https://orcid.org/0000-0002-6076-8623
Figures
Abstract
Analysis of trapped honey bee pollen for pesticide residues is the most widely used method of monitoring the amount of pesticide entering colonies and its change over time. In this study, we collected and analyzed pollen from 70 sites across four bee-pollinated crops over two years to characterize the variation in pesticide detection across sites, crops and at different periods during bloom. Hazard Quotient, HQ, is the most common way that pesticide residues are aggregated into a single pesticide hazard value in the current literature. Therefore, change in pesticide hazard (HQ) was quantified in composite pollen samples collected from pollen traps and in pollen color subsamples separated into pollen from the target crop being pollinated and pollen from other plant species. We used our estimates of the variation in HQ to calculate the number of sample location sites needed to detect a 5% annual change in HQ across all crops or within specific crops over a 5-year period. The number of sites required to be sampled varied by crop and year and ranged between 139 and 7194 sites, costing an estimated $129,548 and $3.35 million, respectively. The HQ values detectable for this cost would be 575 and 154. We identified additional factors that complicate the interpretation of the results as a way to evaluate changes in pest management practices at a state level. First, in all but one crop (meadowfoam), the pollen collected from outside the crop honey bee colonies were pollinating comprised a major percentage of the total pollen catch. Moreover, we found that when the overall quantity of pollen from different pollen sources was taken into account, differences in HQ among crops widened. We also found that while HQ estimates remain consistent across the bloom period for some crops, such as cherry, we observed large differences in other crops, notably meadowfoam. Overall, our results suggest the current practice of interpreting pesticides levels in pollen may come with limitations for agencies charged with improving pesticide stewardship due to the high variation associated with HQ values over time and across crops. Despite the limitations of HQ for detecting change in pesticide hazard, there remains a potential for HQ to provide feedback to regulators and scientists on field-realistic pesticide hazard within a landscape.
Citation: Carlson EA, Melathopoulos A, Sagili R (2024) The power to (detect) change: Can honey bee collected pollen be used to monitor pesticide residues in the landscape? PLoS ONE 19(9): e0309236. https://doi.org/10.1371/journal.pone.0309236
Editor: Wolfgang Blenau, University of Leipzig Faculty of Life Sciences: Universitat Leipzig Fakultat fur Lebenswissenschaften, GERMANY
Received: December 19, 2023; Accepted: August 5, 2024; Published: September 26, 2024
Copyright: © 2024 Carlson et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its supporting files.
Funding: This work was supported by the National Science Foundation (https://www.nsf.gov/) Graduate Research Fellowship (EAC) and the Foundation for Food and Agricultural Research (https://foundationfar.org/), grant no. 549024 (APM and RS) and the Natural Resource Conservation Service(https://www.nrcs.usda.gov/), grant no. NR203A750008G004 (APM and RS). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Reducing the exposure of bees to pesticides is a challenge for regulators and land managers. While pesticides are necessary for protecting crops and have documented benefits in controlling pest populations and increasing food security [1–4], they are also linked to the decline of or a negative impact on pollinating insects [5–11]. In the United States, the risk assessment framework used by the Environmental Protection Agency (EPA) is the first line of defense for reducing pesticide impacts on honey bees [12,13], and specify which pesticide use practices have limited risk to pollinating honey bees during the registration or periodic review of those products [12,13]. These practices are then translated onto instructions on pesticide labels that a pesticide applicator must follow. Although the framework aims to characterize risk under field conditions, there remains a need for post-registration monitoring. This need is urgent because many older pesticides have yet to be evaluated using the EPA’s risk assessment, applicators may be using products in ways not specified by the label, and patterns of pesticide spray-drift may move beyond what is evaluated during registration.
In 2014, the EPA expanded pollinator protections to include both a traditional, regulatory approach [12] and individualized state-level voluntary pollinator protection plans [14,15]. Managed Pollinator Protection Plans (MP3s) are used by 31 states [16] to identify, test, and incentivize best management practices (BMPs) to reduce pesticide exposure to honey bees contracted to pollinate crops. An audit of the MP3 program by EPA noted that there are no robust methods to evaluate the effectiveness of the implementation of BMPs outside of surveys of pesticide applicator intentions [17]. States have looked to pesticide monitoring of bee matrices (pollen, honey, wax, etc) as a way to evaluate their efforts to lower pesticide exposure of honey bee colonies over time. The Association of American Pesticide Control Officials, the national body of state agencies responsible for the implementation of MP3s, report that about one third of states use pesticide monitoring in bee matrices (pollen, honey, wax, etc) as part of their state-level plan [18].
Pesticides hazard to honey bees involved in crop pollination is most frequently estimated from pollen trapped from the corbicula of returning foragers [19]. The detection of pesticides in pollen has long been a tool for researchers to estimate the hazard represented to honey bees in agricultural systems [20–27]. Typically, the aggregate hazard of a colony to a pesticide is estimated by dividing the concentration of each pesticide detected in pollen and by its acute oral toxicity to bees, which is summed into a single, unitless value: Hazard Quotient [28,29] (HQ). While HQ, in various iterations, is the most commonly used metric In literature to condense pesticide residues into a single numerical value [28], there are several reasons to doubt if HQ can adequately represent the hazard to bees from pesticides, including its misalignment with the EPA’s own pesticide risk assessment process [29] and the thresholds of concern associated with HQ values with an unclear connection to hive health metrics or survival [30–32]. Moreover, research into the HQ of chemical residues in trapped pollen from honey bees has suggested other limitations associated with relying on this technique for post-registration monitoring of pesticides.
First, honey bees are generalist pollinators which can forage over large scales [33,34] with preferences for the flowers of different plant species [35]. The heterogeneous nature of agricultural landscapes and mix of floral attractiveness within it leads to a situation where pesticide exposure is not always driven by the applications to the crop that honey bees are contracted to pollinate [25,26,36,37]. The fact that not all pollen originates from the pollinated crop is of significance as both pesticide labels and MP3s focus on pesticide stewardship for specific crops. Second, the mass of pollen collected across cropping systems can be very different, in some cases a 43-fold difference [38]. The difference in pesticide residue concentration within the pollen of each floral source [20] can impact the total pesticide exposure of the colony in ways HQ calculations were not designed for [28]. If, for instance, there are two pollen types which present different quantities of pollen and different HQ values to a hive, there currently is not a way to assess the relative risk of these two types of contamination to the health outcomes of the colony [29]. Consequently, while attempts to determine the HQ associated with a specific crop species by sorting the collected pollen to species may provide specific information for pesticide use on the crop [20,24,36], it may over or underestimate the total exposure a colony is experiencing.
In this study, we estimate the variation in pesticide HQ from pollen trapped from crops pollinated by honey bees in the state of Oregon. We used these estimates to determine the power of a post-registration sampling program to detect moderate changes in HQ (5%) over a five-year period, which is reflective of changes of ±1% in toxic load throughout the Pacific Northwest over the past 15 years [39]. We also estimated the costs required to provide sampling with sufficient power to detect these changes. In doing so, we provide insight into using this method as a post-registration monitoring tool and to state-level programs that are using pesticide detection in pollen to evaluate their progress on their MP3 goals. Moreover, by measuring the weight of pollen entering colonies and comparing it to whether it originated from the crop being pollinated or from other plant species, we were able to evaluate the extent to which such sampling could be used to make inferences back to pesticide stewardship on specific crops. Since most sampling programs only sample for a single period during bloom, we also evaluated variation in HQ across three periods of bloom to determine the adequacy of single points of sampling in estimating pesticide hazard over the course of a pollination. Through these efforts we provide guidance to states embarking on programs of post-registration of monitoring of pesticide in pollen collected by honey bees.
