with external carbon sources (Fierer et al., 2010). However, there are also marked differences with gut microbiome assembly in other invertebrates, such as flies (Drosophila melanogaster) and nematodes (Caenorhabditis elegans). In these hosts, bacterial colonization is highly stochastic and can lead to microbiome compositions that are stably distinct among individuals (Jones et al., 2022; Obadia et al., 2017; Vega & Gore, 2017). These hosts also generally harbour nonhost‐restricted, flexible, environmentally acquired gut microbiomes (Moran et al., 2019). In contrast, the symbiosis between social bees and their gut microbes is ancient and specific (Kwong, Medina, et al., 2017; Kwong & Moran, 2016). By living in dense colonies, social bees enrich their local environment with core symbionts, favouring predictable assembly. Functional redundancy among bacterial species may also be lower in social bees than in flies and nematodes, possibly selecting for stronger host control over microbiome establishment.

Despite predictable assembly at the community level (Figures 1 and 2), we also observe evidence for stochastic colonization at the strain level. Strain‐level diversity is clearly partitioned between the three replicate colonies (Figure 4), a pattern not evident in the ASV (Figure S8) or subspecies (Figure 3) data. Notably, all Gram‐negative bacterial genomes exhibited significant colony partitioning, while only some of the Gram‐positive genomes did so (Figure S9). Gram‐positive bacteria may be more likely to survive outside the host, facilitating dispersal among colonies. Similarly, Gram‐positive gut bacteria of honey bees can be transmitted via hive surfaces, with less reliance on social contact than Gram‐negative species (Powell et al., 2014). Differences in social structuring among mammalian microbiome members have also been linked to bacterial physiology (Moeller et al., 2018; Tung et al., 2015).

There are multiple potential explanations for the origin of the colony‐partitioning pattern. One is an interaction between host and symbiont genotypes (Sauers & Sadd, 2019). There may also be genotype‐by‐environment effects; to give one example, bee colonies of different sizes may have different thermoregulatory capacities and temperatures (Seeley & Heinrich, 1981); this could act as an ecological filter for strains with different thermal tolerances (Hammer, Le, & Moran, 2021). In addition to intrinsic physiological differences between strains, differences in temperature or other environmental factors may explain why the inferred replication rates of Schmidhempelia differed substantially between colonies (Figure 5). A final explanation is founder (or foundress) effects. Bumble bee colonies are initiated by a single foundress queen, who is the source of gut symbionts for her offspring (Hammer, Le, Martin, & Moran, 2021; Su et al., 2021). A diverse pool of strains may be stochastically sorted into a single foundress queen's gut, with the established population resistant to subsequent invasion (i.e., priority effects). This process may be analogous to the neutral bottlenecking described for bacterial strain partitioning among skin pores (Conwill et al., 2022) or the stochastic colonization of individual guts of flies and nematodes (Jones et al., 2022; Obadia et al., 2017; Vega & Gore, 2017).

4.2. Maintenance

Gut microbiome abundance and composition generally stabilize after the colonization phase in newly emerged adults. However, the ratio of two of the core symbiont species, both members of the family Orbaceae, continues to shift with age. The worker gut starts as a near monoculture of Schmidhempelia (Figure 2a). This finding matches previous work: in the B. impatiens genome project, which used DNA from a 1‐day‐old male, the only bacterial genome with substantial representation belonged to Schmidhempelia (Martinson et al., 2014). Over time, Schmidhempelia progressively declines in relative abundance, while Gilliamella increases. This shift is evident in both the amplicon (Figure 2b) and metagenome data sets (Figure 3) and across the three replicate colonies. We observe the same pattern in wild‐derived colonies (Figure 2c), suggesting that it is a common feature of microbiome succession in B. impatiens.

The functional consequences of the switch from Schmidhempelia to Gilliamella are unknown. In honey bees, Gilliamella can ferment pollen cell wall components (Kwong et al., 2014), with products (short‐chain fatty acids) potentially providing bees with a supplementary energy source and lowering gut pH (Zheng et al., 2017). Gilliamella probably perform similar functions in bumble bees, although bumble bee‐derived strains have fewer capabilities for degrading and fermenting pollen components (Hammer, Le, Martin, & Moran, 2021; Zheng et al., 2019). Acidification is thought to limit infection by Crithidia bombi, a trypanosomatid parasite of bumble bees (Palmer‐Young et al., 2018). An increase in Gilliamella may thus contribute to the metabolism and defence of older bees.

