Honey bee colonies change their foraging decisions after in-hive experiences with unsuitable pollen

Animals depend on their perceptual ability to select the most suitable food resources to obtain nutrients. To this end, foragers assess the resources in situ to decide whether to exploit the food source or not. Furthermore, foraging occurs not only to satisfy the immediate needs of foragers, but of other group members (Pierce and Ollason, 1987). Therefore, resource selection can be conflicting if the requirements of the individuals foraging and those using the resources are not the same. Among social insects, foraging is restricted to a subset of workers whose requirements differ from other colony mates (Dussutour and Simpson, 2009). For example, in the honey bee Apis mellifera, pollen is not directly used by foragers, but by nurse bees that consume most of the pollen and process it to feed the larvae and the rest of the colony mates (Moritz and Crailsheim, 1987; Crailsheim et al., 1992). It is not currently known whether pollen selection is based on foragers' assessment at the foraging site or influenced by pollen-related experiences within the nest.

Pollen is the main protein and lipid resource for honey bee colonies and strongly differs in composition among plant species (Roulston and Cane, 2000; Manning, 2001; Somerville and Nicol, 2006). Pollen might also contain deterrent compounds, such as nicotine alkaloid in the tobacco plant or the cyanogenic glycoside amygdalin in almond trees (Detzel and Wink, 1993; London-Shafir et al., 2003). Honey bee foragers (workers over 2 weeks of age) collect pollen with their legs and mouthparts at the food source but are unlikely to ingest it, as they cannot digest pollen (Moritz and Crailsheim, 1987). In this scenario, foragers are unlikely to evaluate pollen composition at the source since detection of many of its constituents requires ingestion. Evidence suggests that foragers' assessment of pollen is limited to pre-ingestive pollen cues available in the outermost layer of the grain (Pernal and Currie, 2002), the pollen kit. The pollen kit also enables foragers to discriminate between resources from different flowering plant species (Dobson and Bergström, 2000; Piskorski et al., 2011) and it might provide information about pollen constituents as its composition could be similar to that of the internal fractions of pollen (Dobson, 1988). Consistently, research on learning and memory shows that honey bees do not need to ingest pollen to reinforce associative learning, as pollen reinforcement occurs through chemo-tactile stimulation of the bees’ tarsi or antennae (Nery et al., 2020; Moreno and Arenas, 2023).

Inside the nest, young workers process pollen into bee bread. Pollen processing includes the addition of sugars, honey and glandular secretions that make the resource easier to digest and probably increases its attractiveness to young bees (Carroll et al., 2017). Nurses (in-hive workers of about 3–10 days of age) consume both fresh pollen and bee bread to incorporate protein into their hypopharyngeal gland secretions that are used for feeding larvae, young workers and queens (Crailsheim et al., 1992). Nurse bees also share jelly with foragers (Crailsheim, 1991) providing them with protein and probably also with information on the nutritional quality of the colony's diet. However, it is unknown whether species-specific information of pollens can also be shared through this mechanism.

During the food selection process, avoiding intoxication is as important as achieving nutrition. Interestingly, nurse bees avoid consuming pollens that have been previously determined to be unsuitable (e.g. adulterated with amygdalin or quinine), which suggests that nurses participate in the process of pollen selection (Lajad et al., 2021). So far, whether and to what extent foragers are able to avoid unsuitable resources solely based on their ability to assess pollen composition at the sources or as a consequence of information acquired within the nest (Hendriksma and Shafir, 2016; Zarchin et al., 2017) remains to be tested. We hypothesise that foraging preferences are adjusted after experiences mediated by the pollens that are being processed inside the nest. One possibility is that foragers might learn to reject pollen cues directly after ingesting small samples of the unsuitable pollen stored in cells (Dreller et al., 1999; Dreller and Tarpy, 2000; Calderone and Johnson, 2002). Alternatively, they could associate pollen-based cues with the effects that the pollen causes on the colony according to its suitability (Seeley, 2009); for example, if the food distributed by the nurses contained toxins that induced malaise in foragers causing the adjustment of pollen collection.

