Linking pollen foraging of megachilid bees to their nest bacterial microbiota

Abstract Solitary bees build their nests by modifying the interior of natural cavities, and they provision them with food by importing collected pollen. As a result, the microbiota of the solitary bee nests may be highly dependent on introduced materials. In order to investigate how the collected pollen is associated with the nest microbiota, we used metabarcoding of the ITS2 rDNA and the 16S rDNA to simultaneously characterize the pollen composition and the bacterial communities of 100 solitary bee nest chambers belonging to seven megachilid species. We found a weak correlation between bacterial and pollen alpha diversity and significant associations between the composition of pollen and that of the nest microbiota, contributing to the understanding of the link between foraging and bacteria acquisition for solitary bees. Since solitary bees cannot establish bacterial transmission routes through eusociality, this link could be essential for obtaining bacterial symbionts for this group of valuable pollinators. Open Research Badges This article has earned an Open Data Badge for making publicly available the digitally‐shareable data necessary to reproduce the reported results. The data is available at https://www.ebi.ac.uk/ena/data/view/PRJEB27223, https://www.ebi.ac.uk/ena/data/view/PRJEB31610, and https://doi.org/10.5061/dryad.qk36k8q

However, only few studies have dealt with their role in larval health McFrederick, Vuong, & Rothman, 2018) or their acquisition routes (McFrederick et al., 2017;Voulgari-Kokota, McFrederick, Steffan-Dewenter, & Keller, 2019). Bees leading a solitary lifestyle with no direct contact between multiple generations are excluded from the benefits of a social structure, where individuals are in constant interaction with their nestmates, and their nest microbiota is more susceptible to environmentally introduced bacteria (Keller et al., 2013;McFrederick et al., 2017;Rothman, Andrikopoulos, Cox-Foster, & McFrederick, 2018;Voulgari-Kokota, Grimmer, et al., 2019;Voulgari-Kokota, McFrederick, et al., 2019). As a result, the establishment of steady host-microbe interactions requires different mechanisms and could be the result of more complex processes, where beneficial microbial interactions are reinforced and negative interactions with pathogens are minimized. These should be able to secure transmission routes over multiple generations through active and passive transmission. Three possible mechanisms are active inoculation of nest structures (as in Kaltenpoth, Yildirim, Gürbüz, Herzner, & Strohm, 2012) including stored pollen, active inoculation of eggs (Hosokawa, Kikuchi, & Fukatsu, 2007), or passive transmission through imported materials and especially food sources (McFrederick & Rehan, 2016;McFrederick et al., 2017;Rothman et al., 2018).
In the present study, we focused on the importance of the imported pollen in the nest environment, as a major source of introducing environmental and foremost floral bacteria in the solitary bee nest. Solitary bees can be polylectic or oligolectic, and their foraging patterns are determined by the period of their flight activity, their morphological traits, the plant availability, and chemical properties of the plants (Michener, 2007;Westrich, 2015). Considering that the multiple aspects of solitary bee ecology can form distinctive nesting conditions for different species, we test the hypothesis that different pollen foraging patterns could establish different routes for bacterial colonization in the nest. Previous studies have undertaken the task of characterizing solitary bee nest microbiota (Keller et al., 2013;Lozo et al., 2015;McFrederick et al., 2017;Voulgari-Kokota, Grimmer, et al., 2019) or have identified the plant composition of pollen provisions (Danner, Keller, Härtel, & Steffan-Dewenter, 2017;Sickel et al., 2015;Villanueva-Gutiérrez & Roubik, 2016).
Simultaneous examination of pollen composition and pollen microbiota from the nests of a wild bee species has shown covariation across different landscapes, even though the causality of this association was not clear (McFrederick & Rehan, 2018). Furthermore, the association between the pollen microbiota and the pollen composition in the nests of a single bee species suggested possible plantmediated host-microbe relationships (McFrederick & Rehan, 2016).
However, the complex relationships in the plant-pollinator-microbe triangle are currently not well understood and multispecies studies are currently lacking.
In order to investigate the association between pollen types and nest microbiota, we conducted a simultaneous pollen and bacterial metabarcoding survey. We used metabarcoding of the ITS2 rDNA to identify plant species which constituted the pollen provisions in 100 nest chambers coming from seven megachilid bee species, and at the same time, we used metabarcoding of the 16S rDNA to separately characterize the bacterial communities of the same pollen provisions and the respective larvae. Our aim was to identify the degree to which introduced pollen bacteria shape the solitary bee nest microbiota, by testing the hypothesis that the plant diversity and composition of pollen provisions influence the bacterial community of pollen in the bee nests, both at the interspecific and at the intraspecific level, and finally that of the respective larvae. Moreover, we investigated the probability of certain plant species serving as reservoirs of specific bacterial taxa, facilitating conserved bee-microbe associations which result from conserved foraging preferences.

