Chronic exposure to lead impairs honey bee learning

39 Pollutants can have severe detrimental effects on insects, even at sublethal doses. 40 Agrochemicals have been identified as important causes of pollinator declines, but the 41 impacts of other anthropogenic compounds, such as metallic trace elements 42 contaminating soils and waters, have received considerably less attention. Here, we 43 exposed honey bee colonies to chronic field-realistic concentrations of lead in food and 44 demonstrate that consumption of this single trace element impaired bee cognition and 45 morphological development. Honey bees exposed to the highest lead concentration had 46 reduced olfactory learning performances and developed smaller heads, which may have 47 constrained their cognitive functions. Our results show that lead pollutants can have 48 dramatic effects on honey bee health and may contribute to the widespread decline of 49 pollinators. 50


Introduction
Metallic trace element (MTE) are naturally present in the environment (Bradl 2005).
However, their widespread uses in industrial and domestic applications has elevated their levels in soils (Wuana & Okieimen 2011), water reservoirs, and plant tissues (Hladun et al. 2015) far above natural baselines around industrialized and urbanized areas, mining sites and agricultural regions (Zhou et al. 2018).Some MTE are wellknown neurotoxins (Mason et al. 2014) and constitute a direct threat for animals inhabiting contaminated environments, including humans (Chen et al. 2016), whose neural development and cognitive functions may be impaired.
Pollinators, such as bees, are in front line.These insects have a severely constrained brain size optimised to perform cognitive operations for exploiting scattered plant resources (Giurfa 2013).Bees can encounter airborne MTE particles while flying (Negri et al. 2015), and may bring back contaminated water, nectar and pollen to their colony nest (Roman 2007;Formicki et al. 2013).MTE bio-accumulate in the bodies of insects (Perugini et al. 2011;Lambert et al. 2012), because of biochemical and structural similarities to non-toxic molecules (Clarkson 1987) and their incorporation in metabolic pathways (Mertz 1981), but they can also contaminate the nest.In honey bee hives, MTE are found in pollen, honey, wax and propolis (Satta et al. 2012;Zhou et al. 2018), and can be transferred to the larvae (Balestra et al. 1992).
Many MTE (e.g.lead, cadmium, selenium, aluminium, manganese) can have detrimental effects on the physiology (Gauthier et al. 2016;Nikolić et al. 2019), foraging behaviour (Søvik et al. 2015;Chicas-Mosier et al. 2017;Sivakoff & Gardiner 2017) and survival (Di et al. 2016;Hladun et al. 2016) of bees.However, potential effects on cognition have been largely unexplored.Recent work indicates that acute exposure to some MTE can reduce the sensitivity of bees to sucrose rewards (Hladun et al. 2012;Burden et al. 2019) as well as elemental forms of learning and memory (Burden et al. 2016); while not necessarily being detected by the sensory organs (Burden et al. 2019).Just as for agrochemical pollution (Gill et al. 2012;Henry et al. 2012), the potential sublethal effects of MTE on bee cognition due to a chronic exposure may have long-term dramatic consequences on populations.In these central-place foraging insects, any alteration in the learning or memory performances of the foragers may reduce food collection and ultimately impair colony development ( Gill et al. 2012;Klein et al. 2017).
Here, we experimentally investigated the effects of chronic lead exposure on honey bee cognition.Lead is of public health significance because of its high degree of toxicity and worldwide distribution (Cameron 1992).Although worldwide lead emissions have been significantly reduced since the phase-out of leaded gasoline (Kierdorf & Kierdorf 2004;Chadwick et al. 2011), they are still important in some countries (Li et al. 2012) and the concentrations of legacy lead remain high in soils (Han et al. 2002;Zhou et al. 2018) and plant tissues (Rashed et al. 2009).We exposed caged honey bee colonies to field-realistic concentrations of lead for 10 weeks and monitored impacts on the morphology and cognition of individual bees, as well as on colony dynamics.We also evaluated the basal relationship between morphological development and cognitive performances in non-contaminated bees from uncaged colonies foraging on natural plant resources.

