Plant identity in pollen pellets
To identify the plants bees collected pollen from, each individual pollen pellet was crushed with a small pestle in a 1.5mL microcentrifuge tube, followed by the addition of a Safarin-O pollen staining solution. Ten µL of homogenate was then placed onto a microscope slide, topped with a coverslip and sealed along the edges using clear nail polish (Jones, 2012). Using light microscopy, we counted and identified 500 pollen grains in each sample at 400x magnification. We verified that this is an appropriate sample size by rarefaction using the ‘rarecurve’ function in the ‘vegan’ package (Oksanen et al., 2019) in R (R Core Team 2013) (Appendix: Fig. S1). We made a library of visually distinct pollen morphotypes, and each pollen grain was identified at the lowest possible plant taxonomic level using reference samples collected from the study site, and online atlases (flickr.com/photos/161453633@N02/collections; paldat.org; blogs.cornell.edu/pollengrains; globalpollenproject.org). Following identification, the proportional abundance of each visually distinct pollen morphotype within a pollen pellet was calculated at both the plant family and pollen morphotype levels. We included a pollen morphotype if it occurred in at least 3% of all pollen identified within a pellet (≥ 15 out of 500 pollen grains).