Bumble bee parasite strains vary in resistance to phytochemicals

Nectar and pollen contain diverse phytochemicals that can reduce disease in pollinators. However, prior studies showed variable effects of nectar chemicals on infection, which could reflect variable phytochemical resistance among parasite strains. Inter-strain variation in resistance could influence evolutionary interactions between plants, pollinators, and pollinator disease, but testing direct effects of phytochemicals on parasites requires elimination of variation between bees. Using cell cultures of the bumble bee parasite Crithidia bombi, we determined (1) growth-inhibiting effects of nine floral phytochemicals and (2) variation in phytochemical resistance among four parasite strains. C. bombi growth was unaffected by naturally occurring concentrations of the known antitrypanosomal phenolics gallic acid, caffeic acid, and chlorogenic acid. However, C. bombi growth was inhibited by anabasine, eugenol, and thymol. Strains varied >3-fold in phytochemical resistance, suggesting that selection for phytochemical resistance could drive parasite evolution. Inhibitory concentrations of thymol (4.53–22.2 ppm) were similar to concentrations in Thymus vulgaris nectar (mean 5.2 ppm). Exposure of C. bombi to naturally occurring levels of phytochemicals—either within bees or during parasite transmission via flowers—could influence infection in nature. Flowers that produce antiparasitic phytochemicals, including thymol, could potentially reduce infection in Bombus populations, thereby counteracting a possible contributor to pollinator decline.

effects of phytochemicals on pollinator infection 22,23,32,33 . Parasites infect new hosts via transmission at flowers 10 and within bee hives 32 , which contain derivatives of nectar, pollen, and other plant materials 21 . Crithidia bombi has not been detected in floral nectar 34 . However, within hosts, C. bombi inhabits the gut lumen, where cells have direct exposure to host-ingested nectar and pollen phytochemicals in the crop, and possibly also in the midand hindgut. In contrast to trypanosomes that infect the circulatory system or organs of their hosts, intestinal C. bombi lacks a physical barrier to shield it from ingested compounds, and may be exposed to phytochemical concentrations that approach those found in nectar and pollen. Hence, oral consumption of phytochemicals by bees could have strong and direct effects on parasites, and the phytochemical concentration that inhibits parasite growth in vitro may provide an estimate of the oral dose that could ameliorate infection in hosts.
Several studies have demonstrated that phytochemical ingestion by B. impatiens and B. terrestris reduces C. bombi infection. Five phytochemicals found in nectar-gelsemine 33 , nicotine 22,23 , anabasine, thymol, and catalpol 22 -reduced C. bombi infection intensities. However, both the magnitude and direction of effects of phytochemicals on C. bombi varied among studies. For example, other studies found that thymol 35 and anabasine 36 did not affect C. bombi infection, and nicotine increased infection intensity 36 . Taken together, these results suggest that phytochemicals have variable effects on C. bombi infection, with effects dependent on the unique combination of parasite strain, host genotype, and abiotic conditions used in each experiment. Therefore, an approach that eliminates host-related variability would help to determine the direct effects of phytochemicals on parasites, and allow comparisons of phytochemical sensitivity among parasite strains.
Both C. bombi strains and floral phytochemical concentrations are variable. Crithidia bombi populations are genetically 37 and phenotypically diverse 32 . Inter-strain variation could determine resistance to phytochemicalsdefined here as the ability to survive, grow, and reproduce when exposed phytochemicals-as has been demonstrated within populations of other pathogenic microbes, such as quinine-and artemisinin-resistant Plasmodium falciparum 38 . Like parasite strains, floral phytochemical concentrations are variable, and have dose-dependent effects on both pathogens and hosts 39 . For example, nectar nicotine and anabasine concentrations spanned multiple orders of magnitude among related Nicotiana species 40 . Within a species, nectar nicotine varied between Nicotiana attenuata plant populations, within populations, and across a six-fold range between flowers of a single inflorescence 41 . Similarly, nectar concentrations of Rhododendron ponticum grayanotoxins varied between native and invasive populations and within patches 42 . Testing a range of parasite strains, phytochemicals and concentrations in a single study could identify candidate medicinal compounds and illustrate the potential effects of phytochemicals on pollinator parasites in nature.
