Proceedings of the Royal Society B: Biological Sciences
You have accessResearch articles

Upper-limit agricultural dietary exposure to streptomycin in the laboratory reduces learning and foraging in bumblebees

,
Elizabeth Dunne

Elizabeth Dunne

Department of Biology, Emory University, Atlanta, GA 30322, USA

Contribution: Data curation, Investigation, Writing – review & editing

Google Scholar

Find this author on PubMed

,
David Hofmann

David Hofmann

Department of Physics, Emory University, Atlanta, GA 30322, USA

Initiative in Theory and Modeling of Living Systems, Emory University, Atlanta, GA 30322, USA

Contribution: Investigation, Software, Writing – review & editing

Google Scholar

Find this author on PubMed

and
Berry J. Brosi

Berry J. Brosi

Department of Biology, University of Washington, Seattle, WA 98195, USA

[email protected]

Contribution: Conceptualization, Formal analysis, Funding acquisition, Methodology, Resources, Software, Supervision, Visualization, Writing – original draft, Writing – review & editing

Google Scholar

Find this author on PubMed

Published:https://doi.org/10.1098/rspb.2021.2514

Abstract

In the past decade, the broadcast-spray application of antibiotics in US crops has increased exponentially in response to bacterial crop pathogens, but little is known about the sublethal impacts on beneficial organisms in agroecosystems. This is concerning given the key roles that microbes play in modulating insect fitness. A growing body of evidence suggests that insect gut microbiomes may play a role in learning and behaviour, which are key for the survival of pollinators and for their pollination efficacy, and which in turn could be disrupted by dietary antibiotic exposure. In the laboratory, we tested the effects of an upper-limit dietary exposure to streptomycin (200 ppm)—an antibiotic widely used to treat bacterial pathogens in crops—on bumblebee (Bombus impatiens) associative learning, foraging and stimulus avoidance behaviour. We used two operant conditioning assays: a free movement proboscis extension reflex protocol focused on short-term memory formation, and an automated radio-frequency identification tracking system focused on foraging. We show that upper-limit dietary streptomycin exposure slowed training, decreased foraging choice accuracy, increased avoidance behaviour and was associated with reduced foraging on sucrose-rewarding artificial flowers. This work underscores the need to further study the impacts of antibiotic use on beneficial insects in agricultural systems.

1. Background

Agricultural antibiotics are used globally to fight bacterial pathogens in crops [1] and livestock [2]. In the past decade, the broadcast application of antibiotics to US crops has increased exponentially [3] (electronic supplementary material, figure S1) due to the expansion of devastating bacterial diseases [4]. This rapid increase is concerning because we lack a full understanding of the off-target impacts to wild and managed organisms present in and around sprayed fields [5,6]. Of particular relevance are the sublethal risks of antibiotic exposure to beneficial insects. For example, bees that perform pollination services might be especially vulnerable as antibiotics are applied at bloom in some crops, a time when bees collect pollen and nectar and provide pollination services. Studies have shown that antibiotics used to treat pathogens within managed hives can indirectly increase honey bee (Apis mellifera) susceptibility to pathogens [7,8] and mortality [7] by impacting the bee gut microbiome. Similarly, antibiotic exposure can affect the bumblebee gut microbiome, decreasing bee immunity to pathogens [9,10]. The impact on bumblebee survival, however, depends on the antibiotic type and concentration [10,11]. Recent evidence suggests that exposure to high antibiotic doses can affect honeybee olfactory learning [12] and delay the onset of foraging in honeybee colonies [13], but much remains unknown about how antibiotics might impact other modes of bee learning, especially when given at the upper limit concentrations applied to crops.

For effective foraging, bees rely on associative learning, the ability to associate a reward (e.g. nectar) with a sensory (e.g. visual or olfactory) stimulus [1416]. Repeated exposure to a rewarding stimulus increases learning accuracy [17] and speed [18]. The ability to associate rewards with stimuli (flower attributes) and to remember such associations permit bees to return to profitable flower patches [19] and avoid unrewarding flowers [20]. Greater associative learning has been associated with higher foraging rates in the field [21], which in turn can boost bee fitness [22] and could potentially lead to increased cross-pollination [23].