| Sampling design
Sampling of the solitary bee nests was conducted with the use of 24 artificial trap nests placed at 10 localities throughout an area with an extent of 30 × 25 km in northern Bavaria, Germany. The sampling sites were shaped by agricultural land use and interspersed seminatural vegetation. The trap nests consisted of 30-50 reed internodes (length = 20 cm, width = 4-10 mm) and were placed in early Spring 2016. They were examined every 2 weeks from May until September 2016, and occupied reed canes with clogged entrances were directly transferred into the laboratory. Nests were opened horizontally, and larvae, pollen clumps, and nest material were separately removed with the use of sterile spatula, transferred into sterile tubes, and directly frozen down to −25°C. Larvae and respective pollen provisions were selected for this study under the condition that they were not visibly affected by pathogens or parasites and that the larvae had not yet completely consumed the pollen provision. In total, 100 larvae and 100 pollen provisions were included in this study. Of the 100 nest chambers included, 35 belonged to  Grimmer, et al., 2019). The prerequisite for sample inclusion in the present study was the sufficient pollen quantity (>10 mg) in the nest chamber at the time of the nest opening for both bacterial and pollen metabarcoding. All specimens were treated exactly the same way through the whole laboratory workflow, as described in the following chapter.

| Laboratory workflow
Genomic DNA isolation from each specimen was conducted with the Macherey-Nagel NucleoSpin kits for soil (Burbach, Seifert, Pieper, & Camarinha-Silva, 2016) and food. As a first step, whole larvae were washed with PBS/EDTA solution. The protocol was modified in order to include an extra step of incubation with proteinase K, to ensure lysis of thick bacterial cell walls. After the whole genomic DNA extraction, we proceeded to the PCR amplification of the 16S rDNA for all larvae and pollen sample and the PCR amplification of the ITS2 rDNA for all pollen samples.
We followed the dual-indexing strategy introduced by Kozich, Westcott, Baxter, Highlander, and Schloss (2013) in order to generate a pooled amplicon library based on the 16S rRNA V4 variable region for the Illumina platform (Illumina, 2017). Primers used to amplify the V4 region were as follows: AATGATACGGCGACCACCGAGATC TACAC TCAGCCGGACTACHVGGGTWTCTAAT (Illumina, 2016a). The same dual index strategy was used to generate the pooled amplicon library for the ITS2 rDNA region used for pollen metabarcoding. We used a combination of plant barcoding primers expanded for Illumina conformity, as described in Sickel et al. (2015). The primer sequences were as fol-