To unveil how foragers obtain information to guide pollen searching, we investigated their abilities to perceive, learn and memorise pollen cues according to the resource quality and in-hive experiences. In dual-choice experiments with colonies confined in flying cages, we first evaluated the avoidance response of foragers to pollens adulterated with amygdalin at the food source. In a second experiment, we quantified preferences for two unadulterated pollens, after one of them was offered adulterated inside the nest. In a third experiment, we measured pollen selection in the field by means of the diversity of pollen samples trapped at the entrance of the hives that were supplied (or not) with amygdalin-adulterated pollen from Diplotaxis tenuifolia, one of the dominant flowering plants in the surroundings. How foragers obtain information for pollen selection is relevant and timely given the increasing alteration of environments by humans that challenges bees' ability to obtain an adequate diet.

We performed experiments during the summer seasons of 2019–2020 and 2020–2021. Experiments I and II were done in the experimental field of the School of Exact and Natural Sciences of the University of Buenos Aires (34°32′S, 58°26′W). We used 8 colonies for experiment I, 8 for experiment IIA and 8 for experiment IIB. All colonies contained a mated queen, 3–4 brood frames and 1–2 frames with food reserves. For these experiments, we confined colonies in flying cages (9×3×3 m) one at a time. The enclosure prevented the interference of foragers with bees from other colonies, while it limited the sources of pollen and artificial nectar available to the experimental colony. Once in the flying cage, we trained foragers to visit a foraging station offering crushed bee-collected multifloral pollen in an ad libitum feeder. Foraging station was located 8 m from the hive entrance. The colonies were confined in the cages for approximately 1 week prior to the onset of the experiments, which was the minimum time necessary to get a constant and relatively high number of pollen foragers visiting the station.

Experiment III was performed in a sunflower crop, located in Coronel Pringles, Buenos Aires, Argentina (68°6′11.94″S, 61°25′46.00″W). We used 22 colonies randomly chosen from a total of 68 colonies settled in 3 different plots within the field. All colonies contained a mated queen, 7–8 brood frames and 2 frames with food reserves. We performed all experiments according to the animal care guidelines of the National Institute of Health (1985) and the current laws of Argentina.

For bees to differentiate pollen according to their related cues (odour, tastes, texture, etc.) and for preferences among pollen options to be comparable within and between experiments, we used monofloral pollens. Monofloral pollens were obtained from multifloral or monofloral bee-collected samples (provided and characterised by Pampero, Amuyen, and CoopSol cooperatives). In the case of multifloral samples, we separated the pollen pellets according to their botanical origin based on their colour, brightness, texture and degree of agglutination. Most pollen samples were obtained in the same season used for data collection. To reduce the quality of pollen, we crushed, weighed and mixed the pollens with amygdalin (Sigma Aldrich, BioXtra, ≥97.0%), a cyanogenic glycoside that naturally occurs in almond pollen (Detzel and Wink, 1993; London-Shafir et al., 2003). We used a concentration of 0.1 mol l⁻¹ previously tested in Lajad et al., 2021. Amygdalin causes malaise-induced behaviours (Hurst et al., 2014) and post-ingestion mortality (Kevan and Ebert, 2005; Ayestaran et al., 2010) when added to a sugar solution fed to individual bees at concentrations higher than it can occur in nature.

The aim of experiment I was to test whether foragers were able to assess pollen composition at the foraging site and, therefore, to perceive amygdalin-adulterated pollen during their foraging visits and make a switch to unadulterated pollen. To this end, we measured foraging preferences of colonies for two different monofloral pollen: one offered adulterated and the other unadulterated. Such preferences were compared with preferences measured in a different group of colonies, whose foragers chose between the same two monofloral pollens offered unadulterated (Fig. 1A).

Here, we used Schinopsis lorentzii versus Salix sp. pollens. At the foraging station, S. lorentzii pollen was always offered unadulterated (Fig. 1A). On the other hand, Salix could be presented as adulterated or unadulterated according to the treatment assigned to the colony. The two pollens were offered in separate feeders located 2 cm apart. The pollen feeders consisted of 360 ml opaque plastic pots with 8 holes (7 mm diameter) in the top (lid) through which the bees could enter and leave loaded with pollen. Inside the pots, 2–3 g of pollen was attached to the bristles of a pipe cleaner (20 cm) rolled into a spiral shape (Arenas et al., 2021). The design of these feeders favours the use of pollen olfactory cues by foragers while it limits the range of visual cues.

