Pesticide residue survey of pollen loads collected by honeybees (Apis mellifera) in daily intervals at three agricultural sites in South Germany

In agricultural landscapes honeybees and other pollinators are exposed to pesticides, often surveyed by residue analysis of bee bread. However, bee bread is a mixture of pollen pellets of different plants collected over a longer time period. Therefore, pesticide content in the hive varies with plant species and time of pollen collection. Hence, the analysis of bee bread is an approximate approach to gain information on detailed pesticide exposure during the agronomic active season. As high-resolution data is missing, we carried out a pesticide residue survey over five years (2012–2016) of daily collected pollen pellets at three agricultural distinct sites in southern Germany. 281 single day pollen samples were selected and subjected to a multi-pesticide residue analysis. Pesticide contaminations of pollen differed between the sites. Intensive pesticide exposure can be seen by high pesticide concentrations as well as a high amount of different pesticides detected. During the five years of observation 73 different pesticides were found, of which 84% are characterized as non-harmful to honeybees. To estimate pesticide risks for honeybees, the pollen hazard quotient (PHQ) was calculated. Even though pesticides were detected in sublethal concentrations, we found substances not supposed to be exposed to honey bees, indicating the necessity for further improvement of seed treatments and increasing awareness of flowering shrubs, field margins and pesticide drift. Additionally, an in-depth analysis of nine pollen samples, divided into sub-fractions dominated by single plant species, revealed even higher concentrations in single crops for some pesticides. We give precise residue data of 1,657 single pesticide detections, which should be used for realistic laboratory and field tests.

Introduction day, were analyzed for nearly 300 active ingredients currently used in agricultural practice. The resulting high-resolution residue data is now available. Furthermore, to get a better understanding of the hazard posed to honeybees by the measured pesticide concentrations, we calculated the pollen hazard quotient (PHQ) following the method by Stoner and Eitzer [31]. In addition, we conducted in depth analyses from nine analyzed samples to identify key plant species within the pollen loads to show plant specific pesticide contamination.

Material and methods
No permits were needed. The beekeepers were voluntarily taken part in this study ontheir personell ground. Due to privacy protection of the beekeepers the exact coordinates of the apiaries will not be given. No endangered species were involved in this study.
Collection sites: "Meadow", "grain" and "fruit" In close cooperation with the respective beekeepers in the State of Baden-Württemberg, representative apiaries for agricultural used landscapes in regions with temperate climate were chosen. Within these areas the voluntarily participating beekeepers had their private apiaries. As honeybees have a vast flight range of a several kilometers radius around the hive, it was not possible to get any information on details of the respective crop protection strategies and intensity by the numerous farmers cultivating the fields in the vicinity of the apiaries. Furthermore, the crop protection intensity usually varies from year to year due to the prevailing weather conditions and pressure by damaging factors. The sites are at least 50 km apart from each other. The "meadow" site (323 m a.s.l.; meters above sea level) is located in the north-east of the town Göppingen. The area around the apiary is characterized by about 60% permanent grassland, i.e. pasture, meadow, orchard meadows and traditional extensive orchards. "Grain" site (569 m a.s.l.) is situated in the east of the village Ertingen, at the edge of a forest. It is characterized by more intensive agriculture with high percentages of small grains (wheat, barley, oat, etc.), grain and silage maize, winter oilseed rape and meadows. "Fruit" site is in the south of Heilbronn (157 m a.s.l.). 40% small grains and maize, as well as 30% permanent crops such as vine, pome, stone and soft fruits, characterize this site. Additional information for each site is given in Table A-C in S1 Tables and in Table A-C in S2 Tables.

