Effect of acute pesticide exposure on bee spatial working memory using an analogue of the radial-arm maze

Pesticides, including neonicotinoids, typically target pest insects by being neurotoxic. Inadvertent exposure to foraging insect pollinators is usually sub-lethal, but may affect cognition. One cognitive trait, spatial working memory, may be important in avoiding previously-visited flowers and other spatial tasks such as navigation. To test this, we investigated the effect of acute thiamethoxam exposure on spatial working memory in the bumblebee Bombus terrestris, using an adaptation of the radial-arm maze (RAM). We first demonstrated that bumblebees use spatial working memory to solve the RAM by showing that untreated bees performed significantly better than would be expected if choices were random or governed by stereotyped visitation rules. We then exposed bees to either a high sub-lethal positive control thiamethoxam dose (2.5 ng−1 bee), or one of two low doses (0.377 or 0.091 ng−1) based on estimated field-realistic exposure. The high dose caused bees to make more and earlier spatial memory errors and take longer to complete the task than unexposed bees. For the low doses, the negative effects were smaller but statistically significant, and dependent on bee size. The spatial working memory impairment shown here has the potential to harm bees exposed to thiamethoxam, through possible impacts on foraging efficiency or homing.

to ingest 37.7mg (≈µl) of the nectar collected during an hour on oilseed rape. This equates to 0.091ng 23 of active ingredient per bee at a 2.4ppb concentration and 0.377ng at 10ppb. This is within the range 24 estimated for one hour of foraging in honeybees (at 10-40% sugar concentrations; EFSA 2012). 25 Feeding 37.7µl of sucrose solution to a bee prior to the task is likely to lower motivation, so this volume 26 was halved and the concentration of sugar and pesticide doubled to keep the amount of active ingredient 27 received by each bee the same. Bees were therefore fed 18.85µl of 43% (w/w) sucrose solution 28 containing either 0ppb (control), 4.8ppb (0.091ng per bee), 20ppb (0.377 ng per bee) and 133ppb (2.5 29 ng per bee) thiamethoxam. As in previous studies into acute effects of pesticides, the full dose was 30 provided in one feed (Henry et al. 2012;Stanley, Smith & Raine 2015). 31 To obtain the required concentrations, we dissolved 100mg thiamethoxam (C8H10ClN5O3S powder; 32 PESTANAL ® analytical standard, Sigma Aldrich, Poole, UK) in 100ml distilled water to produce a 1 33 ppt stock solution. Each fortnight aliquots of 1.2µl, 5µl and 33.3µl of the stock solution were diluted 34 with 250ml 43% Brix sucrose solution to produce 4.8ppb, 20ppb and 133ppb thiamethoxam solutions 35 respectively. 36

Radial Arm Maze (RAM) design 37
The RAM tests working spatial memory by requiring animals to remember which reward locations they 38 have visited and avoid revisits (Foreman & Ermakova 1998). The original RAM was designed for 39 rodents, for which a central chamber is appropriate to prevent animals moving from one reward location 40 to the next in a circle; i.e. it reduces (but not eliminates) the use of stereotypical behaviour so that spatial 41 working memory can be better identified. However, this approach is not perfectly suited to all animals; 42 in particular, flying animals may behave unnaturally in an enclosed arm set-up. As  flowers/inflorescences (Pyke & Cartar 1992), so our RAM apparatus that requires bees to fly attempts 47 to represent a more ecologically relevant foraging decision than one where bees walk through maze 48