Materials and methods
Location and pollen trapping
This study was conducted over two years (n = 30 sites in 2020, n = 40 sites in 2021) on four pollinator dependent crops in the region (meadowfoam, white clover, sweet cherry and hybrid carrot seed). Commercial meadowfoam (Limnanthes alba Hartw. ex Benth., n = 14) and white clover seed (Trifolium repens L., n = 6) are produced in the Willamette Valley region of Oregon, USA (Fig 1). The Willamette Valley has productive soils and a temperate climate with a mean annual rainfall of 96–152 cm and was historically an oak savannah and tall grassland habitat with extensive wetlands; today it is characterized by mixed agricultural production, primarily seed crops, nurseries, forests, livestock pasture, and berries [40,41]. Sweet cherry (Prunus avium L., n = 12) is grown in the Willamette Valley and in the Columbia Plateau region; the Columbia Plateau is characterized by arid sage-brush steppe and grasslands [40,41]. The area is now agriculturally dominated by sweet cherry, pear, and apple production. Carrot seed production (Daucus carota L., n = 9) occurs in the Blue Mountains ecoregion, which was historically desert-like shrubland and grasslands [40,41]; today it is characterized by vegetable seed production, alfalfa, potatoes, and sugar beets. Across all sites, fields averaged 18.9 ha in size (max = 107 ha, min = 5 ha) and were located at least 2 km apart when possible; during bloom, fifteen sites were co-located within 2km of each other (cherry, n = 10; carrot, n = 0; meadowfoam, n = 3; clover, n = 2). Therefore, there is potential for overlap of the foraging radius of honey bee hives. None of the growers used organic farming techniques, but instead relied on standard agrochemicals and agronomic practices. To perform the field work no permits were required as the colonies were privately held, and we secured permission from each commercial beekeeper to enter the site and collect pollen from their hives.
Pollen trapping was conducted as in Topitzhofer et al., 2021; at each experimental site front porch pollen traps were installed on four honey bee hives on a pallet with strong foraging activity, and all alternative entrances to the hive were closed with tape and foam. After an acclimation period, pollen traps were engaged for at least 24 hr but no more than 48 hr during favorable weather (i.e. periods of time without rain or temperatures below 12.8°C). The total pollen samples were then collected and combined from all four hives and transported in coolers with ice and stored in a -20°C freezer until analysis. For each site, pollen was collected between two and five times throughout the bloom period of the crop (n = 78 pollen trapping events in 2020, n = 132 in 2021). If traps failed due to a leaky pollen trap (i.e. bees could evade the pollen trap), then the trapping event was discarded as a “leak” rather than a true zero where no pollen was collected.
Pollen acetolysis and identification
The total pollen samples were separated into color groups using the Pantone Color Guide from a 10 g subsample of the whole [19,20]. Each color group was then acetolyzed with a modified protocol for 0.25 g samples [19,42] which removes the lipid coat and allows identification of the pollen protein exine [43,44]. Pollen grains were then identified as focal crop or non-focal crop pollen with light microscopy using DiscoverLife keys (http://www.discoverlife.org) and then confirmed with PalDat Palynological Database (https://www.paldat.org) and Cornell Pollen Grains Reference Library (https://blogs.cornell.edu/pollengrains). Identification was performed to genus and identified as pollen from the focal crop or pollen that originated from outside the focal crop.
Pesticide residue analysis
Each pollen sample, both the whole sample (composite) and the segregated color groups (color subsamples identified as crop or non-crop), were weighed and placed in 50 mL conical vials with corbicula trapped pollen (1 ≤ 3 g). Pesticide residue analysis was performed by Synergistic Pesticide Laboratory in Portland, Oregon. Composite and color subsamples were shipped overnight in cold storage.
Identity and concentration of pesticide residues were determined through QuEChERS protocol [45] using both LC/MS-MS and GC/MS methods for pollen analysis [10]. Pesticide residues were identified as insecticides, fungicides, or herbicides. Full methodology for pesticide residue analysis is available in (S1 File, pesticide residue testing methodology; S5 File, reporting limits and batch-specific levels of detection for the 292 analytes; and S6 File, pesticide test information including MS polarity).
Hazard Quotient (HQ) was calculated for all samples. HQ is a unitless value which relates the residue detections of a pesticide in bee matrices to the toxicity of that pesticide to individual honey bees, and is used to understand aggregate pesticide entering the colony [28,29]. HQ was calculated for each sample by taking the pesticide residues detected (ppb) and then dividing this value by the LD50 of the pesticide. The LD50 values used for each pesticide were taken from either Traynor et al (2016) supplementary table or the EPA EcoTOX database (https://cfpub.epa.gov/ecotox/) using the oral LD50 for the pesticide to approximate dietary pesticide toxicity. Pesticide label use information was retrieved from the Crop Data Management Systems Database (https://www.cdms.net/Label-Database/Advanced-Search) to determine if the active ingredient was included on an approved label for the crop (note: this database does not include Special Local Need).
Finally, we created a variable which captures the weighted HQ values for different pollen identities entering the colony. For example, HQ > 1000 commonly is used to indicate elevated pesticide levels [28]. We wanted to understand if HQ over 1000 indicates that the same amount of pesticide is entering the colony within each crop. Rather than examining the pesticide hazard through pollen, the Index of HQ weightedness asks if HQ of similar values in each pollen identity provide the same pesticide intake to the colony.
While incoming pollen weight ought to influence the exposure pattern of the colony to a pesticide of a given HQ, it only crudely estimates exposure as pollen consumption is mediated through adult nurse bees [46] and larval consumption rates [29]. Our aim, rather, was to assess the extent to very high or low pollen intake might skew the interpretation of HQ as a proxy for risk.
Statistical analysis
All statistical calculations and visualizations were performed in the R statistical environment [47] and the specific code we used to analyze this data is available in the (S4 File). Instead of a statistical analysis of HQ across systems, we present an exploratory discussion of the differences between HQ and weightedness of HQ. Comparisons between the mass of pollen sources (i.e. crop and non-crop pollen from the same system) were performed using Welch’s two-sample t-test and ANOVA was used to determine if there was a difference between mean masses of color-sorted samples from each crop system.
We conducted a power analysis on log-transformed HQ (log(HQ+1)) [48] to determine the number of sites needed to detect changes in HQ over time. Log-transformation of the data was necessary to meet the assumptions of normality for the rearranged t-test formula. First, the mean HQ value was calculated for each crop for each year of the study (xcurrent). HQ was transformed to normalize the data for use in the power analysis [26,49]. Next, we calculated an annual 5% change in HQ value over 5 years (xfuture) from the untransformed data. This threshold is based on historic change over time; from 1997 to 2012, oral toxic load—defined as the product of the application rate of pesticide and the corresponding acute toxicity to honey bees—increased by 4-fold in one county within the Pacific Northwest, but largely remained unchanged throughout the 15-year time period [39]. We reasoned that a 5% increase or decrease would reflect the worst and best case, respectively, observed historically for our region. We assumed the uniform variance in HQ across all years to compute a pooled standard deviation (sdpooled) [50]. The HQ detectable difference was then log-transformed to put it on the same scale as the pooled standard deviation [51]. This allowed us to determine how many sites were needed to detect differences between two sampling periods and what the minimum detectable HQ value difference would be over the sampling scheme. We avoided pseudo-replication of composite samples by performing separate power analysis separated by year (2020 and 2021) and cropping system, considering peak bloom only. Using these parameters we calculated Cohen’s d as:
Cohen’s d was used to conduct power analyses in R with the package pwr2 using a two-sample power analysis where α = 0.05 and β = 0.2 [48,52,53].