Pollen collection
Pollen traps were set up from spring (March) until late summer (August), depending on the site, at the entrance of one honeybee colony at each of the three locations in Southern Germany. Pollen collection started 2012 and ended 2016 for the "meadow" and "grain" sites. The beekeeper of the "fruit" site collected pollen three years from 2012 to 2014. The place of the colonies remained the same during the whole experimental period (= stationary beekeeping). Bee colonies were maintained according to good apicultural practice. Pollen pellets were collected daily or every second day by each beekeeper and stored separately in labeled plastic bags at -20˚C. Pollen pellets were shipped frozen to the Apicultural State Institute where they were stored under -20˚C until chemical analysis.
Due to financial limitations 9-39 samples for each site and sampling period were chosen and sent to the laboratories for pesticide residue analysis. Depending on the quality (uncontaminated with debris and not slushy) and quantity (more than 5 g pollen per day), samples were selected randomly to cover the entire sampling period of beekeeping. This resulted in varying numbers of samples for each month, year and site, with at least one sample each week. Yet, samples have not been pooled and results represent the pesticide contamination of a single sampling day from one apiary.

Pesticide residue analysis
Residue analysis of a total of 281 samples (over the three sites and three to five years), each > 5 grams, have been performed at two laboratories. The agricultural analytic and research institute in Speyer (LUFA Speyer) performs a multi-residue QuEChERS method following Anastassiades et al. [44] ( § 64 LFGB, BVL L 00.00-115/1:2015-03) following an analysis by GC-MS (/MS) and LC-MS/MS. The Julius Kühn-Institute (JKI) uses a multi-residue method as described by Böhme et al. [45]. Both laboratories were looking for 282 different substances (pesticides and some metabolites) currently used in agricultural practice.
In order to compare all samples from both laboratories, the highest limit of detection (LOD) and limit of quantification (LOQ) for each substance from the analysis by LUFA Speyer had to be chosen and the results obtained from the laboratory of the JKI were considered accordingly. Values or substances below this LOQ were not considered for the comparison.

Residue evaluation
To estimate the hazard to bees emanating from contaminated pollen loads, the pollen hazard quotient (PHQ) was calculated following Stoner and Eitzer [31]. This method was chosen as it provides a simple and comprehensive way to calculate the risk based on LD 50 -values easily available in the internet. The concentration of each pesticide found in a sample (μg/kg) was divided by the LD 50 (honeybee oral; μg/bee) for the respective substance. LD 50 values were obtained from the University of Hertfordshire pesticides properties database [46], the US EPA ecotoxicology database [47] or the Agritox database of the French government [48]. PHQ calculations for thymol based on LD 50 concentrations (honeybee contact) by Dahlgren et al. [49]. For metabolites, the respective LD 50 value of the parent pesticide was used for calculations. Total PHQ per sample (= day; tPHQ day ) was calculated as the sum of all PHQs of the pesticides in the respective sample.
PHQ of > 50 are considered "relevant". Assuming a daily pollen consumption of 9.5 mg by a nurse bee [24,31,35], a PHQ of 50 would correspond to 0.05% of the LD 50 consumed in one day (resulting in 0.5% of the LD 50 in an average 10-day nursing period) [31].
Statistical analysis. PHQ and tPHQ day values were analyzed using the computer software JMP 1 11.1.1 (SAS Institute Inc., Cary, NC, USA) as follows: (i) analysis of variance using a generalized linear model (GLM procedure) for effects by site, year, and month, and (ii) by one-way ANOVA followed by pairwise Student-t test between sites and years.
Each active substance is assigned the respective mode of action class following the classification of the FRAC [50], IRAC [51] and HRAC [52] (Table 1). Since several active substances share the same mode of action, the concentration of each substance at a single day were summed. Grouped pesticides were then analyzed statistically as described above. In the same way total insecticides/day and total fungicides/day were analyzed.

