Taxonomic Characterization of Honey Bee ( Apis mellifera) Pollen Foraging Based on Non-Overlapping Paired-End Sequencing of Nuclear Ribosomal Loci

Abstract

Identifying plant taxa that honey bees ( Apis mellifera) forage upon is of great apicultural interest, but traditional methods are labor intensive and may lack resolution. Here we evaluate a high-throughput genetic barcoding approach to characterize trap-collected pollen from multiple North Dakota apiaries across multiple years. We used the Illumina MiSeq platform to generate sequence scaffolds from non-overlapping 300-bp paired-end sequencing reads of the ribosomal internal transcribed spacers (ITS). Full-length sequence scaffolds represented ~530 bp of ITS sequence after adapter trimming, drawn from the 5’ of ITS1 and the 3’ of ITS2, while skipping the uninformative 5.8S region. Operational taxonomic units (OTUs) were picked from scaffolds clustered at 97% identity, searched by BLAST against the nt database, and given taxonomic assignments using the paired-read lowest common ancestor approach. Taxonomic assignments and quantitative patterns were consistent with known plant distributions, phenology, and observational reports of pollen foraging, but revealed an unexpected contribution from non-crop graminoids and wetland plants. The mean number of plant species assignments per sample was 23.0 (+/- 5.5) and the mean species diversity (effective number of equally abundant species) was 3.3 (+/- 1.2). Bray-Curtis similarities showed good agreement among samples from the same apiary and sampling date. Rarefaction plots indicated that fewer than 50,000 reads are typically needed to characterize pollen samples of this complexity. Our results show that a pre-compiled, curated reference database is not essential for genus-level assignments, but species-level assignments are hindered by database gaps, reference length variation, and probable errors in the taxonomic assignment, requiring post-hoc evaluation. Although the effective per-sample yield achieved using custom MiSeq amplicon primers was less than the machine maximum, primarily due to lower “read2” quality, further protocol optimization and/or a modest reduction in multiplex scale should offset this difficulty. As small quantities of pollen are sufficient for amplification, our approach might be extendable to other questions or species for which large pollen samples are not available.