To determine if time significantly influenced the HQ values, we used the complete dataset in a zero-inflated linear mixed model (R packages lme4, caret, and glmmTMB) on the square root transformed HQ values to meet the needs for normal residuals. This model allows for a combination of both a linear relationship between the continuous model of HQ and the binary model where pesticides are or are not detected [54]. The binary model accounts for those factors which influence detection or non-detection of HQ and the GLMM models the HQ as a continuous variable. Next, we created four different ways to test the concept of time as factors. First, we used bloom period (the bloom state of the crop–early, peak, or late). Next, we generated variables for numeric day of the year (1–365), month (01–12) and year (2020 or 2021). Then, we used AICc testing to determine which variables to include in the final model [55,56] and the deviance residuals were used to determine if the final model was appropriate.
Cost analysis
We calculated cost using a simplified pollen sampling protocol estimating the hours needed to collect and test each composite sample. Travel time to each site was fixed at 2 hours travel (round trip, per visit). Each site requires 3 trips; 1) affix the traps and allows the bees to adjust, 2) engage the traps, 3) collect pollen, release, and remove the traps. Field work on site was estimated to be 1 hour per site, per event. Therefore, the travel and work time for each composite sample obtained is 9 working hours per sample. Each sample also takes between 1–2 hours to process, depending on size. This includes weighing the sample, cleaning it, recording the data, packaging it for transport, and cleaning the traps. We estimate a single sample represents about 10–11 hours of work to collect and prepare. If each pesticide residue panel costs $130 to perform and a qualified field biologist would require $32/hr in wages: $466 is the final cost for collecting, processing, and testing each composite sample. Note that this assumes there is no training required for the field and lab work. Finally, the number of tests needed is twice the number of sites to be tested (2n) because each site needs to be tested in year zero (baseline) and five years later to determine a change.
Results
Over two years, 493 pollen samples were analyzed for pesticide residues for composite (n = 149) and color sorted sub (n = 344) samples. Leaky pollen traps and low pollen collection resulted in the loss of approximately 20 composite samples. Pesticides were detected in 75% of all pollen samples (n = 371) and 87% of composite samples (n = 129). When pesticides were detected, they often resulted in high HQ values, exceeding HQ > 1000 [21,31]. In our study, 30.2% of all contaminated samples (n = 120) and 24% of contaminated composite samples (n = 36) exceeded HQ 1000. Nine total samples and one composite sample exceeded HQ 10,000. Each pesticide detected, the number of times it was detected, and the pollen source are displayed in Table 1.
This includes the number of times that the pesticide was detected within a system, the average residue associated with those detections, and whether the pesticide is approved for use on that crop (excludes Special Local Need).
Pesticides approved for use on the focal crop and pesticides which do not have a registered label use within the focal crop were both found in pollen. Although this provides some insight into pesticide hazard within these floral resources, caution should be used when connecting these detections to spray habits of landowners. Rather, this illustrates the pervasiveness of pesticides in agricultural settings [10,57–59]. Both approved and unapproved active ingredients were detected in all pollen samples (Table 1). By sub-sampling the pollen into color groups, pesticides were also detected in color-sorted sub samples that were not detected in the composite sample.
First, we estimated that the number of sites that would need to be sampled to detect a 5% change in HQ, and these varied considerably by crop and across year of sampling (Table 2). For example, there was a 22-fold difference between the minimum number of sites and test between the crop with the smallest (cherry) and largest (meadowfoam) required number of tests to detect changes in HQ in 2020. Whereas the difference was 13-fold for composite samples collected in 2020 and 2021. In general, estimating HQ without respect to crop required a median number of tests. Based on our estimates of labor costs, detecting a 5% difference in meadowfoam and clover were the most expensive to conduct, whereas cherry was the least expensive to sample. Carrot samples were too few to calculate n in 2021.
The value n represents the number of sites needed for each year that sampling occurs; the number of tests represents 2na, the required sampling effort for both sampling years. The results show how many composite samples would need to be taken to detect a 5% change in HQ values, based on log-transformed data.
We found evidence that HQ values were not consistent across crops in composite samples with cherry registering twice the HQ estimate compared to meadowfoam during peak bloom in 2020. However, this trend did not hold for 2021 composite samples (Fig 2).
The horizontal line indicates the median value (HQ); the boundaries of the box are the 25th and 75th percentiles. The whiskers represent the most extreme datapoints that are no more than 1.5 times the length of the box.
The mass of pollen collected by honey bees at peak bloom varied by crop at peak bloom (Fig 3). Differences occurred both among the total pollen collected, with the amount of crop pollen collected. The total amount of pollen collected from meadowfoam was 4.6 and 4.3 times the amount of pollen collected from carrot seed in 2021 and 2022, respectively. The mass of collected from three systems were dominated by the focal crop pollen, with the exception of carrot seed, where it constituted a minority of the sample, if it was collected at all.
The horizontal line indicates the median value (mass); the boundaries of the box are the 25th and 75th percentiles. The whiskers represent the most extreme datapoints that are no more than 1.5 times the length of the box.
HQ weighted by the mass of pollen collected at peak bloom varied across crops. HQ for total pollen from colonies placed in meadowfoam increased and in carrot decreased relative to the other crops when accounting for mass of pollen collected (Fig 4), with HQ for the latter being 200- and 10,000-fold higher compared to carrot seed (for 2020 and 2021, respectively).
The horizontal line indicates the median value (Acute Toxic Load), the boundaries of the are the 25 and 75th percentiles and whiskers represent the most extreme datapoints that is no more than 1.5 times the length of the box.
To determine if time was related to HQ value, the best fit for the data was a combined GLMM and a zero-inflated binomial model. The dataset was highly skewed and we used a zero-inflated model to determine if there was a separate process effecting zero values. Pollen samples from some sites had consistently low mass over time, meaning that some sites were more likely to have no detectable HQ value associated with them. Therefore, the combined model accounts for both processes that can be causing zero values and processes which influence HQ as a continuous variable. We found that bloom period was not significantly related to transformed HQ values and did not improve the model’s fit (Table 3, Supplementary Information). Site, pollen type, and crop system all influenced if pesticides were detected within a sample or not as random effects. That is, these factors influenced if pesticides were detected at all in a sample (HQ = 0 or HQ ≠ 0). Bloom period and day of the year (1–365) affected the HQ values if HQ was non-zero. That is, these factors influenced the value of HQ when pesticides were detected. HQ values changed over time, but were not associated with a particular trend (Fig 5).
The horizontal line indicates the median value (mass); the boundaries of the box are the 25th and 75th percentiles. The whiskers represent the most extreme datapoints that are no more than 1.5 times the length of the box.
