Planting of neonicotinoid-coated corn raises honey bee mortality and sets back colony development

Experimental setting

Field trials took place in the southern part of Quebec, in Eastern Canada in a region with a historically high level of land use for agricultural purposes. This area is Quebec’s most productive agricultural region and produces over 60% of the province’s corn and soybean crops. Study sites (apiaries) were selected among seasonal commercial bee yards customarily and historically managed by beekeepers. Exposed apiaries (n = 9 for 2012, n = 7 for 2013) were located within a 500 m radius from a treated-corn field, within a broader landscape (3 km radius) dominated by neonicotinoid-treated corn fields (exposed treatment). Control apiaries (n = 3 for 2012, n = 7 for 2013) were located in an area where the main crops are hay, wheat, oats, barley and rye and with minimal distance of 3 km from any crop planted with neonicotinoid seed coating. Minimal distance between same treatment sites was 3 km and 17 km in between exposed-control sites (Fig. 1).

Honey bee colonies

Commercial honey bee hives were placed at every study site by their respective beekeepers/owners according to the usual seasonal timing of such procedures, either before, or at the very beginning, of large-scale corn planting operations. All colonies remained on site for the duration of the experiment and were then moved for pollination purposes by owners. All study sites were factual customary bee yards locations and colonies were entirely managed by owners. However, colony management was standardised between beekeepers: all colonies (Apis mellifera hybrid Italian stock) were reared in Langstroth hives, treated with oxalic acid for varroa mite control and supplied with ad libitum pollen supplements. In each apiary, five colonies were randomly selected from among those with similar colony strength at the beginning of the experiment (five frames of bees and four frames of brood). In 2012 and 2013, four beekeepers managed colonies exclusively in exposed apiaries and two beekeepers managed colonies in both exposed and controled apiaries, for a total of six beekeepers per year.

Mortality index

A one square meter (1 m²) piece of pale beige cloth was placed in front of each experimental hive. Every 48 h, dead honey bees found on the cloth were counted to evaluate mortality levels and then removed. In 2012, the sampling period went from May 3rd to June 15th; in 2013, from May 6th to May 24th. By definition, mortality rate is the measure of the precise number of deaths in a known population, per unit of time. Although colony strength can be estimated, the exact number of living bees it comprises cannot be known, especially in daytime when foragers are out of the hive. Similarly, the number of dead bees cannot be established precisely since honey bees exhibit necrophoric behaviour to limit the spread of disease and parasites and to avoid attracting predators (Visscher, 1983). Furthermore, estimating mortality based on the number of dead bees in front of the hive does not take into account the numerous deaths that might occur away from the hive, for example, when bees die in the field after a lethal exposure to pesticides or when they become disoriented and fail to return to the hive (Porrini et al., 2002). In fact, most of the mortality happens away from the colony and dead bees found in front of the hive only account for 1 to 20% of all deaths (Johansen & Mayer, 1990). As such, the estimate of mortality is referred to subsequently as a “mortality index” and acts as the best proxy for the comparison of mortality rates, while still vastly undervaluing honey bee losses. When mortality exceeded 100 dead bees per colony (minimum amount required for chemical analyses), specimens were collected for ulterior residue analyses.

Chemical analysis

Dead honey bee samples were analyzed using a modified version of the QuEChERS method originally described by Anastassiades et al. (2003). Briefly, 15 ml of acetonitrile was added to 10 mg of a previously ground and homogenized sample of honey bee bodies. Extract was shaken vigorously for 1 min using a Vortex mixer at maximum speed. Afterwards, 0.6 mg of anhydrous magnesium sulfate and 1.5 mg of anhydrous sodium acetate were added to the extract, which was then vortexed for 1 min and centrifuged for 10 min at 3,450 rpm. An aliquot of 1 ml was transferred to a glass vial, and 60 µl of methanol and 20 µl of isoprocarb standard solution (10 ppm) were added. The solution was then strained through a 0.45 µm PTFE filter and 10 µl were analyzed by liquid chromatography/mass spectrometry utilizing a Waters Acquity LC interfaced to a Waters Xevo TQ MS (Halo C-18 columns, 4.6 solid core × 50 mm porous outer shell with 2.7 µm particle). The mass spectrometer was positioned in a positive electrospray mode and utilized a different MS/MS scan for each of the pesticides monitored. Liquid chromatography injections were repeated three times. Parent pesticides and metabolites were identified based on comparisons of their chromatographic retention time with known standards and mass abundance ratios to at least two fragment transitions. The ion ratios between the two transitions had to comply with a maximum difference of 20% with the calibration standard. This multi-residue method allows for detection of over 400 agrochemical compounds at parts per billion concentration levels. Analyses were performed by a team of professional chemists from the Laboratory of food analyses at the Quebec Ministry of Agriculture (MAPAQ).