We used a standardized, high-throughput protocol to test the direct effects of different phytochemicals against multiple parasite strains across a range of chemical concentrations. Cell culture-based methods have been used to quantify the effects of phytochemicals on insect-vectored trypanosome species such as Leishmania donovani 24 , Trypanosoma cruzi 27,43 , and Trypanosoma brucei 24,44,45 that cause disease in humans and are close phylogenetic relatives of C. bombi 46 . Here, we extend a previously described C. bombi cell culturing method 47 to assess variation in the direct effects of nine floral phytochemicals-two alkaloids, one cyanogenic glycoside, a hydroxycinnamic acid, a hydroxycinnamic acid ester, a gallic acid, a phenylpropene, and two terpenoids-on four different C. bombi strains. We also searched published literature to compare phytochemical sensitivity of C. bombi to that of other trypanosome species, animal cells, and insects. To gauge the ecological relevance of each phytochemical's effects in culture, we combined field sampling of five plant species with literature searches to quantify phytochemical concentrations in nectar and pollen.

Results
Cell culture experiments. In comparison to other trypanosome species, C. bombi were remarkably resistant to common phytochemicals, with no growth inhibition at concentrations previously found to lower infection intensity in nectar fed to live bees (Table 1). Among the alkaloids, nicotine at doses of up to 1000 ppm had no effect on growth, and over 1000 ppm anabasine was required for 50% growth inhibition (EC50, Table 1, Fig. 1, Supplementary Figure S1). None of the tested strains were susceptible to the cyanogenic glycoside, amygdalin, nor to the antitrypanosomal phenolics caffeic acid, chlorogenic acid, and gallic acid, even at concentrations that were several orders of magnitude above the inhibitory thresholds of related pathogens ( Table 1). The sesquiterpene βcaryophyllene also did not inhibit growth of any strain at concentrations up to 50 ppb. Of the nine phytochemicals tested, only three-anabasine, eugenol, and thymol-were sufficiently inhibitory to estimate dose-response curves and EC50 values (Fig. 1, Table 1, Supplementary Figures S1-S3).
Eugenol resistance (Fig. 1B) was the most consistent across strains, with all EC50 values between 19.7 and 23.5 ppm, yet the non-overlapping 95% credible intervals (CI) still indicated statistically significant variation. The relative resistance ranks of the four strains were the same as for anabasine and eugenol: Strain VT1 (EC50 = 19.7 ppm, 95% CI: 18.9-20.4 ppm) was again the most sensitive, and strain 12. Resistance to thymol (Fig. 1C) was also variable. As was the case for the other two compounds, strain 12.6 (EC50 = 22.2 ppm, 95% CI: 22.3-21.0 ppm) was the most resistant, with more than three times the resistance of the other three strains, which were not significantly different from one another (VT1, EC50 = 6.26 ppm,  Table 2). In the three plant taxa for which we analyzed both pollen and nectar, concentrations of the chlorogenic acid 5-caffeoylquinic acid were 25-to 30-fold higher in pollen than in nectar (Wilcoxon W-test, M. domestica: W = 25, P < 0.001; V. corymbosum (cultivated): W = 18, P < 0.001; V. corymbosum (wild): W = 0, P < 0.001). Although nectar chlorogenic acid concentrations were lower than pollen concentrations, nectar concentrations were still several orders of magnitude higher than those recorded in honey, with the exception of Leptospermum scoparium honey (Table 2). Similarly, thymol concentrations in the nectar of Thymus vulgaris were over 10-fold above the highest value recorded for natural honey (

Discussion
Crithidia bombi was far less susceptible to the tested trihydroxybenzoic and hydroxycinnamic phenolic phytochemicals than were other, previously studied bloodstream trypanosomes. L. donovani and T. brucei, for example, were inhibited by < 10 ppm of gallic acid 26,48 , whereas concentrations up to 250 ppm had minimal effects on any tested strains of C. bombi. Similarly, caffeic acid, which inhibited L. donovani and T. brucei at < 10 ppm 24 , had no effect on C. bombi strains at concentrations up to 250 ppm. Furthermore, the EC50 for chlorogenic acid against C. bombi was > 2500 ppm, which was 100 times higher than the EC50 for L. donovani (EC50 7-17 ppm 49,50 ) and T. brucei (18.9 ppm 49 ). Although some variation in EC50 estimates could reflect methodological differences between our study and previous investigations, a difference of such magnitude for multiple phytochemicals provides strong evidence of comparatively high phytochemical resistance in C. bombi. This exceptional level of resistance may reflect the evolutionary history of C. bombi. In contrast to L. donovani and T. brucei, which are transmitted by blood-feeding insects and would be expected to have comparatively little direct exposure to phytochemicals, C. bombi may be adapted to chronic phytochemical exposure in the intestine of nectar-and pollen-consuming bumble bees. Bumble bees are generalist pollinators that consume nectar and pollen from a wide range of plant species 11 . Both nectar 51 and pollen 14 contain diverse compound mixtures, to which C. bombi in the gut lumen would be directly exposed 52 , particularly in the proximal parts of the gut, before phytochemicals are absorbed or  Continued metabolized by hosts or commensalists. Study of the mechanisms by which C. bombi withstands such high phytochemical concentrations could offer insight into the evolution of chemical resistance in medically important trypanosomes.