The bees that left the feeders loaded were counted and captured to ensure that they did not carry pollen back to the nest. Here, and in the following experiments, bees already tested were retained in plastic tubes until the end of the evaluation, to ensure that a bee was not measured twice.

Foragers had no previous experience with the pollen offered during testing, so we measured preferences that were not influenced by the experience.

Experiment I consisted of 2 testing events on successive days (24 h interval): the first one (T0) with both pollens unadulterated and the second (T1) with Salix offered either adulterated or not (Fig. 1A). Although all experimental hives came from the same stock and the same apiary, we cannot rule out variability between colonies, probably related to the different types of pollens they experienced before confinement or their particular nutritional needs. Testing trials were carried out at 11:00 h. Trials were extended until we measured the preferences of at least 30 bees or until 1 h had elapsed. From each test (T0 and T1) we obtained the proportion of foragers that preferred Salix as the number of bees foraging on Salix over the number of total foragers collecting pollen in both feeders. We used preferences obtained at T0 as covariate in the analysis (see ‘Statistic analysis’). Then, we compared colony preferences for unadulterated or adulterated Salix pollen presented at the foraging site.

Here, we addressed whether preferences of colonies that had experienced the adulterated pollen differed from preferences of colonies that had experienced the same pollen unadulterated. The objective was for the bees to learn the effects of an adulterated pollen by contrasting/differentiating them from the effects of a different, unadulterated pollen. The decision to perform such a design is based on the fact that the associative experience between a monofloral pollen and an aversive substance failed to establish a memory that biases the consumption preferences of young bees (Lajad et al., 2021). In contrast, when bees were given the opportunity to learn by discriminating the cues of two pollen, one adulterated and one unadulterated, they successfully learned to avoid the adulterated pollen (Lajad et al., 2021).

The experiments consisted of 2 phases and 3 testing events (Fig. 1B,C). In a first test (T0) we measured initial preferences for two monofloral pollens. Again, initial preferences were used as covariate in the analysis to account for variations among experimental colonies (see ‘Statistic analysis’). Immediately after T0, we ran the first phase of the experiment (P1) lasting 2 days, when we offered the adulterated or unadulterated pollen that would establish the in-hive experience. In experiment IIA, we offered adulterated or unadulterated Salix pollen at the foraging station and let the bees collect and introduce it into the nest. Here, Salix pollen was tested against S. lorentzii pollen, which was always presented unadulterated (Fig. 1B). In experiment IIB, adulterated or unadulterated Brassica napus pollen was hydrated to a bee bread-like paste that we poured directly inside the hive (10 ml each day for 2 days), on the top of its central frames. At the foraging station B. napus pollen was tested against Diplotaxis tenuifolia pollen, both unadulterated (Fig. 1C). Note that, despite being similar, experiment IIA includes the possibility of bees foraging on adulterated pollen in P1 (Fig. 1B) may establish aversive associations with the resource over successive visits.

Two days after the onset of the experiment, we measured the effect of in-hive experience on preferences for pollens (both unadulterated) on a second testing trial (T1; Fig. 1B). We then conducted the second phase of the experiment (P2), which ran from day 2 to day 4. In the second phase of the experiment IIA, we let the foragers naturally take the unadulterated Salix pollen from the station back to the nest to assess whether memories established with unsuitable pollens can be reverted on the basis of newly available information. In the second phase of experiment IIB, we also let the foragers take the unadulterated B. napus pollen back to the nest, but at the same time we offered it adulterated inside the hive for the second time. By doing this, we tested the extent to which in-hive pollen mediated experiences resist conflicting information. On day 4, we tested foraging preferences for the third time (T2; Fig. 1C). Proportion of bees at testing trials were compared between colonies treated with unadulterated- and adulterated-pollen.

