Pollination services provided by honey bees are essential for boosting crop yields, supporting the reproduction of plant species, and maintaining global biodiversity (Khalifa et al., 2021 and references therein). Moreover, honey bee products such as honey, collected pollen, propolis, and royal jelly hold significant economic value. However, the survival of honey bees is increasingly threatened by a range of stressors, including pathogens, nutrient deficiencies, and, notably, exposure to agrochemicals (Li et al., 2018; Yang et al., 2023).

Agrochemicals like the herbicide glyphosate (GLY) and the neonicotinoid insecticide imidacloprid (IMI) are widely used in agricultural settings (Simon-Delso et al., 2015; Benbrook, 2016; Eurostat, 2025). While their primary purpose is to control weeds and pests respectively, their application often leaves residues in pollen and nectar, which are essential food resources for the honey bee (Thompson et al., 2014; Bonmatin et al., 2015; Mitchell et al., 2017; Antier et al., 2020). The effects of GLY on bees are diverse, affecting glands and larval development (Vázquez et al., 2018; Faita et al., 2018), navigation (Balbuena et al., 2015), associative learning and gustatory sensitivity (Herbert et al., 2014; Mengoni Goñalons and Farina, 2018; Farina et al., 2019), circadian rhythms (Almasri et al., 2020), and sleep (Vázquez et al., 2020a). GLY exposure disrupts the gut microbiota, increasing bees' susceptibility to pathogens and malnutrition (Dai et al., 2018; Motta et al., 2018; Blot et al., 2019; Motta and Moran, 2020). On the other hand, neonicotinoid insecticides such as IMI act as neurotoxins and affect cognitive abilities of bees when they are exposed to field-realistic levels (Henry et al., 2012; Dively and Kamel, 2012; Osterman et al., 2019; Rortais et al., 2005; Sanchez-Bayo and Goka, 2014; Zhang and Nieh, 2015; but see Williamson et al., 2013).

While bees may avoid resources containing natural aversive or deterrent compounds, potentially reducing or diluting their exposure to harmful substances or metabolites (Amygdalin: Ayestarán et al., 2010; Quinine: Avarguès-Weber et al., 2010; Rodríguez-Gironés et al., 2013; Saponin: Wang et al., 2019), studies remain limited or inconclusive regarding whether honey bees are capable of avoiding pesticide-contaminated resources. Unlike some natural deterrents that bees can detect through their sensory system, for example, quinine because of its bitter taste, synthetic pesticides may evade detection. This appears to be the case for IMI, which does not elicit spiking responses in gustatory receptor neurons (Kessler et al., 2015), leading bees to inadvertently ingest contaminated food (Mengoni Goñalons and Farina, 2018).

If this is also the case for GLY, it is less evident. Foraging bees collect sucrose solution at a consistent rate regardless of its presence (Herbert et al., 2014), however, young bees reduced the consumption of GLY contaminated sucrose solution when confined in cages (Mengoni Goñalons and Farina, 2018). Results under choice conditions were even more intriguing, as they showed that honey bees prefer resources containing low concentrations of GLY (Liao et al., 2017). Conversely, dual-choice experiments showed that bumble bees avoid feeders containing field-realistic levels of Roundup, a GLY-based herbicide (Thompson et al., 2025).

Assessing honey bees' ability to escape or reduce exposure to environmental pollutants is crucial for understanding how they could adapt and survive in human-altered environments, and for exploring new approaches or application strategies that may enable pollinators to minimize their exposure to pesticides. To what extent bees can avoid resources as a mechanism to evade these toxic substances remains unclear. Their ability to do so may depend not only on detecting the pesticide itself, but on experiencing the malaise effects it triggers post-ingestion (Hurst et al., 2014). Furthermore, the resource may provide sensory cues (odour, taste, etc.) specific enough for bees to distinguish it from uncontaminated alternatives. In this regard, nurse bees, 3–4 to 9–10 day old workers that take care of the brood (Lindauer, 1952), provide an excellent opportunity to evaluate avoidance of pesticide-contaminated food, as they consume pollen to obtain proteins to feed the larvae.