Presumably, changes in Schmidhempelia abundance over time also affect hosts, as well as other gut microbial species. Differences between Schmidhempelia and Gilliamella metabolism are evident from genome analyses (Kwong et al., 2014; Martinson et al., 2014). Whereas Gilliamella is a facultative anaerobe with an intact TCA (tricarboxylic acid) cycle, Schmidhempelia is inferred to be an obligately anaerobic fermenter, producing acetate and other short‐chain fatty acids (Martinson et al., 2014). These products would acidify the gut, potentially inhibiting parasites and facilitating subsequent colonization by core symbionts. However, as Schmidhempelia has not been cultured (Hammer, Le, Martin, & Moran, 2021), we lack experimental evidence for its effects on hosts or other microbes.

The temporal dynamics of Schmidhempelia and Gilliamella point to distinct life history strategies, perhaps exemplifying the competition–colonization trade‐off shown in various microbial communities (Livingston et al., 2012; Smith et al., 2018; Yawata et al., 2014). For example, Schmidhempelia may be a pioneer colonizer or ruderal (Grime, 1974), one that is good at dispersing to and exploiting unoccupied gut habitat. Gilliamella may be a better competitor, successfully excluding Schmidhempelia with time. The nature of this competition remains to be determined. Schmidhempelia replication rates appear to be generally stable with age after the colonization phase (Figure 5), suggesting that declines in population size are driven by increased mortality over time, rather than by a dwindling resource supply slowing replication. Increased mortality could be due to interference competition, where Gilliamella directly antagonizes Schmidhempelia, possibly by using type VI secretion systems—possessed by both species (Martinson et al., 2014; Steele et al., 2017)—or other means. It could also be due to apparent competition, where Gilliamella growth induces increased expression of host immune responses (Figure 6b) that are more harmful to Schmidhempelia than to Gilliamella (Kwong, Mancenido, & Moran, 2017). Although the mechanisms are unknown, our data support the existence of variation in life history strategies within the gut microbiome. Such differences are likely to be important drivers of coexistence and community function.

As with the microbiome, gene expression profiles in the hindgut are dynamic up to ~3–6 weeks of age (“middle age”), with many differentially expressed genes between newly emerged, young and middle age (Figure 6a). Multiple genes involved in production of AMPs and ROS, key components of gut epithelial immunity (Engel & Moran, 2013; Lemaitre & Hoffmann, 2007), increase in expression over this time frame (Figure 6b). In contrast, components of the Imd and Toll signalling pathways either decrease in expression or remain stable with age (Figure S10). Pathogen infection induces these pathways, which then activate immune effectors (Buchon et al., 2014; Lemaitre & Hoffmann, 2007). In this experiment, noncore microbes are almost entirely absent from the hindgut (Figure 2a), so induction by pathogens is expected to be minimal. Potentially, the temporal patterns we observe could be due to a shift from low (but more inducible) effector expression to high (and more constitutive) expression with age. These patterns contrast with systemic (haemolymph) immune defences, which decrease with age in bumble bees (Doums et al., 2002; Moret & Schmid‐Hempel, 2009). Differing selective pressures on defence could underlie this discrepancy; for example, gut infection may be more likely to occur or more likely to spread to nestmates (via feces) than haemolymph infection. Currently, however, comparisons between data sets are complicated by the fact that colony age may influence immunity independently of individual age (Moret & Schmid‐Hempel, 2009).

Changes in gut immunity (Figure 6b; Figure S10) appear to be an intrinsic property of ageing in B. impatiens workers, as they occur despite continuous food availability, static environmental conditions in the laboratory, and an apparent lack of pathogen infection. As hypothesized for systemic immunity (Moret & Schmid‐Hempel, 2009), they may represent a plastic adjustment of host defence. For example, increases in constitutive expression of AMPs and ROS may have evolved in response to heightened infection risk with age. An alternative hypothesis is that increases in immune effectors represent unregulated inflammation, a common feature of animal immunosenescence (Peters et al., 2019). In D. melanogaster, increased AMP expression with age is linked to increased gut bacterial load and to deteriorating gut integrity (Erkosar & Leulier, 2014; Rera et al., 2012). However, total gut bacterial load in bumble bees is stable (Figure 1a), and the only taxon that increases in abundance is Gilliamella (Figures 2b and 3; Figure S6), one of the core bumble bee‐specialized symbionts (Hammer, Le, Martin, & Moran, 2021). While Gilliamella may induce bee AMP expression (Kwong, Mancenido, & Moran, 2017), such a response with age could be interpreted as a sign of strengthening, as opposed to deteriorating, immunity.