Intracolonial population dynamic model

The commercial beekeeping context in which this study was carried did not allow for controlled conditions after corn planting as colonies were immediately moved for pollination services of different crops (e.g., lowbush blueberry, cranberry), under different climates, at varying distances from our observation locations and for different durations. Therefore, population modeling was used to simulate colony dynamic. The model is based on the mathematical models of Khoury, Myerscough & Barron (2011) and operates under different scenarios based on the initial size of the colony’s population, queen’s egg-laying rate and mortality levels while only considering the population of female worker bees. In all simulations, the evolution of a typical colony was considered during the first three months of the beekeeping seasons and encompasses the corn planting period in our area (roughly the whole month of May) and the blooming period of the two most important pollinator-dependant crops. The simulation-modelling period starts when colonies emerge from their wintering confine, around mid-April. Population size is usually between 10,000 and 14,000 female worker bees, i.e., the average colony size at the beginning of the season in our study area. Forager bees were set to account for 25% of the colony’s population (Fahrbach & Robinson, 1996). The daily egg-laying rates were set to three realistic levels, namely a rate allowing for fast (1,750 eggs/day), typical (1,500 eggs/day) and slow (1,250 eggs/day) colony development (Winston, 1991). Natural forager death rate was set to 0.08 and 0.14 forager bees per day as reported in scientific literature (Rueppell, Kaftanouglu & Page Jr, 2009; Russell, Barron & Harris, 2013; Betti, Wahl & Zamir, 2014). The term w reflects the emergence rate from brood to adult and was set to 16,000 to simulate a fast-growing colony typical of spring and early summer (Cresswell & Thompson, 2012). Other dynamic parameters remained unchanged from Khoury, Myerscough & Barron (2011) initial model. Simulations were run under the hypotheses of (1) typical and constant forager death rate with no or insignificant foragers’ exposure, and (2) forager death rate increased by pesticide exposure during a 30-days corn planting period. In the later configuration, exposed foragers death rate was raised by a factor of 3.51 as determined by our mortality survey (see ‘Results’).

Statistical analyses

The honey bee mortality index data set was analysed using a generalized randomized block design (GRBD) with repeated measures. The experimental units consisted of a group of five colonies located in honey bee yards. GRBD refers to the randomization of study sites within treatments (exposed and control) and within blocks comprised of both study years (2012 and 2013). In this model, treatments and dates were considered as fixed factors, while years and study sites were considered as random factors. Dead honey bees were collected from the same colonies at repeated intervals throughout the corn planting season. Repeated measures design was used to address the longitudinal nature of the data set. The correlation structure for observations within the same experimental unit was selected based on Akaike Information Criterion (AIC). The normality assumption was tested using the Shapiro–Wilk’s statistic, and the homogeneity of variances was verified using the usual residual plots. Box–Cox power transformation was used to remedy violation of the model’s assumptions. Statistical analyses were performed with the lme() function in the nlme package of R software (V 3.0.2) with the significance level set at 0.05.

Honey bee mortality index

Compilation and statistical analysis of numbers of dead honey bees revealed a significant difference (F_1,22 = 19.440897, p = 0.0002) in observed mortality during the corn planting period between apiaries from areas with intensive corn cultivation and those from corn-free environment, respectively 32.6 and 9.3 dead bees per colony per 48 h (Fig. 2). Daily mortality was on average 3.51 times higher for colonies located in the vicinity of corn fields planted with neonicotinoid-treated seeds. Mortality levels were not statistically different between sites or years within same treatment, and did not vary as a function of time (Figs. 3A and 3B) during the corn planting season.

