The gut parasite Nosema ceranae impairs olfactory learning in bumblebees

Pollinators, such as bees, rely on a rich cognitive repertoire to collect pollen and nectar on flowers. These include associative learning and memories of floral traits such as odours, shapes, colours and textures, to identify the most profitable resources (Giurfa, 2015; Menzel, 2012), and spatial cues for navigation (Collett et al., 2013). Any disruption of these cognitive abilities by environmental stressors can considerably reduce the foraging performance of bees, ultimately compromising brood development and survival (Klein et al., 2017).

In particular, foraging bees are exposed to a number of parasites that can affect their physiology and behaviour (Gómez-Moracho et al., 2017). The microsporidium Nosema ceranae is one of the most prevalent parasites of bees worldwide with a large range of hosts including honey bees (Higes et al., 2006), bumblebees (Plischuk et al., 2009), solitary bees (Ravoet et al., 2014), but also other flower visitors such as wasps (Porrini et al., 2017). Insects become infected by ingesting parasite spores from contaminated water or pollen (Higes et al., 2008), or during physical contact with contaminated individuals (Smith, 2012). The spores invade the gut epithelial cells of the hosts, where they develop (Holt et al., 2013). In honey bees, N. ceranae degenerates the gut epithelium (Higes et al., 2007), alters metabolism (Mayack and Naug, 2009) and disrupts the immune response (Antúnez et al., 2009). This causes a disease (nosemosis) believed to contribute to colony collapse (Cox-Foster et al., 2007).

Nosema ceranae-infected honey bees also show impaired navigation (Wolf et al., 2014) and increased flight activity (Dussaubat et al., 2013), suggesting that their cognitive abilities are affected by the parasite. Recent studies have explored this possibility from a mechanistic point of view using Pavlovian olfactory conditioning of the proboscis extension reflex (PER; Takeda, 1961) in which harnessed bees are trained to associate an odour, or a combination of odours, with a sucrose reward (for recent reviews, see Lavond and Steinmetz, 2003; Matsumoto et al., 2012). However, their results are mixed, presumably because of important variations in parasite exposure protocols, age of bees and parasite post-exposure duration in the different studies (Bell et al., 2020; Charbonneau et al., 2016; Gage et al., 2018; Piiroinen and Goulson, 2016; Piiroinen et al., 2016). Only two studies explored these effects in bumblebees. One study suggests a slight impairment of absolute learning (Piiroinen and Goulson, 2016), and both report no effect on memory (Piiroinen et al., 2016; Piiroinen and Goulson, 2016). Note however that in these two studies less than 3% of the bumblebees exposed to N. ceranae were indeed found to be contaminated by the parasite (i.e. PCR positive) after the behavioural tests.

Given the expanding geographical distribution of N. ceranae worldwide (Klee et al., 2007), its increasing prevalence in wild bees (Plischuk et al., 2009; Porrini et al., 2017; Ravoet et al., 2014), and the potential high fitness costs incurred by bees with impaired cognition (Henry et al., 2012; Klein et al., 2017; Perry et al., 2015), clarifying its influence on host learning and memory is important for risk assessment. In particular, other critical forms of learning, such as the ability to associate one of two odours with a reward (differential learning) and reverse this association (reversal learning), have so far been unexplored. These types of learning are essential in the everyday life of bees, to discriminate flowers, olfactory landmarks and social partners, and require different brain centres [e.g. functional mushroom bodies are necessary for the acquisition of non-elemental associations but not for elemental associations (Boitard et al., 2015; Devaud et al., 2007, 2015; Giurfa and Sandoz, 2012)]. If the effects of the parasite are specific, these types of learning may be more or less impacted. Conversely, if the effects of the parasite are general, all learning types may be impacted.

