No evidence for negative impacts of acute sulfoxaflor exposure on bee olfactory conditioning or working memory

Subjects and harnessing

Five bumblebee (Bombus terrestris audax) colonies, each with approximately 150 workers, were purchased (Koppert Ltd, Haverhill, UK) and moved into wooden colony boxes (28 × 10 × 18 cm) connected to flight arenas (100 × 70 × 50 cm) that contained an ad libitum supply of sucrose solution (50° Brix) and pollen. Only individuals that had been observed foraging on the feeder within flight arenas were subsequently used in the experiment (Martin, Fountain & Brown, 2018). Previous studies suggest that bumblebees are more responsive when starved for a period of time (Stanley, Smith & Raine, 2015b), and consequently prior to all PER experiments involving bumblebees, we collected and harnessed all potential subjects before leaving them overnight, and testing them the following morning.

Returning foraging honey bees (Apis mellifera) were collected from the entrance of five hives from a research apiary at Royal Holloway University of London. Honeybee mortality is high when individuals are harnessed for a sustained period of time, and as a result we collected and harnessed honeybees in the same day, leaving them for one hour after harnessing, before randomly assigning them to a treatment group (see below) and conducting the experiment. Bumblebees and honeybees were tested on different days, and on any single test day, sixteen to forty bumblebees and honeybees were collected and harnessed.

Insecticide exposure

Sulfoxaflor has been developed for a range of different crops, including as a seed treatment for bee attractive crops, but its most common application is currently as a foliar spray (Centner, Brewer & Leal, 2018). Foliar spray applications result in short-term bursts of high insecticide residues in the nectar of sprayed crops (United States Environmental Protection Agency, 2016) and any concurrently flowering weeds. We thus based our estimates for acute exposure on data for the residue levels found in honeybee-collected nectar of cotton sprayed with sulfoxaflor from an Environmental Protection Agency (EPA) study, which demonstrated that over an 11 day period nectar concentrations ranged from 5.41–46.97 ppb (United States Environmental Protection Agency 2016; application rate: 0.045 pounds (0.020 kg) of active ingredient per acre, applied twice). We derived our sulfoxaflor treatments from a stock solution of 1 g dm⁻³ sulfoxaflor (Greyhound Chromatography and Allied Chemicals) in acetone, which was combined with sucrose solution (50° Brix) to make three treatment groups: 2.4 µg dm⁻³ (2.4 ppb), 10 µg dm ⁻³ (10 ppb), 250 µg dm⁻³(250 ppb; positive control) and the negative control (sucrose with acetone only). Before training, we placed each bee horizontally (held in place with modelling clay) and pipetted a 10µl droplet of sucrose solution containing the randomly-assigned treatment solution onto a plastic surface, from which the bees could feed. Bees that did not immediately drink were encouraged to extend their proboscis by antennal stimulation with sucrose. Bees that did not consume the full quantity of sucrose solution were not used in the experiment (excluded bumblebees N = 55, control = 13, 2.4ppb = 16, 10 ppb = 16, 250 ppb = 10; honeybees N = 17, control = 2, 2.4 ppb = 6, 10 ppb = 5, 250 ppb = 4). After feeding, the bees were placed upright and left for an hour (Stanley, Smith & Raine, 2015b).

Training protocol

We used an absolute conditioning proboscis extension reflex (PER) procedure in which lavender scent (conditioned stimulus; CS) was forward paired with antennal stimulation by sucrose solution (unconditioned stimulus; US; 50° Brix). The subjects were placed three cm away from the odour tube that contained filter paper soaked in 4 µl of the lavender essential oil. A programmable logic controller computer was used to blow a constant stream of air containing the odours towards the subjects from the odour tube. The odour tube was replaced every 20–30 trials to ensure that the odour was consistently strong throughout conditioning. Bees were exposed to 5 s of clean airflow (no odour), followed by 10 s of the odour. Six seconds after the start of odour exposure, the subject was presented with 0.8 µl of untreated sucrose solution (50° Brix) from a syringe. A positive response was recorded if the bee extended its proboscis in the first six seconds of odour presentation, before antennal stimulation with the US, and was always rewarded by immediate delivery of the sucrose solution. In the event of a negative response, we additionally recorded whether the bee responded to the antennal stimulation (to ascertain that the subject was motivated to extend its proboscis). Each subject received fifteen trials with an inter-trial interval of approximately 12 min. To ensure that the bees were learning about the odour and not other aspects of the experimental protocol, three non-scented probe trials were randomly distributed between the 5th and 15th learning trials. Bees that responded to the unscented stimulus in any probe trial were not included in the analysis (excluded bumblebees n = 10; honeybees n = 1). Each animal thus received 18 trials in total (15 test trials and 3 probe trials).