Fractions of pollen samples
After receiving the results of the residue analysis, nine samples were chosen for in depth analysis to show the contribution of plant specific pesticide residues to the pesticide contamination of the whole sample. The pollen pellets remaining after taking five grams for residue analysis (correspond here to composite sample), were separated according to their color (correspond here to fraction). A subsample of the color fraction was ground, dissolved in water with a droplet of a tensioactive agent (soap), spread on a microscopic slide, left to dry and embedded with No data regarding ecotoxicological analysis concerning bees are available; no tolerance levels for picaridin in food are given as residues are not expected due to its use as topical insect repellent https://doi.org/10.1371/journal.pone.0199995.t001 Kaiser's glycerol gelatin and a cover glass. Light microscopic palynological analyses were performed at the honey laboratory at the Apicultural State Institute before fractions were sent to LUFA Speyer for pesticide residue analysis.

Pesticide residues
Altogether, 281 honeybee corbicular pollen load samples from three locations were analyzed in a multi-residue screening. One fifth of all samples were found uncontaminated. In total 73 different pesticides, comprising herbicides, fungicides and insecticides (including varroacides and one insect repellent), and some of their metabolites were detected, of which 31 substances have systemic properties (Table 1). At "fruit" 58 different substances were detected in only three sampling years, followed by "grain" with 37 and "meadow" with 24 different substances in five sampling years. At all sites and years fungicides account for more than 50% of the detected substances (Table 2). Generally, the least intensive chemical crop protection activity was identified at the "meadow" site, considering number of detected pesticides, number of pesticide detections and maximum pesticide concentrations. Most of the samples free of pesticide contamination were collected at "meadow" (43.9%) with a mean pesticide load of 2.0 pesticides per sample, followed by "grain" (13.3%) and 3.8 pesticides per samples (Fig 1). All pollen samples collected at the "fruit" site contained, on average 9.8 pesticides per sample. The lowest maximum concentrations were measured at "meadow", followed by "grain" and further exceeded by "fruit" with 294.6, 1,496.4 and 7,177.7 μg/kg, respectively. (Table A-C in S3 Tables  and S4 Tables). In all years at "meadow" and "grain", the months April and May revealed the majority of contaminated samples as well as most pesticide detections, whereas most of the uncontaminated or less contaminated samples were found in the summer months June and July. At "fruit" site all months showed high pesticide frequency, with high contamination also in June or July ( Table 2, Fig 1).
Due to the classification of the respective plant protection products according to German authorization, 83.6% of the measured substances are likely to appear in pollen, due to permitted application into the flower. Twelve of the substances found are not supposed to appear in pollen. Eight substances (and some of their metabolites) are classified harmful to adult bees or bee brood (i.e. methiocarb, clothianidin, imidacloprid, dimethoate, fenoxycarb, pirimicarb, indoxacarb, nicotine) ( Table 1).   The "relevant threshold of 50" [31] has been exceeded six times by the pesticides methiocarb and dimethoate at "grain" site (max. 164.41) and twelve times at "fruit" site by the pesticides dimethomorph, fenhexamid, fluazifop and indoxacarb. Clothianidin and imidacloprid exceeded at "fruit" site four times even a threshold of 500 (max. 600; Table 2) during the three years of observation.
Statistical analysis of PHQ values for some pesticide groups (DMI-fungicides, QoI-fungicides and pyrethroids) did not significantly differ between "meadow" and "grain", yet both sites differed significantly in all groups from "fruit" site (Table 3).
Considering the sum of all PHQs per sample and day, which represents the total pesticide load (tPHQ day ), seven samples of "grain" (4.9%), and 13 samples of "fruit" (23.2%) exceed a tPHQ day value of 50, of which values of four days even exceed a threshold of 500 (Table 2). In general, statistical differences were found amongst the three sites but not between years and months (GLM, S 2 = 34.265133, d.f. = 2, 11, p < 0.0001). However, looking at each year, significant differences among sites were found only in 2013 and 2014, with "fruit" site being different from the other sites (2013: F = 6.6353, d.f. = 2, 69, p = 0.0023 and 2014: F = 10.8483, d.f. = 2, 38, p = 0.0002, respectively).
Taking the sum of all fungicides and insecticides into account, the factor site shows significant differences, while year and month do not differ, with the "fruit" site showing a significant heavier pesticide load compared to both other sites (GLM, S 2 = 34.974217, d.f. = 2, 11, p < 0.0001 and GLM, S 2 = 18.83227, d.f. = 2, 11, p < 0.0001, respectively).