This table displays the results of the linear model and the binomial model, the value associated with each variable and the p-value. The AIC for this model is 3338.5 and BIC is 3384.8 with 482 degrees of freedom on the residuals. The factors of Site and Crop contain 39 and 4 groups respectively; n = 493.
Discussion
We describe the first effort to detect the power of sampling pollen as proxy for changes in pesticide hazard from agricultural cropping systems at state level. Pesticides were detected in the majority of samples, and HQ values were frequently above commonly reported thresholds of concern [28,31], especially within cherry systems. We found that power to detect a modest change in pesticide hazard (5% across a five-year period) varied considerably by the crop surveyed, with the lowest number of sampling sites being 190 sites (at a cost of $177,080) for clover (using 2021 sampling data) to 7,194 sites (at a cost of $3.35 million) for meadowfoam (2021 sampling data). The potential for detecting modest changes in pesticide hazard is grounded in the historic changes in bee toxic load over the past two decades; counties in the pacific northwest experienced a maximum change of 5% from the late 1990s to the mid-2010s [39]. Our study determines the number of sites needed to detect a 5% change every year for five years, however the value of HQ detectable by each sampling scheme was not consistent. For example, if sampling cherry in 2020, it would take 278 sampling events to determine differences between years with a minimum HQ value of 575. However, if sampling meadowfoam in 2021, it would take 7194 sampling events and the detectable difference in HQ would be one third of the cherry samples (HQ = 154). The interpretation of the results was complicated by the fact that in some cases almost half of pesticide detections were from products not registered on the crop being pollinated. The levels of quantification (LOQ) for each pesticide fluctuated from sample to sample as mass was not consistent. This allowed for the detections of pesticides in the color-sorted subsamples which were not found in the composite samples. Moreover, pollen trapped from colonies pollinating carrot seed was heavily weighted to non-crop pollen, a situation which was less pronounced in the other three cropping systems. The problem of hazard was also complicated by the fact that the total mass of pollen collected by honey bees at peak bloom varied across crops. As a result, the total daily exposure to pesticides to the colony (a product of the mass of pollen collected and HQ value of the pollen) resulted in further separation between colonies with the highest pollen collection (meadowfoam) as compared to the colonies with the lowest collection (carrot seed). Notably, however, the time of sampling was an important factor in understanding changes HQ as values changed with the day of the year and the bloom period of the crop. Overall, our findings suggest that post-registration monitoring of pesticides by state agencies may be feasible, but difficult to interpret.
Through our study of the variation in HQ within and across crops we were able to estimate the costs of monitoring for pesticides in pollen with sufficient power to detect a 5% change in HQ for the state of Oregon would cost between $129,548 to $3.35 million. To help contextualize the magnitude of sampling that would be needed, it is helpful to consider the number of colonies used to pollinate these crops in the state, as well as the area dedicated to these crops. For example, Oregon grows approximately 12,000 acres of sweet cherry each year in the Mid-Columbia and Willamette Valleys [60], which means finding sites for every ~86 to ~48 acres of cherry production. However, this becomes much more difficult in crops like meadowfoam which comprise only 4,000 acres in the state and would require samples to be taken every ~1 to 1.3 acres of meadowfoam production [61]. Moreover, some of these crops are pollinated by relatively small numbers of colonies, which would require sampling every colony used for pollination in crops such as meadowfoam (2,584 colonies are estimated to be contracted to pollinate meadowfoam [62]) and it would still be deficient by 528 to 1,013 colonies. Increasing the percent change in HQ over a five year period would help lower the total number of samples need, although are decision to use finer rates of change are based on previous recommendations [63]. Cost, however, may ultimately drive efforts to implement a sampling system sufficiently powerful to detect these changes. Currently, the state of Oregon spends approximately $750,000 on water quality monitoring annually [64], however, the range in cost for detecting differences in HQ was wide ($2,410,152 in 2020 and $175,216 in 2021). Water quality monitoring is also ranked as the most important environmental quality to stakeholders, [65], above pollinator protection.
Hazard Quotient estimates from the total pollen (composite) collected by bees significantly differed across crops in only one year of our study (2020). HQ estimates from pollen collected from colonies contracted to cherry pollination being twice those of colonies contracted to meadowfoam. However, this trend did not hold in 2021. Furthermore, pesticides which are approved for use within the focal crop system and active ingredients which do not appear on approved labels were frequently found within all pollen types. By separating pollen into color-sorted subsamples, we also found pesticides that were not detected within the composite samples. It is common to detect pesticides in bee collected pollen [23,24,26,36,66–68], and so the frequency of detection is not surprising. Although it was not possible to trace the source of pesticides detected in color sorted sub-samples, it should be noted that drift [69–72] rather than the use of a pesticide not registered for the pollinated crop is likely. Furthermore, our methodology of sorting pollen which has been stored in a freezer could potentially allow time for diffusion of pesticides across the lipid-rich pollen pellets [57,73].
By examining the incoming mass of pollen to the hive, we were able to highlight the problem with estimates of HQ when considering the quantity of pollen being collected by colony from a landscape. HQ is a unitless value which is connected to the LD50 of individual adult honey bees [21,31]. However, as Thompson points out [29], estimates of HQ based on pesticide detection in pollen do not account for the consumption rate of that pollen by different castes in a honey bee colony. Overlooked in this critique of the application of HQ to understanding the risk of pesticide residues in pollen to honey bee colonies, is the fact that some pollen is collected at exceptionally low quantities, while others are collected in such large quantities that it is not immediately consumed, but accumulated in the comb in reserve. The implications of this imbalance on risk to the colony have not been previously explored, adding to the gap in our knowledge of how exposure to pesticides in pollen manifests itself in meaningful endpoints [46,74]. Unlike the EPA’s Risk Quotient (RQ), which estimates the amount of pollen reaching each individual bee, HQ does not estimate how much pollen is consumed, and thresholds which imply critical levels of individual bee hazard are largely unvalidated and understudied [30–32]. This leads to an issue with current reliance on HQ: RQ and HQ are often not aligned [29] and while HQ in its many forms dominates the literature, it does not have strong validation for the impacts on individual bees or consequences for colonies [28]. This study highlights one potential underlying reason that HQ may not be connected to colony endpoints. The extent of this problem is best illustrated in our study by the difference between cherry and meadowfoam HQ when the total mass of pollen entering the colony at peak bloom is considered. On the basis of traditional HQ calculations, colonies contracted to cherry pollination face the highest risk from pesticide contamination in pollen. Colonies in meadowfoam, however, yielded 1.9 to 3.4 times the amount of pollen. Consequently, when we weight HQ by the increased pollen intake in meadowfoam, we are no longer able to detect a separation in HQ between these two crops. The nearly double HQ value of cherry over meadowfoam, is eroded by the nearly double intake of pollen in meadowfoam. Our findings of the differential intake of pollen may help contextualize other studies that have documented that floral resources outside of the focal crop contribute to pesticide hazard through increased contamination [20,24,26,27,75,76] or to dilution with clean pollen [77,78]. By taking into consideration the mass of pollen brought into the colony when calculating HQ a shortcoming with a reliance on this metric can be brought into focus. What is neglected by weighting HQ by the mass of incoming pollen is the rate and duration of the consumption of contaminated pollen by the colony. More than anything our findings draw attention to the fact that the current practice of estimating HQ from trapped pollen may not fully capture the exposure of pesticides to individuals in the hive, which may explain why the metric is poorly linked to hive endpoints [31].