Multi-residue analyses of dead honey bees

Chemical analyses of dead honey bees indicated that they were exposed to various agricultural chemicals throughout the corn planting period. A total of 38 different pesticides and metabolites, 19 insecticides, 11 fungicides, seven herbicides and one synergist, were found in the 74 dead honey bee samples (≥100 specimens) that were analysed throughout the study. In the 16 dead honey bee samples collected in control areas (apiaries in corn-free agricultural areas), three pesticides were identified (thiabendazole, clothianidin and thiamethoxam), with some samples containing two pesticides and an average of 1 chemical per sample (Table 1). Of these three pesticides, the fungicide thiabendazole was detected at levels above the limit of quantification, but levels of the neonicotinoid insecticides clothianidin and thiamethoxam were consistently below the limit of quantification, and therefore well below concentrations known to cause acute bee mortality (Lundin et al., 2015). Analysis of the 58 exposed samples collected from apiaries in neonicotinoid-treated corn intensive areas identified 32 pesticides and metabolites, with an average of 3.48 chemicals per sample and up to nine different compounds in a single sample (Table 2). Visualisation of pesticide detections in the form of a network highlights that the three most frequent pesticide co-occurence are clothianidin-thiamethoxam (unsurprisingly, as thiamethoxam is a known precursor of clothianidin), clothianidin-thiabendazole and thiamethoxam-thiabendazole and the most frequent triads of chemicals, which are clothianidin-thiamethoxam-thiabendazole in 30% of samples and atrazine-desethyl atrazine-metolachlor in 15% of samples (Fig. 4). Pesticide risk assessment based on the acute risk quotient (RQ) indicates that the vast majority of compounds score below the threshold level for concern (0.4) (EFED, PMRA & CDPR, 2012). RQ is expressed as the ratio of exposure estimate, in this case, concentrations of active ingredient found in dead bees, to point estimates of effects, as established by the acute contact and oral lethal dose to 50% of the organisms tested (LD50). For example, if clothianidin’s concentration in a given sample was of 19 µg/kg, its oral LD50 at 24 h is of 0.00368 µg/bee and considering a mean honey bee body weight of 0.128 g (EFED, PMRA & CDPR, 2012), then the corresponding RQ value would be of 0.66 ($19\times \left(0.128÷1,000\right)÷0.00368$). The neonicotinoid clothianidin was the only compound whose acute RQ value exceeded the acceptable limit and is therefore an identified source of concerns (Fig. 5).

Intracolonial population dynamic model

Our model reveals that increased mortality levels linked with the planting of neonicotinoid-treated corn has the potential to strongly impact colony dynamics (Fig. 6). Regardless of the queens’ egg laying rate and typical mortality levels, populations from exposed colonies would follow a marked decline during the planting period of neonicotinoid-treated corn, after which colony growth would start anew. Simulations show that exposure to planting of neonicotinoid-treated corn by itself should not be sufficient to cause the collapse of exposed colonies. However, the worst-case scenario predicts colony collapse before the end of the corn planting period (Fig. 6F). Also, population sizes of exposed and surviving colonies are predicted to be considerably reduced at key moments during the Quebec’s beekeeping season, which are during the lowbush blueberry and cranberry crop pollination service periods. For instance, at the beginning of the lowbush blueberry pollination season beginning around early June, exposed colonies are predicted to be at best 2.33 times (Fig. 6A) and at worst 3.47 times (Fig. 6E) smaller than typical unexposed colonies. As for the cranberry pollination season, usually beginning in early July, colonies are still thought to be affected even if they partially recovered. According to our simulations, predicted populations of exposed colonies would still be 1.65 to 2.16 times smaller (respectively Figs. 6A and 6E) than expected populations of unexposed colonies. After the 100-days period, population sizes of exposed colonies are still predicted to be smaller than expected population of typical colonies by several thousand individuals, between 9,000 (Fig. 6A) to 5,600 (Fig. 6E) fewer bees.

Honey bee mortality index

Throughout the entire corn planting season, honey bee colonies established near neonicotinoid-treated fields showed a 3.51 times higher daily mortality index (Fig. 2). In temperate climates, this increased loss of foragers can be a cause for concern as spring is a particularly challenging time for colonies (Mattila & Otis, 2006; Betti, Wahl & Zamir, 2016). Resumption of intensive brood rearing in late winter and early spring may exhaust the colony’s pollen reserve before additional or sufficient resources become available in the environment (Mattila & Otis, 2006). Increased bee mortality would further impair the colony’s foraging efficiency and intensify pollen shortages. Should that happen, nutritional deficiencies would trigger a decline in brood production, which could even come to a complete halt (Mattila & Otis, 2006). Declining honey bee populations would result in seasonal vulnerability which would leave the colony more susceptible to other hazards (Betti, Wahl & Zamir, 2016).

