Characterizing the Floral Resources of a North American Metropolis Using

METHODS, TOOLS, AND TECHNOLOGIES

Characterizing the ﬂoral resources of a North American metropolis using a honey bee foraging assay 1, 2 3 1 DOUGLAS B. SPONSLER, DON SHUMP, RODNEY T. RICHARDSON, AND CHRISTINA M. GROZINGER

1Department of Entomology, Huck Institutes of the Life Sciences, Center for Pollinator Research,Pennsylvania State University, University Park, Pennsylvania 16802 USA 2Philadelphia Bee Company, Philadelphia, Pennsylvania 19125 USA 3Department of Biology, York University, Toronto, Ontario M3J 1P3 Canada

Citation: Sponsler, D. B., D. Shump, R. T. Richardson, and C. M. Grozinger. 2020. Characterizing the ﬂoral resources of a North American metropolis using a honey bee foraging assay. Ecosphere 11(4):e03102. 10.1002/ecs2.3102

Abstract. Roughly a third of described insect species visit flowers, making the flower–insect interface one of the chief pillars of global biodiversity. Studying flower–insect relationships at the scale of communities and landscapes has been hindered, however, by the methodological challenges of quantifying landscape-scale floral resources. This challenge is especially acute in urban landscapes, where traditional floral surveying techniques are ill-suited to the unique constraints of built environments. To surmount these challenges, we devised a “honey bee foraging assay” approach to floral resource surveying, wherein continuous colony weight tracking and DNA metabarcoding of pollen samples are used to capture both the overall availability and taxonomic composition of floral resources. We deploy this methodology in the complex urban ecosystem of Philadelphia, Pennsylvania, USA. Our results reveal distinct seasonality of floral resource availability, with pulses of high availability in May, June, and September, and a period of pro- longed scarcity in August. Pollen genus richness mirrored this pattern, with peak richness in May and June. The taxonomic composition of pollen samples varied seasonally, reflecting underlying floral phenology, with especially strong turnover between May and June samples and between August and September samples delineating well-defined spring, summer, and fall floral resource communities. Trait analysis also revealed seasonal structure, with spring samples characterized by trees and shrubs, summer samples including a stronger presence of herbaceous “weeds”, and fall samples dominated by woody vines. Native flora predominated in spring, giving way to a preponderance of exotic flora in summer and fall. At a basic level, this yields insight into the assembly of novel urban floral resource communities, showcasing, for example, the emergence of a woody vine-dominated fall flora. At an applied level, our data can inform urban land management, such as the design of ecologically functional ornamental plantings, while also providing practical guidance to beekeepers seeking to adapt their management activities to floral resource seasonality. Methodologically, our study demonstrates the potential of the honey bee foraging assay as a powerful technique for landscape-scale floral resource surveying, provided the inherent biases of honey bee foraging are accounted for in the interpretation of the results.

Key words: biodiversity; hive scale; Internal transcribed spacer; metabarcoding; pollination; trnL; urban ecology.

Received 13 November 2019; revised 23 January 2020; accepted 29 January 2020. Corresponding Editor: Mary L. Cade- nasso. Copyright: © 2020 The Authors. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. E-mail: [email protected]

❖ www.esajournals.org 1 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

INTRODUCTION morphology (excluding, for example, flowers requiring sonication to release pollen from It has been estimated that roughly a third of all anthers). described insect species are either directly or Here, we present a two-year study of the floral indirectly dependent on flowers for food (Ward- resource composition and dynamics of Philadel- haugh 2015). This ecological centrality of flow- phia, Pennsylvania, USA. Founded in 1682, the ers-as-food extends to systems in which historic city of Philadelphia straddles the interface of the floral communities have been dramatically Appalachian Piedmont and Atlantic Coastal altered, such as urban landscapes characterized Plain physiographic provinces (Fenneman and by novel assemblies of native and exotic flora Johnson 1946) and hosts over 2500 plant species (Aronson 2014). Urban landscapes can host (Clemants and Moore 2003). Its patchwork diverse communities of flowers and flower visi- heterogeneity of sociological composition is both tors, sometimes even functioning as refugia for a product and driver of concomitant ecological rare species (Baldock 2015) [though see (Geslin variation in abiotic substrate and biotic commu- et al. 2013)], but little is known about the overall nities, earning Philadelphia the apt moniker composition of urban floral resources and how “City of Neighborhoods”. To characterize the flo- they fluctuate phenologically through the grow- ral resources of this complex urban environment, ing season. we employ a “honey bee foraging assay” using a Studying floral resources at the landscape scale network of sentinel apiaries distributed through- is technically daunting in any context (Frankl out the city. Combining DNA metabarcoding of et al. 2005), but the challenge becomes especially pollen samples with continuous colony weight acute in urban landscapes where extreme land- monitoring, we characterize both the taxonomic scape heterogeneity, limited land access, and the composition and overall availability of floral physical obstacles of the built environment render resources in our study system, together with traditional approaches to floral surveying imprac- their temporal dynamics throughout the foraging ticable. Flower-visiting insects might be har- season. We conclude by comparing Philadel- nessed as efficient environmental samplers of phia’s floral resource landscape with patterns landscape-scale floral resources, provided the described in other study systems and discussing spatial and taxonomic scope of their interaction the implications of our data for understanding with the landscape is sufficiently understood and the ecological function of urban flora. the materials they collect can be analyzed informatively (Wood et al. 2018). While not an unbi- METHODS ased representation of local flora, such an approach would manifestly be relevant to the Sites and study years question of trophic function at the flower–insect We conducted fieldwork in 2017 and 2018. Our interface. study sites included 13 apiaries in Philadelphia, The western honey bee (Apis mellifera L.) is Pennsylvania, USA, owned and managed by the arguably the organism best suited for such a Philadelphia Bee Company (Fig. 1). Twelve api- sampling approach. As an extreme generalist, aries were used in each study year, with 11 honey bees have a diet breadth that overlaps shared across years, 1 used only in 2017 (RB), considerably with that of many other nectar- and and 1 used only in 2018 (NE). Each of our pollen-feeding insects (Butz Huryn 1997). A research apiaries included three honey bee colo- honey bee colony surveys the landscape around nies designated as research colonies for our its nest at a range routinely extending several study, resulting in a total of 36 research colonies kilometers, dynamically allocating foraging across 12 sites in each year. Some sites also effort to the most rewarding patches (Visscher included additional colonies not involved in our and Seeley 1982). Thus, the materials it collects study. Research colonies were initiated in April– are informatively biased toward the richest May of each study year from 4- or 5-frame resources within the colony’s foraging range and nucleus colonies installed in 10-frame Langstroth compatible with honey bee behavior and hives. Nucleus colonies were purchased together

❖ www.esajournals.org 2 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Fig. 1. Apiary locations (circles) plotted over land cover raster. The City of Philadelphia is demarcated by a dashed black line. Land cover data are from the Chesapeake Conservancy’s Land Cover Data Project (Chesa- peake Bay Conservancy 2017); only major cover types are shown in legend. Area east of the Delaware River in the State of New Jersey is not shown because it fell outside the scope of the land cover dataset. each year from a single supplier (Swarmbustin’ spikes (i.e., additive outliers), and persistent Honey, West Grove, Pennsylvania, USA). During weight shifts (i.e., level-shift outliers) caused by colony installation, each research hive was colony manipulations involving the addition or equipped with a Sundance I bottom-mounted removal of material from the hive. Because our pollen trap (Ross Rounds, Canandaigua, New colonies were part of a working apicultural busi- York, USA) and a Broodminder hive scale ness, it was not possible to avoid or standardize (Broodminder, Stoughton, Wisconsin, USA). these management artifacts. To address both types of artifacts, we first took Weight monitoring and pattern characterization the first-order difference of each colony weight We set our Broodminder hive scales to record time series, which turns level-shift outliers into hourly weights, beginning at colony installation additive outliers. We then applied a filter to each in the spring of each year and continuing time series where data points (i.e., weight through the end of our sampling season in the changes between consecutive readings) with an fall of each year. During preliminary data visual- absolute value of 2.5 kg or more were identified ization, we identified two types of artifacts in our as outliers and set to zero. The threshold of weight data: apparently random misreadings, 2.5 kg was chosen because histograms of first- resulting in transient (single reading) weight order differenced time series showed consistent