| Data analysis
After data from the 16S rDNA library were acquired, we used fastq-join v1.3.1 (Aronesty, 2013) to join paired ends of forward and reverse reads. Paired joined reads were accepted if longer than 250bp. We used USEARCH v10.02.240 (Edgar, 2013(Edgar, , 2016 for length truncating, quality filtering, and file conversion. Chimera filtering, operational taxonomic unit (OTU) clustering to a minimum identity of 97%, and OTU table construction were performed also with USEARCH (Edgar, 2013(Edgar, , 2016. We filtered low-quality reads after setting the maximum number of expected errors at E max = 1 (Edgar & Flyvbjerg, 2015). Reads with ambiguous characters or singletons were excluded from the downstream analyses.
In the case of the acquired ITS2 rDNA dataset, we kept only the forward reads for downstream analysis, as reverse reads showed less satisfying quality. We filtered reads with high expected error rate or ambiguous characters following the same parameters as described above for the bacterial dataset, and trimmed low-quality bases at the read ends (<Q30). Reads were accepted if longer than 150 bases. We assigned taxonomy for the de novo picked OTUs of the 16S rDNA library using the RDP v16 reference database up to genus level. The ITS2 rDNA reads were directly mapped against a Bavarian floral reference database for ITS2  derived from the ITS2-database (Ankenbrand, Keller, Wolf, Schultz, & Förster, 2015)  Beta diversity was visualized with nonmetric multidimensional scaling (NMDS) ordination. The bacterial community distances between samples were based on Bray-Curtis matrices. The distance matrices for pollen composition were based on presence/absence data, since ITS2 rDNA data are not ideal for quantitative community analysis (Bell et al., 2016). PERMANOVA/Adonis was used to test the pollen composition homogeneity between group levels, by subsequently setting the bee species, the sampling site, and the sampling period as independent factors. Pollen and bacterial OTU distance matrices were compared with Mantel tests based on Pearson's product correlation to explore the degree of association between them (Legengre & Legendre, 2012). Ordination and diversity graphs were constructed with ggplot2 v3.0.0 (Wickham, 2009) and reshape2 v1.4.3 (Wickham, 2007).

Descriptions of bacterial and plant communities related to each
host species were based on the bacterial OTU taxonomic assignments and identified plant species, respectively. Revealed bacterial and plant communities for each bee species were visualized as bipartite networks after low abundance filtering (>1%) with the package bipartite v2.11 (Dormann, Fruend, & Gruber, 2009). Furthermore, we investigated co-occurrence patterns between bacterial and plant taxa with the use of a probabilistic approach with the package cooccur v1.3 (Griffith, Veech, & Marsh, 2016), which uses the observed frequencies of co-occurrence between each pair of taxa and the distribution of each taxon to calculate the respective expected frequencies and return the probabilities that a more extreme value of co-occurrence could have been obtained by chance. The algorithm defines the observed frequency of co-occurrence as positive, negative, or random association (Veech, 2013). The analysis was based on matrices of plant species and bacterial taxa agglomerated up to genus level, containing binary data to indicate the absence or presence of each taxon in each sample over a relative abundance threshold of 1%.
Samples were assigned to seven groups according to the pollen composition of their respective provision with the package GMD v0.3.3 (Zhao & Sandelin, 2012), to form groups with samples showing similar foraging patterns. Classification for clustering the plant species into groups was based on generalized minimum distance functions using k-means after cluster number selection according to the Elbow method which takes into account the percentage of variance explained by the clusters against the number of clusters (Kodinariya & Makwana, 2013). We used FENNEC v1.0.5 (Ankenbrand, Hohlfeld, Weber, Förster, & Keller, 2018) to add plant trait information to our pollen composition data and investigate possible phenological differences between the formed plant clusters.
Subsequently, we used random forest analysis to assign bacterial communities of pollen and larvae (a) to host bee species and (b) to the defined pollen composition clusters and estimate the significance of these variables for correct classification (Junker & Keller, 2015;Prasad, Iverson, & Liaw, 2006) with the packages varSelRF v0.7-8 (Diaz-Uriarte, 2007) and randomForest v4.6-14 (Liaw & Wiener, 2002). The produced confusion matrices from the random forest analysis include the number of correctly assigned communities to either species or plant clusters as well as class error and total out-of-basket (OOB) error rate. To identify indicator bacterial OTUs per tested group, we used variable selection with the OOB error rate estimate set as a minimization criterion.