Here, we investigated whether in-hive experiences mediated by unsuitable pollen affected pollen selection of commercial hives that foraged in an agricultural setting. To this, we monitored a first group of colonies that we supplied with amygdalin-adulterated pollen of D. tenuifolia, a dominant flowering plant highly available in the surrounding prior and during the first day of the experiment. In addition, we surveyed a second group of colonies supplied with unadulterated D. tenuifolia pollen. Finally, we also monitored a third group of colonies that received no extra pollen. We offered D. tenuifolia pollen as in experiment IIB.

We used the floral diversity of collected pollen as a proxy of the colony's pollen selection. Pollen samples were collected using conventional pollen traps (frontal-entrance trap), consisting of a wooden structure with a removable metal mesh inside. We sampled colonies for 4 h, 1 day before (T0) and 2 days after the beginning of in-hive pollen offering (T1; Fig. 1D). We separated pollen pellets from the samples based on their colour. We weighed them to calculate the relative proportion of each pollen type and in turn, we used those proportions to calculate the Shannon diversity index (Hill, 1973). We used the diversity of samples taken at T0, as a covariate to control for differences among colonies before manipulation.

All statistical analyses were performed in R (https://www.r-project.org/). Differences in forager choices (proportions) of experiment I were addressed by means of generalised linear model (GLM) and in experiment II, differences were addressed by means of generalised linear mixed model (GLMM), both experiments with binomial distributions (Crawley, 2012). Model assumptions were checked with the DHARMA library (https://CRAN.R-project.org/package=DHARMa) and the variation in the number of individuals tested in each event was accounted for through the weights. In experiment I, we explored the impact of treatment (unadulterated and adulterated) as a fixed effect and we added initial preferences as covariate. By adding initial preferences as a covariate, we controlled for intrinsic variation among hives and thus prevented them from influencing post-treatment results. Here, we used the glm function of the Stats package. In experiments IIA and IIB, we analysed the effect of treatment (unadulterated and adulterated) and test (T1 and T2) as fixed effects and its interaction. We added the colony and the month when the colony was studied, as random factors and the initial preferences as covariate. Here, we used the glmer function of the lme4 package (https://cran.r-project.org/package=lme4) that uses Wald Z-tests to approximate P-values for GLMMs (Bolker et al., 2009). Post hoc contrasts were conducted on models to assess effects and significance between fixed factors using emmeans v.1.5 (https://CRAN.R-project.org/package=emmeans) with a significance level of 0.05.

Differences in the Shannon diversity index (H) of experiment III were assessed using general linear models with normal distribution. We explored the effect of treatment (unadulterated, adulterated, no-pollen) and we used initial preferences as covariate. Homoscedasticity and normality assumptions were checked (Levene and Shapiro–Wilk tests, respectively). Tukey's tests were used for contrasts using the multcomp package in R (https://CRAN.R-project.org/package=multcomp).

Colonies did not change their foraging preferences when amygdalin was added to the pollen at the feeder. Our results showed that foraging preferences of colonies exposed to adulterated (0.567±0.025) or unadulterated pollen (0.586±0.026) were similar (treatments: χ²=0.291, d.f.=1, P=0.590; Fig. 2).

Colonies significantly reduced their foraging preferences for the pollen that had been offered adulterated inside the nest, either if it was incorporated by foragers themselves (experiment IIA; Treatment×Time: χ²=12.606, d.f. =1, P<0.001; Fig. 3A) or artificially, on the top of the frames (experiment IIB; Treatment: χ²=17.999, d.f.=1, P<0.001; Treatment×Test: χ²=0.076, d.f.=1, P=0.782; Fig. 3B). Both experiments (IIA and IIB) showed similar results for T1. In experiment IIA, foragers from amygdalin-treated colonies showed a lower preference for Salix pollen (adulterated: 0.479±0.049, N=4) than bees from unadulterated-pollen treated colonies (0.682±0.043, N=4) (unadulterated–adulterated: z-ratio=3.023, P=0.003). Similarly, in experiment IIB, bees from amygdalin-treated colonies choose B. napus pollen less (adulterated: 0.255±0.049, N=4) than controls (unadulterated: 0.459±0.055, N=4) (unadulterated–adulterated: z-ratio=3.387, P<0.001).