Antibiotics could negatively affect associative learning and foraging via several possible mechanisms, including (i) gut microbiome impacts (which in turn can affect the gut-brain axis), (ii) direct neurotoxicity and (iii) gustatory deterrence. First, a growing body of work suggests that the gut microbiome plays a key role in insect cognition and behaviour [24]. For example, microbe-free flies (Drosophila melanogaster) take longer to associate an aversive reward with a stimulus relative to control flies with a gut microbial community [25]. Germ-free honeybees (A. mellifera) are less likely to display the proboscis-extension reflex when exposed to low-concentration sucrose solutions [26] and to associate odors with rewards [12], compared to bees with a normal microbiome. The abundance of a core gut microbial taxon (Lactobacillus Firm-5) is also positively correlated with bumblebee memory of rewarding flowers [27]. Given that gut symbionts are key for the digestive process in bees [26], gut microbiome disruption could also reduce nutritional uptake and energetic budgets, decreasing bee activity levels and willingness to engage with external stimuli [28]. Second, while insects have a haemolymph–brain barrier [29,30], there is virtually no information on whether antibiotics or their metabolites can cross it and, if so, generate neurotoxic impacts sufficient to affect behaviour [31]. Third, bees may avoid foraging on nectar with antibiotic residues. For example, bumblebees are deterred from feeding (i.e. consuming smaller volumes and retracting their proboscis) on bitter-tasting (quinine) sucrose solution compared to a plain sucrose solution [32]. Irrespective of the mechanism, a disruption of foraging-associated behaviours due to antibiotic exposure could have detrimental consequences for bee populations and their functional roles as pollinators.

Here, we tested the impact of an upper-limit dietary exposure to an antibiotic used in broadcast spraying of crops (streptomycin) on individual-level bumblebee associative learning via operant conditioning assays in which bees can interact with and modify their behaviour in response to the stimuli [33]. We hypothesized that antibiotic exposure would negatively impact associative learning, given the potential mechanisms described above. Specifically, we predicted that antibiotic-fed bees would (i) take longer to train, (ii) avoid stimuli more often during training, (iii) display lower accuracy in remembering the rewarding association (stimulus/reward) after training and (iv) obtain fewer rewards overall compared to unexposed bees (forage less in a laboratory foraging arena with artificial flowers).

2. Material and methods

(a) General experimental set-up and antibiotic dose

We used commercially sourced Bombus impatiens (Koppert Biological Systems, Howell, MI, USA) which have well-characterized associative learning and foraging behaviour [34]. All colonies were queenright, healthy (with brood, emerging foragers and no visual abnormalities such as moldy pollen, dead larva, etc.), and were kept in dark conditions at room temperature (22 °C). Two days before each behavioural assay, 20 to 30 naive (never-before exposed to the conditioned stimulus) bees from the same colony were split into two sub-colonies housed inside disinfected plastic containers (16.20 L × 23.16 W × 13.97 H cm) altered with an aluminium insect screen flooring (0.011 gauge, Adfords). The sub-colonies were fed an ad libitum sucrose diet (control, boiled 0.5 Mol sucrose in 0.2 µm double filtered water) or the same diet with 200 ppm streptomycin added (streptomycin sulfate, Acros Organics, New Jersey, USA).

After two days of exposure, we ran free movement proboscis extension reflex assays (figure 1a) [34] to assess (i) training ability, (ii) avoidance of visual stimuli and (iii) associative learning accuracy in a choice test with unharnessed bees. We also quantified the impact of the dietary antibiotic exposure on (iv) individual foraging success, inside a RFID automated foraging arena (figure 1b). Finally, we compared individual nectar consumption rates among bees in the antibiotic-exposed versus control groups in experiments parallel to those described above, but in which individual bees were isolated (i.e. not with other individuals in a sub-colony) and their feeding rates quantified via weighing of sucrose feeders.

Figure 1.

Figure 1. Graphical depiction of behavioural assays. (a) Free movement proboscis extension protocol assay with training and choice test to assess the impact of bee treatment (control versus antibiotic) on associative learning (short-term memory). (b) Bee foraging assay where bees are fitted with RFID tags and allowed to forage for 2 hours inside an automated RFID foraging chamber to assess the impact of the antibiotic treatment on bee foraging. (Online version in colour.)