Bee colonies
We ran the experiments with caged hives from 14/06/2019 (day 1) to 23/08/2019 (day 70) using nine colonies of Apis mellifera (Buckfast) maintained in 5 frame hives (Dadant).Each colony was placed in an outside tent (3m x 3m) at our experimental apiary (University Paul Sabatier, France) to control the food intake and the foraging experience of bees.Each tent contained a feeder with sucrose solution (with or without lead, see below) and a water supply.Both were located 1m apart, 2m in front of the hive entrance.Caged colonies were given pollen patties (Icko, Bollène, France) once a week directly into the hives.We also ran experiments with bees from uncaged hives, by randomly collecting bees from colonies in the same apiary.These non-contaminated bees had free access to natural plant resources.

Lead exposure
We assigned the caged colonies to one of three lead treatments (three colonies per treatment): unexposed (hereafter 'control bees'), exposed to a low (0.075 mg.L -1 ) concentration of lead ('L bees'), exposed to a high (0.75 mg.L -1 ) concentration of lead ('H bees').Bees were exposed to lead by ingesting 50% (w/v) sucrose solution from the feeder, to which we added lead (II) chloride (PbCl2) (Sigma-Aldrich, Lyon, France).
The low and high lead concentrations fell within the range of concentrations measured in floral nectar (Eskov et al. 2015) and honey (Lambert et al. 2012) in contaminated environments.Both concentrations are sublethal to adult honey bees (Di et al. 2016).
Control hives were fed 50% (w/v) sucrose solution.We maintained the hives in these conditions for 70 days, during which colonies consumed on average 8.5±0.6 (SE) kg of sucrose solution and 616±25 (SE) g of pollen (N=9).We kept track of the foraging experience of the nectar foragers (number of days since the onset of foraging) from each colony by paint-marking bees with a colour code while feeding on the sucrose solution feeder (Posca pen, Tokyo, Japan).

Learning assays
We assessed the cognitive performances of bees from caged and uncaged colonies using olfactory conditioning of the proboscis extension reflex (PER) (Giurfa & Sandoz 2012).
For caged colonies, we tested bees exposed during their whole life (foragers collected between days 46 and 70 after the start of the exposure) from 8 of the 9 colonies (one control hive showed low foraging activity).
All bees were submitted to a reversal learning task, a two-task assessing the cognitive flexibly of bees in response to changes in flower rewards (Raine & Chittka 2007).Phase 1 is a differential learning phase, in which the bees must learn to differentiate an odour A reinforced with sucrose (50% w/v in water) and an odour B not reinforced (A+ vs. B-).Phase 2 is a non-elemental learning phase, in which the bees must learn the opposite contingency (A-vs.B+).We used pure limonene and eugenol (Sigma-Aldrich, Lyon, France) as odours A or B alternately on successive days, so that each combination was used for about half of the bees for each treatment.
We tested new foragers (between 24 and 48 hours after the onset of foraging) to avoid variations in cognitive performances caused by inter-individual differences in foraging experiences (Cabirol et al. 2018).On the morning of each test, we collected foragers from each hive on the feeders (except from one control hive due to a low foraging activity), cooled them on ice and harnessed them in restraining holders, which allowed free movements of their antennae and mouthparts (Matsumoto et al. 2012) (Fig. 1A).Rotational movements of the head were prevented by fixing the back of the head with melted bee wax.We then tested all bees for PER by stimulating their antennae with 50% sucrose solution, and kept only those that responded for the conditioning (77% of the 543 bees tested in total).These bees were fed 5µL of sucrose solution and left to rest in a dark incubator for 3h (temperature: 25±2°C, humidity: 60%).
Bees were trained using an automatic stimulus delivery system (Fig. 1A; (Aguiar et al. 2018)).Each training phase included five trials with the reinforced odorant and five trials with the non-reinforced odorant in a pseudo-random order with an eight-minute inter-trial interval.Each conditioning trial (37 s in total) started when a bee was placed in front of the stimulus delivery system, which released a continuous flow of clean air (3,300 mL.min -1 ) to the antennae.After 15 s, the odour was introduced to the airflow for 4 s, the last second of which overlapped with sucrose presentation to the antennae using a toothpick (Fig. 1A) and subsequent feeding for 4 seconds for the rewarded trials.For the unrewarded trials, no sucrose stimulation was applied.The bee remained another 15 s under the clean airflow.Bees were kept in the incubator for 1h between the two learning phases (A+ vs. B-and A-vs.B+).
During conditioning, we recorded the presence or absence of a conditioned PER to each odorant at each trial (1 or 0).Each bee was given a learning score for phase 1 (1 if the bee responded to A+ and not to B-in the last trial of phase 1, 0 otherwise) and for phase 2 (1 if the bee responded to B+ and not to A-on the last trial, 0 otherwise) (Cabirol et al. 2018).Short-term memory (1 h) was assessed by comparing the responses at the last trial of phase 1 and the first trial of phase 2. Each bee was given a memory score for the two odorants (1 if the bee still responded appropriately to the A+ and B-on the first trial of the phase 2, 0 otherwise).