❖ www.esajournals.org 3 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL. discontinuities around Æ 2.5 kg, such that the NE hives due to premature death, and SP hive 3 interval À2.5 to 2.5 contained a smooth curve of due to premature death. This left a total of 25 ostensibly “normal” readings, while readings colonies across 9 sites for 2017 and 30 colonies outside this interval were rare and extreme. We across 11 sites for 2018. then reintegrated each time series to reconstruct To characterize weight patterns, we used gen- de-artifacted weight curves and selected only eralized additive modeling (GAM) implemented midnight readings for downstream analysis. For in R (R Core Team 2019) using package mgcv the purposes of our study, we were interested (Wood 2017) to relate colony weight to time and principally in the temporal patterns of resource site. Our goals were (1) to identify major gain availability arising from landscape-scale floral and loss motifs, (2) to determine whether there phenology, and how these patterns might vary was significant site-to-site variation in weight site to site. To isolate these signals from the noise pattern, and (3) to determine whether there was of colony-level differences in amplitude of sufficient similarity across sites to justify a global weight variation (an expected result of differ- weight model. Accordingly, we constructed three ences in colony strength), we normalized (scaled nested hierarchical GAMs, following Pedersen and centered) the de-artifacted weight curves for et al. (2019). Model 1 included only a global each colony. This pipeline is available in Data S1 smooth of weight by time, without site-specific and depicted graphically in Fig. 2. Raw data are smooths. Model 2 included both a global smooth available in Data S2. and site-specific smooths. Model 3 included only Hives subject to known anomalies were omit- site-specific smooths, without a global smooth. ted from downstream analysis. In 2017, we omit- These models are analogous to the G, GI, and I ted all hives from sites CH, EG, and SP due to models, respectively, of Pedersen et al. (2019). late colony installation. We also omitted CC hive Since any effect of site or hive on group means 3 and RB hive 2 due to scale malfunctioning. In was removed by our normalization step, we did 2018, we omitted SG hive 3 and WP hive 3 due to not include site or hive random effects. After user error (failure to replace dead batteries), all optimizing the fit of each model, we used AIC

Fig. 2. Weight data processing workflow illustrated for three hives at one site. Raw weights (A) contain additive and level-shift outliers. To remove these artifacts, time series were first differenced (B), then filtered (C), and reintegrated (D). Reintegrated time series were then subsampled to include only midnight readings (E). Finally, time series were z-normalized and fit with a GAM smooth (F).

❖ www.esajournals.org 4 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL. model selection to identify the most appropriate sample was air-dried in an open conical tube model. GAM smooths were then first-order dif- under a fume hood for 24 h to allow the EtOH to ferenced to enable the visualization of detrended evaporate, yielding the final granularized pollen gain and loss periods. R code for our GAM anal- sample. ysis is available in Data S1. From each granularized sample, we trans- ferred a 10 mg aliquot to a 2-mL beadmill tube. Pollen sampling To each tube, we added 1 mL of either Qiagen Pollen samples were collected at roughly buffer AP1 (for 2017 samples) or ultra-pure water monthly intervals (hereafter “sampling peri- (for 2018 samples) and approximately 0.5 mL of ods”). In 2017, sampling occurred in late May/ 1.3-mm chrome steel beads (BioSpec, BioSpec, early June (hereafter “June sampling period”), Bartlesville, Oklahoma, USA). Samples were then late June/early July (hereafter “July sampling per- subjected to 4 9 1 min run on an Omni Bead iod”), early/mid-August, and mid-September, for Ruptor 24 Elite (Omni International, Kennesaw, a total of 4 sampling periods. In 2018, we Georgia, USA) set to a speed of 6 m/s. Samples expanded our sampling to include early May, were iced between runs to prevent overheating. early June, early July, early August, early For our 2017 samples, we purified beadmill September, and early October, for a total of 6 lysates using a Qiagen DNeasy Plant Minikit sampling intervals. Pollen traps were activated according to the manufacturer’s instructions. In for roughly 3–7 days at the end of each monthly 2018, we streamlined our workflow by using period. Exact sampling dates and durations var- Phire Direct PCR reagents (Thermo Fisher, Wal- ied from site to site due to travel time and tham, Massachusetts, USA) that allowed us to weather constraints, and a detailed sampling avoid the purification step (hence the use of schedule is available in Appendix S1. For the water rather than buffer AP1 in beadmilling). In purposes of downstream analysis, sampling both methods, a 1 lL aliquot of the final product dates were binned discretely by monthly sam- (either purified DNA extract or raw lysate) was pling period. Samples were collected by scooping used as PCR template. the contents of the pollen traps into 50-mL conical tubes. Excess pollen was removed from the Library preparation and sequencing trap and poured into the inside of the top box of Past work has shown that using multiple loci the hive to help the colonies recover from any in pollen DNA metabarcoding improves the pollen deficit caused by our sampling and to robustness of the method, particularly with ensure that the pollen trap would be empty in respect to quantitative reliability (Richardson preparation for the next sampling period. et al. 2015, 2019a). For this study, we selected the plastid intron trnL and the nuclear ribosomal Pollen sample preparation spacer regions ITS1 and ITS2. Each of these loci Pollen processing began with the creation of has been used individually or in combination pooled subsamples by combining 4/n g of pollen with other loci in past applications of pollen from each hive within each site-date, where n DNA metabarcoding (Keller et al. 2015, Kraai- represents the number of hives sampled in a site- jeveld et al. 2015, Smart et al. 2017), though ours date. For most samples, n was equal to 3 hives, is the first study to employ this exact combina- though there were some cases where fewer than tion. We chose ITS1 and ITS2 because these 3 hives at a given site could be sampled, either markers are highly divergent and offer relatively due to colony death or pollen trap malfunction- high taxonomic resolution (Chen 2010, Wang ing. Each 4 g subsample (hereafter “sample”) et al. 2015); trnL is far less divergent but easily was then suspended in 40 mL of 70% EtOH and amplified (Taberlet et al. 2007), and there is some vortexed until fully dispersed. The resulting sus- evidence that single-copy plastid markers might pension was then centrifuged (Eppendorf 5810 be more quantitatively reliable than the ITS R, Hamburg, Germany) at 3000G for 2 min and regions (Richardson et al. 2015). decanted. We then resuspended the pollen sedi- Our PCR methods follow the nested protocol ment in 10 mL of 100% EtOH and repeated the of Richardson et al. (2019a), which aims to mini- vortexing and centrifugation steps. Lastly, the mize PCR biases introduced by interactions