In addition to being less susceptible to phytochemicals than were other trypanosomes, C. bombi showed no growth inhibition at phytochemical concentrations exceeding those documented in honey ( Table 1, Table 2). For example, for the known antitrypanosomal compound caffeic acid, C. bombi was not inhibited by 250 ppm (Table 1), over 9 times the maximum honey value of 26.8 ppm ( Table 2, range 0.76-26.8 ppm for 14 honey types) 53 ; for gallic acid, C. bombi was again robust to 250 ppm (Table 1), or 3 times the maximum reported honey value of 82.5 ppm (Table 2; among 14 honey types, only oak honey exceeded 1 ppm gallic acid) 53 .
There are a number of nonexclusive explanations for the insensitivity of C. bombi to phytochemicals above their natural concentration range. First, the phytochemical concentrations found in honey samples may underestimate naturally occurring concentrations. Fanning of nectar to produce honey 11 , as well as prolonged storage, may evaporate volatile nectar components such as thymol, eugenol, and β -caryophyllene and could promote oxidation of phenolic compounds 54 . The thymol and chlorogenic acid concentrations measured in our field samples (Table 1), which were orders of magnitude higher than the values for honey found in the literature, illustrate this point. Second, in natural settings, phytochemicals are encountered in complex combinations, such that total phytochemical concentrations of biologically active compounds may far exceed the concentration of any one chemical component. Pollen comprises a mixture of phytochemicals, with the sum concentration of all phenolic constituents reaching 1.3-8.2% phenolics by weight (13,000-82,000 ppm) 55 . Even honey may contain up to 12,000 ppm total phenolics (range 1,600-12,000 ppm) 53 . Third, in their hosts, parasites are subject to additional antimicrobial chemicals produced by the host immune system and competing gut microbiota. Multiple antimicrobial peptides produced by bees have synergistic effects with one another 56 , and should be tested for synergy with floral phytochemicals as well. The Bombus gut microbiome includes species that produce ethanol and organic acids 57 , which also inhibit microbial growth 58,59 . Hence, the high resistance of C. bombi that we observed to single phytochemicals may be necessary to tolerate the effects of multiple phytochemicals, antimicrobial peptides, and microbiome-derived toxins acting in concert. Future experiments should explicitly address the interactive effects of multiple phytochemicals in combination. In addition to explaining why C. bombi has such high resistance to individual phytochemicals under optimal conditions, the interactive effects of multiple factors may explain why low concentrations of phytochemicals were sufficient to decrease parasitism in live bees 22 . All tested strains of C. bombi were resistant to phytochemicals at concentrations 100 times higher than those previously shown to be medicinal in B. impatiens and B. terrestris. Our strains were not inhibited by up to 1000 ppm nicotine, or 500 times the 2 ppm previously found to ameliorate infection in bees 22,23 . Our lowest EC50 value for anabasine (628 ppm) was still over 100-fold higher than the 5 ppm previously shown to reduce infection levels 22 . Inhibitory concentrations of thymol, where the minimum EC50 of the four strains was 4.5 ppm, were likewise more than 20-fold the 0.2 ppm medicinal concentration in B. impatiens 22 . These discrepancies far exceed the ~3-fold variation found among strains in our study, indicating that differences between in vitro and in vivo inhibitory concentrations do not merely reflect the use of different strains in our study versus previous live-bee experiments. We suggest that the low phytochemical concentrations necessary to ameliorate host infection may reflect phytochemical-induced changes in hosts, which could complement the direct effects of phytochemicals on parasites. For example, phytochemical ingestion may act indirectly on parasites by modulating the host immune response, as shown in humans 60 and in honey bees, where a honey constituent increased expression of genes that encode antimicrobial peptides 61 . Phytochemicals could also act as antioxidants that scavenge free radicals 62 and reduce the deleterious effects of pathogens 39 . Studies of live bees are needed to define how phytochemicals exert indirect effects on parasite infection via modulation of host immunity or behavior, such as induction of antimicrobial peptides or stimulation of intestinal motility that expels parasites from the gut 63 .