Morphometry
We evaluated developmental differences among bees by conducting morphometric measures on frozen individuals (-18°C) from caged and uncaged hives.To test the effect of lead exposure on morphology, we collected foragers from the caged hives the day before exposure (day 0), during exposure (day 53) and at the end of the experiment (day 70), and measured their head length and head width (Fig. 2A).We also sampled emerging adult bees every week from each of the caged hives (before exposure, during exposure, and at the end of the exposure period), making sure that no Varroa sp.mites were present in the cell.We measured their fresh body weight (±0.001g) (precision balance ME103T, Mettler-Toledo Gmbh, Greifensee, Switzerland) and eight morphometric parameters: head length, head width, forewing length, forewing width, femur length, tibia length, basitarsus length, basitarsus width (Fig. 2A) (Mazeed 2011;De Souza et al. 2015).To test the basal relationship between cognitive performances and morphology, we measured the head length and head width of the non-contaminated bees randomly collected from uncaged hives after the conditioning experiments.We took all measurements (±0.01 mm) using a Nikon SMZ 745T dissecting scope (objective x0.67) with a Toupcam camera model U3CMOS coupled to the ToupView software.

Colony dynamics
We assessed the effect of lead exposure on colony dynamics through continuous measurement of hive parameters in the caged colonies.Colony weight (±0.01 kg) was recorded every hour with an electronic scale (BeeGuard, Labège, France) below each hive.Every two weeks we opened the hives and took pictures of both sides of each frame with a Panasonic Lumix DMC-FZ200 equipped with a F2.8 25-600 mm camera lens.From the pictures, we estimated the areas of capped brood and food stores using CombCount (Colin et al. 2018).We weighted each frame, after gently removing the adult bees, and determined the total weight of adult bees (bee mass) by subtracting the tare of the hive and the weight of the frames from the weight of the hive.Five times during the experiment (day 0, day 24, day 38, day 53, day 70), we estimated Varroa loads by collecting ca.300 bees from uncapped brood frames in a jar and adding 15g of icing sugar.We shook the jar for two minutes, released the bees and counted the number of fallen mites.The number of mites was standardised and expressed as number of Varroa sp.phoretic mites per 100 bees (Dietemann et al. 2013).

Lead quantification
We analysed lead in sucrose solution and bees from caged hives using Inductively Coupled Plasma Emission Spectroscopy (ICP-OES, quantification limit: 5-20 µg.kg -1 , precision measure: 1-5%; AMETEK Spectro ARCOS FHX22, Kleve, Germany).We assessed lead concentration in sucrose using the high concentration sucrose solution (the low concentration solution fell under the ICP-OES detection thresholds).The solution was acidified at 3% of HNO3 with ultra-pure 69% HNO3 to avoid precipitation or adsorption in containers.It was then diluted with a HNO3 3% solution to reduce the spectral interference and viscosity effects.We assessed lead content in bees collected during the 4 th week of exposure.For each sample, we pooled bees in batches of 5.Each batch was rinsed with 5 mL HNO3 at 3% for 30 s. Bees were wet mineralized in 50mL polypropylene tubes using a Digiprep system (SCP Science, Quebec, Canada) with 5 mL of 69% nitric acid.A digestion phase was carried out at room temperature overnight, followed by a second phase of heating at 80°C for 60 min.The nitric acid was evaporated, and the samples were diluted with 9 mL of 3% HNO3.Final solutions were at 3% HNO3 and total dissolved solids below 5%.