❖ www.esajournals.org 5 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL. between template and multiplex indices (Berry Table 1. eDirect queries used to download GenBank et al. 2011). Briefly, PCR1 amplifies barcode tar- reference sequences. get regions (trnL, ITS1, and ITS2, in separate Locus Query reactions), PCR2 adds Illumina read-priming trnL esearch -db nucleotide -query "Viridiplantae [ORGN] oligonucleotides, and PCR3 adds dual multiplex | fi indices (Kozich et al. 2013, Sickel et al. 2015). For AND trnL [ALL]" e lter -query 100:2000[SLEN] ITS1 esearch -db nucleotide -query "Viridiplantae [ORGN] the initial amplification of our target loci (PCR1), AND (its1 OR internal transcribed spacer [ALL])" | we used the trnL-c and trnL-h primers described efilter -query 100:2000[SLEN] by Taberlet et al. (2007) and the ITS1-u1, ITS1-u2, ITS2 esearch -db nucleotide -query "Viridiplantae [ORGN] AND (its2 OR internal transcribed spacer [ALL])" | ITS2-u3, and ITS2-u4 primers described by efilter -query 100:2000[SLEN] Cheng et al. (2016). Following PCR3, we cleaned and normalized our libraries using a SequalPrep normalization kit (Thermo Fisher), pooling libraries within markers so that they could be paired reads were then aligned to our curated sequenced equimolarly. In addition to running reference sequences using the usearch_global negative controls on all gels, we also propagated algorithm in VSEARCH (Rognes et al. 2016), 3 technical replicate negative controls for each with a minimum query coverage of 80% and a marker through the entire nested PCR protocol minimum identity of 75%. Only the top hit for and sequenced these alongside our positive sam- each query was retained for downstream anal- ples, enabling us to detect trace contamination ysis. that did not appear on gels. We also used a 10% After alignments were made, we performed unsaturated matrix of multiplex tags so that criti- further processing in R (R Core Team 2019). cal mistagging rates could be both mitigated and First, alignments were subjected to more strin- measured (Esling et al. 2015, Schnell et al. 2015). gent identity thresholds: 97% for trnL and 95% For a full description of primers, PCR conditions, for ITS1 and ITS2. Then, subject sequence acces- and negative controls, see Appendices S2, S3, S4. sion numbers were joined with taxonomic lin- Prepared libraries were sequenced using eages using the R package taxonomizr (Sherrill- 2 9 300 Illumina MiSeq kits, with 2017 and 2018 Mix 2019). Using the grouping and summariz- samples sequenced separately. Sequencing was ing utilities of the R package dplyr (Wickham performed at the Genomics Core Facility of Penn- et al. 2019), we tallied, for each marker, the sylvania State University. All sequencing output number of reads by genus and sample (i.e., site- has been archived on NCBI Sequence Read date) and expressed each genus tally as a pro- Archive under Bioproject PRJNA548320. portion by dividing its read count by the total number of reads for the sample to which it Bioinformatics belonged. Genera comprising less than 0.05% of Reference sequences were downloaded from the total read count for a given sample were dis- NCBI GenBank on 22 February 2019 using the carded to avoid low-abundance false positives. eDirect API (Table 1). Downloaded reference We then used the table joining utilities of dplyr sequences were then filtered at the genus level to (Wickham et al. 2019) to merge the genus tallies include only genera documented as occurring in across our three markers, and genera detected the states of Pennsylvania, New York, New Jer- by only one of the three markers were discarded sey, Delaware, or Maryland based on the USDA as potential false positives. Finally, we calcu- PLANTS database (USDA and NRCS 2020). We lated the median proportional abundance for then used MetaCurator (Richardson et al. 2019b) each genus across all markers by which it was to trim our reference sequences to the specific detected and rescaled the median proportional regions of our amplicons and taxonomically composition of each sample so that it totaled to dereplicate identical sequences. one. In comparisons with pollen microscopy, MiSeq reads were first mate-paired using this approach has been shown to yield quantita- PEAR (Zhang et al. 2014) and converted from tively informative results (Richardson et al. FASTQ to FASTA format using the FastX 2019a). R script for this workflow is available Toolkit (A. Gordon et al., unpublished). Mate- from the authors upon request.

❖ www.esajournals.org 6 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Taxonomic analysis manually added to the processed dataset by Prior to taxonomic analysis, we excluded any searching the genera in the USDA PLANTS samples in which one of the three libraries was Database without geographical constraints and represented by fewer than 1000 reads to avoid selecting species with distributions including making inferences based on insufficient sequenc- Pennsylvania, Delaware, New Jersey, Maryland, ing depth. In 2017, all 44 libraries were ade- or New York. Using tidyverse (Wickham 2017) quately sequenced. In 2018, we dropped 6 of 71 functions in R, we filtered the resulting species libraries (8%) due to under-sequencing, includ- list by the genera detected in our pollen samples ing June samples for sites CH, EG, and NE, and summarized each trait for each genus by September samples for FF and NE and the Octo- concatenating unique trait values. Where trait ber sample for WP. values were not uniform across species within a Using the proportional abundance data given genus, the trait was considered “undeter- described above, we characterized the taxonomic mined”. For growth habit, we combined “Tree” composition of pollen samples across sites, sam- and “Shrub” classes into “tree/shrub” and the pling periods, and years using principal coordi- “Forb/Herb” and “Graminoid” classes in to nates analysis (PCoA) based on the Jaccard “herb”. Raw data, derived trait tables, and R dissimilarity metric and implemented with the R script are available in Data S1 and S2. package vegan (Oksanen 2019). Proportional abundances were square-root-transformed prior RESULTS to ordination, achieving the equivalent of a Hel- linger transformation (Legendre and Gallagher Weight dynamics 2001) on raw community data. Differences across Weight dynamics were characterized by dis- years, sampling periods, and sites were evalu- tinct gain and loss motifs that were largely con- ated by permutation tests using vegan’s adonis sistent across years, with periods of strong function (Oksanen 2019). We also analyzed sam- weight gain in May and June, weak gain or weak ple genus richness as a function of sampling per- loss in July, strong loss in August, weak to strong iod. Differences in richness across periods were gain in early September, and strong loss from the tested by ANOVA and pairwise differences were second week of September to the end of data col- analyzed with Tukey’s post hoc test. R script for lection in October (Fig. 3). For both years of our this workflow is available in Data S1. study, HGAM analysis with AIC model selection supported the use of both global and site-specific Trait-based analysis smoothers (Model 2), indicating both a site effect To explore the relationship between sampled and an archetypal pattern of weight dynamics relative abundance and genus traits, we classi- for our study system (Table 2). Site-specific fied each genus detected in our samples on the weight curves including integrated data (i.e., not basis of growth habit (tree/shrub, vine, herb), life subject to first-order differencing) can be gener- cycle (perennial, biennial, annual), and native ated using scripts and data in Data S1 and S2. status (native or exotic to contiguous United States). Trait data were obtained by downloading Pollen sample composition from the USDA PLANTS Database (USDA, We detected a total of 119 and 139 genera in NRCS 2020) all species records for Philadelphia, 2017 and 2018, respectively (tabulated data avail- PA, along with annotations for “Growth Habit,” able in Data S2). Pollen diversity (genus richness) “Duration,” and “Native Status.” We opted for a differed significantly across sampling periods narrower geographical scope for trait analysis (2017: F = 6.5, P < 0.001; 2018: F = 16.3, than we did for reference sequence curation P = 0.001), with June having significantly higher because the addition of congeners not likely to be diversity (Tukey test, P < 0.05) than August and present in our local study system would increase September in 2017 and May and June having sig- intrageneric trait heterogeneity, resulting in more nificantly higher diversity (Tukey test, P < 0.05) ambiguous trait assignments. Genera found in than all other months in 2018 (Fig. 4). A sum- our pollen data but absent from the USDA mary of sequencing output by sample is avail- PLANTS Database Philadelphia records were able in Data S2.