We are not aware of any published estimates of streptomycin concentrations and/or persistence in pollen and floral nectar. Therefore, we selected our dosing to represent an upper-limit dietary concentration that could (i) disrupt the bumblebee gut microbiome with minimal impact on organismal functioning, and (ii) be representative of concentrations used to spray crops. First, our dosing was guided by Meeus et al. [11] who tested the same dietary dose of streptomycin (200 ppm) and did not see impacts on worker survival, egg-laying behaviour or drone emergence in a bumblebee species (Bombus terrestris). Our own survival analyses (using mortality data from the nectar consumption experiment; see section 2d below) found that after three days of ad libitum consumption of sucrose with 200 ppm streptomycin, the streptomycin-treated bees displayed a trend toward higher survival relative to controls, though it was not statistically significant (p = 0.061; electronic supplementary material, table S8 and figure S4), consistent with no adverse direct effects of the dose tested. Even higher doses of dietary streptomycin (500 and 1000 ppm) do not seem to impact the survival of adult insects of much smaller mass than bees, such as citrus psyllids Diaphorina citri Kuwayama [35]. In the freshwater invertebrate Daphnia magna—orders of magnitude smaller in mass than bumblebees—acute toxicity after 48 h of exposure was estimated as 487 ppm, without detectable impacts on reproduction [36]. Perhaps due to this lack of impact on invertebrate survival, high streptomycin doses have been used experimentally to deplete the insect gut microbiome and measure behavioural outcomes. For example, 400 ppm streptomycin was used to study whether the gut microbes influence mating behaviour and locomotion of D. melanogaster [3739]. Second, the streptomycin concentration we chose is representative of concentrations used in crop foliar sprays [40]. Much higher concentrations are alternated with oxytetracycline (another antibiotic) in citrus groves in Florida (USA) under emergency-label commercial formulations to control citrus greening [41,42]. Tree trunk delivery of antibiotics is being investigated as an alternative to these foliar sprays [43]. In one of these assays, streptomycin was detected at 200 and 300 ppm in the xylem and phloem of young citrus trees [44]. This is particularly relevant because nectar in most plants is derived from phloem, often concentrated by nectary tissues [45]. When bacterial disease pressure is high, crops that require streptomycin at lower doses (50–100 ppm), such as apple and pears, may receive multiple foliar applications four to five days apart [46,47], which increases the potential for bioaccumulation in pollen, nectar and within bee colonies. Finally, our length of exposure of 48 h is consistent with the apparent residual persistence of streptomycin in floral stigmas, which can reduce bacterial growth on these structures for at least that length of time [48,49].

(b) Free movement proboscis extension reflex assay

To enhance bee appetites, we removed the feeders an hour before the beginning of each trial. We then transferred single bees into individual custom-made rectangular Plexiglass enclosures (13 L × 2.86 W × 3.18 H cm, Plastic-Craft Products, West Nyack, NY, USA). We tested 106 bees (53 antibiotic-exposed and 53 controls) from five different colonies in 12 trials.

(i) Training (via differential conditioning)

We conditioned each bee to two visual stimuli (yellow and blue coloured card strips) saturated with a neutral appetitive reward (water) and a positive appetitive reward (0.5 Mol sucrose) (figure 1a). We permuted the assignment of the colour stimulus and the reward among bees.

The training phase consisted of a series of ‘presentations’, as shown in figure 1a. In each presentation, we inserted the sucrose conditioned stimulus in the enclosure, until the bee touched the stimulus with its antennae [50] and/or proboscis. After three seconds of contact we removed the sucrose stimulus, immediately offered the neutral water stimulus, and repeated the procedure. The presentation ended after one minute, even if the bee had maintained avoidance behaviours (i.e. walking past or not approaching the conditioned stimuli) through the presentation. We waited thirty seconds to begin a new presentation. Most B. impatiens foragers learn to associate visual stimuli with rewards after five presentations [34]. Therefore, the training phase ended if: (i) the bee met the training ‘threshold’ of touching both stimuli with its antennae and/or proboscis in at least one presentation (allowing bees to compare the rewards provided by different coloured stimuli in a single presentation), in addition to touching at least one of the stimuli four other times; or (ii) the bee continued to exhibit avoidance behaviours after 15 presentations and did not meet the training threshold. Only bees that met the training threshold moved to the ‘choice test’ five minutes later.

During the training phase we recorded (i) number of presentations to meet the training threshold and (ii) number of presentations in which the bee avoided the stimuli, later used to estimate avoidance of the stimuli. We also recoded whether the bee touched a sucrose and/or water rewarding stimulus in each presentation, which allowed us to assess if trained bees from both treatments had touched the sucrose and water rewarding stimuli a similar number of times (electronic supplementary material, figure S2).