| RE SULTS
We examined 100 megachilid bee nest chambers, which belong to seven solitary bee species. Sequencing for the 16S rDNA generated on average 8,173 high-quality reads (range from 1,006 to 53,729, SD = 8,644.126), while sequencing for the ITS2 rDNA generated on average 13,791.22 high-quality reads (range from 2,310 to 61,846, SD = 5,298.317), after quality and control filtering. For ITS2 reads, 1,958,852 out of 2,139,677 acquired sequences (91.55%) were successfully mapped to the reference database with an identity threshold of >97%. We found clusters for 2,874 bacterial OTUs and assignments for 415 plant species (Appendix S1).

| Bacterial communities in larvae and pollen
Lactobacillus was the most abundant taxon in larvae and pollen of all three Megachile species included in this study, while it occurred in all sample groups (Figure 1) Chiou et al., 2018) and to the honeybee-associated

| Plant species composition in pollen provisions
The interaction network between bee species and plant species is illustrated in Figure 1

| Co-occurrence of bacterial taxa with plant species
The probabilistic co-occurrence analysis investigated which plant species were significantly associated with specific bacterial taxa in the pollen provisions of all nests. Results showed possible connections between bacterial taxa and plant species, marked as positive, negative, and random interactions (Figure 2).

| Alpha diversity of bacterial OTUs and pollen plant species
Shannon bacterial diversity was higher for pollen samples (gray box plots in Figure 3) compared to larvae (black box plots in Figure 3).
The Wilcoxon test between all larvae and pollen samples returned statistically significant results for Shannon values (p < .001***).
Furthermore, we conducted pairwise Spearman's rank correlations to investigate associations between bacterial diversity for larvae, bacterial diversity for pollen, and pollen-type diversity. Spearman's F I G U R E 2 Probabilistic co-occurrence analysis results for the most abundant bacterial taxa found in pollen and the plant species in pollen provisions. Bacterial taxa shown here are agglomerated up to genus level or up to family level if not better classifiable. Analysis is based on absence-presence data. The relative abundance threshold for a taxon to be considered as existent in a sample is 1%. Taxa showing only random interactions were removed F I G U R E 3 Left: Shannon diversity of bacterial communities in larvae and pollen samples for different solitary bee species based on revealed bacterial OTUs per nest chamber. Shannon bacterial diversity was consistently higher for pollen samples (gray box plots) in comparison with larvae (black box plots). Right: Shannon diversity of pollen species in pollen provisions for different solitary bee species

| Correlation of bacterial and pollen composition
We were able to detect statistically significant Mantel correlations of pollen composition with pollen bacterial communities and also with larval bacterial communities in the whole dataset (Table 1).
When we conducted the same tests within each host bee species,  (Table 1).

| Bacterial and pollen beta diversity
The ordinations for the bacterial communities and the pollen composition of all samples are shown in Figure 4. Adonis/PERMANOVA results indicated that bee species identity was more significant as a discriminative factor for pollen composition in our dataset (R 2 = 0.33,

| Assignment of bacterial communities to host bee species and pollen composition
We further tested with a cluster analysis whether different dietary habits within and between bee species are associated also with differences in the respective bacterial microbiota. Pollen provisions from all samples were divided into seven clusters according to their composition in plant species. These clusters were treated as different whole mixed diets for individual larvae to which they were assigned. The ecological trait analysis showed that only O. bicornis foraged for trees, while all the other bee species visited subshrubs or forbs ( Figure 5).
We tested the efficiency of the pollen clusters and of the host bee species identity to predict the respective pollen bacterial communities with random forests ( Figure 5). Random forest analysis as- to pollen composition cluster.