❖ www.esajournals.org 7 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

ornamental shrub Christmas berry (Photinia vil- losa is the only member of its genus present in our study system) dominated the composition of a single site (CC). Between May and June, a strong phenological shift is evident in the floral community, marked by a transient lull in weight gain (Fig. 3) and a near-total turnover of pollen composition appar- ent both in the raw data and in the wide discontinuity between May and June samples in PCoA ordination (Fig. 5). Following this shift, June pollen samples (Figs. 6, 7) were characterized by smaller trees and herbaceous flora associated with forest edge, mid-successional, or open landscapes, including hawthorn (Crataegus), honey locust (Gleditsia), sumac (Rhus), clover (Trifolium), and sweet clover (Melilotus), along with magnolia trees (Magnolia) that are common in our study system as ornamental plantings. False indigo Fig. 3. First-order difference of global GAM (Amorpha), a shrub that grows abundantly along smooths for 2017 and 2018. Positive and negative y- the banks of the Schuylkill River, was also abun- values reflect weight gain and loss, respectively, in the dant in some samples in 2017. integrated (non-differenced) data. In July (Figs. 6, 7), both weight gain and genus Table 2. AIC output of HGAM analysis. richness were attenuated (Figs. 3, 4), and pollen composition became dominated by clover (Tri- Model by year df AIC DAIC folium) and Virginia creeper (Parthenocissus), 2017 accompanied by a strong presence of sweetclover Model 2 (global + site) 89 4552 – (Melilotus) in 2017 and ornamental crepe myrtle Model 3 (site) 104 4569 17 (Lagerstroemia) in 2018. Mullein (Verbascum) and Model 1 (global) 14 5665 1113 magnolia (Magnolia) were also important pollen sources, particularly in 2017. 2018 Model 2 (global + site) 190 2045 – During the apparently sparse month of August Model 3 (site) 245 2098 53 (Figs. 3, 6, 7), the abundance of clover (Trifolium) Model 1 (global) 24 5992 3948 pollen declined but sweetclover (Melilotus) remained abundant. Spikenard (most likely the invasive understory tree, Aralia elata) emerged as PCoA revealed strong differentiation across sam- a major pollen source. Crepe myrtle (Lager- pling periods in overall pollen composition (Fig. 5). stroemia) continued to be prominent alongside The greatest discontinuities were seen across the another ornamental species, Japanese pagoda May–June and August–September intervals. Permu- tree (Styphnolobium japonicum is the only member tation testing indicated that site (F = 1.5, R2 = 0.09, of its genus present in our study area), which is P = 0.004), year (F = 5.3, R2 = 0.03, P = 0.001), and commonly planted as a street tree in our study sampling period (F = 16.9, R2 = 0.43, P = 0.001) system and has become naturalized in some were all significant predictors of pollen sample com- forested areas. The August–September interval position, but sampling period explained by far the saw another strong discontinuity in pollen com- most variance. position (Fig. 5). May samples (collected only in 2018) (Fig. 6) In September (Figs. 6, 7), clematis (most likely were dominated by wind-pollinated tree genera, sweet autumn clematis, Clematis ternifolia) and including willow (Salix), maple (Acer), oak (Quer- ivy (Hedera helix and/or H. hibernica) dominated cus), ash (Fraxinus), and plane tree (Platanus). pollen samples, accompanied in 2017 by a strong Apple (Malus) was also abundant. The presence of elm (the Chinese elm, Ulmus

❖ www.esajournals.org 8 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Fig. 4. Boxplot of genus richness by sampling period. Letters indicate significant differences according to Tukey’s test. parviflora, is the only member of its genus that goldenrod (Solidago), and clematis remained flowers in September in our study system). In important at some sites. 2017, crepe myrtle (Lagerstroemia) remained a major resource. Boneset (Eupatorium) was found Trait analysis across most sites in both years but was especially Trait-based patterns were generally consistent abundant in 2017. across the two years of our study (Fig. 8). With At the close of the foraging season in October respect to growth habit, we saw a predominance (sampled only in 2018) (Fig. 7), pollen samples of trees and shrubs early in the season, giving from all sites consisted mainly of ivy (H. helix way to herbs in mid-summer, then returning to and/or H. hibernica). Late-season asters were also trees and shrubs in August, and finally shifting common, though, including American asters to vines in September and October. In terms of (Symphyotrichum), snakeroot (Ageratina), Artemi- life cycle, the contribution of perennials appears sia (likely A. vulgaris, common mugwort), and to have dwarfed that of biennials and annuals

❖ www.esajournals.org 9 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Fig. 5. PCoA ordination of pollen data. Pollen samples (site 9 period) are represented by circles (2017) and crosses (2018). Sampling periods are indicated by color, and samples belonging to each period are joined in con- vex hulls. throughout the whole foraging season, with the with the caveat that colony weight is affected by caveat that some sampling periods had a large overall health and honey bee life history in addi- proportion of genera with undetermined life tion to resource availability. We mitigated these cycle due to trait non-uniformity among con- potentially confounding effects by omitting colo- geners. Native genera ostensibly comprised the nies with health anomalies, standardizing colony majority of pollen in May and June while exotic starting conditions, and analyzing detrended genera predominated in July, August, September, weight dynamics (i.e., differenced time series) and October. These patterns should be inter- rather than absolute weight. Under this interpre- preted cautiously, though, since the proportion tation, spring in our study system is a time of of undetermined genera was high for most sam- plenty, characterized by two pulses of resource pling periods, and in some cases, abundant gen- availability, the first beginning in mid-May and era that could represent both native and exotic the second in mid-June. Resource availability species (e.g., Aralia, Clematis, Phragmites) likely tapers off through July and reaches a minimum consisted mainly of their exotic constituents (e.g., in August, a period during which all colonies in A. elata, C. ternifolia, P. australis). our study lost weight. After a brief pulse of resource availability in early September, the DISCUSSION growing season is effectively over, with minimal resource availability from mid-September to the Continuous colony weight monitoring end of our measuring period in mid-October. revealed strong gain and loss motifs. We inter- These results—especially the occurrence of a pret these dynamics as a measure of the overall late-summer/early-fall dearth of floral resources abundance of floral resources in our study area, —are broadly consistent with temporal patterns

❖ www.esajournals.org 10 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Fig. 6. Genus composition of 2017 pollen samples. Genera are ranked by mean proportional abundance across sites. Only genera with a maximum within-site proportional abundance of at least 2.5% are shown. of floral resource availability documented in system comports with this pattern suggests that England (Couvillon et al. 2014), France (Requier urban landscapes, despite the extreme modifica- et al. 2015), Denmark (Lecocq et al. 2015), Ger- tion of their floral communities, conform to many (Danner et al. 2017), and the Midwestern archetypal patterns of floral phenology that are United States (Wood et al. 2018). The notion of a shaped more by climate than by the particulars “summer dearth” is also a ubiquitous anecdote of individual landscapes. among beekeepers in temperate ecoregions [e.g., One of the most salient patterns in our data is Japan: (Hiratsuka 1920), Ireland (Morony 1906), the importance of woody perennials, including the United States (Demuth 1918)]. That such sim- both trees/shrubs and woody vines like Hedera, ilar results should be found in such geographi- Parthenocissus, and Clematis. These findings con- cally disparate study systems might be explained tribute to a growing consensus about the impor- partly by the fact that some of the most importance of woody plants as pollinator resources tant honey bee foraging resources are cosmopoli- (Mach and Potter 2018, Donkersley 2019), per- tan synanthropes (e.g., Trifolium, Melilotus, haps especially in urban environments (Burgett Hedera), but the resource dynamics we observed 1978, Wattles 2017, Lau et al. 2019). In May, the seem to be characteristic of temperate floral com- most important resources included wind-polli- munities even across systems with little overlap nated trees like Salix, Acer, Quercus, Fraxinus, and in major flora. That our Philadelphia study Platanus, reinforcing Saunders’ (2018) call for

