Active ingredients detected in bee collected pollen
All 188 pollen samples had at least 12 active ingredients detected in each sample, with a maximum of 31 AIs and an average of 22.0 ± 0.3 per sample. Over both years, 80 of the 259 screened pesticide active ingredients were detected in the pollen. These included 28 fungicides, 26 insecticides, 21 herbicides, two miticides, and one rodenticide. We also detected one synthetic antioxidant and one pesticide synergist (Table S1). We detected approximately twice as many AIs in pollen collected by honey bees (68 AIs) in 2019 than in pollen collected by bumble bees (32). All AIs detected in pollen from bumble bees were also collected by honey bees in either 2018 or 2019. Honey bee collected pollen also had significantly more AIs on average detected at each site (35.0 ± 0.9 S.E. AIs per site) compared to bumble bees (18.6 ± 0.6) in 2019 (R2m = 0.73; X2 = 68.2, df = 1, p < 0.001).
Farm management strategy (conventional, organic, or unsprayed) influenced the average number of pesticides detected in pollen samples collected from honey bees in 2018, but not in 2019 (Table 1, Fig. 1). For honey bee pollen in 2018, samples from organic farms had more individual pesticide AIs detected on average than that from conventional (Tukey’s HSD: p = 0.031) or unsprayed farms (Tukey’s HSD: p = 0.027) (Table 2, Fig. 1). In 2019, there was no significant difference in the average number of AIs found at conventional or unsprayed farms for either honey bees (p = 0.90) or bumble bees (p = 0.58) (Table 2, Fig. 1).
Average number of active ingredients (AIs) detected at each farm. Dark lines indicate the median, diamonds indicate the mean, boxes represent the upper and lower quartile, whiskers indicate the maximum and minimum number of AIs detected. Data are separated by which bee collected the pollen (HB honey bee, BB bumble bee) and in which year the data were collected. Upper case letters indicate significant differences within the 2018 data, and lower case letters indicate significant differences within the 2019 data. Graph created in R30 v3.6.2 with the package ggplot243.
Pesticide concentrations detected in bee collected pollen
Across both years and bee species, the average concentration of all detected pesticides in individual pollen samples was 477.9 ppb ± 57.1. Contrary to patterns observed in the number of AIs detected, we found significantly higher average concentrations of pesticides in pollen collected from bumble bees (1243.4 ppb ± 231.4) compared to honey bees in 2019 (577.7 ppb ± 95.8) (R2m = 0.023; X2 = 50.94, df = 1, p < 0.001).
Bees on conventionally managed farms had higher concentrations of pesticides in their pollen (Table 1). For honey bee-collected pollen in 2018, conventionally managed farms had significantly higher average concentrations of pesticides compared to unsprayed farms (Tukey’s HSD: p = 0.004), with no significant difference between organic and conventional farms (Tukey’s HSD: p = 0.60) and no significant difference between unsprayed and organic farms (Tukey’s HSD: p = 0.19) (Table 2). For both honey bee and bumble bee collected pollen in 2019, samples from conventional farms had significantly higher average concentrations of pesticides compared to unsprayed farms (Honey bees: p = 0.014. Bumble bees: p < 0.001) (Table 2).
Across all samples, fungicides were detected at significantly higher average concentrations (393.6 ppb ± 53.6) compared to insecticides (56.3 ppb ± 10.9; Tukey’s: p < 0.001) and herbicides (26.5 ppb ± 4.5; Tukey’s: p < 0.001) (R2m = 0.402; X2 = 456.5, df = 2, p < 0.001) (Fig. 2). There was no significant difference between average concentration of herbicides and insecticides (Tukey’s: p = 1.00) (Fig. 2).
Concentrations of pesticides detected in bee-collected pollen from colonies. Each data point represents the concentration of an active ingredient found in an individual sample. Pesticides were detected in pollen collected from honey bees in 2018 (grey), pollen collected from honey bees in 2019 (yellow), and pollen collected from bumble bees in 2019 (blue). Dark lines indicate the median, diamonds indicate the mean, boxes represent the upper and lower quartile, whiskers indicate the maximum and minimum concentration detected (besides outliers), and the dots represent outliers. Letters indicate significant differences between the pesticide types and this pattern was consistent when all samples were combined for analyses, or when samples were separated out by bee and year for analyses. Graph created in R30 v3.6.2 with the package ggplot243.
Fungicides also represented the highest individual detections in the pollen samples (Table 1). Of the top ten highest detections across all samples, nine of them were fungicides (range 1365.9–5753.6 ppb). The other was an insecticide (methoxyfenozide, 1406.2 ppb) (Table 1). The highest detection of an herbicide (diuron) was 34th in overall detection ranking at 475.6 ppb (Table 1).