We found that temporal variation is an important factor when sampling for HQ. The model of best fit was a zero-inflated model as, like many biological datasets [54,79], there were many points where the response variable (HQ) was zero. Pollen identity, pollen mass, site, and cropping system all influenced if HQ was likely to be zero. That is, some sites (and therefore also the crop associated with that site) had HQ values of zero more frequently. The GLMM was then modeled in a negative binomial distribution, and the final model of best fit included both the day of the year (1–365) and the bloom state of the crop. Taken together, this demonstrates that both the focal crop and the surrounding landscape are influencing the HQ value at the pollen trap. This is consistent with other studies which find that landscape context matters to HQ values [22,23,26,58,77,80]. This is particularly pronounced in a heterogeneous landscape like that of Oregon [40,41]; bees have the opportunity to collect from a wide variety of available pollen. The difference between pesticide hazard within the foraging area of the bees and the crop of interest may be greater in heterogenous landscapes than compared to regions which rely on a few key monocrop products. This variation in HQ values could be due to several factors. Due to the episodic nature of pesticide exposure, pesticide contamination in pollen is easily missed. This could weaken the connection between pesticide residues and crop-specific pesticide use practices. Shifting from a bloom-state focused pollen sampling strategy (i.e. during peak bloom) to monitoring which documents changes in HQ over long periods of time [23,81] may be the most effective. Adding landscape use within the foraging radius of the honey bee colony would be likely to improve the fit of the model; however, these type of models are not in use by states seeking to monitor pesticides.
Honey bees are highly studied and relatively easy to transport throughout a region, while also already being incorporated into pesticide risk assessment processes. Leveraging their usefulness as an agricultural pollinator to collect information about pesticide contamination and the health of other bees within a landscape is promising. However, despite their generalist foraging behavior, caution should be used honey bee collected pollen to monitor for changes in pesticide residues over time. Bee species with smaller or larger foraging radii, for example mason bees (Osmia) or bumble bees (Bombus) may be better suited to capture field to field variation [82,83], but these may also distort the hazard patterns experienced by honey bees. There are other potential ways to reduce costs of monitoring schemes and increase predictive power. There could be merits, for example, to limiting the number of pesticides tested within a sample. By nature of their high LD50 values, insecticides contribute disproportionately to the HQ value and focusing on just these chemistries could reduce testing costs. Traynor et al (2016) also found that high fungicide HQ contributions were associated with hive loss and queen events over the course of the pollination circuit in monitored hives [31]. Incident reporting on bee poisonings and associated HQ values could be one way to validate currently common thresholds, as has been done with application-rate HQ estimations [84].
Conclusion
In current monitoring schemes, honey bees are used as ecological sensors, collecting pollen from within a foraging radius and enabling testing at the hive, much in the same way that water quality monitoring is used to assess stream health. Such monitoring plays an important role in ground-truthing voluntary actions (like MP3s and BMPs) as well the assessment of risk by pesticide regulators. However, the connection between pesticide contaminated pollen collected by bees and the actual risks of these pesticides to a honey bee colony is more complicated than the connection between the water sample and aquatic insect health. In aquatic systems, pesticides are applied to fields and then wash into surface water which can be collected at specific points [85]. The pesticides within the water can be assessed for their impacts on the health of stream organisms [86,87], chemical residues can be related to application rates of chemicals [88], and the overall state of the watershed determined. Finally, different streams and sites can be targeted for restoration and mitigation efforts [70,89–91] based on the results of sampling.
While other studies have noted the limitations of HQ metrics [28,29,31] we expand this criticism by highlighting a number of challenges that may make estimates of HQ from trapped pollen samples less informative than water quality samples. First, connecting honey bee collected pollen to pesticide practices in a cropping system are more nuanced compared to water quality sampling. Pollen is not collected evenly from the surrounding floral resources [92,93] and pesticide contamination from the focal crop pollen may be amplified or diluted. As in numerous other studies [25–27,94,95], we found that the amount of focal crop pollen in a pollen trap was not always a large proportion of the total pollen collected by the colony. This represents a key challenge for using pollen sampling as a post-registration monitoring tool, namely the inability to reliably link pesticide contaminating pollen back to a pesticide use practices in a specific crop where honey bees are being contracted for pollination.
A second challenge that we discovered was the high variation in HQ within sites, across sites, within a crop, across time periods within a crop and across crops. We estimated that the high variability required relatively high sample effort to detect interannual changes in hazards that are above those that have been experienced historically. To this end, it is likely that existing state-level pesticide monitoring conducted by may be well below the levels to detect meaningful changes.
Finally, we demonstrated that HQ values from trapped pollen do not currently account the fact that the mass of incoming pollen can vary widely across pollen trapping periods. This difference in pollen mass can lead to HQ estimations representing distortions in the pesticide exposure of the colony. The incoming pesticide load may differ, even when HQ values are equivalent.
In spite of the challenges we found in using pollen sampling as post-registration tool, such sampling may prove useful in characterizing risk when pesticides are being registered or reviewed. EPA employs full scale field studies (Tier III) when risk cannot be characterized by either lab studies on individual bees (Tier I) or on studies where bees are forced to forage on plants that have been treated with the pesticide (Tier II). Tier III studies, however, often are expensive, costing an estimated $1.2 million, compared to Tier II ($75,000–150,000) or Tier I ($30,000–50,000) [96]. Estimates of spatiotemporal variation of HQ from studies such as ours, could help regulators model field variability to supplement Tier III studies to refine estimations of risk. A major challenge, however, remains in interpreting HQ values from trapped pollen in terms of risk. Although we demonstrated that the importance of accounting for the mass of pollen collected by bees in calculating HQ, this represents part of the picture of exposure, as it does not account for the rate in which pollen is being consumed by different bee castes or pollen stored within the colony and consumed over time [46,74]. More research needs to be done to better characterize how contaminated pollen translates to actual exposure of bees to the pesticides. Such work may resolve the issue of establishing meaningful thresholds for HQ levels, which are commonly used [21,31] but appear at odds when compared to models of pollen consumption, such as BeeREX [29], which are used in risk assessment.
Supporting information
S1 File. Synergistic Pesticide Labs Summary of pollen testing methodology.
https://doi.org/10.1371/journal.pone.0309236.s001
(PDF)
S2 File. A table displaying every detection of pesticide from the study as an individual row in the dataset.
https://doi.org/10.1371/journal.pone.0309236.s002
(CSV)
S3 File. A table displaying the total HQ values and summary information for each SampleID.
https://doi.org/10.1371/journal.pone.0309236.s003
(CSV)
S4 File. R code for creating all of the figures.
https://doi.org/10.1371/journal.pone.0309236.s004
(RMD)
S5 File. Reporting limits from Synergistic Pesticide Labs that includes the batch-specific level of detection, levels of quantification, and the 292 pesticides each sample was tested for.
https://doi.org/10.1371/journal.pone.0309236.s005
(XLSX)
S6 File. Pesticide test information on the empirical formula, isotopic mass, MS polarity, MS ion, retention time, and precursor for pesticide tests performed.
https://doi.org/10.1371/journal.pone.0309236.s006
(XLSX)
S7 File. Power analysis on untransformed HQ data.
https://doi.org/10.1371/journal.pone.0309236.s007
(DOCX)
Acknowledgments
The authors would like to thank the members of the honey bee and pollinator health labs at Oregon State University who helped collect and sort pollen samples. We would also like to thank Brian Van Ee, our statistical consultant who helped create the zero-inflated model used in determining if time is relevant to sampling pollen.