Here, we built on a recently established method yielding high rates of experimental infection by N. ceranae (Gómez-Moracho et al., 2021) to study the impact of the parasite on different cognitive tasks in bumblebees. We used PER conditioning to compare the olfactory learning and memory performance of control bumblebees, bumblebees exposed to the parasite and bumblebees contaminated by the parasites (PCR positive) at 7 days post-exposure.

We used bumblebee workers, Bombus terrestris (Linnaeus 1758), from 14 commercial colonies acquired from Biobest (Westerlo, Belgium). Before the experiments, we verified the absence of N. ceranae (Martín-Hernández et al., 2007), and other common parasites [Nosema bombi (Klee et al., 2006); Crithidia bombi (Schmid-Hempel and Tognazzo, 2010)] in a PCR using 15 bumblebees from each colony. We maintained bumblebees in their original colonies with ad libitum access to the syrup provided by the manufacturer and germ-free pollen (honey bee-collected pollen exposed to UV light for 12 h), in a room at 25±1°C under a 12 h:12 h light:dark photocycle, until parasite exposure.

We obtained fresh spores from naturally infected honey bee colonies (Apis mellifera) maintained at our experimental apiary (University Toulouse III, France). To prepare spore solutions, we dissected the gut from 15 honey bees and crushed them in 15 ml of distilled H₂O. We confirmed the presence of N. ceranae and the absence of Nosema apis (another common parasite of honey bees) in each homogenate by PCR (Martín-Hernández et al., 2007), and purified them following standard protocols (Fries et al., 2013). We centrifuged homogenates in aliquots of 1 ml at 5000 rpm for 5 min and re-suspended the pellet in 500 µl of dH₂O by vortexing. This was repeated 3 times to obtain a spore solution of 85% purity (Fries et al., 2013). We counted N. ceranae spores in an improved Neubauer haemocytometer (Cantwell, 1970) in a light microscope (×400) and adjusted the spore inoculum to 15,000 spores µl⁻¹ in 20% (w/w) of sucrose solution. Spore solutions were used within the same week they were purified.

We exposed bumblebees to N. ceranae as described in Gómez-Moracho et al. (2021). Briefly, we confined individual bumblebees in a Petri dish for 5 h without food. We then exposed some bumblebees to a 20 µl drop of 20% sucrose solution containing 300,000 N. ceranae spores. Control bumblebees received 20 µl of sucrose solution (20% w/w). We only used bumblebees that consumed the entire drop of sucrose within the next 2 h. We then allocated bumblebees into microcolonies of 20–25 individuals, containing a gravity feeder with ad libitum access to food (Kraus et al., 2019). As diet can affect host–parasite relationships (Frost et al., 2008), we provided bumblebees with an artificial diet with a protein to carbohydrate ratio of 1:207, which has previously been shown to elicit highest N. ceranae prevalence in bumblebees (Gómez-Moracho et al., 2021). The diet was made with a fixed total amount of nutrients of 170 g l⁻¹ (protein+carbohydrates) and 0.5% vitamin mixture for insects (Sigma). Carbohydrates were supplied as sucrose (Euromedex). Proteins consisted of a mixture of casein and whey (4:1) (Nutrimuscle) (Gómez-Moracho et al., 2021). We kept bumblebee microcolonies in a room at 25±1°C with a 12 h light:12 h dark photoperiod until the behavioural tests. Every day, we renewed the diet and removed dead bumblebees.

