Investigating the use of pollen DNA metabarcoding to quantify bee foraging and effects of threshold selection

DNA metabarcoding of pollen is a useful tool for studying bee foraging ecology. However, several questions about this method remain unresolved, including the extent to which sequence read data is quantitative, which type of sequence count removal threshold to use and how that choice affects our ability to detect rare flower visits, and how sequence artefacts may confound conclusions about bee foraging behavior. To address these questions, we isolated pollen from five plant species and created treatments comprised of pollen from each species alone and combinations of pollen from multiple species that varied in richness and evenness. We used ITS2 and rbcL metabarcoding to identify plant species in the samples, compared the proportion of pollen by mass to the proportion of sequencing reads for each plant species in each treatment, and analyzed the sequencing data using both liberal and conservative thresholds. We collected pollen from foraging bees, analyzed metabarcoding data from those samples using each threshold, and compared the differences in the pollinator networks constructed from the data. Regardless of the threshold used, the relationship between the proportion of pollen by mass and sequencing reads was inconsistent, suggesting that the number of sequence reads is a poor proxy for pollen abundance in mixed-species samples. Using a liberal threshold resulted in greater detection of original plant species in mixtures but also detected additional species in mixtures and single-species samples. The conservative threshold reduced the number of additional plant species detected, but several species in mixtures were not detected above the threshold, resulting in false negatives. Pollinator networks produced using the two thresholds differed and illustrated tradeoffs between detection of rare species and estimation of network complexity. Threshold selection can have a major effect on conclusions drawn from studies using metabarcoding of bee pollen to study plant-pollinator interactions.