The current decline of pollinators may disrupt ecosystems and ecosystem services with potentially harmful effects on nature and human society. While the importance of habitat loss and fragmentation, pollution and increased disease risk in driving pollinator decline has been clearly demonstrated, the impact of resource diversity is less well understood. In this study, we investigated the effect of pollen diversity and composition on reproductive success and fitness of Bombus terrestris colonies. We asked the question whether a higher plant diversity results in a more diverse diet, lower pathogen incidence and a higher colony fitness. To answer these questions, colonies of lab-reared bumblebees were placed in species-poor heathland and species-rich semi-natural grasslands that strongly differed in plant community composition and diversity. We examined pollen loads on the bodies of foragers and identified the plant taxa present in the realized diet via DNA metabarcoding of the ITS2 marker. Liquid chromatography-mass spectrometry (LC-MS) was used to compare peptide composition of pollen samples from both habitats. Colony fitness was assessed by counting the number of sexuals produced by the colony at the end of its cycle. At the same time, colonies were examined for parasite incidence. Pollen composition and diversity on pollinators’ bodies differed significantly between bees foraging in grasslands and heathlands. Concomitantly, peptide composition differed significantly between pollen samples from grasslands and heathlands. Colonies developed significantly better in heathland sites than in grasslands. In addition, colony fitness was only weakly related to pollen diversity and effects in some cases depended on the habitat where the bees were foraging. Pathogen incidence was very low and not affected by habitat. Overall, our results indicate that plant diversity is not necessarily a good predictor of colony fitness, and that vegetation composition and associated differences in both the quantity and quality of pollen are more important than pollen diversity per se.

Colony performance

After collection of the hives at the end of the experiment, all bumblebees were killed and stored at -20°C to conserve the hives’ structure. All hives were dissected, and individual caste members were counted. The number of queens, workers, males, regular pupae and queen pupae were counted to assess colony development. The production of sexuals is mainly indicative for colony performance and directly related to fitness. Therefore, the total number of queens (emerged and pupae), the number of males, and the ratio of gynes:males was used as proxy for colony performance. A higher ratio in the latter indicates better colony performance because of the higher overall investment in queens than males (Duchateau et al, 2004). In addition, we recorded other information such as colony net weight gain, presence of involucrum, and the weight of the pollen mass stored in the pollen pots inside the colony.

DNA metabarcoding of pollen samples

Pollen DNA was extracted using the Norgen Biotek Corp Plant/ Fungi DNA Isolation Kit following the manufacturer’s instructions. Given the powdery nature of pollen, no grinding in liquid nitrogen was performed. DNA quantity and quality were assessed using the Thermo Scientific NanoDrop Spectrophotometer (Desjardins & Conklin, 2010). Species identification was done via DNA metabarcoding of the internal transcribed spaced 2 nuclear ribosomal fragment (ITS2) (Suchan et al., 2019), using the primer pair ITS-S2F and ITS-S3R (Kozich et al., 2013; Sickel et al., 2015). The PCR consisted of 1 minute at 95 °C for initial denaturation, followed by 30 cycles of 15 seconds denaturation at 95°C, 15 seconds at 55 °C for hybridization and 15 seconds of elongation at 72 °C.  The ALLin HiFi DNA Polymerase (HLE0205 van highQu) master mix was used.

The PCR product was purified following the Agencourt® AMPure ® XP kit, with a single 70% ethanol wash. DNA yield quantification was done on the Invitrogen Qubit Fluorometer (Thermofisher). The aliquot DNA library was sent to Genomics Core Leuven for Illumina MiSeq sequencing. Data processing consisted of matching forward and reverse Illumina reads followed by omitting low quality samples. Reads with a similarity of 97% or higher were clustered in Operational Taxonomic Units (OTU). From each OTU, a reference sequence was computed and Blasted against the NCBI Gen database for species identification. If no species match was found, the higher taxonomic level was assigned to each OTU (genus or family). Only OTUs with a frequency larger than 0.0005% were kept. This way, the majority of rare OTUs from potential mismatches and high intraspecific variation for the marker were omitted from the data.

The relative number of Illumina sequence reads of a plant species or OTU per sample was considered to be a good approximation of the relative abundance of that taxonomic entity in the diet of Bombus terrestris (Deagle et al, 2018). All OTU abundances were transformed from number of sequence reads to relative abundances per sample.