As anticipated, mean mortality levels of control colonies were particularly stable and did not differ throughout the corn planting season (Fig. 3A). However, contrary to expectations, worker mortality of exposed colonies did not significantly differ as a function of time (Fig. 3B). We initially believed that mortality levels would be higher at the beginning of corn planting and gradually subside during spring as a similar pattern was observed with acetylcholinesterase expression in bumblebees (Samson-Robert et al., 2015). Our results did suggest that mean mortality levels tended to decrease over the season but the magnitude of variability in mortality between days and colonies of exposed sites did not allow for the detection of a significant effect (Fig. 3B). We hypothesize that weather conditions are a key driver of pesticide related bee deaths which would explain the variability observed in mortality levels between days. Since honey bees do not venture outside when it is raining and corn cannot be planted when heavy or frequent precipitation occurs, both pesticides exposure and honey bee deaths should be minimal. When weather conditions become favourable again, corn planting and honey bee foraging resume and mortality should consequently rise. We also hypothesize that proximity of on-going plantings with honey bee hive and bees foraging areas, rather than regional progression of plantings, play a key role in pesticide exposure and honey bee deaths, which would explain the observed variability in mortality levels between colonies of exposed sites.

Multi-residue analyses of dead honey bees

Our results also indicate that honey bees from bee yards established in the vicinity of agricultural fields are not only widely exposed to neonicotinoid insecticides during corn planting, but also to a wide array of agricultural pesticides that play an important role in impairing colony health (Tables 1 and 2). Pesticide residue concentrations in dead bees were generally low and seldom exceeded LD₅₀. However, many pesticides are rapidly metabolized in live bees, which may explain the general lack of high residue concentrations in recently dead bees. Nicotinic insecticides are particularly problematic in that regard since they are known to be metabolized within a few hours (Suchail, Debrauwer & Belzunces, 2004; Suchail et al., 2004; Cresswell et al., 2014; Blacquière & Van der Steen, 2017), making their detection especially tricky. Moreover, metabolites of neonicotinoids, most of which were not targeted by the chemical analyses, have also been shown to be highly toxic for bees (Suchail, Guez & Belzunces, 2001; Nauen et al., 2003; Blacquière et al., 2012) increasing the long-term toxicity of these compounds. Most pesticide detections also revealed the co-occurrence of many active ingredients within the same samples (Fig. 4), which would be expected to exhibit enhanced toxicity if they share the same mode of action or target site (Yu, 2014; Zhu et al., 2014) or if compounds compete for a limited pool of detoxification enzyme (Berenbaum & Johnson, 2015; Johnson, 2015). Neonicotinoids thus pose a particular threat, as they are known to be additively toxic when they occur together (Andersch, Jeschke & Thielert, 2010; Pavlaki et al., 2011; Palmer et al., 2013), this was the case for 31% of our samples (Fig. 4). Neonicotinoids are also known to interact additively with various other agrochemicals. As such, combined neonicotinoid exposure to certain pyrethroids, like λ-cyhalothrin, has been related to increased toxicity (Gill, Ramos-Rodriguez & Raine, 2012; Chen et al., 2015). Combined exposure with fungicides like boscalid (Tsvetkov et al., 2017), trifloxystrobin (Wachendorff-Neumann et al., 2012) and ergosterol inhibitor (Iwasa et al., 2004; Thompson et al., 2014; Johnson, 2015; Sgolastra et al., 2017) also result in synergistic toxicity. In the last six years, 12 patent applications have also claimed or demonstrated additive toxicity between the fungicide thiabendazole and thiamethoxam (Donley, 2016), which occurred together in 30% of our samples. Herbicidal compounds have also been shown to be additively toxic when occurring together with neonicotinoids, as was the case with atrazine in 19% of our samples. Finally, combined exposure with in-hive acaricides, such as coumaphos, results in enhanced toxicity in addition with impaired memory and olfactory learning (Williamson & Wright, 2013; Berenbaum & Johnson, 2015; Johnson, 2015). Of all the insecticides identified in dead honey bee bodies, those of the neonicotinoid family, clothianidin and thiamethoxam, were detected far more frequently than others. Furthermore, clothianidin was the only pesticide whose acute RQ exceeded the threshold for concern (Fig. 5) and thereby poses the highest risk for honey bees, which is in accordance with the findings of Sánchez-Bayo & Goka (2014). Taking into account (1) the known high toxicity of nitroguanidine neonicotinoid insecticides (Goulson, 2013); (2) concentration levels found in dead bees in our study; (3) the possible synergistic interactions with other agrochemicals identified in collected dead bees; and (4) the synchronism between the timing of honey bee deaths and concurrence with planting of treated corn in the vicinity, our study concludes that exposure to neonicotinoid compounds, mostly clothianidin, was the most likely cause of the bee mortality observed at the hives during our study.