Our four C. bombi strains varied in resistance to the three phytochemicals that inhibited growth, spanning a five-fold range for thymol and a three-fold range for anabasine. Overall, strain "12.6" exhibited both the fastest growth ( Supplementary Figures S1-S4) and the highest phytochemical resistance (Fig. 1). Strains with a high rate of growth might be able to form biofilms that provide protection from growth-inhibiting chemicals, or metabolize the chemicals before deleterious effects are realized. Studies that use a greater number of strains are needed to test for positive correlations between phytochemical resistance and growth rate, both in cell cultures and in live bees, where C. bombi exists within a diverse microbial community 64 . Alternatively, negative correlations could reflect trade-offs between resistance and growth or infectivity. Variation in phytochemical resistance among parasites could be a target and possibly a result of natural selection. At the landscape scale, regional parasite and plant sampling, combined with cell culture experiments, could establish whether parasites show evidence of adaptation to phytochemicals characteristic of their local plant community. These correlative studies could be complemented by experiments that test how parasites respond to chronic phytochemical exposure, and whether resistance can evolve over time.
Our sampling data show that thymol inhibited C. bombi at concentrations found in T. vulgaris nectar. The range of EC50 values for C. bombi (4.5 to 22 ppm) spanned the natural range of thymol concentrations in T. vulgaris nectar (5.2-8.2 ppm). Although nectar concentrations did not completely inhibit growth, 50% growth inhibition could meaningfully decrease the intensity of infection and its negative effects on bees. Also, because it is likely that some thymol was lost during sample processing, our measurements may provide a conservative estimate of thymol-mediated inhibition by Thymus nectar. Thymol is used prophylactically to combat Varroa mite infestations 65 , and inhibited Nosema infection in A. mellifera 19 and Crithidia infection in B. impatiens 22 . Although it is possible that nectar thymol is absorbed or metabolized by bees or their gut commensalists, or diluted through combination with nectar of other species, phytochemicals are detectable in the lumen post-ingestion 52 , and even very low nectar concentrations (0.2 ppm) can reduce C. bombi infection intensity in B. impatiens 22 . Because individual bumble bees generally forage from only one or several floral species 66 , consumption of medicinally relevant amounts of thymol would seem plausible in the wild. Our study builds on prior results by reporting concentrations of thymol in floral nectar for the first time, and documenting the direct activity of this phytochemical against multiple parasite strains at naturally occurring concentrations.