❖ www.esajournals.org 11 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Fig. 7. Genus composition of 2018 pollen samples. Genera are ranked by mean proportional abundance across sites. Only genera with a maximum within-site proportional abundance of at least 2.5% are shown. increased attention to be given to wind-polli- adventitious shrubs and woody vines like Aralia, nated plants as resources for pollinators. Later in Parthenocissus, Clematis, and Hedera, account for the season, ornamental trees and shrubs like the majority of pollen foraging. Only in July did Lagerstroemia and Styphnolobium, along with herbaceous plants consistently account for the

❖ www.esajournals.org 12 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Fig. 8. Trait composition by sampling period. Traits were scored as undetermined for a given genus when they were non-uniform across constituent species. majority of pollen foraging, and this was attribu- shift of pollinator visitation to edge and open table mostly to a single dominant genus (Tri- habitats. The late-summer discontinuity in our folium). These findings highlight the importance data probably reflects the effective (though not of accounting for trees and other woody plants absolute) end of Trifolium and Melilotus bloom in studies of plant–pollinator ecology, particu- (which may also account for the observed weight larly in urban areas where these resources seem loss of colonies in August) and the shift to the fall to be of prime importance, despite the technical flora dominated by Clematis and Hedera. challenges of canopy sampling that have An unexpected—and perhaps characteristi- prompted some authors to focus exclusively on cally urban—pattern in our data is the over- low vegetation (Matteson et al. 2013, Lowenstein whelming abundance of woody vines in summer et al. 2019). and fall pollen samples, beginning in July with The strongest discontinuities in the composi- Parthenocissus and peaking in September and tion of our pollen samples, particularly evident October with Clematis and Hedera. In our study in clustering patterns seen in PCoA ordination region, Parthenocissus is represented by both (Fig. 5), occurred between the May and June native P. quinquefolia and exotic P. tricuspidata sampling periods and between the August and (USDA, NRCS 2020). The only species of Clematis September sampling periods. As observed by abundant in our study system that flowers in the Robertson (Robertson 1895), the late-spring dis- fall is exotic C. ternifolia (D. Sponsler, personal continuity reflects the closure of the tree canopy, observation). Hedera is represented by H. helix and which marks the end of flowering for most tree H. hibernica, both exotic (USDA, NRCS 2020). and understory species, with a corresponding With the exception of the native P. quinquefolia,

❖ www.esajournals.org 13 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL. all these vines were introduced to North America the season, corresponding to the flowering of as ornamentals and have since become widely native trees and shrubs. In July and August, a sea- naturalized in urban and suburban areas, where son of overall dearth in floral resources (Fig. 3), they thrive ubiquitously in both seminatural and exotic plants predominate. This pattern could be developed areas. It is puzzling, in light of our interpreted not as a displacement of native floral findings, that neither Clematis nor Hedera is men- resources but a compensation for a period of natu- tioned in classic surveys of North American pol- ral scarcity. The ornamental trees Lagerstroemia linator-attractive flora, and Parthenocissus is and Styphnolobium, absent from most previous deemed at best a secondary nectar source of surveys of North American floral resources regional importance (Lovell 1918, Pellett 1920, (Lovell 1918, Pellett 1920, Ayers and Harman Lieux 1972, Ayers and Harman 1992). This raises 1992; though see Lau et al. 2019), deserve special the question of whether the dominance of these mention since they are not generally regarded as vines as summer and fall floral resources was invasive. Moreover, the evident function of these simply overlooked in the past—reflecting the ornamental trees as resources for flower-visiting general neglect of urban landscapes in 20th-cen- insects suggests a caveat to Quigley’s (2011) dis- tury literature (Pickett et al. 2016)—or whether it missal of designed greenspaces as “Potemkin gar- is the result of more recent invasive success. In dens” devoid of ecological function. While the either case, if these taxa dominate flower–insect annual ornamentals Quigley probably had in interactions in other cities to the extent observed mind appear, indeed, to be of negligible impor- here, it would be a striking case of urban biotic tance as pollen resources (Fig. 8), their woody homogenization (McKinney 2006) propagated counterparts (e.g., Lagerstoemia, Styphnolobium, from the level of community composition to that Ligustrum, Hydrangea, Photinia, Magnolia, Phel- of interaction network structure. Indeed, at least lodendron) can be major pollen resources. Notably, in the case of Hedera, its comparable importance while annuals were virtually absent from our pol- as a bee plant in our study system and in len samples, there is some evidence that annual Brighton, UK (Garbuzov and Ratnieks 2014), ornamentals can be significant late-season nectar together with its widespread distribution in sources (Erickson et al. 2019). As discussed above, North American cities and suburbs (Jones and however, the predominance of exotic flora— Reichard 2009), suggests that its role as a homog- specifically Clematis and Hedera—continues enizer of urban ecosystems may already be well through September and October, a period tradi- established. It should also be noted that Clematis tionally associated with a pulse of floral resources and Hedera dominate the composition of our pol- driven by native asters (e.g., Solidago, Symphy- len samples at a time of year traditionally associ- otrichum) (Lovell 1918, Pellett 1920, Ayers and ated with the flowering of native asters (e.g., Harman 1992). The eclipse of these flora by Clema- Ageratina, Cichorium, Eupatorium, Eutrochium, Sol- tis and Hedera can be interpreted as a consequence idago, Symphyotrichum) (Lovell 1918, Pellett 1920, of urban environmental filtering (Aronson 2016), Ayers and Harman 1992) and still dominated by with asters being suppressed by urban distur- native asters in rural areas (Sponsler et al. 2017, bance regimes (e.g., mowing) and exotic woody Wood et al. 2018). While detected in our samples, vines being favored both by the ample substrate these native flora were dwarfed in abundance by of the built environment and by the high propag- Clematis and Hedera. ule pressure stemming from human cultivation. The special importance of the exotic vines in The relationship between colony weight our data reflects a larger pattern in which exotic dynamics and pollen sample composition plants play a prominent role in the floral resource requires careful interpretation. First, any relation- community of our study system. With the caveat ship between our weight data and pollen data is that our genus trait analysis was hampered by partly obscured by the fact that our weight data trait non-uniformity in many genera, our data are continuous but our pollen data represent suggest a seasonal structure to the relative impor- only a several-day interval per sampling period. tance of native and exotic flora (Fig. 8), closely This is a limitation that could theoretically be mirroring the findings of Wood et al. (2018) in overcome by higher sampling frequency. A more Michigan, USA. Native flora predominate early in fundamental constraint on the joint interpretation