Comparison of pesticides in pollen between years
We found that patterns of pesticide detection in honey bee-collected pollen were highly variable between farms and across years, with no consistent trends detected at the seven longitudinal farms. Sixty-six total AIs were detected in honey bee collected pollen at the longitudinal farms over the 2 years of sampling. Of those, forty-two active ingredients were detected in both years, while ten AIs were only detected in 2018 and 14 AIs were only detected in 2019. Additionally, the average concentration of pesticide residues detected in honey bee pollen at a given farm was not significantly correlated between years (Pearson’s r = 0.45, t = 1.12, df = 5, p = 0.31), indicating that exposure is not consistent year to year at individual farms. While average pesticide concentrations were higher at the longitudinal farms in 2019 (428.1 ppb ± 91.7) compared to 2018 (203.3 ppb ± 28.9), this was largely driven by one farm (Fig. S2, Farm 4). Farm was significant in the linear model (R2 = 0.56; F7,83 = 10.39, p < 0.001), while Year was not (F1,83 = 3.54, p = 0.06). Though the interaction of Year and Farm was significant in the model (Year*Farm: R2 = 0.56; F5,83 = 5.31, p < 0.001), making it hard to interpret the role of individual factors.
Source of pesticides in pollen
Based on spray records and management guides, of the 80 AIs detected in bee-collected pollen, 12 AIs (15% of AIs detected) are commonly applied to blueberry fields during bloom (Table S1), and even fewer were actually sprayed on focal fields; nine AIs were applied to focal fields in 2018 and seven in 2019. For honey bees, the majority of AIs detected are not registered for use on blueberry at any time. This was the case in both 2018 and 2019, and for all field management types, with AIs not registered for use in blueberry averaging between 54.1 and 57.5% of the AIs detected at each farm (Fig. 3A). Far fewer of the AIs were sprayed on the focal farms, 8.4% of AIs detected on conventional farms in 2018 and 8.9% in 2019. For bumble bees, AIs not registered for use on blueberries accounted for less of the AIs detected, averaging 36.8% of the AIs detected at conventional farms and 37.8% at unsprayed farms (Fig. 3A), with the majority of AIs either being sprayed on the focal farm or likely other blueberry farms during bloom (conventional—51.5% of AIs, unsprayed—51.0%). However, only 16.7% of AIs detected in bumble bee pollen collected from colonies on conventional farms were those sprayed at the focal farm.
Average percent contribution of pesticides to the (A) number of active ingredients (AIs) detected at a site, and (B) overall sample concentration. Contributions were determined by spray records and registration status. Active ingredients are separated into those that were either applied on the focal field during bloom (black), registered for use on blueberries during bloom but not sprayed in the focal fields (dark grey), registered for use on blueberries outside bloom (light grey) or not registered for use on blueberries at any time (white). Graph created in GraphPad Prism 931.
Eight active ingredients were detected in over 90% of all samples—atrazine, azoxystrobin, boscalid, chlorpyrifos, fluopyram, imidacloprid, metolachlor, and pyraclostrobin (Table S1). Only three of these were applied during bloom on the focal farms in 2018 and 2019: the fungicides azoxystrobin, boscalid and pyraclostrobin. In 2019, fluopyram was also applied. Although not reported in the application history of focal farms, metolachlor is commonly applied in conventionally managed blueberry fields for weed control. Atrazine and chlorpyrifos were not applied at any time in blueberry fields, and imidacloprid was only applied on some non-focal farms after bloom. The frequency of AIs detected in sampled varied somewhat by bee species and focal farm management (Table 1), though fungicides applied for management of blueberry pathogens during bloom were common across all sample types. Conversely, for insecticides, AIs not registered for use on blueberry farms during bloom were more common, including carbaryl, chlorpyrifos, clothianidin, and imidacloprid. Imidacloprid is registered for use on blueberry after bloom, carbaryl is registered for use outside bloom though is rarely used, and the others are not registered for use at any time on blueberry (Table S1). Methoxyfenozide, which is used for blueberry pest control during bloom was also common (Table 1). For herbicides, atrazine and metolachlor were found in every sample (Table 1), and while metolachlor is commonly used on blueberry farms during bloom, atrazine is not registered for use at any time (Table S1).
The overall highest concentrations of pesticide residues were associated with blueberry pest management during bloom. Six pesticides had detections with concentrations that were above the upper limit of linearity: boscalid (10 detections above ULOL), pyraclostrobin (5 detections above ULOL), pyrimethanil (5 detections above ULOL), azoxystrobin (3 detections above ULOL), carbendazim (3 detections above ULOL), and methoxyfenozide (1 detection above ULOL) (see Table S2 for ULOL concentrations). Of these active ingredients, all were applied in conventional blueberry fields used in this study during bloom (though not on all farms), except carbendazim, which is not used in blueberry pest management (Table S1).
The contribution of AIs from blueberry pest management to the overall sample concentration varied based on focal farm management and bee species but followed somewhat similar trends as the contribution to the number of AIs (Fig. 3B). For honey bees at unsprayed fields, the majority of the overall sample pesticide concentration came from AIs not registered for use on blueberry (conventional/organic) at any time of the year, with much less contribution from blueberry AIs applied during bloom or post-bloom on conventional fields (Fig. 3B). In contrast, AIs sprayed on conventional blueberry farms during bloom contributed the most to the pesticide concentrations for bumble bees on unsprayed farms, with exposure likely happening at neighboring conventional blueberry farms. Similar exposure occurred for honey bees at organic farms.