arms. 49
The baffles between flowers fulfil a similar role to the rodent RAM central chamber in greatly reducing 50 but not eliminating stereotypical behaviour (bees can fly over baffles, but cannot move directly from 51 flower to flower), and preventing bees from seeing any other flowers when on a particular flower, 52 meaning that each represents an independent reward location (like arms of the rodent RAM) and that 53 bees fly out of the array to see and subsequently visit the other flowers, to some degree mimicking the 54 return from an arm to the centre of the rodent RAM (Olton & Samuelson 1976). The apparatus was 55 located in a cue-rich laboratory environment with constant lighting during testing; cues such as arena 56 walls and baffles did not differ between treatments and no additional landmarks were provided in the 57 foraging arena as cue use was not being explicitly tested. 58 Behaviour on the RAM 59 The vast majority of the time, bees fed on the full 10µl of sucrose solution once the proboscis made 60 contact. On the rare occasions some solution remained, the flower was exchanged for a clean, empty 61 one as usual once the bee had left. A revisit to the most recent flower visited was only counted if >20 62 seconds of flying occurred between visits. Exchanging visited flowers for clean ones was done from the 63 back of the maze while the bee was feeding on its next flower to minimise disturbance. 64 For the first bout, 20µl droplets of sucrose solution were used to increase motivation; for the subsequent 65 nine training bouts and the final testing bout 10µl drops were used to ensure all flowers could be visited 66 before satiation. On the last flower, the experimenter increased the size of the drop to allow the bee to 67 fill its crop and return to the colony (Burmeister, Couvillon & Bitterman 1995). 68 Access to the nest box was blocked while a bee was in the arena to encourage it to visit all eight flowers; 69 however, if the bee made three attempts to return before visiting all eight flowers the entrance was 70 unblocked and the bee allowed to return to the nest box to minimise loss of motivation. The entrance 71 was also unblocked after a bee had been in the arena for longer than 20 minutes; if 30 minutes elapsed 72 the bee was guided towards the entrance with a plastic pot. In the final bout, recording of visits was 73 stopped after 30 minutes, as a longer period in the maze may not be an accurate test of working memory. 74 To administer the pesticide dose after the tenth bout, the bee was intercepted on its way to the arena by 75 manipulating the tunnel doors to direct the bee into a plastic pot (diameter: 60mm). After allowing the 76 bee to acclimatise for approximately one minute, 18.85µl of sucrose solution containing the relevant 77 dose of pesticide was fed to the bee using a pipette inserted through a hole in the pot. 78

Statistical analysis 79
We employed the AIC-IT approach as this does not only test the support for the null hypothesis but also Before the experiment began, 14 bees from a separate colony were tested on the RAM to confirm that 93 bees would interact with the maze and feed from the artificial flowers and to ascertain the number of 94 training bouts needed before asymptotic performance was reached. Ten training bouts were used 95 because the data for both correct choices in first the eight choices and total revisits showed performance 96 reached an asymptote by bout ten (Fig. S2). 97 Six bees were also tested on a partially baited version of the maze to ensure bees were not able to detect 98 the presence of a reward behind the flowers (e.g. by olfaction). Four flowers were randomly selected to 99 be baited and the first four unique (non-repeated) flower visits recorded. Mean number of visits to baited 100 flowers in the first four unique choices was 1.83, which was not significantly different from the chance 101 expectation of 2 (one-sample t-test, t=-0.30715, df=5, p=0.7711). 102

103
Repeating the analysis excluding bees that did not visit all eight flowers (n=13) did not affect the 104 outcome so the final analysis includes all 61 bees. 105

Total revisits 106
Analysing the full dataset but omitting the positive control still showed an effect on total revisits at 107 field-realistic doses. Removing the high dose reduced support for the model containing a size * 108 treatment interaction but increased support for the other models including treatment (Table S2) Table  124 S3). 125

Monte Carlo simulation and behaviour on the RAM 126
Holding the bee in the tunnel for 45 minutes following pesticide exposure did not visibly decrease 127 motivation (bees flew directly to the array on entering the arena) and had no effect on maze performance 128 flowers were visited were included in the total revisits analysis (n=30). Every arrival at a flower ("visit", 138 "revisit", "land" and "approach") was included in the transition probability matrix (Fig. S1a). 139 Stereotypical behaviour improved simulated RAM performance in terms of correct choices before the 140 first revisit and correct choices in the first eight choices (see Fig. 5 in main text), due in part to the 141 moderate contiguity ("nearest neighbour") preference and a low probability of revisiting the flower last 142 visited. However, stereotypical behaviour increased total revisits; the most likely explanation for this is 143 the unequal total frequencies with which each flower was visited (Fig. S1a). The contiguity preference 144 and tendency to travel upwards is consistent with known bumblebee foraging behaviour (Pyke 1978). The success of applying our modified RAM to bumblebees is likely due to their ecological requirements 153 as nectarivores, as the apparatus has been shown to be most relevant in species for whom the test mimics