❖ www.esajournals.org 14 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL. of weight and pollen data is that honey bee colony Conclusions and future research weight is driven mainly by changes in stored Frankl et al. (2005) pose the question of honey (i.e., the product of nectar foraging), but whether the landscape-scale quantification of flo- pollen trap samples represent only pollen forag- ral resources is possible. While there may be ing. For taxa thought to be major sources of both multiple ways to answer this question in the pollen and nectar (e.g., Trifolium, Melilotus) (Ayers affirmative, including the approach presented by and Harman 1992), relative abundance in pollen its original authors (Frankl et al. 2005), we subsamples may approximate simultaneous contri- mit that our “honey bee foraging assay”—the bution to nectar foraging. Other taxa, such as the combination of continuous weight monitoring wind-pollinated trees that dominate our May and pollen metabarcoding—constitutes a land- samples, produce copious pollen but little or no scape-scale quantification of floral resources. We nectar (Ayers and Harman 1992). Conversely, taxa acknowledge that honey bee foraging is not an such as knotweed (Reynoutria) are prolific nectar unbiased sampling of floral resources, and we do sources but not thought to contribute significantly not interpret our results as an objective census of to pollen foraging (Ayers and Harman 1992). the floral community. Nevertheless, an objective Thus, we recommend the conservative approach census of floral resources would arguably be less of interpreting pollen samples as qualitative indi- relevant to questions of ecological function than cators of flora in bloom during contemporaneous the “biased” sampling of a generalist florivore. patterns of weight dynamics, with further inter- The most serious limitation of our approach is pretation based on ancillary data or expert knowl- the imperfect representation of the total insect edge of the study system. florivore community by honey bee proxy. Honey While the focus of our study is on the use of bees have the broadest described diet breadth of honey bees as a model generalist forager, our any single flower-visiting insect species, often data have practical relevance for beekeepers in visiting a hundred or more floral species in any study region. Colony weight dynamics can given region, but their foraging activity in any inform hive inspection regimes and allow bee- particular time and place tends to be focused on keepers to plan management interventions such relatively few species (Butz Huryn 1997), with a as “supering” (the addition of boxes to accom- preference for those occurring at high abundance modate colony growth), honey harvest, supple- (Percival 1947, Leonhardt and Bluthgen€ 2012). mental feeding, or colony splitting (Gary 1992). Studies that focus on the dietary overlap between This of particular importance to urban beekeep- honey bees and other bee species vary widely in ers who often contend with difficult access to their findings, with percent overlap ranging from their apiaries (Alton and Ratnieks 2016) and 19% (Leonhardt and Bluthgen€ 2012) to 97% must schedule their management judiciously. It (Paini and Roberts 2005) when comparing honey is also important to note times of dearth so that bees to individual non-Apis bee species and from inter-colony robbing can be mitigated, for 33% (de Menezes Pedro and De Camargo 1991) example, through the use of entrance reducers to 45% (Steffan-Dewenter and Tscharntke 2000) and the minimizing of open-hive manipulations when comparing honey bees to an aggregated (Gary 1992). The clear demarcation of spring, non-Apis bee community. We are unaware of any summer, and fall floral communities in our studies focused on comparing honey bee forag- data is also of apicultural significance. Tradi- ing to that of non-bee florivores, though such tionally, apiarists in our study region recognize data could be extracted from comprehensive spring and fall nectar flows, and honey is often plant–pollinator network studies (e.g., Chacoff marketed in terms of these seasonal varieties. et al. 2012). These uncertainties notwithstanding, Our data show a more nuanced, three-season we maintain that the honey bee’s dietary general- pattern of nectar flows. Knowing this pattern ism, long foraging range, and amenability to would potentially allow beekeepers to harvest a standardized sampling techniques warrant its third seasonal honey variety, a significant asset preliminary use as a sampler of landscape-scale to urban beekeepers who lack access to single- floral resources, and the results of the methods species cropping needed to market floral vari- we employ can provide an informative baseline etal honeys. against which to compare the floral resource use

❖ www.esajournals.org 15 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL. by other florivores. In principle, though, the for- longer whether cities can support such communi- aging assay approach to landscape-scale floral ties (Baldock 2015) but rather how they are struc- resource surveying could be implemented with tured by the heterogeneous habitats of which model organisms other than the honey bee, and cities are composed (Baldock 2019). Our work we urge the development of such methods. Bum- begins to shed light on this question, revealing the ble bees (Bombus spp.), in particular, are amen- emergence of weedy vines as dominant floral able to both pollen barcoding (Pornon 2016) and resources, the seasonal structuring of the relative colony weight monitoring (Goulson et al. 2002), importance of trait-based floral guilds (e.g., native and their choice of floral resources would be vs. exotic, woody vs. herbaceous), and the dra- expected to differ from that of honey bees, matic temporal dynamics of overall floral resource though with substantial overlap (Leonhardt and availability. While we urge caution in extrapolat- Bluthgen€ 2012). ing our findings beyond our Philadelphia study The use of continuous colony weight monitor- system, the general consistency of our findings ing to infer resource availability is a field in its with comparable datasets (Garbuzov and Rat- infancy. While it is self-evident that changes in a nieks 2014, Couvillon et al. 2014, Requier et al. colony’s weight reflect the net flux of material 2015, Danner et al. 2017, Wood et al. 2018, Lau between colony and environment, there is no et al. 2019) suggests that the patterns seen in our established proportionality between “true” data are shaped by ecological and social drivers resource availability (however one chooses to that are not unique to our locality. Future work define this) and a given change of weight in a sen- should focus on elucidating these ecological dri- tinel colony. With this limitation in mind, we are vers and formulating testable hypotheses regard- conservative in our interpretation of colony ing the relationships between landscape, floral weight dynamics, focusing on the direction of resources, and florivores. weight change (i.e., positive or negative) rather than its absolute magnitude. We encourage future ACKNOWLEDGMENTS work aimed at clarifying the relationship between fi the linked dynamics of resource availability and T. Jones, D. Brough, K. Oxman assisted and eld and colony weight. Furthermore, the biological inter- laboratory work. K. Frank, K. Wattles and members of the Philadelphia Beekeepers Guild and the Philadel- pretation of the resource fluctuations inferred phia Botanical Club provided informative conversation from colony weight patterns must vary with the regarding local floral resources. M. McIntyre, Weavers life history of the focal taxa in question. A peren- Way Coop, Congregation Rodeph Shalom, Hotel Sof- nial organism, like the honey bee colonies in our itel, Shane Confectionery, Greensgrow Farm, The study, cannot “escape in time” from periods of Philadelphia Insectarium, the Philadelphia Business scarcity. The same would be true of an annual and Technology Center, Mt. Moriah Cemetery, Share organism with a long active season, such as a Food Program, W. B. Saul high school, Paradiso Restau- bumble bee colony. In contrast, organisms with rant, and A. Pfeffer provided apiary locations for this relatively short active seasons—a category that study. This work was funded by a USDA-NIFA post- includes the vast majority of flower-visiting insect doctoral fellowship for D. Sponsler (grant 2017-07141), species—depend only on the availability of a Penn State Apes Valentes research award to D. Spon- sler, and USDA–NIFA– SCRI grant 2016–51181–235399 resources during select periods. An analogous to C. Grozinger. point can be made with respect to the spatial scale of an organism’s activity. A landscape that is, on LITERATURE CITED the whole, resource poor from the perspective of a foraging honey bee colony, could contain a small patch of rich resources that could suffice for the Alton, K., and F. Ratnieks. 2016. Roof top hives: Practi- cal beekeeping or publicity stunt? Bee World provisioning of insects with small foraging ranges 93:64–67. able to subsist within the resource patch. Aronson, M. F. J., et al. 2014. A global analysis of the Finally, it is not incidental that our methodology impacts of urbanization on bird and plant diversity was implemented in a complex urban landscape. reveals key anthropogenic drivers. Proceedings of With respect to the ecology and conservation of the Royal Society B: Biological Sciences plant–pollinator communities, the question is no 281:20133330.