For honey bees at conventional farms, the contribution of AIs to the sample concentration was split between AIs that were sprayed on the focal farm, those that were likely sprayed on neighboring conventional farms, and AIs that are not registered for use on blueberry. Much lower contribution came from AIs used on blueberry farms post-bloom. In contrast, for bumble bees the majority of the pesticide concentration in samples came from AIs sprayed on the focal farm, with the rest primarily being AIs sprayed on neighboring blueberry farms during bloom. Much less contribution came from AIs not registered for use on blueberry or those sprayed on blueberry post-bloom (Fig. 3B).
For pollen collected by honey bees in 2018 and bumble bees in 2019, the majority of high detections came from AIs that were either sprayed on the focal farm during bloom or likely from neighboring blueberry farm management during bloom (Table 1). However, for honey bees in 2019, the highest detections came from an AI not registered for use in blueberry pest management at any time of the year. Carbendazim was the highest detection at both conventional (5753.6 ppb) and unsprayed (2333.2 ppb) farms in 2019 (Table 1). Carbendazim and thiophanate-methyl, of which carbendazim is a metabolite, are AIs of fungicides not registered for use in blueberries so it is assumed these residues came from farms growing other crops. Conversely, for bumble bees at the same farms, the highest detections (top 20 highest detection concentrations for bumble bees) were all AIs of products registered for use in blueberries during bloom, including the fungicide boscalid (highest: 1757.6 ppb) and the insecticide methoxyfenozide (1406.2 ppb) (Table 1).
Blueberry pollen collection
On average, pollen trapped from honey bee colonies in 2019 had 1.8% ± 3.2 blueberry pollen, while pollen collected from bumble bees had 25.9% ± 3.2 blueberry pollen. Blueberry pollen was collected from honey bee pollen traps at eight out of the 14 farms in 2019. At these eight farms, blueberry pollen made up between 0.04 and 16.7% of the total pollen collected. Pollen collected by bumble bees included blueberry pollen in all 15 farms in 2019, ranging from 3.9% of total pollen gathered to 45.6%.
Although the amount of blueberry pollen explained only 6% of the variation in concentration of pesticides used during blueberry bloom, there was a positive correlation between pesticide concentration and the amount of blueberry pollen collected from honey bee pollen traps (R2 = 0.06; F1,95 = 5.54, p = 0.02). This relationship was not significant for bumble bees (R2 = 0.11; F1,14 = 1.53, p = 0.24).
Effects of landscape composition on pollen collection and pesticide exposure
Across all sampled fields in 2019, the average (± SE) percent blueberry fields in the surrounding landscape was 22% ± 5 at the 500 m scale, 12% ± 3 at 1000 m, and 8% ± 2 at 2000 m. The average amount of blueberry pollen collected by honey bees increased with the percent of blueberry in the landscape at 500 m (R2 = 0.37, F1,12 = 7.18, p = 0.02) and at 1000 m scale (R2 = 0.59, F1,12 = 16.44, p < 0.01), but not at 2000 m. No significant relationship was found between the amount of blueberry pollen collected by bumble bees and percent blueberry at the three tested spatial scales (p > 0.05) (Fig. S3). There was a significant positive correlation between percent blueberry in the landscape and pesticide concentration in the pollen collected by bumble bees (500 m: R2 = 0.49; F(1,14) = 12.70, p = 0.003; 1000 m: R2 = 0.34; F(1,14) = 6.79, p = 0.022) (Fig. 4). Pesticide concentration detected in honey bee collected pollen was only significantly correlated with percent blueberry area at the 1000 m scale (R2 = 0.34, F1, 13 = 6.13, p = 0.029) and not at the 500 m or 2000 m scales (Fig. 4; Fig. S3). There was also no significant relationship between pesticide concentration in pollen collected by both bee species and percent blueberry located at 2000 m scale, as well as percent other agriculture in the landscape across all scales (p > 0.05).
Relationships between percent of the landscape in blueberry production within 1000 m of bee colonies and the concentration of pesticides detected in pollen collected by bumble bees (red) and honey bees (blue). Lines indicate smoothed linear regression lines with 95% confidence intervals. Graph created in R30 v3.6.2 with the package ggplot243.
There was a significantly larger average percent of blueberry located within the 500 m scale landscapes surrounding conventional farms (N = 10, 31.4% ± 5.6) compared to unsprayed farms (N = 5, 7.7% ± 3.5) in 2019 (R2 = 0.38; F1,14 = 7.83, p = 0.015), but no significant difference was found at the larger scales (1000 m and 2000 m; p > 0.05). There was also no significant difference in percent other agricultural lands between the two farm management types at any spatial scale (p > 0.05).
Source: Ecology - nature.com