While older adult bees may be directly exposed to contaminants when foraging on crop resources (Macri et al., 2024), younger adults, which performed in-hive tasks, are at risk due to the consumption and processing of contaminated resources stored within the nest (Blacquière et al., 2012; Mengoni Goñalons and Farina, 2015; 2018; Farina et al., 2019). Glyphosate has been detected in honey and beebread from hives near treated crops (Rubio et al., 2014; Thompson et al., 2014; Berg et al., 2018; El Agrebi et al., 2020; De Souza et al., 2021; Lace et al., 2025). Reported concentrations in honey averaged 69 µg/L and reached up to 1.6 mg/L. Average concentrations detected in beebread samples range from 0.059–0.69 mg/kg, which maximum values can reach 1.9 mg/kg (Lace et al., 2025). Although levels of IMI in pollen or beebread remain unknown, values measured in a global collection of honey can reach up to 6.3 µg/kg (Cresswell, 2011).

Inside the hive, pollen is mixed with honey and other components to be processed into beebread (Johnson, 2023). Nurse bees consume large amounts of pollen and beebread to support the development of their mandibular and hypopharyngeal glands, whose secretions are used to feed the larvae (Winston, 1987; Crailsheim et al., 1992).

Recent study indicates that nurse bees can adaptively change their pollen preferences to avoid unsuitable pollens, selecting the resources that have been incorporated into the nest (Lajad et al., 2021). Consumption preferences of nurse-aged bees were reduced for pollens adulterated with natural deterrents such as amygdalin or quinine, likely because bees associated pollen cues with malaise, forming memories that both reduced consumption and discouraged orientation toward unsuitable pollen (Lajad et al., 2021). Given the ability to avoid pollen experienced with noxious compounds, we wonder whether bees could also avoid pollen contaminated with pesticides to mitigate the deleterious effects of these compounds in colonies. So far, whether young bees can establish avoidance memories when exposed to beebread or pollen contaminated with the agrochemicals remains to be tested. If so, reducing the consumption of contaminated resources could decrease the likelihood of bees being exposed to harmful resources and prevent the spread of pesticides to nestmates, including larvae.

In this study, we aimed to assess whether young bees exhibit avoidance behaviour toward pesticide-contaminated pollen food or pollen food that bees had experienced as contaminated on previous days. Specifically, we measured the consumption preferences of nurse-aged bees for two monofloral pollen food, before, during and after one of them was adulterated with concentrations of GLY or IMI that had been reported in bee bread matrix or pollen.

The experiment lasted for 6 days and consisted of three different phases: initial, training, and testing (Fig. 1). The first 2 days (initial phase), bee cages received the two pollen options. For the training phase (days 3 and 4), we contaminated one of the two pollen options with GLY at the concentrations of 0 mg/l (control), 1.25 mg/l or 2.5 mg/l (Fig. 1) as described in the previous section.

To control for the effect of botanical origin on learning, we used two different pollen combinations. In a first series, cages received monofloral pollens from B. napus and D. tenuifolia (henceforth, B-D series). In a second series, cages received S. lorentzii and D. tenuifolia pollens (henceforth, S-D series). We contaminated B. napus in the B-D series and S. lorentzii in the S-D series. During the testing phase (days 5 and 6; Fig. 1), caged bees received the same two pollen types as in the training phase but unadulterated. Based on the volume (ml) of pollen consumed from each syringe during the final 24 h of the training phase (day 4), and the first 24 h of the testing phase (day 5) we calculated the Standardized Consumption (SC). For the B-D series, SC was defined as the volume of B. napus pollen consumed relative to the total pollen consumed (B. napus + D. tenuifolia pollen). For the S-D series, SC was defined as the volume of S. lorentzii pollen consumed relative to the total pollen consumed (S. lorentzii + D. tenuifolia pollen). We compared the SCs obtained for the three concentrations of GLY, taking into account both the experimental phase (training or testing) and the identity of the pollens offered in the cage (B. napus vs. D. tenuifolia: B-D series or S. lorentzii vs. D. tenuifolia: S-D series). The SCs calculated during the initial phase provided an estimate of the bees' initial preferences for the two pollens and were subsequently used in the statistical analysis to account for variations among bee cages (see the 'Statistics' section).