Statistics
We ran all analyses with R Studio v.1.2.5033 (RStudio Team 2015).We compared lead content of bees using a Kruskal-Wallis test (package FSA; (Ogle et al. 2019)).We evaluated the effects of lead exposure on colony parameters with a multi-model approach (MMI), with treatment, exposure duration (standardised using rescale function, package arm; (Gelman & Su 2013)) and their interaction as fixed effects, and hive identity as random factor.We ran a model selection (package MuMIn; (Barton 2020)) and applied a conditional model average to evaluate the effects of the different factors on the response variables.To evaluate the effect of Varroa sp.load, we used a subset of the colony data for the days when mites were counted.We ran a MMI followed by a conditional model average to assess the effects of treatment, time, Varroa counts and their interactions on brood area (square-root transformed), food stores area and bee mass.
For learning assays, we tested colony effects by comparing the proportions of bees with learning or reversal scores of 1 between hives in each treatment group.We used proportion tests, followed by pairwise comparisons with a Bonferroni correction (package RVAideMemoire; (Hervé 2020)), to evaluate whether lead exposure changed sucrose responsiveness (i.e.proportions of unresponsive bees across treatments).We extracted individual learning slopes from the raw data (using a linear mixed effect model (LMM, package nlme; (Pinheiro et al. 2019)) with individual nested in test day as a random effect and trial as random slope).We then compared learning slopes between treatments, during each learning phase, by performing generalized linear mixed-effects models (GLMM) (package lme4; (Bates et al. 2015)), with hive identity as random factor and treatment and standardized duration of exposure as fixed effects.
We compared proportions of successful responses during the 5 th trial of each learning phase using a binomial GLMM, with odorants, treatments, standardized duration of exposure and their interactions as fixed effects, and bee identity nested in the hive identity as random factors.We ran a similar GLMM to compare the learning, reversal and memory scores.We used a MMI and model average to evaluate the effect of treatment and standardised duration of exposure, and their interactions, on the behavioural variables (PER responses, learning, reversal and memory scores).
For the morphometric analyses on caged bees, we used LMMs for each parameter, considering treatment, duration of exposure and their interaction as fixed effects, and hive identity as random factor.To assess the global effect of lead, we collapsed the nine parameters into the first component of a principal component analysis (PCA) (package FactoMineR, (Lê et al. 2008)).Bees were clustered into subgroups based on PCA scores, and clusters were compared with a permutational multivariate analysis of variance (PERMANOVA) (package vegan; (Oksanen et al. 2019)).We ran a LMM on individual coordinates from the PCA, with treatment, exposure duration and their interaction as fixed effects, and hive identity as random factor.To assess the effect of head size on cognitive performance of uncaged bees, we collapsed the head width and length measures into the first component of a PCA and ran a binomial GLMM on learning, memory and reversal scores, with individual coordinates from the PCA as fixed effect, and test day as random factor.
Exposure to high lead concentration reduced learning performance.
We assessed the effect of lead exposure on cognition using an olfactory reversal learning task bees from caged hives.The proportion of bees that responded to the antennal stimulation of sucrose was similar across treatments (control bees: 74% N=113, L bees: 69% N=122, H bees: 76% N= 132; χ²=1.423,df=2, p=0.491), indicating that lead exposure did not affect appetitive motivation or sucrose perception.
Comparing the responses of bees between the last trial of phase 1 and the first trial of phase 2, one hour later, gives insight about short-term memory.We only found a tendency for H bees to show a reduced percentage of correct responses between the two phases (Conditional average model: H bees, p=0.078) (Tables 1 and S2).The duration of exposure to lead did not affect the response levels of bees, nor their learning scores, at the end of either learning phases (Tables 1 and S2).
Bees exposed to the high lead concentration were shorter with smaller heads.
We assessed the effect of lead exposure on bee development by measuring morphological parameters on foragers and emerging adults from the caged hives.
Foragers collected on the day before the beginning of lead exposure (day 0), had similar head measurements irrespective of treatment (LMM: Treatment effect: p=0.529 for head width, p=0.509 for head length).However, foragers collected on the 8 th or 10 th week of lead exposure had significantly smaller heads than controls (L bees: head length p=0.017;H bees: head width p=0.040, head length p<0.001;Table S3).This effect was independent of the duration of exposure (LMM: Treatment:Time effect: p=0.711 for head width, p=0.663 for head length).
We further explored the effects of lead treatment on the morphology of adults (Fig. 2A) using a PCA on the nine morphometric parameters measured on emerging bees (Fig. 2B, Table S4).Two PCs explaining 57% of the variance were sufficient to separate control bees and H bees into two distinct clusters, while L bees remained in between (PERMANOVA: Pseudo-F=3.313, p=0.004; control bees vs. L bees: p=1; C bees vs. H bees: p=0.021;L bees vs. H bees, p=0.177).PC1 explained variations in general body size while PC2 explained variations in wing and basitarsus width, and femur length.PC1 was negatively correlated with the interactive effect between concentration and duration of exposure (LMM: Treatment:Time effect: p<0.001).
Overall, H bees displayed a decrease in the same order of magnitude for most of the parameters, except for weight (-12.5%) and head width (-2.99%) (Table S5).This indicates that bees exposed to the high concentration of lead during larval development had a lighter body with shorter legs and wings, as well as smaller heads, compared to unexposed bees (Table S6).