| D ISCUSS I ON
In the present study, we investigated the association between the pollen composition and the nest bacterial microbiota for seven megachilid bee species by comparisons of solitary bee nests both between bee species and within the same bee species. Although a degree of the discovered microbiota variation between different bee species could be attributed to the bee species traits and the possible ability of the mother bee to actively transfer bacteria to her eggs, we showed that the bacterial diversity and composition of the nest is also related to the plant diversity and composition of the provided

| Bacterial composition in the nest environment
The investigation of the solitary bee nest bacterial microbiota revealed diverse bacterial communities. Additionally, we found bacterial taxa which were prevalent only in the pollen provisions of some bee species, while they were absent from others. This is the case with Lactobacillus spp. in particular; the genus is closely connected with the pollen provisions as well as with the larvae of Megachile bees (Figure 1). At the same time, Lactobacillus spp.   Sickel et al., 2015). Yet, to avoid potential problems with abundance estimations in pollen metabarcoding (Bell et al., 2016;Richardson et al., 2019), we in this study concentrate on the presence/absence-based analyses.

| Pollen composition in nest provisions
On the bee side, if plants act as reservoirs or transfer hubs for bacteria McFrederick et al., 2017), then specialized bee-plant interactions in a landscape could secure specialized bee-bacteria relationships. Thus, plant availability and specialized interactions with plants would be significant not only for the nutrition of the bees, but also for the maintenance of their nest microbiota.

| Co-occurrence of plant species with bacterial OTUs in the pollen
Co-occurrence analysis enabled us to look into bacterial taxa from pollen, which might be associated with specific plants. On the one hand, we expected to see bacterial taxa which are commonly associated with plants (Gnanamanickam, 2006;Junker & Keller, 2015;Junker et al., 2011) to be part of co-occurrence relationships with various plant species. The genus Pseudomonas, for instance, was associated with most plant species as randomly distributed (Veech, 2013), while the genus Erwinia co-occurred particularly with plants

| Association between pollen composition and bacterial microbiota
Our dataset consisted of seven solitary bee species sampled across numerous sites of a geographical region. Therefore, their revealed foraging patterns could be the result of spatial and temporal fac- In general, the pollen composition was significantly correlated with the bacterial community in pollen, as well as with the bacterial community in larvae through our whole dataset (Table 1)

| CON CLUS IONS
Specialized plant-bee relationships could lead to specialized bee-microbe interactions, making passive transmission through imported pollen an important driver of the natural bee nest microbiota. In the present study, we were able to reveal distinctive foraging patterns for seven bee host species, as well as to describe the bacterial communities of their nests. The parallel investigation of the nest bacterial communities and of the pollen composition allowed us to propose host-microbe interactions, which might be secured through floral resource exploitation ( Figure 6). Pollen alpha diversity was correlated with the respective bacterial diversity in the pollen provisions. Also, the pollen community structure was significantly correlated with that of the pollen bacterial communities for the whole dataset and as well within some bee species. Moreover, pollen composition was a successful predictor for the bacterial microbiota at an intraspecific level. Future studies should include more bee species, F I G U R E 6 The parallel investigation of the nest bacterial communities and of the pollen composition suggests that floral resource exploitation plays a role in bacterial acquisition and host-microbe interactions larger sample sizes, and investigation of the floral microbiota of actual bee-visited plants for a direct comparison with the acquired larval bacterial community in the nests. Such dedicated experiments will increase our understanding of the causality behind the observed covariance between pollen composition and bacterial microbiota in the nests of solitary bees and to interpret its importance for their health and fitness.

ACK N OWLED G M ENTS
We appreciate allowance to use sequencing devices of the Department of Human Genetics, University of Wuerzburg. This

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
AK, AV-K, and IS-D designed the study. AV-K and GG performed laboratory and fieldwork. AV-K, MJA, and AK analyzed data. AV-K drafted the manuscript, and all authors contributed to the final version.