(ii) Associative learning accuracy (a choice test)

We saturated both coloured stimuli in water and offered them to the bee simultaneously. We recorded the colour of the stimulus to which the bee extended its proboscis, to assess if it had formed an association with the sucrose-rewarding stimulus in the training phase.

(c) Bee foraging assay

To test whether the reduced performance in associative learning translated to bumblebee foraging deficits, we performed separate trials with antibiotic-exposed and control bees inside an experimental array with artificial flowers. Two days prior to a trial, we glued ultra-lightweight (approx. 3 mg) radio frequency ID or RFID tags (mic3-TAG 16 K, Microsensys GmbH, Erfurt, Germany), to the bee's thorax. This technology was designed specifically for bees and has been previously used to assess the impact of pesticides on bee foraging [51]. Two days later, we removed the bees from the sub-colonies (antibiotic-exposed and control), deprived them of feeders for two to four hours, and subsequently placed them inside a foraging arena built around computer-controlled artificial flowers, each with a short-range (approx. 5 mm) RFID antenna and tracking system (figure 1b) (Microsensys GmbH, Erfurt, Germany). We ran five parallel arenas, where each arena within the enclosure had two identically sized and shaped artificial flowers differing only in colour (yellow or blue). Each flower dispensed either water or a sucrose reward of a similar size (2 to 3 µl). The flower would dispense a droplet only if the bumblebee had been gone for at least 30 seconds; if it arrived before that time the clock would reset to 0, thus standardizing two important factors that influence foraging rates in bumblebees: reward volume variance [52] and the expected time between reinforcements [53]. The reward was released and timed when the RFID antenna detected a bee crawling into the artificial flower. The bumblebees were individually kept in the foraging enclosures for two hours which allowed acclimation and onset of foraging activity. Although bees were not able to return to their nest during this period, any bias in feeding due to fullness and inability to offload the crop was identical for all bees, regardless of treatment. Out of 169 bees tested from four colonies, 156 bees foraged (80 antibiotic-exposed and 76 controls). We analysed the foraging data only for the bees that had foraged on both flowers (water and sucrose-rewarding) at least once (n = 75).

(d) Nectar consumption

To assess the potential for differential consumption of the control and the antibiotic-treated sucrose diet, in eight separate trials, bees from four colonies (two trials per colony) were housed individually and fitted with either a sucrose control (n = 78 bees) or a sucrose plus 200 ppm streptomycin (n = 80 bees) feeder. The enclosures consisted of disinfected plastic containers (118 mL Ziploc mini-square containers) with perforated orifices for gas exchange and an inverted sterile 1.5 ml flip-top tube (centrifuge tubes) filled with the solution accessible through a small opening at the top. As this set-up was compliant with guidelines for oral acute toxicity tests in bumblebees [54], we also measured bee survival, and report that analysis in electronic supplementary material table S8 and figure S4. We assessed nectar consumption by first calculating the difference between the weight of the filled feeders at the start of the experiments and three or 4 days later (day variation did not have an impact on the statistical analysis and, therefore, was excluded from statistical models), and subtracting the average weight change from two evaporative controls per treatment. For the nectar consumption analysis, we excluded bees that died before the fourth day (n = 57 antibiotic-exposed bees and n = 43 control bees).

(e) Data analysis

We carried out all analyses in R statistical programming language, v. 3.6.0 (R Development Core Team). We implemented generalized linear mixed models (GLMMs) with the ‘glmmTMB’ package [55]; error models are specified for each response below, including beta-binomial and negative binomial models for overdispersed proportion and count data. We assessed the effect of dietary treatment (antibiotic versus control, fixed effect) on (1) training speed (number of presentations required to meet the training threshold, Poisson), (2) avoidance of visual stimuli (proportion of presentations in which the bees avoided the stimuli, beta-binomial), (3) learning accuracy in the choice test (selection of sucrose- rewarding versus water-reward strips, beta-binomial), (4) foraging success in the artificial flower arena (count of total flowers visited, count of sucrose-rewarding flowers visited, both negative binomial) and (5) nectar consumption (weight change of feeders, Gaussian). In addition, for (1) and (2) we included stimulus colour (yellow or blue) as a fixed effect. For (3) we included as fixed effects the number of avoided presentations during training, the total number of presentations during training in which the bee touched (with antenna or proboscis) the sucrose-rewarding stimulus, and the total number of presentations during training in which the bee touched the water-rewarding stimulus (electronic supplementary material, figure S2). We added a random effect (random intercept) of colony of origin to all models (1 to 4), and in the foraging success models (4) we included a second random effect of foraging enclosure number, to control for any potential variability in the foraging arena environment and the artificial flowers within. For (5) we ran a linear model with colony as a fixed effect as a previous GLMM with colony as random effect produced non-normal residuals, even after square root, log and box-cox transformations on the raw data. We assessed the statistical significance of explanatory variables using the ‘Anova’ function from the ‘car’ package [56].