Bees with bigger heads showed better learning performance.
We explored the basal relationship between head size and learning performance by testing bees unexposed to lead from uncaged hives to a reversal learning task (N=149).
The first component (PC1) of the PCA collapsing head width and length explained 73.6% of the variance and represented the global head size.In phase 1, the proportion of learners (79% N=118) increased with head size (Fig. 3A), while short-term memory recall was unaffected (Fig. 3B).In phase 2, the proportion of learners (16% N=23) was also affected by head size (Fig. 3C).Therefore, bees with bigger heads showed better learning performances in absence of any cage confinement or lead treatment.

Discussion
Recent studies suggest that MTE can have sublethal effects on individual pollinators, with potential detrimental consequences for populations (Søvik et al. 2015;Burden et al. 2016Burden et al. , 2019;;Skaldina & Sorvari 2019).Here, we demonstrate that honey bees chronically exposed to field-realistic concentrations of lead in food have reduced body sizes and learning abilities.The positive correlation between head size and learning performances in unexposed bees suggests that consumption of this single trace element affects bee development, constraining brain size and cognitive function, thus constituting a major source of environmental stress for bees.
The impact of lead was independent of the duration of exposure.While our experimental setup allowed contamination through foraging over several weeks, similar effects were observed in adults exposed for varying durations.Yet, all bees had undergone their larval stage during the exposure period, suggesting that most of the detrimental effects of lead is caused by larval ingestion of contaminated food brought by foragers.Lead alters larval development in flies and bees (Cohn et al. 1992;Safaee et al. 2014;Di et al. 2016) and, in mammals, alterations in neural development are correlated with cognitive impairments, including learning and memory (Grandjean & Landrigan 2006;Giordano & Costa 2012;Mason et al. 2014).
In our experiments, lead did not generally affect olfactory or gustatory perception or appetitive motivation, as previously observed in more systematic sucrose sensitivity tests (Burden et al. 2019), nor differential learning (first learning phase) or short term memory.Rather, we found a specifically decreased ability in reversal learning (second learning phase), a task involving cognitive plasticity that allows bees to update their memories about relevant associations between flower cues and food (Ferguson et al. 2001).A higher sensitivity to lead in reversal learning has been documented in rats (Hilson & Strupp 1997;Garavan et al. 2000), monkeys (Bushnell & Bowman 1979) and humans (Evans et al. 1994;Finkelstein 1998).Most probably, this reflects a stronger impact of lead intoxication on the development and/or function of brain circuits that are more specifically involved in a non-elemental task, such as reversal (e.g.orbitofrontal cortex in mammals (Schoenbaum et al. 2000)), than in simpler forms of learning like differential learning that do not involve the resolution of ambiguities (Giurfa 2013).
In honey bees, effective reversal requires the normal function of the mushroom bodies (MBs), that are brain centers otherwise dispensable for differential conditioning (Boitard et al. 2015).The specific reversal impairment of lead-exposed bees might thus be understood if neural circuits would be more affected in the MBs than in other brain regions.Since reversal learning requires GABAergic transmission in the MBs (Boitard et al. 2015), alterations in GABA release observed in mammals exposed to lead (Lasley et al. 1999) might contribute to the specific impairment of reversal learning in exposed bees.More generally, lead exposure is known to alter the maturation of the brain excitation/inhibition balance during development, through multiple effects such as loss of GABAergic interneurons (Stansfield et al. 2015), altered maturation of GABAergic neurons (Wirbisky et al. 2014;Neuwirth et al. 2018), decrease in GABA and glutamate release (Lasley et al. 1999;Xiao et al. 2006) or transport (Strużyńska & Sulkowski 2004), inhibition of post-synaptic glutamatergic action (Neal & Guilarte 2010).In insects, although no specific effect on GABAergic signalling has been demonstrated yet, the known effects of lead exposure on synaptic development (Morley et al. 2003), presynaptic calcium regulation (He et al. 2009) and acetylcholinesterase activity (Nikolić et al. 2019) are compatible with a disruption of the excitation/inhibition balance.We proposed earlier that reaching an optimal value for such balance in MB circuits is what determines efficient reversal learning in mature adults, and particularly in the so-called calyces of the MBs, where a sparse coding of odorants is believed crucial for efficient reversal (Cabirol et al. 2017(Cabirol et al. , 2018)).Developmental lead exposure could jeopardize this balance in many ways, like in mammals, and result in impairment of the reversal learning phase only.Indeed, early environmental conditions shape MB circuits and reversal learning performances of adult bees (Cabirol et al. 2017).
Further evidence supports the hypothesis of a developmental effect of lead, since bees exposed to the highest concentrations developed lighter bodies, with shorter legs and wings and smaller heads.Interestingly, unexposed bees with smaller and shorter heads had a lower learning performance (albeit not specifically in the reversal phase).In bees, head width tend to be correlated with the volume of the brain (Honey Irrespective of the detailed mechanisms by which lead impairs bee learning, our results on an ecologically relevant cognitive task raise the concern of pollinators exposed to environmental metallic pollution.Importantly, lead contents measured in the bodies of exposed bees in our experiments ranged within the measurements from bees in natural conditions (Goretti et al. 2020).The two concentrations of lead in the sucrose solutions used for chronic exposure (0.075 and 0.75 mg.L -1 ) fell below the maximum level authorized in food (3 mg.kg -1 ; Codex Alimentarius 2015) and irrigation water (5 mg.L -1 ; Ayers & Westcot 1994), and the lowest concentration was under the threshold set for honey by the European Union (0.10 mg.kg -1 ; Commission Regulation 2015).This suggests that bees foraging on flowers in contaminated environments also exhibit cognitive and developmental impairments.Ultimately, these effects can alter colony function, for instance by reducing the collective foraging efficiency and gradually causing population declines, as found with the effects of pesticides at sublethal concentrations (Klein et al. 2017).Although our experiment, and recent similar approaches (Hladun et al. 2016), did not capture these effects at the population level, differences in colony performances are expected over the longer term.Our results thus call for future studies to better characterize the impact of lead exposure in bee populations, including in combination with other MTE as such cocktails are often found in contaminated areas (Badiou-Bénéteau et al. 2013;Goretti et al. 2020).More generally, a better assessment of the contribution of heavy metal pollutants to the widespread declines of insects has become an urgent necessity for preserving ecosystem services.1 and S2).Bar plots show the proportions of learners (black) and non-learners (white) in the last trial of phase 1 (B) and phase 2 (C), with sample size displayed.Statistical comparisons were obtained with p-values from the model average following MMI procedure (Table 1).(ns: non-significant, p>0.05; *p<0.05;**p<0.01;***p<0.001)