The same procedure was followed to test the effect of IMI on young bees' consumption preferences (Fig. 1). We followed the same procedure as in the previous experiment but used IMI instead of GLY to adulterate the pollen. We adulterated B. napus pollen in the first B-D series and S. lorentzii pollen in the S-D series. We used the following concentrations for IMI: 0 µg/l (control), 1µg/l or 10µg/l (Fig. 1).

In addition, we also calculated the total pollen consumption during the 6 days of the GLY and IMI experiments.

To assess if GLY or IMI exposures affect survival, every day, we counted the number of dead bees in the cage. Survival was calculated as the total number of bees that remained alive in the cage minus the total number of dead bees. To further challenge the bees and increase the resolution of the variable, we extended the experiment to 8 days. During these additional two days (days 7 and 8), bees continued to be fed unadulterated pollen.

The effects of GLY and IMI were addressed in separate analyses. All analyses were performed in R v. 4.2.1 (http://www.R-project.org/). To assess differences in survival we used the Kaplan-Meier method. We explored the impact of the concentrations of the compound (3 levels consisting of null, low, and high concentrations) and the pollen identity (2 levels consisting of B-D series and S-D series) as fixed effects, as well as their interaction. We performed pairwise comparisons within each series. To this end we used the log-rank test with Holm's correction for multiple comparisons, implemented through the pairwise_survdiff() function in the survival package (Therneau and Grambsch, 2000).

To assess differences in SCs, we used linear mixed models (LMM). Homoscedasticity and normality assumptions were checked (Levene's and Shapiro–Wilk's tests, respectively). In both GLY and IMI experiments we considered the phase of the experiment (2 levels consisting of the training and the testing phase), concentration and pollen identity as fixed factors. We also analysed their interaction. We added SCs obtained during the 'initial phase' and total pollen consumption (the consumption during the 6 days of experimentation) as covariates and included 'cage' as a random factor. The model was specified as: lme(SC ~ phase*concentrations*pollen_identity + SC_initial_phase + total_consumption, random = ~ 1|cage, data). The significance of effects was assessed using analysis of variance (ANOVA) on the fitted model. Tukey's post hoc tests were used for contrasts using the multicomp package in R.

To analyse the effects of GLY and IMI exposures on total pollen consumption during the six days of experimentation, a general linear model was fitted including concentration and pollen identity as fixed factors, and their interaction. The model was as follows: lm (total_consumption ~ concentrations*pollen_identity, data) The significance of effects was assessed using ANOVA on the fitted model. Tukey's post hoc tests were used for contrasts using the multicomp package.

Figure 4 shows the standardised consumption (SC) we used to quantify the consumption preferences of nurse-aged bees as a function of GLY exposure concentration, pollen identity and experimental phase (see Table S2). Pollen identity and experimental phase did not affect the SC (F_{1, 29} = 0.112; P = 0.741; F_{1, 29} = 0.037; P = 0.848). In addition, no interactions were statistically significant (Table S3A), although the interaction between phase, concentration, and identity showed a marginal trend (F₂,₂₉ = 3.30, P = 0.051). Among the covariates, total consumption was significant (F₁,₂₉ = 5.78, P = 0.023), while SC at the initial phase was not (F₁,₂₉ = 0.004, P = 0.950). This indicates that total consumption varied among cages, although such differences were not detectable during the initial phase of the experiment. Significant differences in SC were detected for the pesticide concentration (F_{2, 29} = 6.590; P = 0.004; Fig. 5), with values of 0.56 ± 0.02 at 0 mg/l, 0.50 ± 0.02 at 1.25 mg/l, and 0.43 ± 0.02 at 2.5 mg/l. SCs obtained in cages exposed to the highest concentration of GLY (2.5 mg/l) differing from those of the Control (0 mg/l) groups (t-ratio = 3.722; P = 0.002; Fig. 5). Estimated marginal means (± SE) were 0.559 ± 0.024 at 0 mg/l, 0.501 ± 0.024 at 1.25 mg/l, and 0.434 ± 0.024 at 2.5 mg/l, showing a clear reduction with increasing dose. Thus, the decrease in SC observed in cages exposed to the highest concentration could be solely explained by the effect of offering GLY-contaminated pollen during the training phase, which extended into the testing phase, regardless of pollen identity.