References
- 1. Cooper J, Dobson H. The benefits of pesticides to mankind and the environment. Crop Prot. 2007;26: 1337–1348.
- 2. Popp J, Pető K, Nagy J. Pesticide productivity and food security. A review. Agron Sustain Dev. 2013;33: 243–255.
- 3. Damalas CA. Understanding benefits and risks of pesticide use. Sci Res Essays. 2009;4: 945–949.
- 4. Hatfield RG, Strange JP, Koch JB, Jepsen S, Stapleton I. Neonicotinoid Pesticides Cause Mass Fatalities of Native Bumble Bees: A Case Study from Wilsonville, Oregon, United States. Environ Entomol. 2021;50: 1095–1104. pmid:34145877
- 5. Sánchez-Bayo F, Wyckhuys KAG. Worldwide decline of the entomofauna: A review of its drivers. Biol Conserv. 2019;232: 8–27.
- 6. Pettis JS, Lichtenberg EM, Andree M, Stitzinger J, Rose R, vanEngelsdorp D. Crop Pollination Exposes Honey Bees to Pesticides Which Alters Their Susceptibility to the Gut Pathogen Nosema ceranae. PLoS One. 2013. pmid:23894612
- 7. Rundlöf M, Andersson GKS, Bommarco R, Fries I, Hederström V, Herbertsson L, et al. Seed coating with a neonicotinoid insecticide negatively affects wild bees. Nature. 2015;521: 77–80. pmid:25901681
- 8. Gill RJ, Ramos-Rodriguez O, Raine NE. Combined pesticide exposure severely affects individual-and colony-level traits in bees. Nature. 2012;491: 105–108. pmid:23086150
- 9. Goulson D, Nicholls E, Botías C, Rotheray EL. Bee declines driven by combined Stress from parasites, pesticides, and lack of flowers. Science. 2015. pmid:25721506
- 10. Mullin CA, Frazier M, Frazier JL, Ashcraft S, Simonds R, Vanengelsdorp D, et al. High levels of miticides and agrochemicals in North American apiaries: implications for honey bee health. PLoS One. 2010;5: e9754. pmid:20333298
- 11. Potts SG, Biesmeijer JC, Kremen C, Neumann P, Schweiger O, Kunin WE. Global pollinator declines: Trends, impacts and drivers. Trends Ecol Evol. 2010;25: 345–353. pmid:20188434
- 12.
US EPA, EAD, CDPR. White Paper in Support of the Proposed Risk Assessment Process for Bees. 2012; 1–275.
- 13.
US EPA. Data Requirements for Pesticide Registration. In: A US Environmental Protection Agency [Internet]. 2016 pp. 1–11. www.epa.gov/pesticide-registration/data-requirements-pesticide-registration#nto.
- 14. Obama B. Presidential memorandum: Creating a federal strategy to promote the health of honey bees and other pollinators. Improv Heal Honey Bees Other Pollinators Natl Strateg Res Action Plan. 2014; 221–225.
- 15.
State FIFRA Issues, Research and EG. Final Guidance for State Lead Agencies for the Development and Implemenation of Managed Pollinator Protection Plans. 2015; 1–7. https://obamawhitehouse.archives.gov/the-press-office/2014/06/20/presidential-memorandum-creating-federal-strategy-promote-health-honey-b.
- 16. Whitney KS, Stringer BB. Evaluation of US state pollinator plans using 3 evidence-based policymaking frameworks. bioRxiv. 2021.
- 17.
The Office of the Inspector General, United States Environmental Protection Agency. EPA Needs to Determine Strategies and Level of Support for Overseeing State Managed Pollinator Protection Plans OFFICE OF INSPECTOR GENERAL Ensuring the safety of chemicals. 2019. www.epa.gov/oig.
- 18.
2022 AAPCO Pollinator Survey Results. 2022. https://aapco.org/wp-content/uploads/2022/06/Survey-National-pollinator-plan37.pdf.
- 19. Topitzhofer E, Lucas H, Carlson E, Chakrabarti P, Sagili R. Collection and identification of pollen from honey bee colonies. J Vis Exp. 2021;2021: 1–23. pmid:33554968
- 20. Stoner KA, Cowles RS, Nurse A, Eitzer BD. Tracking Pesticide Residues to a Plant Genus Using Palynology in Pollen Trapped from Honey Bees (Hymenoptera: Apidae) at Ornamental Plant Nurseries. Environ Entomol. 2019;48: 351–362. pmid:30753426
- 21. Stoner KA, Eitzer BD. Using a Hazard Quotient to Evaluate Pesticide Residues Detected in Pollen Trapped from Honey Bees (Apis mellifera) in Connecticut. PLoS One. 2013. pmid:24143241
- 22. Drummond FA, Ballman ES, Eitzer BD, Du Clos B, Dill J. Exposure of honey bee (apis mellifera L.) colonies to pesticides in pollen, a statewide assessment in Maine. Environ Entomol. 2018;47: 378–387. pmid:29509899
- 23. Böhme F, Bischoff G, Zebitz CPW, Rosenkranz P, Wallner K. Pesticide residue survey of pollen loads collected by honeybees (Apis mellifera) in daily intervals at three agricultural sites in South Germany. PLoS ONE. Public Library of Science; 2018. pmid:29979756
- 24. Favaro R, Bauer LM, Rossi M, D’Ambrosio L, Bucher E, Angeli S. Botanical Origin of Pesticide Residues in Pollen Loads Collected by Honeybees During and After Apple Bloom. Front Physiol. 2019;10: 1–18. pmid:31620006
- 25. Urbanowicz C, Baert N, Bluher SE, Böröczky K, Ramos M, McArt SH. Low maize pollen collection and low pesticide risk to honey bees in heterogeneous agricultural landscapes. Apidologie. 2019;2011: 379–390.
- 26. Graham KK, Milbrath MO, Zhang Y, Baert N, Mcart S, Isaacs R. Pesticide risk to managed bees during blueberry pollination is primarily driven by off-farm exposures. Sci Rep. 2022;12: 1–16. pmid:35504929
- 27. Mcart SH, Fersch AA, Milano NJ, Truitt LL, Böröczky K. High pesticide risk to honey bees despite low focal crop pollen collection during pollination of a mass blooming crop. Sci Rep. 2017;7: 1–10. pmid:28422139
- 28. Carlson EA, Melathopoulos A, Sagili R. The Value of Hazard Quotients in Honey Bee (Apis mellifera) Ecotoxicology: A Review. Front Ecol Evol. 2022;10.