Thymol and eugenol have been shown to possess broad-spectrum antimicrobial activity against bacteria 39 , fungi 67,68 , and trypanosomes 25 . These hydrophobic compounds readily penetrate and disrupt cell and mitochondrial membranes, thereby disrupting ionic gradients and causing leakage of reactive oxygen species 69 . Reactive oxygen species can oxidize monoterpenes and phenylpropenes like thymol and eugenol, which both contain double bonds and free hydroxyl groups. Oxidized phytochemicals can then initiate a free radical cascade that damages cell lipids and proteins 69 , leading to disruptions of organelle function and energy production in trypanosomes 25 . Rapidly dividing cells are especially susceptible, because they are easily penetrated during cell division 69  Phytochemicals such as thymol and eugenol, which display strong antimicrobial activity but are relatively benign to bees 70 , could have high medicinal value for both wild and managed bees that have access to plants containing these compounds. In general, bees are less susceptible than are microbes to toxic effects of essential oils 70 , and can be attracted to relevant antimicrobial concentrations 71 , which would increase the likelihood of voluntarily ingesting medicinally significant amounts of these phytochemicals under natural conditions. Eugenol, which has been found in over 400 plant species from 80 families 72 , has been shown to stimulate bee foraging and pollen collection in bumble bees 73 ; 50 ppm eugenol in sugar water was attractive to honey bees 74 , whereas only 19.7-23.5 ppm inhibited C. bombi growth in our study. Similarly, the A. mellifera 14-day LD50 for thymol exceeded 1000 ppm 70 , far higher than the 4.5-22.3 ppm thymol that inhibited our C. bombi. Future studies should test whether availability of flowers containing thymol (such as T. vulgaris) or eugenol is sufficient to reduce bee parasitism in the field; such plant species could be recommended to gardeners and as hedgerow species in agricultural areas. Additional studies that examine correlations between plot-and landscape-level plant species composition and pollinator parasite loads will yield additional ecological insights.
Our field sampling revealed higher levels of phytochemicals in nectar and pollen compared to previous reports of the same phytochemicals in honey. For example, the 5.2-8.2 ppm nectar thymol measured in this study is more than ten times greater than the highest reported concentration in natural honey ( Table 2). For chlorogenic acid, we identified three species with pollen concentrations > 400 ppm, which is 50 times the highest value previously reported for honey ( Table 2). Our findings highlight large differences between the phytochemical composition of nectar and honey, and indicate the need for more comprehensive sampling of nectar and pollen, including volatile compounds such as eugenol, to establish the types and concentrations of phytochemicals to which parasites might be naturally exposed. Sampling bumble bee honey in addition to honey bee honey may also reveal differences in chemical composition due to variation in foraging preferences or post-collection processes. Future sampling efforts will identify candidate antimicrobial phytochemicals for future testing in bees and other pollinators, and also document which floral species are sources of known antiparasitic compounds. Given the relatively unexplored nature of nectar and pollen relative to leaf phytochemistry, further sampling has significant potential to uncover new compounds of ecological and potentially medical significance.
Collectively, our experiments demonstrate the ecological and evolutionary relevance of direct effects of phytochemicals on a pollinator parasite. We show that the bumble bee parasite C. bombi is less susceptible to phytochemicals than are bloodstream trypanosomes, is inhibited by some nectar and pollen phytochemicals at naturally occurring concentrations, and exhibits inter-strain variation in resistance. Our results emphasize the importance of inter-strain variation and concentration-dependent responses in explaining the effects of phytochemicals on pollinator diseases, and highlight the need for additional analysis of nectar and pollen to profile the full range of phytochemicals and concentrations that occur in nature.  47 with the addition of 2% antibiotic cocktail to combat bacterial and fungal contaminants (penicillin 6 mg mL −1 , kanamycin 10 mg mL −1 , fluorcytosin 5 mg mL −1 , chloramphenicol 1 mg mL −1 as described 47 ). The remaining strains were isolated by flow cytometry-based single cell sorting of homogenized intestinal tracts (strain VT1) or bee feces (C1.1 and IL13.2) as described previously 47 . All strains were isolated directly from wild bees with the exception of VT1, which was first used to infect laboratory colonies of B. impatiens (provided by Biobest, Leamington, ON, Canada). The cell used to initiate the parasite culture was obtained from an infected worker of one of the commercial colonies. Cultures were microscopically screened to identify samples with strong Crithidia growth and absence of bacterial or fungal contaminants, then stored at − 80 °C in a 2:1 ratio of cell culture:50% glycerol until several weeks before the experiments began. Thereafter, strains were incubated at 27 °C and propagated weekly in 5 mL tissue culture flasks (300-500 μ L cultured cells in 5 mL fresh culture medium) 47 .