Medium- and long-term memory tests, whereby the subjects were presented with the conditioned odour in isolation for a single trial, were conducted with the same subjects 3 h and 24 h after the last learning trials, respectively. Once the experiment was finished, bees were frozen and their size recorded by measuring thorax width with electronic callipers (Mitutoyo), three times, from which a mean value was taken. We recorded size because it may influence the rate at which the insecticide is absorbed; larger bees empty their gut at a faster rate (Fournier et al., 2014) and previous studies have correspondingly found size-dependent effects of acute insecticide exposure on cognition (Samuelson et al., 2016).

In total, we tested 240 bumblebees and 174 honeybees. Bees that did not extend their proboscis in response to antennal sucrose stimulation in at least 5 learning trials were not used (bumblebee N = 64, control = 17, 2.4 ppb = 14, 10 ppb = 16, 250 ppb = 17; honeybee N = 6, control = 1, 2.4 ppb = 1, 10 ppb = 2, 250 ppb = 1). A further 3 bumblebees were removed from the experiment because they extended their proboscis before the odour was presented. Five bumblebees died, as did 46 honeybees. One bumblebee was harnessed poorly and thus not included, as were 10 honeybees. This resulted in final sample sizes of 102 bumblebees and 94 honeybees (bumblebees: control = 23, 2.4 ppb = 26, 10 ppb = 24, 250 ppb = 29: honeybees: control = 29, 2.4 ppb = 22, 10 ppb = 22, 250 ppb = 21).

Statistical analysis

We followed an information theoretic model selection approach. The initial model set included a full model and all subsets, including a null model that contained solely the intercept and “Colony” as a random factor. We selected a 95% confidence set of models based on Akaike weights derived from AICc values. In cases where the 95% confidence set contained more than one model, the models were averaged (Burnham & Anderson, 2002) (including the null if it was included within the confidence set) to produce parameter estimates and 95% confidence intervals. Data collected for bumblebees and honeybees were analysed separately due to potential differences between the species (see Siviter et al., 2018b).

Following Stanley, Smith & Raine (2015b), we analysed three dependent variables to identify sulfoxaflor effects on PER performance: (i) whether the bee responded to the CS in the absence of antennal stimulation (hereafter: “positive response”) in at least one trial overall (ii) the total number of positive responses (hereafter learning level) from bees that learnt the association, and (iii) the trial that the bee first exhibited a positive response. We used generalised linear mixed effect models with binomial or Poisson error structures, or mixed effect Cox models, respectively, where treatment, bee size and their interaction were specified as fixed factors, and colony as a random factor (see Tables S1 & S2). For medium- and long-term memory, we analysed whether or not the bee exhibited a positive response to the CS following the same method (binomial error structure). We used the packages, lme4 (Bates et al., 2015), MuMin (Barton, 2016), Coxme (Therneau, 2018), Hmisc (Harrell & Dupont, 2018) and pscl (Jackman, 2017).

Subjects

Seven bumblebee colonies (B. terrestris audax), each with approximately 150 workers, were obtained from Biobest (Agralan Ltd, Swindon, Wiltshire, UK) and upon arrival each was transferred into a plastic bipartite nest box (28 × 16 × 10.5 cm, with a central divider that allowed access between compartments). When transferring bees into the nest box individual bees were tagged with unique number disks, allowing the identification of individuals. During experiments, the nest box was attached to the radial arm maze (RAM; description below), with access controlled using sliding trap-doors. When the bees were not being tested, gravity feeders were placed in the RAM with an ad libitum supply of 43° Brix sucrose solution. Colonies were provided with approximately 7 grams of pollen in the nest box 3 times a week. Colonies were used in succession rather than simultaneously, and newly emerged bees were tagged daily throughout the experiment.

Radial arm maze

A radial arm maze classically consists of 8 arms, each of which contains a hidden food reward (Foreman & Ermakova, 1998). Animals forage within the maze and search for the food rewards, whilst avoiding re-visiting arms that they have already depleted, and Samuelson et al. (2016) have previously confirmed that bumblebees use working memory to minimize such revisits. We based our design on the set-up used by Samuelson et al. (2016) but modified their original vertical design to create a horizontal version. The aim of this modification was to reduce reliance on learnt movement rules by forcing subjects to return to the centre of the maze between choices, as is usually the case for rodent versions of the RAM (Olton & Samuelson, 1976; Foreman & Ermakova, 1998). Our horizontal maze was constructed from acrylic plastic, sealed together with non-toxic grey silicone (Bondit). Each of the 8 arms contained a removable platform (7.2 × 2.6 × 0. five mm) upon which the bees could land to access a small hole in the wall. By extending the proboscis through this hole, bees could access a sucrose reward (43° Brix) that was not visible from the platform (volume varied between stages; see below). After visitation, the platform could be rapidly replaced with a clean alternative to prevent the use of scent marks to identify visited arms. The availability of visual global landmarks (often a view of the laboratory) has been shown to contribute to performance in a RAM for rodents and other animals (Foreman & Ermakova, 1998; Wilkinson, Chan & Hall, 2007), but (a) our laboratory regularly changes in appearance and (b) light control was important for our video software. Thus, our maze walls were opaque, but papered with a black and white panoramic photo of the laboratory to allow bees to orientate.