The results of experiments IIA and IIB differed by the third testing event (T2). In experiment IIA, where foragers switched from carrying adulterated to carrying unadulterated Salix pollen, the differences detected at T1 between amygdalin-treated colonies and controls were lost (adulterated: 0.565±0.049; unadulterated: 0.523±0.050; unadulterated–adulterated: z-ratio=−0.608, P=0.543). In contrast, when we let the foragers incorporate unadulterated B. napus pollen while we continued offering it inside the hive (experiment IIB), preference for B. napus pollen remained lower in the treated group than in the control (unadulterated: 0.495±0.062; adulterated: 0.265±0.049; unadulterated–adulterated: z-ratio=3.405, P<0.001).

Colonies showed differences in the diversity (i.e. composition and/or relative abundance) of pollen selected after the addition of the adulterated pollen inside the hives (Treatments: F_2,17=4.671, P=0.024; Fig. 4). Colonies that did not receive extra pollen showed the lowest diversity index (no pollen: 1.1±0.29, N=6) while colonies that received the adulterated D. tenuifolia pollen showed the highest index value (1.88±0.5, N=8). Intermediate values of diversity were observed in colonies that received unadulterated D. tenuifolia pollen (1.43±0.59, N=7). Statistical analysis detected significant differences in the H-index between colonies treated with adulterated and colonies with no added pollen (adulterated–no pollen: t-ratio=−2.584, P=0.048). No differences were detected between colonies with no added pollen and colonies supplied with unadulterated pollen (no pollen–unadulterated: t-ratio=−0.976, P=0.601) or between colonies supplied with unadulterated and amygdalin-adulterated pollen (unadulterated–adulterated: t-ratio=−1.725, P=0.225).

Many questions about the process by which honey bees evaluate and select pollen remain unanswered. Here, we addressed how honey bee colonies deal with deterrents in pollen. We found that foragers do not respond to the addition of amygdalin at the pollen source during the first encounter with the unsuitable resource. Instead, the offering of amygdalin-adulterated pollen within the hive was enough to change bees’ foraging preferences. Thus, foragers can make decisions that are not solely based on resource perception, but on resource-related information acquired in the hive environment. The impact of in-nest experiences on foraging responses have been well documented among social insects, including the honey bees (e.g. Farina et al., 2020), stingless bees (Mc Cabe and Farina, 2009), bumblebees (Dornhaus and Chittka, 1999), yellow jacket (Jandt and Jeanne, 2005); and ant colonies (Arenas and Roces, 2017), within both appetitive (Arenas et al., 2007, 2008) and aversive contexts (Arenas and Roces, 2016, 2017). Here, we extend our knowledge of the behaviours that are influenced by in-nest experiences, showing that the incorporation of unsuitable pollen affects pollen foraging.

Pollen is a complex and chemically variable resource that provides the nutrients for the development of the colony. Pollens differ greatly in composition depending on their botanical origin, with some pollens containing low levels of protein and/or lacking essential nutrients (Herbert et al., 1970; Wille et al., 1985; Roulston et al., 2000; Rasmont et al., 2005; Somerville and Nicol, 2006; Weiner et al., 2010). Recollection is therefore expected to be biased towards pollen with the higher nutrient value at a given moment, such as protein or lipids. Preferences for protein-rich pollen have been observed in other bee species (Vaudo et al., 2016), but to a lesser extent in the honey bee (Solberg and Remedios, 1980; Schmidt, 1984; van der Moezel et al., 1987; Pernal and Currie, 2001; Pernal and Currie, 2002; Leonhardt and Blüthgen, 2012; Ruedenauer et al., 2015; Beekman et al., 2016). As expected, we found that honey bees foragers were unable to spontaneously reject amygdalin at the foraging site, both when bees collected finely crushed (i.e. powdered) pollens (experiment I) and in the presence of hydrated pollens (Fig. S1). The lack of preferences is consistent with bees being unable to distinguish between a sugar solution adulterated with amygdalin and an unadulterated sugar solution (Wright et al., 2010).