We tested the cognitive performance of bumblebees of unknown age using PER at day 7 after parasite exposure (Fig. 1A). The day before the behavioural tests, we kept diets to low levels (∼200 µl per bumblebee) to keep bumblebees motivated for the PER experiments. Three hours before the behavioural tests, we collected bumblebees from the microcolonies, chilled them on ice for 5 min and restrained them in a modified 2 ml Eppendorf tube (hereafter, capsule) that we cut in length to fit each bumblebee (adapted from Toda et al., 2009; Fig. 1B). Bumblebees were tested in the horizontal position and could move forward and backward inside the tube, and therefore retract their head (Toda et al., 2009). We found these conditions better suited to perform PER experiments with bumblebees than the classical vertical harnessing used for honey bees (Giurfa and Sandoz, 2012), in which bumblebees appeared paralysed (T.G.-M., T.D. and M.L., unpublished data). In this approach, we obtained comparable to better learning performance than in previous studies [e.g. 58.1% in our study versus 44% in Laloi et al. (1999) or 57% in Piiroinen et al. (2016)]. Once the bumblebees were in the capsule, we kept them in the dark, in an incubator at 28°C, with no access to food. Bumblebees were left in the capsules for 3 h before the experiments, and for the whole duration of each conditioning protocol [i.e. a total of 4 h for sucrose responsiveness test, 5 h for absolute learning with short-term memory (STM) test, 6 h for reversal learning, and 28 h for absolute learning with 24 h memory (long-term memory, LTM) test]. All bumblebees that finished the conditioning protocols were kept at −20°C for later analysis of their infection status through PCR.

We tested the sucrose responsiveness of bumblebees to control for potential influences of N. ceranae on their reward perception or feeding motivation. We presented seven sucrose solutions to each bumblebee, from concentrations of 0% (pure water) to 60% (w/w), with increments of 10% (Graystock et al., 2013). For each concentration, we touched the antennae of the bumblebee with a toothpick soaked in the corresponding sucrose solution to elicit the PER. We presented solutions in an increasing concentration gradient with an inter-trial interval of 5 min between concentrations. We discarded all the bumblebees responding to water (i.e. 0% sucrose solution) to avoid the effect of thirst on sucrose responsiveness (Baracchi et al., 2018), and those showing an inconsistent response (i.e. bumblebees responding to lower but not higher sucrose concentrations; Table 1) (Scheiner et al., 2003). As sucrose concentrations were systematically presented in the same increasing order to the bumblebees, we calculated a gustatory score for each bumblebee based on the lowest concentration at which they responded. Bumblebees whose first response was observed following exposure to 10% sucrose had a score of 1, whereas bumblebees that responded for the first time to 60% sucrose had a score of 6. Therefore, the lower the score, the lower the sucrose sensitivity threshold of the bumblebee.

All experiments shared the same general protocol (Fig. 1A). An encapsulated bumblebee (Fig. 1B) was placed 1.5 cm in front of an automated conditioning setup (described in Aguiar et al., 2018) delivering a continuous stream of odourless air at 1.2 ml s⁻¹ to which specific odours were selectively added (Raiser et al., 2017). We used two odorants as conditioned stimulus (CS): nonanal and phenylacetaldehyde (Palottini et al., 2018; Sommerlandt et al., 2014), at a 1:100 dilution in mineral oil. Before conditioning, we tested the responsiveness of bumblebees to sucrose by touching both antennae with a toothpick soaked in 50% (w/w) sucrose solution without allowing them to lick. Bumblebees extending their proboscis were considered motivated and kept for the experiments. Conditioning trials (Fig. 1Aiii) consisted in 15 s of odourless airflow, followed by 6 s of CS, and 3 s of unconditioned stimulus (US, i.e. 50% sucrose solution applied with a toothpick on the bumblebee's antennae), with 2 s of overlap between CS+US, and 20 s of odourless airflow (Aguiar et al., 2018). The inter-trial interval was 10 min. An air extractor was placed behind the bumblebee to prevent odorant accumulation during CS delivery. Bumblebees extending their proboscis within 3 s of US presentation (i.e. 2 s CS+US and 1 s US) were allowed to lick the toothpick soaked in sucrose (50% w/w). In unrewarded trials (see reversal learning and memory tests, below) no US was applied. We scored a conditioned response if the bumblebee extended its proboscis to the odour delivery before sucrose presentation. Bumblebees that responded to the odour in the first conditioning trial were discarded from the analysis. We used conditioned responses to calculate individual scores for each bumblebee describing its performance during conditioning (i.e. acquisition score) (Monchanin et al., 2021; see details below). Exposed and control bumblebees were always conditioned in parallel.