In depth residue analysis of color-sorted pollen pellet fractions
In depth analysis of nine previously analyzed samples (each corresponds here to a composite sample) revealed pesticide contamination caused by predominantly one plant species. Each of Table 3. Honeybee risk exposure to pesticide groups as expressed by PHQ. PHQ values of single detections in the same sample of the same pesticide group were added (mean ± SEM). One-way-ANOVA, followed by a pairwise Student t-test at α = 0.05.  Table 2 c) Acetamiprid, thiacloprid; clothianidin and imidacloprid were detected only at "fruit" site https://doi.org/10.1371/journal.pone.0199995.t003

Pesticide groups a
Pesticide residues in pollen loads of honeybees the nine chosen samples was sorted according to pollen pellet color resulting in five to twelve fractions of plant species (correspond here to fraction). In many cases, active substance concentrations in the fractions exceeded the concentrations found of the same substances in the composite sample (e.g. up to 1,600 x for fenhexamid in the sample from June 14, 2012) ( Table A-I in S5 Tables). Some substances were detected in low concentrations in the fractions that were not found in the composite samples before. Frequently, the same substances are detected in multiple fractions and varying widely in the range of their concentrations. The  Pesticide residues in pollen loads of honeybees range of concentrations is shown in a box-and-whisker-plot exemplarily for one date (06.06.2013, "fruit" site), for pesticides being detected in at least eight sub-fractions (Fig 3). Furthermore, single crops, such as Brassicae sp. or grapevine (Vitis vinifera L.) show high pesticide pollution (Figs 4 and 5). Some tree or weed species, such as false acacia (Robinia pseudoacacia L.), maple (Acer L.) or Ranunculus L. or dandelion (Taraxacum F. H. Wigg.), were also contaminated, even though in lower concentrations. However, in some commonly bee pollinated weed species, such as Achillea L., high concentrations of pesticides could be detected (Fig 6).