Impaired colony growth and contributions to collapse

Unusually high spring forager mortality (Fig. 2) might represent a heavy burden for colonies, assuredly rendering them more vulnerable to other threats and potentially causing their collapse. When implementing the increased in-field mortality into a honey bee population dynamic model, all scenarios exhibited a major deviation from the expected colony dynamic (Fig. 6). When considering the 100-day period, all exposed colonies suffered an important developmental setback, which has the potential to, in extreme scenarios (Fig. 6F), cause the collapse of the colonies. In most cases however, simulations show that colonies should survive the increased mortality and recover over the course of the season, given the proper care and favourable conditions. Nonetheless, colonies seldom encounter such ideal conditions and in reality must usually deal with several pathogens, parasites and predators (Potts et al., 2010; Vanbergen et al., 2013; Goulson et al., 2015; Gill et al., 2016; Klein et al., 2017). In bad-case scenarios (Figs. 6D and 6E ), when the corn planting period ends, colonies are expected to dwindle just above 5,000 individuals, which represents the lowest viable colony size (Henry et al., 2012; EFSA, 2013). With such small populations, even a slight disruption of the colony, caused by any type of stress (dietary, parasites, pathogens, weather, transhumance, etc.), might be sufficient to cause its collapse. Recently, Henry et al. (2015) confirmed that field exposure to neonicotinoid was indeed associated with increased forager mortality. They also concluded that colonies compensate for the excess mortality without impairing colony growth or honey production. These results appear to be in contradiction with our own conclusions. However, Henry et al. colonies were exposed to oilseed rape grown from seeds treated with neonicotinoid and research on honey bees have repeatedly shown little adverse effects at colony-level in this particular crop (Cutler & Scott-Dupree, 2007; Pohorecka et al., 2012; Pilling et al., 2013; Cutler & Scott-dupree, 2014). Being particularly attractive to honey bees, it seems counter intuitive that this crop would not adversely impact colonies. However, oilseed rape produces abundant sugar-rich nectar (Kevan, Lee & Shuel, 1991; Pernal & Currie, 1997) and good quality pollen (Somerville, 2001) allowing for fast and sustained colony development. We believe these benefits outweigh the costs of exposure to neonicotinoid residues from treated seeds.

Pollination services

A recent study by Stanley et al. (2015) showed, for the first time, that exposure to neonicotinoid pesticides induced colony-level reduction in pollination services provided by bumblebees to apples. In the case of honey bees, impaired colony development as shown in our population dynamic models may negatively impact both pollination services and honey production, resulting in important economic losses for beekeepers and berry growers in Quebec. For one, to be economically profitable for beekeepers, or even considered for pollination services, honey bee colonies need to meet population size standards in Quebec (FAQ, 2015). Immediately following chronic exposure from corn plantings, predicted honey bee populations of bad-case scenarios (Figs. 6D and 6E) would not meet the minimal required standard of 7,000 bees and could simply not be rented for pollination services in Quebec. Even mid-summer blooming crops would potentially deal with the after effects of this chronic exposure, as predicted population of colonies would still only be a fraction of the required strength. One common way of dealing with weak colonies is to combine them together so the resulting colonies, although fewer in numbers, would then have the required strength for commercial pollination. As chronic exposure to neonicotinoid compounds reduces the number of honey bee colonies available for rent, pollinator-dependent crops might suffer from this pollination service shortage. For example, lowbush blueberry and cranberry crops are two of Quebec’s fastest growing productions and annually monopolise 90% of the colonies on rent throughout the province (ISQ, 2014). Over the last 10 years, honey bee colony rental prices for crop pollination has increased by more than 60% due to the increasing pollination demand from these two crops (Belzile & Li, 2014; ISQ, 2014). Chronic exposures to toxic chemicals thus forces beekeepers to deal with weakened colonies, more vulnerable to parasites and pathogens, worth less for pollination services, sometimes not rentable at all and in all cases producing less honey. Crop growers also bear the consequences as weaker bee colonies are not as efficient at pollinating their crops and rental prices are likely to increase if supply of bee hives becomes a negative issue.