We checked final model residuals for collinearity (‘performance’ package [57]), variance inflation (VIF < 3, ‘car’ package [58]), and heteroscedasticity and normality of errors (‘DHARMA’ package [59]). All final models presented here met assumptions. All datasets and a fully reproducible analysis report in R Markdown, including assessment of model assumptions, are available at Figshare doi:10.6084/m9.figshare.13471983.

3. Results

(a) Free movement proboscis extension assays

Antibiotic exposure delayed training and increased avoidance behaviour towards the stimuli relative to control bees. Specifically, antibiotic-exposed bees required more presentations to meet the training threshold relative to controls (figure 2a; electronic supplementary material, table S1; incidence rate ratio, IR = 1.39, n = 102, p < 0.0001), and antibiotic exposed bees were more likely to exhibit avoidance behaviours toward the stimuli relative to controls (figure 2b; electronic supplementary material, table S2; odds ratio, OR = 3.46, n = 102, p < 0.0001).

Figure 2.

Figure 2. Box plots of response variables during the free movement proboscis extension reflex protocol used to assess the impact of the dietary treatment on Bombus impatiens individual associative learning. Bees were fed a 0.5 Mol sucrose solution with antibiotic (200 ppm streptomycin) or without it (control). The circles representing measurements have been jittered to avoid overlap. (a) Number of presentations to meet the training threshold (min 5 presentations and max 15 presentations). A presentation consisted of sequentially offering a sucrose and a water reward to the test bee. Bees meet the training threshold by touching both stimuli in one presentation and touching at least one stimulus four other times. Asterisks represent statistically significant differences (n = 102 bees from five colonies, Wald Χ2 = 18.82, p < 0.0001). (b) Proportion of presentations where the test bee showed avoidance of the visual stimuli. Proportions were estimated by dividing the number of presentations in which the bee avoided the stimuli by the total number of presentations received during training phase. Asterisks represent statistically significant differences (n = 102 bees from five colonies, Wald Χ2 = 17.90, p < 0.0001); (c) Plotted proportion of bees that advanced to the choice test and made the correct choice by selecting the sucrose-rewarding stimulus. Proportions were estimated dividing the number of bees that made the correct choice by the total number of bees tested in the trial, separately for each treatment (12 trials of 5 to 8 bees per trial, 3 to 5 bees per treatment, from five colonies). Asterisks indicate statistically significant differences from a binomial test (electronic supplementary material, table S3; n = 94 bees from five colonies, Wald Χ2 = 7.6250, p = 0.0057). (Online version in colour.)

Dietary antibiotic exposure also reduced the associative learning accuracy. Of those bees that advanced to the single choice test—a short term memory-based choice between water and sucrose-associated coloured strips—antibiotic-exposed bees were less likely to select the correct (sucrose-rewarding) stimulus compared to the control bees (figure 2c and electronic supplementary material, table S3, OR = 0.19, n = 94, p = 0.0057), even after controlling for the number of avoided presentations and number of presentations in which the bee tasted the sucrose or water-rewarding stimulus during training. The negative impact was such that the rate at which antibiotic-treated bees selected the sucrose reward was statistically no different than chance (χ2=3.13, n = 46, p = 0.0768), whereas control bees selected the sucrose reward 87.5% of the time, significantly more than expected by chance alone (χ2=27.00, n = 48, p < 0.0001).

(b) Bee foraging assays

Antibiotic-exposed bees also had reduced foraging success. While antibiotic exposure did not affect the number of total visits to both sucrose and water-rewarding flowers (figure 3a; electronic supplementary material, appendix table S4; IR = 0.84, n = 75, p = 0.0917), antibiotic-exposed bees visited fewer sucrose-rewarding flowers relative to control bees (figure 3b; electronic supplementary material, table S5; IR = 0.73, p = 0.0176).

Figure 3.