Phytochemical treatment media were prepared by dissolving stock chemicals either directly in medium followed by sterile filtration (for the more soluble nicotine, anabasine, amygdalin, chlorogenic acid, and eugenol) or by pre-dissolving compounds in ethanol (for the less soluble caffeic acid, gallic acid, β -caryophyllene, thymol). Treatment concentrations were chosen to span the range of concentrations known to occur in plant nectar and pollen (Table 1) and/or inhibit trypanosomes (Table 2), with maximal concentrations limited by compound solubility. For experiments using dilutions prepared from an ethanol-based stock, we equalized the ethanol Scientific RepoRts | 6:37087 | DOI: 10.1038/srep37087 concentration in each treatment by adding ethanol (up to 1% by volume, depending on the phytochemical) to the treatments of lesser concentrations. Experimental design. We conducted 9 experiments, each testing all 4 parasite strains in parallel against a single phytochemical. Cell cultures (1 mL) were transferred to fresh medium (5 mL) and allowed to grow for 48 h in tissue culture flasks. Immediately before the assay, cultures were transferred to 50 mL centrifuge tubes and centrifuged for 10 min at 10,000 g. The supernatant was removed and the cells were resuspended in 3 mL fresh medium. Cell density of the resulting suspension was calculated by counting parasite cells at 400x magnification using a Neubauer hemocytometer. Each strain was adjusted to a cell density of 1,000 cells μ L −1 .
A separate 96-well plate was prepared for each strain, i.e., 4 plates per experiment, one for each of the four strains. Each plate contained eight replicate wells at each of six phytochemical concentrations, with each concentration assigned to columns 3-10 of a given row to minimize edge effects. To each well, 100 μ L of 1,000 cells μ L −1 cell suspension was added to 100 μ L of the phytochemical-enriched treatment medium using a multichannel pipette, resulting in a starting cell density of 500 cells μ L −1 . The outer wells of the plate (columns 1, 2, 11, and 12, plus the remaining wells in rows A and B) were filled with 100 μ L treatment medium (8 wells per concentration) and 100 μ L control medium; these wells were used to control for changes in optical density (OD) unrelated to cell growth. Plates were incubated for 5 d at 27 °C on a microplate shaker (250 rpm, 3 mm orbit). OD readings (630 nm) were taken at 24 h intervals, as described previously 75 , immediately after resuspending the cells (40 s, 1000 rpm, 3 mm orbit) using the microplate shaker. We calculated net OD (i.e., the amount of OD resulting from parasite growth) by subtracting the average OD reading of cell-free control wells of the corresponding concentration, plate, and timepoint. For analysis of assays using the volatile phytochemicals eugenol and thymol, we excluded the replicates closest to the control wells that contained highest phytochemical concentrations (2 per treatment for eugenol, 3 per treatment for thymol). These replicates had markedly reduced growth compared to other samples in the same treatments; we attributed this growth reduction to exposure to phytochemicals that volatilized from the neighboring control wells.

Statistical analysis of cell culture experiments.
Dose-response curves for each strain and phytochemical were computed for the three phytochemicals for which the highest tested concentration resulted in complete inhibition of growth-near-complete inhibition is necessary for accurate estimation of the concentration that inhibits growth by 50% (EC50). All statistical analysis was carried out using the open source software R v3.2.1 76 following methods used for antimicrobial peptides 56 . For each sample, the growth integral (i.e., area under the curve of net OD vs. time) was calculated by fitting a model-free spline to the observed OD measurements using grofit 77 . The relationship between phytochemical concentration and growth integral was modeled with a Markov chain Monte Carlo algorithm using Just Another Gibbs Sampler 78 in combination with the R-package rjags 79 . We used the following model to describe the relationship between phytochemical concentration (c) and growth integral (g): where r denotes growth in the absence of the phytochemical, E max represents the maximum effect at high concentrations, and C 50 is the phytochemical concentration at which 50% of the maximum effect is reached. The parameter h, the Hill coefficient, indicates how steeply the effect increases around the concentration C 50 . From this model, we derived parameter estimates and 95% highest posterior density credible intervals (CI) of the EC50 for each phytochemical. We defined strains as having significant differences in resistance when their 95% CI's did not overlap. Each strain's dose-response curve and EC50 were calculated independently of the other strains; in other words, the EC50 represents the phytochemical concentration resulting in 50% of maximal inhibition for a particular strain.  Supplementary Table S1 for sampling locations, dates, and cultivars). We quantified thymol in Thymus vulgaris nectar and chlorogenic acids in Malus domestica (domestic apple), wild and cultivated Vaccinium corymbosum (blueberry), Prunus dulcis (almond), and Persea americana (avocado). Up to 10 samples of each tissue were collected, typically from each of three cultivars for agricultural species. For Thymus vulgaris cv. Silver, few plants were in flower at the time of collection, so it was only possible to collect enough nectar for a single nectar sample.