As already proposed for the amygdalin, honey bee foragers might learn in a delayed manner through the effects of consuming adulterated food. The idea that amygdalin causes malaise post-ingestion in bees was proposed by observing that an adulterated sucrose solution impairs learning toward the end of the olfactory conditioning (Wright et al., 2010). More evidence comes from bees decreasing their level of responses to the learned odour after associating the sucrose reward with an aqueous solution of amygdalin (Ayestarán et al., 2010). Therefore, changes in foraging preferences after the incorporation of adulterated pollen into the nest might involve learning of pollen-based cues (odour, taste, texture, etc.) with the post-ingestive consequences of amygdalin consumption.

Whether pollen ingestion that triggers avoidance learning occurs at the food source or exclusively inside the nest is not entirely clear. Although evidence suggests that mature honey bees would not consume raw pollen (Moritz and Crailsheim, 1987; Crailsheim et al., 1992), we cannot completely rule out bees ingesting some very small amounts while collecting it. Consequently, our results cannot exclude that foragers may have learned the association during successive visits to the food source (only in experiment IIA). Even if this would be the case, these experiences seem not to improve avoidance of the unsuitable pollen compared to those taking place entirely within the nest (experiment IIB), exposing that avoidance can occur exclusively by the information gained inside the hive. Given that we used different monospecific pollens, and that the experimental design was not balanced, the lack of rejection [e.g. in the second trial (T2) in experiment IIA] may be explained, at least in part, by differences in the nutritional demands of the colonies, as Salix may offer nutrients that S. lorentzii pollen does not and vice versa.

Having found that learning about pollen toxicity could take place directly from experiences with adulterated pollen within the hive, we wondered how this might occur. On the one hand, foragers might assess resource constituents from partially digested pollens that are available in the cells that they inspect before and after their foraging trips (Dreller et al., 1999; Dreller and Tarpy, 2000). Similarly, we proposed that for learning at the source foragers might ingest small samples of pollens that are being processed into bee bread (Dreller et al., 1999; Dreller and Tarpy, 2000; Calderone and Johnson, 2002). Newly prepared bee bread is easily digestible and more attractive to the young bees than raw pollen and stored pollen reserves (Carroll et al., 2017). Partially digested pollen in the cell would improve access to compounds that were trapped inside the pollen grain, then less available at the source to be evaluated during foraging. Furthermore, the bee bread is likely to maintain pollen cues, which could be associated with the consequences of consuming an unsuitable resource. Pollen cues associated with jelly distributed by nurses containing toxins that induce malaise in foragers could also be learned. Evaluation of pollen while being processed could be adaptive, preventing foragers from collecting and accumulating resources that are harmful to larvae and young bees.

Another possibility is that pollen selection in honey bees does not require the individual ability of mature workers to detect resource suitability on their own, but could be achieved through feedback from larvae and young bees, so foragers would be biased to revisit the same food source that enabled proper nourishment of the brood. Thus, foragers might learn about pollen suitability not from the resource itself (either in the food source or inside the hive), but through interactions with nurses or the changes that they generate in the nest. In this regard, we recently showed that nurse bees are sensitive to both toxic and nutritional pollen components (Lajad et al., 2021). Nurse-aged bees could associate pollen-based or pollen-related cues with malaise experiences triggered by pollens adulterated with amygdalin or quinine, leading to aversive memories that reduce pollen consumption and discourage orientation to pollen-related cues (Lajad et al., 2021). Under this scenario, unsuitable pollen would be consumed less and probably accumulates in the nest, providing information for other colony mates. For example, volatiles or tastes of unsuitable pollen that accumulate, age and/or become ‘entombed’ in the cells (van Engelsdorp et al., 2009) may induce latent inhibition among foragers that translates such experiences into avoidance responses during searching (Fernández et al., 2009).

In addition, we must consider the impact of interactions between foragers and in-hive bees which, having learned about the noxious properties of certain pollens, may discourage foragers by not following their dances or by not sharing food when they begged.