Figure 3. Box plots of the response variables during the automated foraging chamber assays (2 h) of Bombus impatiens individuals (n = 75 from four colonies). Bees were fed a 0.5 Mol sucrose solution with an antibiotic (200 ppm streptomycin) or without it (control). The circles representing measurements have been jittered to avoid overlap. (a) Combined number of rewards (water + sucrose) obtained by individual bees (Wald Χ2 = 2.8435, p = 0.0917). (b) Number of sucrose rewards obtained by individual bees (Wald Χ2 = 5.6337, p = 0.0176). (Online version in colour.)

(c) Nectar consumption

Bees consumed less streptomycin-containing sucrose solution relative to sucrose-only solution. Although different colonies consumed statistically distinct quantities (electronic supplementary material, tables S6 and S7; main effect of colony F[3,91]=12.00, p < 0.0001), on average, the bees ingested slightly less antibiotic-laced sucrose (0.765 ± 0.246 g) compared to sucrose-only (0.869 ± 0.343 g) solution over 3–4 days (figure 4; electronic supplementary material, table S7, F[1.91]=5.35, p = 0.0230). There was not an interaction of treatment and colony (electronic supplementary material, table S7; F[3.91]=0.79, p = 0.5189).

Figure 4.

Figure 4. Box plots of the amount of sucrose solution consumed by Bombus impatiens individuals (n = 99 from four colonies). Bees were fed a 0.5 Mol sucrose solution with an antibiotic (200 ppm streptomycin) or without it (control) in 1.5 mL feeders. The filled feeders were weighed at the beginning of the trial and three to four days later to quantify the solution consumed. The circles represent measurements that have been jittered to avoid overlap. Asterisks represent statistically significant differences (F[1,91]=5.3485, p = 0.0230). (Online version in colour.)

4. Discussion

Our results show that a potential upper-limit agricultural dietary exposure to streptomycin was detrimental to several metrics of bee learning and foraging success. We documented (i) delays in bee training speed (figure 2a), (ii) increased avoidance of stimuli (figure 2b), (iii) decreases in choice accuracy (figure 2c), (iv) reduced number of sucrose rewards obtained by individual bees (figure 3b) and (v) reduced nectar consumption (figure 4).

Distinguishing the mechanisms driving these results is beyond the scope of this study. We speculate that the patterns we documented were driven by the impact of antibiotics on the bee brain–gut axis. It is well established that microbes play a role on insect behaviour [60] and that microbial depletion via antibiotics can lead to behavioural changes in animals [61]. In honeybees, antibiotic exposure can impair individual olfactory learning, potentially driven by gut microbial depletion [62], and antibiotic applications within hives can delay the onset of foraging in workers [13]. It is thus plausible that disrupting the gut–brain axis led to learning delays (i.e. decreased accuracy in the choice test). Nonetheless, Leger & McFrederick [63] did not find differences in visual associative learning between germ-free and control B. impatiens foragers, but the lack of microbes in the germ-free bees was not confirmed, and as has been shown in D. melanogaster, germ-free organisms may still contain microbes [38]. In summary, more work is needed to understand the mechanisms behind the impacts we report here.

From a broad perspective, negative effects of antibiotic dietary exposure on key aspects of bee foraging behaviour may lead to detrimental consequences for bee fitness and pollination outcomes. Successful colony foraging emerges from the allocation of individuals to food exploration and food exploitation [64]. Thus, the negative impact of dietary exposure to antibiotics on bees' willingness to explore visual stimuli (i.e. avoidancebehaviour) could be detrimental to foraging initiation, which occurs in bumblebees exposed to neurotoxic pesticides [65,66]. When bees are motivated to forage, associative learning allows them to match rewards with floral cues [14,17,67]. In aggregate, individuals' associative learning abilities over their lifespan can have implications for colony fitness [68]. For example, bumblebee individuals that fail to learn an association between an olfactory cue with a reward have shorter foraging careers [69], and colonies with slow learning bees have been shown to collect less nectar compared to colonies with faster-learning bees [21], but see [70,71]. Considering that nectar and pollen stores lead to increased queen production in bumblebees [72] and that higher colony-level foraging results in greater brood production and honeybee colony survival [73], the potential negative impacts of antibiotics on individual learning could lead to colony-level impacts. In addition to potential impacts on bee fitness, the reduction in associative learning could translate to hindered pollination. The ability of bees to associate rewarding and unrewarding flowers with visual cues can enhance cross-pollination [74]. It is also likely that associative learning leads to a greater proportion of conspecific visits within a foraging bout [23,75,76]. Because conspecific pollen is needed for successful ovule fertilization, a decline in associative learning due to antibiotic exposure could ultimately hamper cross-pollination and thus plant fitness. For example, there is evidence that exposing bumblebee colonies to field-realistic doses of neonicotinoids, which affect individual learning [77], can lead to significant decreases in crop cross-pollination [78]. Additionally, based on our sucrose consumption results, floral nectar tainted with antibiotics could be less attractive to bees, with further implications to crop pollination.

