Wildflower habitats planted along field borders are a widely promoted strategy for supporting bees in agricultural landscapes. However, honeybees (Apis mellifera), which are often stocked at high densities in crop lands, can compete with wild bees for pollen and nectar, potentially limiting the successfulness of wildflower plantings in supporting diverse bee communities.

Using weekly samples of five study sites in Northern California we assessed how plants in pollinator-friendly seed mixes varied in their ability to provide bees with abundant and nutritious pollen under intense honeybee competition.

We quantified pollen production, protein and lipid content, and end-of-day pollen availability for different plant species. We also sampled bee visits to flowers and assessed the composition of pollen on bee bodies. Using these data, we investigate how the nutritional quality of pollen in wildflower plantings and honeybee abundance impacted native bee pollen nutrition.

Bees collected more nutritious pollen (i.e., pollen with more protein) from plantings with more nutritious plant species (i.e., sites with more high-protein plants). However, as honeybee abundance increased, the nutritional quality of native bee diets declined. We also detected important interactions between honeybee abundance and the nutritional quality of flowers in plantings, such that, for some bee taxa, there was no impact of competition on pollen diet quality in high-nutrition plantings.

Please see the meta data file (README.md) for more detailed information about variable names and their interpretation. 

Study sites and pollinator surveys - We conducted this work in the California Central Valley at five replicated wildflower plantings neighboring conventionally managed Almond orchards. Sites varied in honeybee abundance due to spatial variation in the placement of commercial apiaries. In 2017 and 2018, we surveyed bees and their visits to flowering plants over four sample rounds from April – May. We netted insects actively visiting flowers during 10-minute walks of two 100 m² transects sampled once in the morning and once in the afternoon (40 minutes total). Netted pollinators were collected individually in separate vials to minimize pollen contamination and euthanized using dry ice, except for bumblebee queens, which we identified, swabbed for pollen, and released. We collected up to twenty honeybees during netting transects and counted any additional honeybees. Bees were identified to species or morphospecies by expert taxonomists (Skyler Burrows, USDA Bee Lab, Logan, Utah, and Joel Gardner, University of Manitoba, Canada). In statistical analyses (and in data provided here), we grouped bees into five categories: Apis mellifera, Bombus spp., Megachilidae, Halictidae, and “Other bees” (comprising Andrenidae, Colletidae, and non-corbiculate Apidae). The datafile "beeflow_df.csv" contains information on pollinator species identifications, associated taxonomic groupings used in statistical analyses, the identity of the flowering plant species a given pollinator was visiting when collected, and site-level information on honey bee abundance.

Pollen availability in flowers - At the end of each sampling day, we assessed pollen availability for the most abundant and well-represented plant species: Clarkia unguiculata, Clarkia williamsonii, Collinsia heterophylla, Eschscholzia californica, Lupinus densiflorus, and Phacelia californica. We measured pollen availability as the proportion of dehisced anthers with pollen visible to the naked eye, using one to three flowers on 10-20 “open-pollinated” (i.e., unmanipulated) plants. The datafile "pollen_avail_df.csv" contains information on the number of dehisced anthers with and without pollen for flowers of different plant species sampled at different sites. 

Nutritional quality of pollen from different plant species -  We grew monospecific flower plantings for each of the ten most-visited plant species (representing 96% of floral visits and 92% of bee-collected pollen). We covered each planting with organza fabric prior to anthesis to prevent insect visitation, collected fresh pollen from flowers, and assessed pollen protein and lipid content. As we harvested pollen for macronutrient analyses, we also collected data on the number of florets sampled and the mg of pollen extracted from those florets, giving us an estimate of pollen production per floret.

Assessing pollen diet composition  – In the lab, we used fuchsin-tinted jelly cubes to remove and stain pollen from bee bodies, which we then melted onto microscope slides. When swabbing bees, we focused our effort on scopae (i.e., pollen from specialized pollen-collection hairs), but also lightly brushed the head and thorax. For Apis and Bombus, we used tweezers to dislodge small clumps of pollen from multiple areas of corbiculae (i.e., pollen baskets). We identified pollen to species using light microscopy and a pollen reference collection. In 2018, we assessed pollen from corbiculae and scopae separately from pollen removed from other body parts, but ultimately pooled pollen from both sources for each specimen in our analyses. Using data on the composition of pollen on bee bodies and information on the protein and lipid content of pollen from different plant species, we calculated the mean protein content and protein-to-lipid (P:L) ratios of pollen collected by each bee specimen sampled. In the datafile "beeflow_df.csv", these variables are named 'bee.protein.prop' (estimated protein content of pollen on bee bodies) and 'PL.ratio' (estimated P:L ratio of pollen on bee bodies). When possible, we separately estimated the protein content and P:L ratios for bee scopal/corbicular loads (variable names: "load.protein.prop" and "load.PL.ratio"). 

Estimating nutritional quality of wildflower plantings at time of sampling - We estimated nutrient availability in wildflower plantings for each site sample by calculating the mean protein-to-lipid (P:L) ratio and protein content of available pollen. In the datafile "beeflow_df.csv", these variables are named 'protein.prop.by.mg' (estimated protein content of pollen available in plantings at time of sampling) and 'PL.ratio.by.mg' (estimated P:L ratio of pollen available in plantings at time of sampling). To generate these estimates, we used data on the composition of floral resources available in netting transects and data on the protein and lipid content of pollen from these plant species. On the same day and in the same transects where bees were collected, we assessed flowering species composition by counting and identifying all flowers in 10 evenly spaced 1m² quadrats. To account for the fact that not all plants produce the same amount of pollen, we weighted the contribution of different plant species by their pollen abundance, where pollen abundance was calculated as the number of florets of a given species counted during a site sample multiplied by the pollen production per floret of that same species.