Discussion
This long-term study provides for the first time detailed information on the daily pesticide intake into honeybee colonies through pollen foragers during the agronomic active season. We thereby compared three agricultural landscapes with different crops and different chemical crop protection intensity over a period of five years. We were able to show the changes during the seasons, and the range of pesticide concentrations within the months, as well as the changing pesticide combinations. Remarkable differences in residue levels between sites depending on the agricultural intensity were found. Surprisingly at the "meadow" site with huge portions of grassland and meadow orchards we found different pesticides in all observation years. However, in comparison to the other sites it showed the lowest intensity of pesticide use. We found many samples without any detectable contamination and the few positive samples only with low concentrations of pesticides. The "grain" site shows already a higher pesticide use with greater portions of grains and maize in the surrounding fields. More substances and higher concentrations of pesticides were measured, especially in spring months. Finally, the "fruit" site, with high portions of permanent crops, such as grapes, is the site with the highest pesticide loads, throughout all months and years of monitoring and 100% contaminated samples, although this site was sampled only three years.
Even though we found the highest concentrations at "fruit", in comparison with other studies, these concentrations are still on a low level. The highest concentrations found in our study, also being in mg/kg, were up to 14   Pesticide residues in pollen loads of honeybees evaluation by Johnson et al. [22] who affirms that it is not unusual to find mg/kg residue levels in hive matrices or collected goods of honeybees when foraging in conventionally farmed land or as pollinators in monocultures with no alternative flowers.
Consequently, PHQ calculations reflected the same image. The higher the cultivation intensity of a field, the more pesticides are used, considering frequency of applications and amount of different substances used. Hence, PHQ are higher at sites with more chemical crop treatments, due to higher concentrations and a higher frequency of comparatively more bee toxic substances. Accordingly, the concentrations measured at the "meadow" site and the toxicity of the detected substances resulted in lower PHQ values, not even coming up the relevant threshold of PHQ = 50. PHQ of 50 was exceeded only at "grain" and fruit", when substances are categorized harmful to honeybees (methiocarb, dimethoate, clothianidin, imidacloprid) or when measured concentrations were in mg/kg (fluazifop-butyl, dimethomorph, fenhexamid). In general, PHQ values calculated from our pollen samples, are low compared to other studies. Here, PHQs of 500 are exceeded only four times, with 600 being the highest. Traynor et al.
[24] calculated in bee bread samples from citrus a PHQ of > 2,000, McArt et al. [57] calculated from beebread from apple orchards a PHQ > 4,000, and Stoner and Eitzer [31] calculated maximum PHQ values of > 40,000. However, to estimate the possible toxicity for bees, PHQs are only estimates. Stoner and Eitzer [31] were concerned that the picture might be misleading, as honeybee larvae are exposed to more pollen in a very susceptible stage of development [35]. For example, the insecticide fenoxycarb has a low toxicity for adult bees (LD 50 = 200 μg/bee) which sums up for a low PHQ. Yet studies have shown, that this pesticide is very toxic when ingested by larvae [58][59][60][61]. Also, Traynor et al.
[24] noted, that PHQ calculations are only simplistic reflections, as pesticide interactions, such as synergism or antagonism are not taken into account, as well as potential metabolism or detoxification of the substances. For this reason, Sanchez-Bayo and Goka [62] included also synergistic effects and frequency of pesticide occurrence into their risk calculations.
Synergism or antagonism of substances may occur, when different substances are exposed to honeybees at the same time. Here, too, we found multiple different substances in one sample, adding up to dozens of different substances during the year inside the hive, as other authors reported previously [23,24,33,34]. Negative side-effects have been proven for e.g. tank mixtures of pyrethroids with sterol biosynthesis inhibiting fungicides (SBI-fungicides) or SBI -fungicides and neonicotinoids [63][64][65][66][67][68].
In addition to synergistic or additive effects, side-effects by sublethal pesticide concentrations are possible as well. Even though we have shown, that pesticide concentrations exceeded mg/kg in few samples, pesticide detections occurred in sublethal concentrations. Sublethal effects on pesticides, i.e. insecticides have been investigated in honeybees or other pollinators. It is known, that behavior and cognitive performance is impaired, such as learning, memory, homing or flight behavior [36,37,[69][70][71][72][73]. In addition, there are changes in biochemistry, growth and development, gene expression [39][40][41][42]74,75] or social interactions [38,76]. Even though most effects are determined in worker bees, effects on queen fecundity or sperm viability are reported, too [77,78].