Protein analyses

Complementary to DNA metabarcoding, liquid chromatography-mass spectrometry (LC-MS) was used to unravel possible compositional differences in protein content and to assess the quality of the pollen diet. Twenty milligrams of fresh pollen were grounded with pestle and mortar in liquid nitrogen and transferred to a clean Eppendorf tube. Total pollen protein content was extracted using a phenol extraction for recalcitrant plant tissue (Carpentier et al., 2005; Faurobert et al., 2007), because plants protect the pollen genetic material with the sporopollenin protein, one of the most inert biological polymers (Mackenzie et al., 2015) and the method also efficiently deals with polysaccharides, lipids and other phenolic compounds (Faurobert et al., 2007). Whilst working on ice to inhibit protease activity, 500µl of extraction buffer was mixed with the ground pollen. Next, 500µl buffered phenol was added to each sample and vortexed thoroughly on ice. The mixture was then centrifuged for 10 min at 12 000 rpm at 4 °C. The phenolic phase containing the proteins was collected and an equal volume of extraction buffer was added to the phenolic phase. The mixture was centrifuged for 5 min at 12,000 rpm at 4 °C. Proteins were precipitated overnight at -20 °C from the residual phenol phase using 100 mM ammonium acetate in cold methanol (5*Volume of phenolic phase) followed by centrifuging 60 min at 13,000 rpm at 4 °C. The supernatant was carefully removed to not resuspend the protein pellet. The protein pellet was rinsed twice in 2 ml rinsing solution without resuspension. Each rinse with an incubation for 1 hour at -20 °C, followed by centrifuging for 30 min on 13,000 rpm at 4 °C. After rinsing, the pellet was dried under the fume hood before resuspension in 90 µl in lysis buffer, without overheating to avoid carbamylation. Samples were centrifuged for 30 min at 13 000 rpm at 30 °C and stored at -80 °C until further purification and mass spectrometry preparation. After the last extraction step, peptide quantification was done using the procedure for the 2-D Quant Kit (GE Healthcare Life Sciences).

Proteins were selectively digested between lysine and arginine with the endoprotease Pierce™ Trypsin Protease (MS Grade) from ThermoFisher Scientific. Cleavage to polypeptides reduced the polypeptides' length to an average of 700–1500 dalton, which is a suitable range for MS. Cleaved samples were resuspended and 1,4-Dithiothreitol (DTT, Sigma, final working concentration 20 mM) was added followed by 15min incubation. Next, iodoacetamide (IAA, final working concentration 50 mM) was added to each sample as an irreversible cysteine peptidase and incubated in the dark for 30min. Ammonium bicarbonate (150mM, 3* sample volume) was added for protein reduction. 0.2 µg trypsin was added to each sample followed by overnight incubation at 37 °C. After overnight digestion, 1/10th of sample volume of 1% trifluoroacetic Acid (TFA) was added to stop trypsin activity (SyBioMa, 2019). Samples were purified using Pierce® C18 Spin Columns (Thermo Scientific). Cleaned samples were dried in a SpeedVac vacuum drier (Thermo Scientific) for approximately 40 min.

Peptide separation was performed on the Ultimate 3000 HPLC Liquid Chromatography (Thermo Fisher Scientific). Peptide charging was done with Electrospray Ionization (ESI) followed by mass analysis with Thermo Fisher Scientific Orbitrap Q. In total, 32 pollen samples from the first and second sampling (10/08 & 13/08 and 18/08 & 19/08/2019) were analyzed. 27 samples went successfully through the analysis pipeline and yielded usable results.

Incidence of parasites

Parasite incidence can strongly influence colony fitness. At the end of the five-week field experiment, bumblebees were therefore checked for the presence of endo- and ectoparasites. Infection of queens by Parasitellus mites was noted for each hive and was assigned a rank. Four categories were distinguished: ‘no mites’ meaning that not a single queen was found to be carrying mites, ‘few mites’ for low infection rate (<25%) or low number of mites on the queens, ‘medium infection’ for up to 50% of queens affected or medium densities per queen and ‘high infection’ for everything above the previous.

From each hive, three workers were randomly selected for the detection of endoparasites, yielding to 216 dissected workers. Because workers are the most motile cast, the chance of encountering parasites is higher in workers than in the other castes. The workers were screened for incidence of Nosema bombi and N. ceranae, Crithidia sp. and Apicystis sp. using quantitative PCR (qPCR). The hindgut, possibly with some remains of trachea or the poison gland, of three workers from the same hive were pooled in a clean Eppendorf tube and stored at -20°C before qPCR. qPCR was performed using the primers described by Huang et al. (2016) and the protocol specified in Bosmans et al. (2018). Each reaction was performed in duplicate. In each analysis, a positive (pathogen DNA) and negative control (template DNA replaced by sterile water) were included. Additionally, six gut samples from wild bumblebees that were known to be either contaminated or free of pathogens (screened by microscopy and regular PCR) were included as additional controls. To verify amplification specificity, a melting curve analysis was performed. CT-values were used to calculate DNA concentration and to assess parasite load. Quantification was based on standard curves from amplification of cloned target sequences in a TOPO-TA vector (Invitrogen). All samples exceeding the DNA concentration of the negative control (water) were considered positive. From three of the 72 hives, the qPCR was not successful (2 heathland and 1 grassland hive).

The presence of Physocephala sp. was not expected prior to dissection of the workers but was frequently encountered by visual encounter of the larvae or pupae (Schmid-Hempel and Schmid-Hempel, 1996). Thorough colony infection rates could not be extrapolated from the incidence data of the five randomly selected workers nor can the parasitoids be transferred horizontally. Incidence per colony was therefore only categorically (‘present’ vs ‘absent’) recorded.

See README file. There are separate files for each section above. R code refers to any of these sections and datasets.