Unusually, these changes in immunity (and other endogenous processes) decelerate with age. No genes are differentially expressed in the hindgut between middle‐aged and old bumble bees (Figure 6a). In contrast, transcriptomic changes in old age have long been observed in Drosophila, C. elegans, mice and humans (Berchtold et al., 2008; Doroszuk et al., 2012; Lee et al., 1999; Song et al., 2020). In a fish model, the gut transcriptome is also markedly different toward the end of the lifespan, and is associated with upregulated immunity and an enrichment of potentially pathogenic bacteria (Smith et al., 2017).

Gut immunity and microbiomes are likely to covary, and we find that microbiome dynamics also slow as bees enter old age. This stability contrasts with the major microbiome changes observed between life stages (Figure S2) and earlier in the adult stage (Figures 1 and 2). Total microbial abundance is stable in old bees (Figure 1a), and there is no evidence of microbiome disruption—with the exception of a single male from a wild‐derived colony (Figure S5)—or loss of any symbionts besides Schmidhempelia (Figure 2). All bees were reared indoors, and indoor‐reared bumble bees have been shown to have lower gut microbiome diversity (Hammer, Le, Martin, & Moran, 2021; Meeus et al., 2015; Parmentier et al., 2016). However, the bees studied here were exposed to noncore microbes in their food and rearing environment, and previous work has documented occasionally large numbers of Enterobacteriaceae and other noncore bacteria in indoor‐reared B. impatiens when exposed to stressors (Palmer‐Young et al., 2019; Rothman et al., 2019, 2020). The microbiome stability we observe in old bees indicates a lack of intrinsic senescence processes that would disrupt core symbionts and allow invasion, rather than simply a lack of exposure to noncore microbes. Bumble bees therefore contrast with humans (Biagi et al., 2010; Claesson et al., 2011, 2012; Wilmanski et al., 2021), as well as other animals such as flies, mice and fish (Erkosar & Leulier, 2014; Smith et al., 2017; Stephens et al., 2016; Thevaranjan et al., 2017), which exhibit microbiome senescence (or at least community‐wide shifts during ageing) even when reared in the laboratory. Our data also weigh against the hypothesis that individual senescence underlies the microbiome disturbance observed in wild bumble bee populations. As mentioned above, it is the youngest bees that appear to be the most vulnerable. These results support previous work finding microbiome disruption to be concentrated in young bumble bees (Parmentier et al., 2016).

There are many potential proximate causes of microbiome stability in old age. Communal living may buffer microbiome disturbances by providing a continuous source of microbes that can be transmitted between individuals or via a shared social environment, such as a nest (Diez‐Méndez et al., 2022; Moeller et al., 2018; Tung et al., 2015; Yarlagadda et al., 2021). In our experiment, diet was kept constant, and bees appear to consume pollen even in old age based on the presence of plant DNA in metagenomes (Figure S7) and observations of gut colour. Our transcriptomic data also suggest that the gut microenvironment stabilizes after bees reach middle age (Figure 6a). In addition to inoculation from nestmates, a steady resource supply, structural integrity, and maintenance of immune responses in the gut (Figure 6b) probably help maintain stable core microbiomes. One caveat is that these bees were not able to fly, a factor that should be addressed in future work. Bee flight is metabolically costly, reduces lifespan and affects systemic immune responses (Doums & Schmid‐Hempel, 2000; Ellington et al., 1990; Schmid‐Hempel & Wolf, 1988). Free‐foraging honey bee workers do exhibit changes in gut microbial abundance, composition and replication rates with age (Ellegaard & Engel, 2019; Kapheim et al., 2015; Kešnerová et al., 2020; Maes et al., 2021; Powell et al., 2014). On the other hand, as noted earlier, temporal changes in the honey bee gut microbiome may be primarily driven by a shift from performing in‐nest tasks to foraging (Jones et al., 2018). In overwintering honey bee workers, which do not forage much, if at all, the gut microbiome is largely stable into old age (Maes et al., 2021).