- 29. Thompson HM. The use of the Hazard Quotient approach to assess the potential risk to honeybees (Apis mellifera) posed by pesticide residues detected in bee-relevant matrices is not appropriate. Pest Manag Sci. 2021;77: 3934–3941. pmid:33899320
- 30. El Agrebi N, Wilmart O, Urbain B, Danneels EL, de Graaf DC, Saegerman C. Belgian case study on flumethrin residues in beeswax: Possible impact on honeybee and prediction of the maximum daily intake for consumers. Sci Total Environ. 2019;687: 712–719. pmid:31412474
- 31. Traynor KS, Pettis JS, Tarpy DR, Mullin CA, Frazier JL, Frazier M, et al. In-hive Pesticide Exposome: Assessing risks to migratory honey bees from in-hive pesticide contamination in the Eastern United States. Sci Rep. 2016;6. pmid:27628343
- 32. Lee K V., Goblirsch M, McDermott E, Tarpy DR, Spivak M. Is the brood pattern within a honey bee colony a reliable indicator of queen quality? Insects. 2019;10: 1–17. pmid:30626029
- 33. Couvillon MJ, Riddell Pearce FC, Accleton C, Fensome KA, Quah SKL, Taylor EL, et al. Honey bee foraging distance depends on month and forage type. Apidologie. 2015;46: 61–70.
- 34. Beekman M, Ratnieks FLW. Long-range foraging by the honey-bee, Apis mellifera L. Funct Ecol. 2000;14: 490–496.
- 35.
USDA. Attractiveness of Agricultural Crops to Pollinating Bees for the Collection of Nectar and/or Pollen United States Department of Agriculture. 2007. http://www.ascr.usda.gov/complaint_filing_file.html.
- 36. Graham KK, Milbrath MO, Zhang Y, Soehnlen A, Baert N, McArt S, et al. Identities, concentrations, and sources of pesticide exposure in pollen collected by managed bees during blueberry pollination. Sci Rep. 2021;11: 1–13. pmid:34413379
- 37. Mcart SH, Fersch AA, Milano NJ, Truitt LL, Böröczky K. High pesticide risk to honey bees despite low focal crop pollen collection during pollination of a mass blooming crop. Sci Rep. 2017;7: 1–10. pmid:28422139
- 38. Topitzhofer E, Lucas H, Chakrabarti P, Breece C, Bryant V, Sagili RR, et al. Assessment of Pollen Diversity Available to Honey Bees (Hymenoptera: Apidae) in Major Cropping Systems during Pollination in the Western United States. J Econ Entomol. 2019;112: 2040–2048. pmid:31237612
- 39. Douglas MR, Sponsler DB, Lonsdorf E V., Grozinger CM. County-level analysis reveals a rapidly shifting landscape of insecticide hazard to honey bees (Apis mellifera) on US farmland. Sci Rep. 2020;10: 797. pmid:31964921
- 40. Pater DE, Bryce SD, Thorson TD, Kagan J, Chappell C, Omernik JM, et al. Ecoregions of Western Washington and Oregon. US Geological Survey; 1996. pp. 1987–1987.
- 41. Thorson TD, Bryce SD, Lammers DA, Woods AJ, Omernik JM, Kagan J, et al. Ecoregions of Oregon. Reston, Virginia, US Geol Surv (map scale 11,500,000). 2003; 1.
- 42. Lau P, Bryant V, Rangel J. Determining the minimum number of pollen grains needed for accurate honey bee (Apis mellifera) colony pollen pellet analysis. Palynology. 2018;42: 36–42.
- 43. Jones GD. Pollen analyses for pollination research, acetolysis. J Pollinat Ecol. 2014;13: 203–217.
- 44. Jones GD. Pollen extraction from insects. Palynology. 2012;36: 86–109.
- 45. A M, L SJ, S D, S FJ. Fast and easy multiresidue method employing acetonitrile extraction/partitioning and “dispersive solid-phase extraction” for the determination of pesticide residues in produce. J AOAC Int. 2003;86: 412–431. pmid:12723926
- 46. Sponsler DB, Johnson RM. Mechanistic modeling of pesticide exposure: The missing keystone of honey bee toxicology. Environ Toxicol Chem. 2017;36: 871–881. pmid:27769096
- 47.
R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2021. https://www.r-project.org/.
- 48. Cohen J. Statistical Power Analysis for the Behavioral Sciences. Stat Power Anal Behav Sci. 2013.
- 49. Graham KK, Gibbs J, Wilson J, May E, Isaacs R. Resampling of wild bees across fifteen years reveals variable species declines and recoveries after extreme weather. Agric Ecosyst Environ. 2021;317: 1–39.
- 50. Lebuhn G, Droege SAM, Connor EF, Potts SG, Minckley RL, Griswold T, et al. Detecting Insect Pollinator Declines on Regional and Global Scales. Conserv Biol. 2013;27: 113–120. pmid:23240651
- 51. Wolfe R, Carlin JB. Sample-size calculation for a log-transformed outcome measure. Control Clin Trials. 1999;20: 547–554. pmid:10588295
- 52. Lan L, Lian Z. Application of statistical power analysis—How to determine the right sample size in human health, comfort and productivity research. Build Environ. 2010;45: 1202–1213.
- 53. Potvin PJ, Schutz RW. Statistical power for the two-factor repeated measures ANOVA. Behav Res Methods, Instruments, Comput. 2000;32: 347–356. pmid:10875184
- 54. Fávero LP, Hair JF, Souza R de F, Albergaria M, Brugni T V. Zero-inflated generalized linear mixed models: A better way to understand data relationships. Mathematics. 2021;9: 1–28.
- 55. Zuur AF. Mixed Effects Models and Extensions in Ecology with R. 2013.
- 56. Stoica P, Selén Y. Model Order Selection: A review of information criterion rules. IEEE Signal Process Mag. 2004;21: 36–47.
- 57. vanEngelsdorp D, Evans JD, Donovall L, Mullin C, Frazier M, Frazier J, et al. “Entombed Pollen”: A new condition in honey bee colonies associated with increased risk of colony mortality. J Invertebr Pathol. 2009;101: 147–149. pmid:19361513
- 58. Calatayud-Vernich P, Calatayud F, Simó E, Pascual Aguilar JA, Picó Y. A two-year monitoring of pesticide hazard in-hive: High honey bee mortality rates during insecticide poisoning episodes in apiaries located near agricultural settings. Chemosphere. 2019;232: 471–480. pmid:31163323
- 59. Calatayud-Vernich P, Calatayud F, Simó E, Picó Y. Occurrence of pesticide residues in Spanish beeswax. Sci Total Environ. 2017;605–606: 745–754. pmid:28679118
- 60.
Thompson A. Estimated Impact of X-disease and Little Cherry Disease in Washington and Oregon from 2015 to 2020 Estimated Impact of X-disease and Little Cherry Disease in Washington and. 2022;60.
- 61. Zasada IA, Weiland JE, Reed RL, Stevens JF. Activity of Meadowfoam (Limnanthes alba) Seed Meal Glucolimnanthin Degradation Products against Soilborne Pathogens. 2012.
- 62. Sagili R, Chief B. Pacific Northwest (PNW) 2011 Beekeeper Pollination survey. 2011.
- 63. Douglas MR, Baisley P, Soba S, Kammerer M, Lonsdorf E V., Grozinger CM. Putting pesticides on the map for pollinator research and conservation. Sci Data. 2022;9: 1–15. pmid:36114185
- 64.
Pesticide Stewardship Council. In: Oregon Environmental Council [Internet]. 2015. https://oeconline.org/the-pesticide-stewardship-partnership-program-one-year-later/.