The SC consumption preferences of the nurse bees in the IMI experiment are shown in Fig. 6. The SCs varied according to pollen identity (F_{1, 27} = 8.366, P = 0.007; Table S3B; Fig. S1A) and the phase of the experiment (F_{1, 27} = 11.440, P = 0.002; Table S3B; Fig. S1B). Interestingly, SC in cages that experienced B. napus as the contaminated pollen exhibited values significantly higher than those cages that experienced S. lorentzii as the contaminated pollen, (B-D vs. S-D series_t-ratio = 2.679; P = 0.012; Fig. S1A). In addition, SC calculated at the testing phase (when both pollens were uncontaminated) were lower than those obtained at the training phase (t-ratio = 2.596; P = 0.015; Fig. S1B). Interestingly, the analysis detected a significant effect of IMI exposure on the SCs (concentration_F_{2, 27} = 3.644, P = 0.039), with values of 0.55 ± 0.03 in controls, 0.48 ± 0.03 at 1 µg/L, and 0.44 ± 0.03 at 10 µg/L, showing a decline for higher concentrations. Pairwise contrasts showed that the highest IMI concentration significantly reduced SC compared with the control (t-ratio = -2.550; P = 0.043; Fig. 7). No significant two-way interactions were found (Table S3B). SC at the initial phase (F₁,₂₇ = 0.35, P = 0.560) and total consumption (F₁,₂₇ = 0.11, P = 0.750) were not significant. In summary, consumption preferences were affected by contamination of pollen with both GLY and IMI showing that young bees were able to change their preferences to avoid pollens that had been experienced as less suitable for consumption.

We detected differences in pollen consumption in both GLY and IMI experiments (Table S2). Exposure to GLY affected total consumption according to the pollen identity (concentration*identity_F_2,30 = 4.895, P = 0.014). Bee cages in B-D series exposed to both concentrations of GLY consumed less than control (0 vs. 1.25 mg/l t-ratio = 3.257, P = 0.007; 0 vs. 2.5 mg/l t-ratio = 2.652, P = 0.033; Fig. S2). No differences were detected among groups in the S-D series (Table S4A).

For the IMI experiment, total pollen consumption did not differ by concentration, experimental phase, pollen identity or any of their interactions (Table S4B).

Bees are the most relevant pollinators in agricultural environments and are probably the most exposed to the presence of agrochemicals in terrestrial ecosystems. Here, we investigated honey bees' ability to avoid exposure to these environmental pollutants to understand to what extent they can cope in human-altered environments. Despite the multiple effects of pesticide exposure on bee behaviours, our results indicate that young honey bees can recognise pollen contaminated with GLY or IMI and selectively reduce their consumption. In general, the highest levels of avoidance occurred on the second day of training, which supports the hypothesis that bees did not immediately avoid contaminated pollen when GLY or IMI were present in food, but instead showed avoidance after several hours of exposure. Avoidance also occurred when pollen was no longer contaminated, indicating that young honey bees can actively reduce the pollen consumption based on previous experience.

Testing consumption preferences with two uncontaminated options allow us to support the hypothesis that young workers associate pollen-related cues with experiences of agrochemical-induced malaise, leading to memories that influence bee responses for at least 24 hours. This finding is consistent with the ability of young worker honey bees to avoid pollen they have learned to be noxious (Lajad et al., 2021). Thus, other than spontaneous responses to chemicals such as GLY or IMI, bees may notice the presence of the agrochemical through its delayed effect on their physiological state (Ayestarán et al., 2010; Hurst et al., 2014; de Brito Sanchez et al., 2015; Lajad et al., 2021). In particular, honey bees may associate specific cues of the food, like its scent or taste, with the malaise experienced after consuming pollen or nectar contaminated with noxious compounds.

Most studies examining the effects of GLY on honey bees have applied them diluted in sucrose solutions, primarily focusing on the behaviour of mature workers (Decourtye et al. 2004a; 2004b; Bortolotti et al., 2003; Eiri and Nieh, 2012; Williamson et al., 2013; Fischer et al., 2014). To a lesser extent, research has explored their impact on young adult workers (Mengoni Goñalons and Farina, 2018) or after incorporating agrochemicals into a pollen matrix (Dively et al., 2015; Azpiazu et al., 2023). From these studies, we can recognise that GLY acts on different timescales. In the short- or intermediate-term, GLY administration has been shown to cause negative effects on orientation and navigation (Balbuena et al., 2015), changes in taste sensitivity and learning (Herbert et al., 2014; Mengoni Goñalons and Farina, 2018) and alterations in the sleep pattern (Vázquez et al., 2020a). In the long-term, GLY exposure has been primarily associated with disruptions in the microbial flora (Motta et al., 2018; 2020; Vázquez et al., 2023). Bacterial symbionts in the gut are beneficial for enhancing digestion, detoxification of food, and defence against pathogens and parasites (Engel et al., 2016; Zheng et al., 2018; Romero et al., 2019). Consequently, exposure to glyphosate increases susceptibility to opportunistic pathogens (Motta et al., 2018; 2020).