❖ www.esajournals.org 16 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Aronson, M. F. J., et al. 2016. Hierarchical filters deter- de Menezes Pedro, S. R., and J. M. F. De Camargo. mine community assembly of urban species pools. 1991. Interactions on floral resources between the Ecology 97:2952–2963. Africanized honey bee Apis mellifera L and the Ayers, G. S., and J. R. Harman 1992. Bee forage of native bee community (Hymenoptera: Apoidea) in North America and the potential for planting for a natural “ cerrado” ecosystem in southeast Brazil. bees. Pages 437–493 in J. M. Graham, editor. The Apidologie 22:397–415. hive and the honey bee. Dadant & Sons, Hamilton, Demuth, B. M. 1918. Bees for the winter cluster. Glean- Illinois, USA. ings in Bee Culture 46:462–463. Baldock, K. C. R., et al. 2015. Where is the UK’s polli- Donkersley, P. 2019. Trees for bees. Agriculture, nator biodiversity? The importance of urban areas Ecosystems & Environment 271:79–83. for flower-visiting insects. Proceedings of the Royal Erickson, E., S. Adam, L. Russo, V. Wojcik, H. M. Society B: Biological Sciences 282:20142849. Patch, and C. M. Grozinger. 2019. More than meets Baldock, K. C. R., et al. 2019. A systems approach reveals the eye: the role of ornamental plants in supporting urban pollinator hotspots and conservation oppor- pollinators. Environmental Entomology 49:178- tunities. Nature Ecology & Evolution 3:363–373. 188. Berry, D., K. Ben Mahfoudh, M. Wagner, and A. Loy. Esling, P., F. Lejzerowicz, and J. Pawlowski. 2015. 2011. Barcoded primers used in multiplex ampli- Accurate multiplexing and filtering for high- con pyrosequencing bias amplification. Applied throughput amplicon-sequencing. Nucleic Acids and Environmental Microbiology 77:7846–7849. Research 43:2513–2524. Burgett, M., et al. 1978. Urban apiculture. Pages 187– Fenneman, F., and D. M. Johnson. 1946. Physiographic 219 in G. W. Frankie and C. S. Koehler, editors. Per- divisions of the conterminous. U.S. United States spectives in urban entomology. Academic Press, Geological Survey, Reston, Virginia, USA. Cambridge, Massachusetts, USA. Frankl, R., S. Wanning, and R. Braun. 2005. Quantita- Butz Huryn, V. M. 1997. Ecological impacts of intro- tive floral phenology at the landscape scale: Is a duced honey bees. Quarterly Review of Biology comparative spatio-temporal description of “flow- 72:275–297. ering landscapes” possible? Journal for Nature Chacoff, N. P., D. P. Vazquez, S. B. Lomascolo, E. L. Conservation 13:219–229. Stevani, J. Dorado, and B. Padron 2012. Evaluating Garbuzov, M., and F. L. W. Ratnieks. 2014. Ivy: an sampling completeness in a desert plant-pollinator underappreciated key resource to flower-visiting network: sampling a plant-pollinator network. insects in autumn. Insect Conservation and Diver- Journal of Animal Ecology 81:190–200. sity 7:91–102. Chen, S., et al. 2010. Validation of the ITS2 region as a Gary, N. E. 1992. Activities and behavior of honey novel DNA barcode for identifying medicinal plant bees. Pages 269–372 in J. M. Graham, editor. The species. PLOS ONE 5:e8613. hive and the honey bee. Dadant & Sons, Hamilton, Cheng, T., C. Xu, L. Lei, C. Li, Y. Zhang, and S. Zhou Illinois, USA. 2016. Barcoding the kingdom Plantae: new PCR Geslin, B., B. Gauzens, E. Thebault, and I. Dajoz. 2013. primers for ITS regions of plants with improved Plant pollinator networks along a gradient of universality and specificity. Molecular Ecology urbanisation. PLOS ONE 8:e63421. Resources 16:138–149. Goulson, D., W. Hughes, L. Derwent, and J. Stout. Chesapeake Bay Conservancy. 2017. Land cover data 2002. Colony growth of the bumblebee, Bombus ter- project. http://chesapeakeconservancy.org/conser restris, in improved and conventional agricultural vation-innovation-center/high-resolution-data/la and suburban habitats. Oecologia 130:267–273. nd-cover-data-project/ Hiratsuka, Y. 1920. Honey plants in Japan. American Clemants, S., and G. Moore 2003. Patterns of species Bee Journal 60:100. richness in eight northeastern United States cities. Jones, C. C., and S. Reichard. 2009. Current and poten- Urban Habitats 1:4–16. tial distributions of three non-native invasive Couvillon, M. J., R. Schurch,€ and F. L. W. Ratnieks. plants in the contiguous USA. Natural Areas Jour- 2014. Waggle dance distances as integrative indica- nal 29:332–343. tors of seasonal foraging challenges. PLOS ONE 9: Keller, A., N. Danner, G. Grimmer, M. Ankenbrand, K. e93495. von der Ohe, W. von der Ohe, S. Rost, S. Hartel,€ Danner, N., A. Keller, S. Hartel,€ and I. Steffan-Dewenter. and I. Steffan-Dewenter. 2015. Evaluating multi- 2017. Honey bee foraging ecology: season but not plexed next-generation sequencing as a method in landscape diversity shapes the amount and diver- palynology for mixed pollen samples. Plant Biol- sity of collected pollen. PLOS ONE 12:e0183716. ogy 17:558–566.