There are several caveats to our findings. While a dose–response assay would be ideal to test the impact of antibiotics on bee learning and foraging, field-realistic residual concentrations of streptomycin in floral nectar and pollen need to be determined first. Such work is beyond the scope of this study. Instead, our findings represent the impacts of a potential upper limit scenario of agricultural exposure to streptomycin, but one that has not definitively been confirmed in the field. Another caveat to our results is that the lower consumption of streptomycin-containing sucrose solution (relative to streptomycin-free) may indicate taste deterrence [32]. Consequently, the antibiotic-treated bees could have potentially formed a negative association with the sucrose. This is relevant because such a negative association could confound bee learning outcomes (e.g. bees avoiding engaging with a sucrose-rewarding stimulus). However, several lines of evidence suggest that negative association with sucrose in antibiotic-exposed bees either was not present or was not relevant for learning outcomes. First, in all behavioural assays, we used stimuli consisting of sucrose (with no antibiotics added) versus water, and bees from both treatments tasted both these stimuli. Second, all bees from both treatments that advanced to the associative learning test (choice test) had met a training threshold. Third, bees from both treatments that advanced to the learning test had touched the sucrose-rewarding stimulus a similar number of times during the training phase (electronic supplementary material, figure S2). Fourth, while statistically significant, reduced feeding on antibiotic-containing sucrose solution had a small effect size, a mean difference of 104 mg over three days (or just 35 mg difference per day), which may not be a biologically meaningful difference to bumblebees.

Our findings bring into focus the need for a comprehensive research agenda to fill pressing knowledge gaps related to the use of broadcast-spray antibiotic use in crops [6]. Further work should establish whether the antibiotic-related behavioural impacts we show here are of relevance for managed and wild insects and plants present in crop fields sprayed with antibiotics. Much is unknown in this realm, including the residual antibiotic concentrations in floral pollen and nectar, exposure routes (i.e. dietary, topical) and exposure time, as well as the possibility for interactions with other agrochemicals [79]. The laboratory evidence for antibiotic-associated declines in bee immunity and survival [7] make field studies even more imperative. Additionally, future studies should identify the mechanism(s) by which antibiotics impact the bee brain-gut axis, as this would allow us to better understand the interplay between symbionts and toxins in modulating insect behaviour.

Data accessibility

All datasets and a fully reproducible analysis report in R Markdown, including assessment of model assumptions, are available from Figshare: https://doi.org/10.6084/m9.figshare.13471983. The supplementary figures and tables are provided as electronic supplementary material [80].

Authors' contributions

L.A.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, supervision, validation, visualization, writing—original draft, writing—review and editing; E.D.: data curation, investigation, writing—review and editing; D.H.: investigation, software, writing—review and editing; B.J.B.: conceptualization, formal analysis, funding acquisition, methodology, resources, software, supervision, visualization, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This research was funded by a grant from the Eva Crane Trust (to L.A. and B.J.B.). L.A. postdoctoral training was supported by a Fellowship in Research and Science Teaching (FIRST)—Institutional Research and Academic Career Development Awards (IRACDA—K12GM000680–21) from the National Institute of General Medical Sciences (NIGMS). Grants from the National Science Foundation to B.J.B. (DEB-1120572 and DEB-1501928) and funds from Emory University supported the construction of the automated foraging chamber.

Acknowledgements

We thank Nicole Gerardo for facilitating the nucleic acid quantification, and Shawnee Boyd, Kenneth Zamora and Sean Parker for laboratory assistance. We thank Carolyn Ayers, Emily Dobbs, Travis Dynes, Lowell Ramsey and the Emory Department of Physics workshop for contributions to the automated foraging chamber installation and maintenance.

Footnotes

Electronic supplementary material is available online at https://doi.org/10.6084/m9.figshare.c.5823546.

Published by the Royal Society. All rights reserved.

References