Figures
bees: Gronenberg & Couvillon 2010; Bumblebees : Riveros & Gronenberg 2010), the MBs and the MB calyces (Honey bees: Mares et al. 2005; Bumblebees: Smith et al. 2020).Lead exposure during larval development is thus likely to have impaired aspects of head and brain development, that affect preferentially MB-dependent cognitive tasks such as reversal learning.In our experimental conditions, continuous exposure to environmentally realistic amounts of lead resulted in bioaccumulation of the metal in the bees' bodies, which may explain the dose-dependent effects observed on cognition and morphology.

Figure 1 :
Figure 1: Learning and memory performances of bees from caged hives.A) Photo

Figure 2 :
Figure 2: Morphometric analysis of bees from caged hives.A) Details of the

Figure 3 :
Figure 3: Relationship between head size and cognitive performance in unexposed

Table 1 : Parameter estimates from the conditional model average for response levels at the end of both learning phases, and for learning, reversal and memory score models in bees from caged hives.
Significant p-values (<0.05) are shown in bold.SE=conditional standard errors.

Table S2 : Model selection with Akaike's information criterion corrected for small 767 sample size (AICc) for the response level at the end of both learning phases, as well 768 as for the learning, reversal and memory scores for bees in caged hives.
Full and  769null models are presented with the three best models considered for each analysis.770

Table S5 : Median, minimum and maximal values of each morphological parameter of emerging bees from caged hives, per treatment and percentage of variation between medians compared to unexposed bees.
Percentage of variationfor each parameter was the percentage of variation of median values between treatments.We considered variation of the same order of magnitude as isometric scaling, if not as allometric scaling.