Stage 1—Group training

The objective of this stage was to identify motivated foragers. Each morning before testing, the RAM was set up with 10 µl 43° Brix sucrose solution on each landing platform. All bees were then allowed into the RAM to forage on the landing platforms (platforms were continuously reloaded with sucrose solution when drained). Only bees that were observed foraging within the maze at this stage (by inserting the proboscis into the holes at the end of each arm) proceeded to Stage 2 (Individual training).

Stage 2—Individual training

The objective of the individual training stage was for bees to learn the win-shift nature of the RAM task, over the course of 10 training bouts. During each bout, bees were required to visit all eight artificial platforms and then return to the nest box to empty their crop. At the onset of each bout, each platform contained 10 µl of sucrose solution (20 µl for the first bout, to increase motivation). Rewards were not refilled after visitation, but landing platforms were replaced with identical but clean replacements. Once the bee found the final reward, we increased the amount of sucrose solution in that arm (from outside the maze) so that the bee’s crop was full, encouraging her to return to the nest box. Choices in the RAM were recorded as either: Correct –feeding from platforms that had not yet been visited, or Incorrect- attempting to feed from platforms that had already been depleted.

If a bee attempted to return to the nest box three times prior to visiting all landing platforms, or if the trial exceeded 20 min, the bee was permitted to return to the nest box. As with Samuelson et al. (2016), each bee completed 10 training bouts.

Stage 3—Pesticide exposure

Our pesticide exposure regime differed from that used in the PER regime because our RAM experiment was designed to allow direct comparison with the results described by (Samuelson et al., 2016) for thiamethoxam. Samuelson et al. (2016) aimed to mimic the dosage received during one hour of foraging for nectar, whilst overcoming the problem that feeding on a large volume of sucrose may reduce a bee’s motivation to participate in the RAM. To that end, bees were only fed half of the volume of nectar that would normally be consumed during such a foraging bout (0.5 × 37.7 mg), with a doubled concentration of sulfoxaflor. To allow for direct comparison, we followed the same approach here (and bumblebees thus received a higher dosage than those in the PER treatment groups described above). Each test subject was intercepted as it was returning to the RAM after emptying its crop following the 10th training bout. They were placed into a plastic beaker, and fed 18.85 µl of sucrose solution from the randomly assigned treatment group. We included four treatment groups, intended to mimic foraging on crops with nectar containing either 0 ppb (control), 5 µg dm⁻³ (5 ppb), 10 µg dm⁻³ (10 ppb) or 250 µg dm⁻³ (250 ppb or positive control) of sulfoxaflor, so bees from each treatment group received 0, 0.045, 0.091 & 2.5 ng respectively. After consumption, the bees were held in the plastic beaker for 45 min before being returned to the nest (Samuelson et al., 2016). 60 bees were originally trained on the RAM but 2 failed to re-commence foraging after the exposure stage (N values, control = 14, 5 ppb = 15, 10 ppb = 15, 250 ppb n = 14).

Stage 4—Test trial

Following exposure, the bees were presented with the exact set up they had experienced in the training phase of the experiment (phase 2) and tested one final time. After completing the task bees were collected and frozen and, at a later date, we measured their thorax width.

Statistical analysis

As with experiment 1, we used an information theoretic model selection approach when analysing each dependent variable and, as in previous work (Olton & Samuelson, 1976; Foreman & Ermakova, 1998; Samuelson et al., 2016), three different measures were chosen to assess performance; (i) total revisits to platforms which have been previously visited, (ii) the number of correct choices made before making a revisit and (iii) the proportion of correct choices in the first eight visits. For all dependent variables, treatment, bee size and the interaction between them were included as fixed factors with colony included as a random factor. To account for overdispersion, we used a generalised linear model with a negative binomial distribution error structure (glm.nb) to analyse total revisits, and a generalised linear model (glm) with a Poisson distributed error structure to analyse the number of correct choices in the first 8 visits. A mixed effect cox model (coxme) was used to analyse correct choices before first revisit. All analyses were conducted in R studio (version 1.1.419) using the R packages lme4 (Bates et al., 2015), MuMin (Barton, 2016), Coxme (Therneau, 2018), AER (Kleiber & Zeileis, 2008), MASS (Ripley & Venables, 2002), Hmisc (Harrell & Dupont, 2018).