However, as eusocial honeybee colonies are considered to be a superorgansm, they are able to tolerate or buffer stressors such as pesticides or sublethal effects better in comparison to solitary living pollinators [79]. Wild bees use few pollen pellets for provisioning single larvae and these consume pollen directly without a nurse bee in between. Nonetheless, residue analysis of single color fractions revealed sometimes even higher concentrations in certain plant species in comparison to the whole daily collected pollen sample. Generally speaking, pollen of single crops, such as grapes (V. vinifera) or Brassica sp. show more pesticides and higher residue values. Furthermore, pesticides found in pollen pellets of trees such as horse chestnut (Aesculus hippocastanum L.), maple (Acer) or false acacia (R. pseudoacacia) or wild flowers and companion plants, respectively, indicate applications during flowering or drift. In addition, from high pesticide concentrations in weed species, such as Achillea, direct applications in the field, aiming probably towards fungal diseases in grains and Achillea may be inferred. Hence, solitary bees, that only use small amounts of a certain pollen source for the provision of their nests, are generally more endangered in such intensively cultivated agroecosystems [9,80]. Wild bees not only contribute to biodiversity but are also important indicators of environmental health and should not be harmed by pesticides [5,81]. Therefore, not only honeybees but other species of bees should be taken into consideration when crop protecting substances are tested for registration and risk assessments of pesticides should be revised [82].
Not only in the daily collected samples but also in their color sorted fractions, we found 12 substances that are either categorized harmful to honeybees or are not supposed to be exposed to honeybees. For example, picaridin is not a crop protection agent, rather than a common ingredient of commercially available insect repellents. On inquiry, the beekeeper confirmed using a repellent to keep away ticks when working outside at the apiary. This substance on legs and arms probably could contaminate the pollen load during pollen collection. However, this compound is not known to be harmful to honeybees [83]. Another example is seed coating, where an exposure to bees is not expected. However, during sowing process, substances used for seed treatments may be abraded, emitted by machinery and cause dust-drift to adjacent flowering plants [12,84]. This could explain the frequent detection of methiocarb, fuberidazole or pencycuron in our spring samples at the time of sowing. Two other substances, coumaphos and thymol, are registered as varroacides for in-hive applications. These substances might volatilize inside the hive and settle on the hairs of the honeybee. When combing through the hair while making pollen packages the substances could be incorporated into the pollen loads. Toxic neonictinoids, i.e. clothianidin and imidacloprid, have been found at the "fruit" site as well a few times with a maximum concentration of 2.4 μg/kg and therefore also below the level considered for acute honey bee toxicity. Our data do not reveal a clear source for the contamination of pollen with these systemic pesticides. As these neonicotinoids are persistent in soil, water and non-target plants, the accumulation in the environment from earlier use cannot be excluded [19,85,86]. Pollen from oilseed rape may be contaminated. However, clothianidin was not found in Brassica sp. sub-fractions in our analyses. Dust abrading from seed treated fodder beet or beet root, which is sown at that time, seems unlikely due to abrasion-proof seed pilleting, but can also not be excluded. We can only speculate, from which plants these detections originate, and further in-depth analyses should be conducted. As clothianidin is registered as spray application in some crops, an unintentional contamination of flowering weeds or shrubs is imaginable. Thus, single contaminated pollen loads might have contaminated the composite sample.
"Cocktails" of pesticides in pollen might raise concern for human health, too, as pollen pellets are used as nutritional supplements [87]. Looking at maximum residue limits for apicultural products, single samples at every site exceeded the limits for some substances leading to an exclusion of the respective pollen samples from further sales. MRL are based on residue data when pesticides are applied correctly according to use instructions [54]. Nonetheless, directly collected pollen by honeybees in the field from treated plants were not considered when thresholds for apicultural products were defined. Therefore, the current agricultural practice makes it impossible for beekeepers to merchandise pollen collected in intensive landscapes-independent of whether there is an impact on human health or not.
Nonetheless, concluding from the majority of concentrations and pesticides found, we assume no misuse of pesticides by the farmers at our three sites and in the observation period, which would lead to direct lethal effects. However, further improvements of seed coatings or sowing technique should be considered. In any case, flowering shrubs, weeds or ruderal plants should be considered while applying chemicals, as well as drift of substances to flowering plants nearby.
The study at hand supports results of other researchers handling pesticide residues in pollen or bee bread samples. We further contribute more details on changes on daily and monthly levels of pesticide exposure. As the detected pesticide concentrations are in sub-lethal levels, we find no reason to believe, that the data published here are to be blamed for higher winter colony losses. Nonetheless more research is necessary in this regard, as they raise concern on chronic effects of these sub-lethal concentrations as well as effects of pesticide mixtures or interactions of pesticides with diseases or pathogens. Some studies already dealt with these questions [45,[88][89][90][91] even though most investigations still engage in only one endpoint and not in a combination of all three aspects. However, further research needs to be done in order to assess possible risks of chronic ingestion of pesticide mixtures in sub-lethal concentrations on honeybees and other pollinators. Our study gives detailed information to be used to design experiments based on field realistic data.