The hypothesis that pollen decision-making is influenced by the access to information about the different resources being processed in the nest is supported by an additional experiment with colonies deprived of pollen intake for 3 consecutive days (Fig. S2). Using hives with permanent pollen traps that were fed artificial protein supplements, we observed a decrease in the pollen diversity of their samples compared with colonies that freely incorporated pollen (Fig. S2). We speculate that the lack of information about the different pollens inside the nest led foragers to make decisions based on their innate preferences for pollen-related cues rather than on the effects of pollen on the nutritional or health status of the colony.

By contrast, foraging patterns of the colonies supplied with amygdalin-adulterated pollen (experiment III) showed a shift towards diversification of pollen sources. This finding is consistent with evidence that colonies exposed to pollens low in protein content, lacking essential nutrients and/or containing toxins, increase the diversity of pollen types they collected (Di Pasquale et al., 2013, 2016), likely as a strategy to dilute the effect of any unsuitable pollen by mixing different types. In addition, colony responses to nutritional deficits include the selection of pollens with complementary nutrients (Hendriksma and Shafir, 2016; Zarchin et al., 2017). Diversification might be achieved by reallocation of foragers to novel resources, which is likely to be induced by the avoidance response to pollens that had been experienced as unsuitable inside the hive. We detected no changes in the total weight of pollen samples collected in colonies treated with no-pollen, unadulterated or adulterated pollen (Fig. S3), suggesting that pollen foraging activity did not change after offering an unsuitable pollen. This evidence rules out the idea that diversification is due to a generalised increase in foraging activity that enables the incorporation of pollen from less-represented floral species in the colony's diet because of better exploration/exploitation of the environment.

In an attempt to measure changes in the proportion of D. tenuifolia pollen collected after being offered adulterated within the colonies, we noted important fluctuations in the availability of D. tenuifolia pollen during the experimental period that prevented detecting any effect among treatments. Diplotaxis tenuifolia was abundant during the first day of the experiment (28 January 2021); however, its availability decreased during the following days (31 January onwards), resulting in very little D. tenuifolia pollen in the pollen samples of all treatments (Fig. S4A). Interestingly, a preliminary study conducted in 2020 (11–15 January), where D. tenuifolia remained abundant throughout the experimental period revealed that amygdalin-treated colonies showed a reduction in the proportions of D. tenuifolia collected compared with levels in control colonies (Fig. S4B). In the 2020 season, treated colonies that foraged less pollen from D. tenuifolia, also exhibit a decrease in the diversity of collected pollen samples (Fig. S5), suggesting that despite differences in experimental conditions between confined hives (collecting only two pollen types inside the flying cage; experiment II) and hives collecting freely in the field (experiment III), the colony's response to unsuitable pollens is the same. Furthermore, this approach was complemented using a third group of colonies treated with pollen adulterated by quinine, a naturally bitter-tasting alkaloid that bees detect (and reject) prior to ingestion (Wright et al., 2010; Lajad et al., 2021). Thus, we confirmed that the changes in pollen selection patterns not only occur with amygdalin-adulterated pollen but extends to pollens adulterated with other aversive compounds that may act before or after ingestion.

Altogether, our results highlight the importance of resource-mediated experiences within the social environment of the honey bee hive and goes further, suggesting that social learning, which occurs within the colony (Giurfa, 2012), is critical for selecting resources whose constituents cannot be perceived prior to ingestion at the food source. While regulation of pollen foraging through colony feedback fits well in Apis mellifera, it is less likely in other bee colonies such as bumblebee colonies, where interactions between in-nest workers and foragers (e.g. trophallaxis) are scarce (Goulson, 2003). For nutrient selection, bumblebees would rely on individual gustatory perception, which is consistent with their fine abilities to discriminate and assess the quality of pollen at the source (Nicholls and de Ibarra, 2014; Ruedenauer et al., 2015, 2016, 2017).

With increasing land use and extensive production practices that threaten the diversity of resources used by the bee, our finding is timely and opens the door for further research into the mechanisms and strategies that honey bees display for selecting their pollen sources.

We thank M. J. Corriale for her help with statistical analyses and W. Farina for the fruitful comments at the early stage of this paper. Thanks to L. Martinez for his collaboration in the field experiments. We also thank the anonymous reviewers for their positive comments and suggestions.