Although IMI operates through a completely different mode of action, the same principle of malaise could apply to bees that become intoxicated with food contaminated with this insecticide. Imidacloprid belongs to the neonicotinoid class of insecticides, which disrupt neural activity by binding to nicotinic acetylcholine receptors (Matsuda et al., 2001). Among young workers, IMI affects gustatory responsiveness and olfactory learning (Mengoni Goñalons and Farina 2015; Mengoni Goñalons and Farina, 2018), showing that the cholinergic nicotinic pathway is particularly sensitive to neonicotinoids during early adulthood (Guez et al., 2001; 2003). Electrophysiological approaches revealed that bees cannot taste neonicotinoids (Kessler et al., 2015), suggesting they are unable to reduce food consumption based on changes in the resource's palatability due to IMI. Then, if exposure to neonicotinoids at field-realistic concentrations (Cresswell, 2011; Blacquiere et al., 2012) is not lethal, individuals may be able to recall firsthand the effects of the neurotoxin and link them to the feeding event that occurred before the intoxication.

While reducing the consumption of contaminated food may help mitigate exposure to agrochemicals, it may also lead to a reduction in overall intake, which can adversely affect bee nutrition and health, ultimately compromising colony survival. Mengoni Goñalons and Farina (2018) reported that young honey bees reduce the consumption of GLY-contaminated sucrose solutions. This phenomenon was further observed in our experiments. Caged bees exposed to GLY in the B-D series exhibited lower levels of total pollen consumption than the control group (Fig. S2 and Table S4A). While we cannot rule out the possibility that GLY reduces pollen palatability, the experimental design with two alternative (i.e. contaminated and uncontaminated pollen options) supports the hypothesis that the overall reduction in pollen consumption was due to malaise induced by the agrochemical (Hurst et al., 2014). However, no significant changes in bee survival were detected in B-D series, suggesting that in this case, the reduction in consumption did not significantly affect workers' nutritional status.

In a simplified scenario involving only two pollen types with distinct nutritional profiles, shifts in feeding preferences induced by contamination could enhance or impair bees' nutritional status, independently of their total pollen intake. B. napus pollen is known for its high lipid content (Evans et al. 1991; Somerville 2005), intermediate-quality crude protein (Evans et al. 1991; Roulston et al. 2000; Pernal and Currie 2001; Somerville and Nicol 2006), and a high proportion of essential amino acids (Cook et al. 2003; Somerville and Nicol 2006). A comparable profile was found in D. tenuifolia pollen, though with a slightly higher lipid content (Tourn, 2013). Thus, although GLY treatment reduced overall pollen consumption, it also induced a shift in the relative intake of the two pollen types, which nonetheless provided sufficient nutrition to meet bees' requirements. In the S–D series, bees did not reduce their overall consumption; however, survival was lower in the control cages than in the treated ones. Once again, differences in mortality could be explained by GLY-induced changes in feeding preferences, which reduced the intake of a highly palatable pollen in favour of a more nutritious one. While D. tenuifolia pollen is recognized as highly nutritious, information on the nutritional composition of S. lorentzii is still lacking to substantiate this hypothesis (see Changazzo and Salgado Laurenti, 2019 for total protein).

In the IMI experiments, the situation may be somewhat different, as the higher mortality might result from an increased consumption of the contaminated pollen. Bees in the B–D series, but not in the S–D series, showed lower survival than their respective controls, probably because they ingested relatively more B. napus pollen, likely more highly valued by the bees, than S. lorentzii in the S–D series.