❖ www.esajournals.org 17 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Kozich, J. J., S. L. Westcott, N. T. Baxter, S. K. High- Paini, D. R., and J. D. Roberts. 2005. Commercial honey lander, and P. D. Schloss. 2013. Development of a bees (Apis mellifera) reduce the fecundity of an Aus- dual-index sequencing strategy and curation pipe- tralian native bee (Hylaeus alcyoneus). Biological line for analyzing amplicon sequence data on the Conservation 123:103–112. MiSeq Illumina sequencing platform. Applied and Pedersen,E.J.,D.L.Miller,G.L.Simpson,andN.Ross. Environmental Microbiology 79:5112–5120. 2019. Hierarchical generalized additive models in Kraaijeveld, K., L. A. Weger, M. Ventayol Garcıa, H. ecology: an introduction with mgcv. PeerJ 7:e6876. Buermans, J. Frank, P. S. Hiemstra, and J. T. Dun- Pellett, F. C. 1920. American honey plants; together nen. 2015. Efficient and sensitive identification and with those which are of special value to the bee- quantification of airborne pollen using next-gener- keeper as sources of pollen. Dadant & Sons, Hamil- ation DNA sequencing. Molecular Ecology ton, Illinois, USA. Resources 15:8–16. Percival, M. 1947. Pollen collection by Apis mellifera. Lau, P., V. Bryant, J. D. Ellis, Z. Y. Huang, J. Sullivan, D. R. New Phytologist 46:142–173. Schmehl, A. R. Cabrera, and J. Rangel. 2019. Seasonal Pickett, S. T. A., M. L. Cadenasso, D. L. Childers, M. J. variation of pollen collected by honey bees (Apis Mcdonnell, and W. Zhou. 2016. Evolution and mellifera) in developed areas across four regions in future of urban ecological science: ecology in, of, the United States. PLOS ONE 14:e0217294. and for the city. Ecosystem Health and Sustainabil- Lecocq, A., P. Kryger, F. Vejsnæs, and A. Bruun Jensen. ity 2:e01229. 2015. Weight watching and the effect of landscape Pornon, A., et al. 2016. Using metabarcoding to reveal on honeybee colony productivity: investigating the and quantify plant-pollinator interactions. Scien- value of colony weight monitoring for the beekeep- tific Reports 6:27282. ing industry. PLOS ONE 10:e0132473. Quigley, M. F. 2011. Potemkin gardens: biodiversity in Legendre, P., and E. D. Gallagher. 2001. Ecologically small designed landscapes. Pages 85–91 in J. meaningful transformations for ordination of spe- Niemela,€ editor. Urban ecology: patterns, pro- cies data. Oecologia 129:271–280. cesses, and applications. Oxford University Press, Leonhardt, S. D., and N. Bluthgen.€ 2012. The same, but Oxford, UK. different: pollen foraging in honeybee and bumble- R Core Team. 2019. R: a language and environment for bee colonies. Apidologie 43:449–464. statistical computing. R Foundation for Statistical Lieux, M. H. 1972. A melissopalynological study of 54 Computing, Vienna, Austria. Louisiana (U.S.A.) honeys. Review of Palaeobotany Requier, F., J.-F. Odoux, T. Tamic, N. Moreau, M. and Palynology 13:95–124. Henry, A. Decourtye, and V. Bretagnolle 2015. Lovell, J. H. 1918. The flower and the bee: plant life Honey bee diet in intensive farmland habitats and pollination. Scribner’s Sons, New York, New reveals an unexpectedly high flower richness and a York, USA. major role of weeds. Ecological Applications Lowenstein, D. M., K. C. Matteson, and E. S. Minor. 25:881–890. 2019. Evaluating the dependence of urban pollina- Richardson, R. T., H. R. Curtis, E. G. Matcham, C.-H. tors on ornamental, non-native, and “weedy” floral Lin, S. Suresh, D. B. Sponsler, L. E. Hearon, and R. resources. Urban Ecosystems 22:293–302. M. Johnson. 2019a. Quantitative multi-locus Mach, B. M., and D. A. Potter. 2018. Quantifying bee metabarcoding and waggle dance interpretation assemblages and attractiveness of flowering reveal honey bee spring foraging patterns in Mid- woody landscape plants for urban pollinator con- west agroecosystems. Molecular Ecology 28:686– servation. PLOS ONE 13:e0208428. 697. Matteson, K. C., J. B. Grace, and E. S. Minor. 2013. Richardson, R. T., D. B. Sponsler, H. McMinn-Sauder, Direct and indirect effects of land use on floral and R. M. Johnson. 2019b. MetaCurator: a hidden resources and flower-visiting insects across an Markov model-based toolkit for extracting and urban landscape. Oikos 122:682–694. curating sequences from taxonomically-informa- McKinney, M. L. 2006. Urbanization as a major cause tive genetic markers. Methods in Ecology and Evo- of biotic homogenization. Biological Conservation lution 11:181–186. 127:247–260. Richardson, R. T., C.-H. Lin, J. O. Quijia, N. S. Riusech, Morony, W. 1906. The Month’s Work. Irish Bee Journal K. Goodell, and R. M. Johnson. 2015. Rank-based 5:35. characterization of pollen assemblages collected Oksanen, J., et al. 2019. vegan: community ecology by honey bees using a multi-locus metabarcoding package. R package version 2.5-6. https://CRAN.R- approach. Applications in Plant. Sciences 3: project.org/package=vegan 1500043.

❖ www.esajournals.org 18 April 2020 ❖ Volume 11(4) ❖ Article e03102 METHODS, TOOLS, AND TECHNOLOGIES SPONSLER ET AL.

Robertson, C. 1895. The philosophy of flower seasons, bees and wild bees in central Europe. Oecologia and the phaenological relations of the ento- 122:288–296. mophilous flora and the anthophilous insect fauna. Taberlet, P., E. Coissac, F. Pompanon, L. Gielly, C. Miquel, American Naturalist 29:97–117. A. Valentini, T. Vermat, G. Corthier, C. Brochmann, Rognes, T., T. Flouri, B. Nichols, C. Quince, and F. and E. Willerslev. 2007. Power and limitations of the Mahe. 2016. VSEARCH: a versatile open source chloroplast trnL (UAA) intron for plant DNA bar- tool for metagenomics. PeerJ 4:e2584. coding. Nucleic Acids Research 35:e14. Sherrill-Mix, S. (2019). taxonomizr: Functions to Work USDA, NRCS. 2020. The PLANTS database. National with NCBI Accessions and Taxonomy. R package Plant Data Team, Greensboro, North Carolina, version 0.5.3. https://CRAN.R-project.org/packa USA. http://plants.usda.gov ge=taxonomizr Visscher, P. K., and T. D. Seeley. 1982. Foraging strat- Saunders, M. E. 2018. Insect pollinators collect pollen egy of honeybee colonies in a temperate deciduous from wind-pollinated plants: implications for polli- forest. Ecology 63:1790–1801. nation ecology and sustainable agriculture. Insect Wang, X.-C., C. Liu, L. Huang, J. Bengtsson-Palme, H. Conservation and Diversity 11:13–31. Chen, J.-H. Zhang, D. Cai, and J.-Q. Li. 2015. ITS1: Schnell, I. B., K. Bohmann, and M. T. P. Gilbert. a DNA barcode better than ITS2 in eukaryotes? 2015. Tag jumps illuminated–reducing sequence- Molecular Ecology Resources 15:573–586. to-sample misidentifications in metabarcoding Wardhaugh, C. W. 2015. How many species of arthro- studies. Molecular Ecology Resources 15:1289– pods visit flowers? Arthropod-Plant Interactions 1303. 9:547–565. Sickel, W., M. J. Ankenbrand, G. Grimmer, A. Wattles, K. 2017. Trees and shrubs for bees. https:// Holzschuh, S. Hartel,€ J. Lanzen, I. Steffan-Dewenter, phillybeekeepers.org/wp-content/uploads/2017/02/ and A. Keller. 2015. Increased efficiency in identify- BToS_3rd-ed_20170204_5pp.pdf ing mixed pollen samples by meta-barcoding Wickham, H., R. Francßois, L. Henry, and K. Muller,€ with a dual-indexing approach. BMC Ecology et al. 2019. dplyr: a grammar of data manipulation. 15:20. R package version 0.8.3. https://CRAN.R-project. Smart, M. D., R. S. Cornman, D. D. Iwanowicz, M. org/package=dplyr McDermott-Kubeczko, J. S. Pettis, M. S. Spivak, Wickham, H. 2017. Package tidyverse: easily install and C. R. V. Otto. 2017. A comparison of honey and load the “Tidyverse”. cran.microsoft.com. bee-collected pollen from working agricultural Wood, S. N. 2017. Generalized additive models: an lands using light microscopy and ITS metabarcod- introduction with R. Second edition. Chapman and ing. Environmental Entomology 46:38–49. Hall/CRC, Boca Raton, Florida, USA. Sponsler, D. B., E. G. Matcham, C.-H. Lin, J. L. Lanter- Wood, T. J., I. Kaplan, and Z. Szendrei. 2018. Wild bee man, and R. M. Johnson 2017. Spatial and taxo- pollen diets reveal patterns of seasonal foraging nomic patterns of honey bee foraging: a choice test resources for honey bees. Frontiers in Ecology and between urban and agricultural landscapes. Jour- Evolution 6:210. nal of Urban Ecology 3:juw008. Zhang, J., K. Kobert, T. Flouri, and A. Stamatakis. Steffan-Dewenter, I., and T. Tscharntke. 2000. Resource 2014. PEAR: a fast and accurate Illumina Paired- overlap and possible competition between honey End reAd mergeR. Bioinformatics 30:614–620.

SUPPORTING INFORMATION

Additional Supporting Information may be found online at: http://onlinelibrary.wiley.com/doi/10.1002/ecs2. 3102/full

❖ www.esajournals.org 19 April 2020 ❖ Volume 11(4) ❖ Article e03102