Pollen samples were collected using clean forceps by pinching off anthers, avoiding as much filament as possible. We collected at least 5 mg per sample, consisting of pollen, the pollen sac, and a small amount of filament. We collected from mature, undehisced or newly dehiscing anthers only. In most species, pollen was pooled across flowers within plants, but not across plants. Nectar samples were collected using separate glass microcentrifuge tubes. Care was taken to avoid contaminating samples with pollen. Depending on the plant species, we collected nectar through the corolla opening, or by removing and gently pressing the corolla to produce nectar at the flower base. Each nectar sample contained at least 5 μ L but typically 20 μ L nectar, added to 80 μ L EtOH to prevent spoilage. Nectar was often pooled across individual plants to obtain sufficient volumes per sample. Samples were kept on ice in the field and then stored at − 20 °C until lyophilization. Alcohol from Thymus nectar samples was evaporated at room temperature. We acknowledge that some thymol, which is volatile, may have been lost from the samples during evaporation, which we deemed necessary to prevent spoilage during shipping. As a result, our results may underestimate true nectar concentrations of this phytochemical. Analysis of chlorogenic acids. Pollen samples were extracted in methanol following previously published methods 80 . Unground pollen (5-50 mg) was sonicated for 10 min with 1 mL methanol in a 2 mL microcentrifuge tube, then incubated without shaking for an additional 24 h at room temperature. Samples were centrifuged for 5 min at 12,000 rpm, and the supernatants analyzed by liquid chromatography (LC) using High Resolution Electrospray Ionisation Mass Spectroscopy (HRESIMS). Chlorogenic acids were identified based on spectral comparisons with authentic standards in the library at Royal Botanic Gardens, Kew, UK. HRESIMS data were recorded using a Thermo LTQ-Orbitrap XL mass spectrometer coupled to a Thermo Accela LC system performing chromatographic separation of 5 μ l injections on a Phenomenex Luna C18(2) column (150 mm × 3.0 mm i.d., 3 μ m particle size) with a linear mobile phase gradient of 10-100% aqueous MeOH containing 0.1% formic acid over 20 min. The column temperature was maintained at 30 °C with a flow rate of 0.5 ml min −1 . Spectra were recorded in positive and negative modes at high resolution (30,000 FWHM (full width at half maximum)) and compared to authentic standards from the laboratory's compound library including the three chlorogenic acid isomers: 3-caffeoylquinic acid, 4-caffeoylquinic acid and 5-caffeoylqunic acid.
Lyophilized nectar (original volume ~10 μ L) was extracted in 50 μ L methanol and injected directly onto an LC-MS system with a ZQ LC-MS detector on a Phenomenex Luna C18 (2)  Statistical comparison between pollen and nectar. Within each of the three plant types for which we measured chlorogenic acids in both pollen and nectar-M. domestica, wild V. corymbosum, and cultivated V. corymbosumwe compared pollen and nectar 5-caffeoylquinic acid concentrations using an unpaired, two-sided Wilcoxon signed-rank test.
Analysis of thymol in Thymus vulgaris nectar. For analysis of thymol, dried nectar from a sample of known volume (~10 μ L) was extracted in 250 μ L of chloroform to which was added 500 ng of decyl acetate (50 μ L of a 10 ng μ L −1 solution) as an internal standard. The extract was injected directly onto an Agilent 6890 gas chromatograph coupled to an Agilent 5973 mass spectrometer with a DB-5 fused silica capillary column (30 m length, 0.25 mm diameter, 0.25 μ m film thickness) (Agilent). The column temperature was held at 50 °C for 2 min, then heated to 240 °C at 6 °C min −1 . The ion source was held at 150 °C, and the transfer line was held at 250 °C. Thymol was identified by comparison to a thymol standard (Sigma Ltd) and quantified using the fragment ion m/z = 135 relative to the Total Ion Chromatogram (TIC) for the decyl acetate internal standard. This ratio was corrected using a response factor, which was obtained by analyzing a standard sample containing equal concentrations of thymol and decyl acetate.