GLY in pollen samples has been detected at maximum concentrations of 0.024 mg/kg (Jesús et al., 2024), 0.025 mg/kg (Lace et al., 2025), and up to 0.162 mg/kg (Macri et al., 2024) depending on the location from which the samples were collected. Reported maximum GLY levels in beebread range from 0.54 mg/kg (Bergero et al., 2021) to as high as 1.9 mg/kg (Lace et al., 2025), clearly exceeding the European Union maximum residue limit of 0.05 mg/kg. Levels of IMI in pollen or beebread remain unknown, but maximum concentrations measured in a global collection of honey reached 6.3 µg/kg (Cresswell, 2011), providing a reference point for potential exposure in honey bees. Considering these values, it can be argued that the concentrations used in our experiments (see Methods for equivalences), although high, still fall within the range reported for hive products.

Exposure to agrochemicals under laboratory conditions rarely reflects real environmental circumstances, yet these studies remain crucial for understanding the strategies organisms employ to cope with toxic compounds. Under natural conditions, the way in which the different types of pollen are presented to in-hive bees may not be as conducive as it is in our laboratory trials. Inside the nest, foragers may deposit pollen from different plant species within the same cells, which could make it challenging for nurse bees to distinguish between the pollen types. However, under certain circumstances, such as during mass flowering events (Bänsch et al., 2020; Macri et al., 2024) or in agricultural settings with low floral diversity, one or a few species may constitute almost all of the colony's resources, making it easier for nurse bees to associate the cues of a particular pollen type with any potential malaise effect.

The ability of bees to avoid agrochemical compounds likely arises as an adaptation to avoid harmful pollens. While this ability may help improve colony productivity and development in environments where only certain plant pollens are toxic, it is still unclear how much bees actually benefit from avoiding pesticides in agricultural settings, where all resources within a bee's foraging area could potentially be contaminated. In agroecosystems, the availability of contaminated versus uncontaminated resources depends on several factors, including the farmer's management practices and the phenology of cultivated and wild plants, which may bloom before, during, or even after pesticide application. Although agrochemicals that affect the bees are commonly applied through spraying (Giesy et al., 2000; Macri et al., 2025), other methods might favour contaminated and non-contaminated resources to coexist within the same honey bee foraging area, making sense of the bees' strategy of selecting non-contaminated pollen. Our results suggest that bees can reduce their exposure if the context is suitable. Therefore, it is crucial to explore new strategies that facilitate the separation of contaminated from uncontaminated resources. In this regard, agricultural practices, pesticide application methods, timing of applications, and the phenology of both crop and wild surrounding plants should be carefully considered and integrated to create conditions that allow bees to access less contaminated resources, even in landscapes where pesticides are used.

While estimating the impact of avoiding contaminated pollen on the development and performance of honey bee colonies is challenging, it is evident that brood would benefit from such avoidance. Larvae fed food containing GLY traces are likely to experience a delayed moulting, reduced weight and survival, and changes in microbiota as well (Vázquez et al., 2018; 2023; Dai et al., 2018, Li et al., 2024). Then, the avoidance of contaminated pollen by nurse bees would positively affect the quality of the jelly fed to the larvae, thereby not affecting their development. To what extent the avoidance of contaminated pollen by young worker bees also influences the decisions of bees engaged in other food-related tasks remains to be determined. The possibility that the accumulation of unsuitable or low-quality pollen, resulting from nurse avoidance behaviour, becomes a source of information for foragers has recently been discussed (Lajad and Arenas, 2024; 2025), after showing that unsuitable pollen offered within the hive led foraging bees to avoid the resource at the food source.

This study constitutes the first comparative assessment of young honey bees' ability to discriminate between pesticide-contaminated and uncontaminated pollen at field-realistic concentrations, and to subsequently learn to avoid the contaminated resource. We show that bees can discriminate between the two pollen types in a delayed manner, likely after associating pollen-related cues with malaise induced by agrochemical ingestion. The scope of our finding, and the extent to which this avoidance response contributes to bees' ability to cope with contaminated environments remains unclear. Nonetheless, given the limited prospects for reducing agrochemical use, it is crucial to understand the mechanisms that allow bees to avoid exposure to these agents to propose effective management strategies. So far, our results highlight that access to uncontaminated alternative pollen resources may provide a significant advantage for enhancing colony performance and adaptability within agroecosystems.