- 65. Schulte LA, Niemi J, Helmers MJ, Liebman M, Arbuckle JG, James DE, et al. Prairie strips improve biodiversity and the delivery of multiple ecosystem services from corn–soybean croplands. 2017;114: 11247–11252. pmid:28973922
- 66. Barmaz S, Vaj C, Ippolito A, Vighi M. Exposure of pollinators to plant protection products. Ecotoxicology. 2012;21: 2177–2185. pmid:22752801
- 67. Tosi S, Costa C, Vesco U, Quaglia G, Guido G, Simone T, et al. A 3-year survey of Italian honey bee-collected pollen reveals widespread contamination by agricultural pesticides. Sci Total Environ. 2018;615: 208–218. pmid:28968582
- 68. Humann-Guilleminot S, Binkowski ŁJ, Jenni L, Hilke G, Glauser G, Helfenstein F. A nation-wide survey of neonicotinoid insecticides in agricultural land with implications for agri-environment schemes. J Appl Ecol. 2019; 1502–1514.
- 69. Sanchez-Bayo F, Goka K. Pesticide residues and bees—A risk assessment. PLoS One. 2014;9. pmid:24718419
- 70. Reichenberger S, Bach M, Skitschak A, Frede HG. Mitigation strategies to reduce pesticide inputs into ground- and surface water and their effectiveness; A review. Sci Total Environ. 2007;384: 1–35. pmid:17588646
- 71. Otto S, Lazzaro L, Finizio A, Zanin G. Estimating ecotoxicological effects of pesticide drift on nontarget arthropods in field hedgerows. Environ Toxicol Chem. 2009;28: 853–863. pmid:19391688
- 72. Goebel KM, Davros NM, Andersen DE, Rice PJ. Tallgrass prairie wildlife exposure to spray drift from commonly used soybean insecticides in Midwestern USA. Sci Total Environ. 2022;818: 151745. pmid:34801497
- 73. Krupke CH, Hunt GJ, Eitzer BD, Andino G, Given K. Multiple routes of pesticide exposure for honey bees living near agricultural fields. PLoS One. 2012;7. pmid:22235278
- 74.
Sponsler DB, Grozinger CM, Hitaj C, Rundlöf M, Botías C, Code A, et al. Pesticides and pollinators: A socioecological synthesis. Science of the Total Environment. Elsevier B.V.; 2019. pp. 1012–1027.
- 75. Botías C, David A, Horwood J, Abdul-Sada A, Nicholls E, Hill E, et al. Neonicotinoid Residues in Wildflowers, a Potential Route of Chronic Exposure for Bees. Environ Sci Technol. 2015;49: 12731–12740. pmid:26439915
- 76. Simon-Delso N, S Martin G, Bruneau E, Delcourt C, Hautier L. The challenges of predicting pesticide exposure of honey bees at landscape level. Sci Rep. 2017;7: 1–10. pmid:28630412
- 77. Colwell MJ, Williams GR, Evans RC, Shutler D. Honey bee-collected pollen in agro-ecosystems reveals diet diversity, diet quality, and pesticide exposure. Ecol Evol. 2017;7: 7243–7253. pmid:28944014
- 78. Centrella M, Russo L, Moreno Ramírez N, Eitzer B, van Dyke M, Danforth B, et al. Diet diversity and pesticide risk mediate the negative effects of land use change on solitary bee offspring production. J Appl Ecol. 2020;57: 1031–1042.
- 79. Gotelli , Nicholas J and Ellison AM. A Primer of Ecological Statistics. Am Stat. 2005;59: 350–350.
- 80. Friedle C, Wallner K, Rosenkranz P, Martens D, Vetter W. Pesticide residues in daily bee pollen samples (April–July) from an intensive agricultural region in Southern Germany. Environ Sci Pollut Res. 2021;28: 22789–22803. pmid:33432407
- 81. Ostiguy N, Drummond FA, Aronstein K, Eitzer B, Ellis JD, Spivak M, et al. Honey bee exposure to pesticides: A four-year nationwide study. Insects. 2019;10: 1–34. pmid:30626027
- 82. Jha S, Kremen C. Resource diversity and landscape-level homogeneity drive native bee foraging. Proc Natl Acad Sci U S A. 2013;110: 555–558. pmid:23267118
- 83. Greenleaf SS, Williams NM, Winfree R, Kremen C. Bee foraging ranges and their relationship to body size. Oecologia. 2007. pmid:17483965
- 84. Thompson HM, Thorbahn D. Review of honeybee pesticide poisoning incidents in Europe–evaluation of the hazard quotient approach for risk assessment. Julius-Kühn-Archiv. 2010; 103. http://pub.jki.bund.de/index.php/JKA/article/view/140.
- 85. Chow R, Scheidegger R, Doppler T, Dietzel A, Fenicia F, Stamm C. A review of long-term pesticide monitoring studies to assess surface water quality trends. Water Res X. 2020;9: 100064. pmid:32995734
- 86. Castillo LE, Martínez E, Ruepert C, Savage C, Gilek M, Pinnock M, et al. Water quality and macroinvertebrate community response following pesticide applications in a banana plantation, Limon, Costa Rica. Sci Total Environ. 2006;367: 418–432. pmid:16643988
- 87. Sánchez-Bayo F, Goka K, Hayasaka D. Contamination of the Aquatic Environment with Neonicotinoids and its Implication for Ecosystems. Front Environ Sci. 2016;4: 1–14.
- 88. Hunt JW, Anderson BS, Phillips BM, Tjeerdema RS, Richard N, Connor V, et al. Spatial relationships between water quality and pesticide application rates in agricultural watersheds. Environ Monit Assess. 2006;121: 243–260. pmid:16758283
- 89. Morton PA, Cassidy R, Floyd S, Doody DG, McRoberts WC, Jordan P. Approaches to herbicide (MCPA) pollution mitigation in drinking water source catchments using enhanced space and time monitoring. Sci Total Environ. 2021;755: 142827. pmid:33097257
- 90. Gustavsson M, Kreuger J, Bundschuh M, Backhaus T. Pesticide mixtures in the Swedish streams: Environmental risks, contributions of individual compounds and consequences of single-substance oriented risk mitigation. Sci Total Environ. 2017;598: 973–983. pmid:28468125
- 91. Quaglia G, Joris I, Broekx S, Desmet N, Koopmans K, Vandaele K, et al. A spatial approach to identify priority areas for pesticide pollution mitigation. J Environ Manage. 2019;246: 583–593. pmid:31202826
- 92. Garbuzov M, Couvillon MJ, Schürch R, Ratnieks FLW. Honey bee dance decoding and pollen-load analysis show limited foraging on spring-flowering oilseed rape, a potential source of neonicotinoid contamination. Agric Ecosyst Environ. 2015;203: 62–68.
- 93. Couvillon MJ, Schürch R, Ratnieks FLW. Waggle dance distances as integrative indicators of seasonal foraging challenges. PLoS One. 2014;9: e93495. pmid:24695678
- 94.
Topitzhofer E. Effects of pollen collected by honey bees from pollination dependent agricultural cropping systems on honey bee nutrition. Thesis. 2014; 99.
- 95. Zawislak J, Lorenz G, Adamczyk J, Wiedenmann R, Joshi NK. Proportion of commodity crop pollens and pesticide contamination in honey bee diets in two different landscapes. Environ Adv. 2021;5: 100116.
- 96. Whitford F, Steeger T, Feken M, Krupke CH, Hunt G, Johnson R, et al. The Complex Life of the Honey Bee: Environmental, Biological, and Chemical Challenges to Colony Health. Perdue Ext Serv. 2017.