In the bumble bee gut, senescence of the microbiome and of endogenous processes (such as immunity) appears to be either absent, or compressed into such a short window that we did not observe it. We hypothesize that this is explained by the unique selection pressures that accompany eusociality. Evolutionary theories of ageing suggest that in a nonsocial host organism, (i) selection against late‐acting, deleterious variants—either host alleles or microbes—should be weak, and (ii) such variants may trade off with early‐life, prereproductive benefits (Hamilton, 1966; Kirkwood & Rose, 1991; Williams, 1957). The situation is different in bumble bee workers, which often complete their entire life cycle before colony reproduction occurs at the end of the season (Goulson, 2003). According to theory, the strength of selection should be maximal up until the onset of reproduction (Hamilton, 1966). In eusocial insects, what counts is the colony's production of sexual offspring (Kramer et al., 2016), as workers are usually sterile. Hence, for most of the colony lifespan, maintenance of microbiomes and immunity in workers should be under strong selection even in old age, given their expected effects on inclusive fitness (i.e., overall colony reproductive success). Core gut symbionts may contribute indirectly (e.g., via nutrition) to worker performance—brood care, foraging, defence, etc.—which in turn will affect production of new queens and males at the end of the colony cycle. Workers may also benefit their reproductive siblings (the new queens and males) by acting as a vector for core symbionts, and not for pathogens or parasites. Microbiomes of at least some other highly social animals do not appear to become destabilized in old age (Maes et al., 2021; Reese et al., 2021; Risely et al., 2021), raising the question of whether group living contributes to differences in microbiome senescence. In general, organisms display diverse patterns of mortality and reproduction with age (Jones et al., 2014), and such diversity appears to extend to microbiome dynamics.

Variation in microbiome dynamics may also be expected within species, especially in eusocial insects, which contain castes subject to unique selection pressures (Hölldobler & Wilson, 2009). Our study focused exclusively on the nonreproductive worker caste, but future work should examine how microbiomes change with age in reproductives. In honey bees, these dynamics differ between queens and workers (e.g., Anderson et al., 2018; Copeland et al., 2022; Tarpy et al., 2015). Queen–worker differences may also apply to bumble bees, even though—unlike honey bees (Kwong & Moran, 2016)—bumble bee queens acquire gut bacterial communities compositionally similar to those of workers (Koch et al., 2013; Su et al., 2021). In Bombus lantschouensis, prediapause queens show large decreases in core gut symbionts with age (Wang et al., 2019), strongly contrasting with the stability we observe in B. impatiens workers. Potentially, only a small number of core symbionts are needed for successful transmission, favouring a reduction in titre before diapause (Vigneron et al., 2014). Queen–worker differences in microbiome dynamics may also be related to immunity. For example, queens have been reported to exhibit stronger resistance to gut parasite infection, and distinct immune activity in haemolymph, relative to age‐matched workers (Ruiz‐González et al., 2022).

5. CONCLUSIONS

Even in the relatively simple gut microbial communities of laboratory‐reared worker bees, we see a complex assortment of temporal patterns that differ between symbiont taxa, vary with phylogenetic scale and decelerate as hosts age. Some of these patterns are convergent with those in other hosts. At the level of symbiont species and genera, assembly is predictable, with dynamics similar to those of human infant gut microbiomes. At the strain level, assembly resembles the stochastic colonization dynamics observed in flies and nematodes. We also find unique temporal patterns that contrast with those in other hosts: in bumble bee workers, neither gut microbiomes nor gut immunity appear to senesce. This stability may be due to the important contributions of each to inclusive fitness, even in old age. Temporal dynamics differ markedly among bacterial symbiont species, suggesting distinct ecological strategies within the microbiome for colonization and persistence. Many of the patterns we observe would be undetectable by 16S rRNA gene sequencing, emphasizing the need to use quantitative and higher‐resolution methods to study microbiome dynamics. We also characterize the transcriptomic landscape of the bumble bee gut, finding that expression of genes involved in immunity (and other processes) changes in similar ways to the microbiome over host age—probably due to bidirectional feedbacks or to common selection pressures acting on both. A priority for future work is to determine the mechanisms underlying these microbial and immunological dynamics, and to assess functional consequences for bumble bee health and pollination services.

AUTHOR CONTRIBUTIONS

TJH and NAM designed the research. TJH conducted the commercial bee rearing and sampling, molecular methods, bioinformatics and statistical analyses. AEC reared and sampled bees from the wild‐queen‐derived colonies. The manuscript was drafted by TJH and subsequently revised and approved for submission by all authors.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Supporting information

Appendix S1:

Hammer, T. J. , Easton‐Calabria, A. , & Moran, N. A. (2023). Microbiome assembly and maintenance across the lifespan of bumble bee workers. Molecular Ecology, 32, 724–740. 10.1111/mec.16769

DATA AVAILABILITY STATEMENT

Raw reads from 16S rRNA gene sequencing, metagenomics and RNAseq are deposited in the NCBI SRA (BioProject PRJNA849590). Sample metadata, qPCR data, processed 16S data (ASV tables and sequences), gene‐level counts of mapped reads, other raw data files and R code are available from the Dryad repository (https://doi.org/10.5061/dryad.gb5mkkws9).