We tested the effects of N. ceranae on the ability of bumblebees to associate an odour with a reward. This form of learning primarily requires peripheral brain centres, (i.e. the antennal lobes) as it can be observed in bees with a non-functional central brain (Giurfa and Sandoz, 2012). We trained bumblebees in a spaced 3-trial absolute conditioning learning protocol (Fig. 1Ai) that was shown to generate robust LTM in bees (Menzel et al., 2001). We used the same rewarded odour (CS+) during training of a given bumblebee, but both nonanal and phenylacetaldehyde were used as a CS+ for different bumblebees. For each of these bees, we calculated an acquisition score (sum of PER responses, i.e. 0–2), and compared the learning curve to assess the increase of PER responses over trials. Bumblebees that did not respond to the US in at least one trial were considered not motivated and were removed from the learning analysis (i.e. 30.3% control and 29.6% exposed bumblebees; Table 1).

We tested memory retention of bumblebee responders that showed at least one conditioned response in either of the last two trials (Simcock et al., 2018; Wright et al., 2015) (i.e. 41.3% of bumblebees conditioned for STM and 30–68% of bumblebees conditioned for LTM; see details in Table 1). We performed tests either 1 h (i.e. STM) or 24 h (i.e. LTM) after the last acquisition trial. Bumblebees were tested for either 1 h or 24 h memory, but never for both because of the risk of reconsolidation or extinction of memory when the CS is presented several times without a reward to bees (Bouton and Moody, 2004). For tests performed after 24 h, bumblebees were fed until satiation with 50% (w/w) sucrose solution right after conditioning; unfortunately, 9–20% of responder bumblebees died before the test (Table 1). In bees, LTM is dependent on protein synthesis whereas STM is not (Menzel, 2001). Studying these two types of memory was thus a means to explore whether exposure to N. ceranae interfered with protein synthesis. We presented bumblebees with the two odorants without any reward: one odour used as a CS+ to test for memory formation, and the second odour as a novel odorant (NOd), to control for potential generalization (Matsumoto et al., 2012). For example, when nonanal was used as the CS+, phenylacetaldehyde was used as a NOd, and vice versa. Bumblebees responding only to the CS+, and not to the NOd, were considered to have generated a specific memory to the rewarded odour. Other responses registered were those to the NOd alone (inverted response), both odours (general response) or neither of the odours (no memory). Just after the memory test, we tested the motivation of bumblebees by touching their antennae with a toothpick soaked with 50% (w/w) sucrose solution. Bumblebees that did not respond to the US were discarded from the analysis (i.e. 14.06% of alive responder bumblebees in LTM; Table 1). Sample sizes dropped between learning and memory tests because of the selection protocol of responders, mortality in tubes and loss of motivation by the bumblebees (see details in Table 1).

We tested the effects of N. ceranae on the ability of bumblebees to learn to discriminate two odours and reverse the task. This form of learning involves two phases. The differential conditioning phase (phase 1) requires the antennal lobes but is not dependent on central brain centres (i.e. mushroom bodies) (Boitard et al., 2015; Devaud et al., 2007). The reversal learning phase (phase 2) requires both functional antennal lobes and mushroom bodies to be observed (Boitard et al., 2015; Devaud et al., 2007). In phase 1, we trained bumblebees to discriminate between two odours. This consisted in 10 trials (Fig. 1Aii), five with each odour that was either paired with the US (A+) or unpaired (B−), presented in a pseudo-random order. The rewarded and unrewarded odours were randomized on different training days. Bumblebees that did not respond to the US in two or more trials were discarded from the analysis and for the reversal phase. In phase 2, we trained bumblebees to invert the first learnt contingency. This reversal phase started 1 h after the end of the differential phase. Here, we trained bumblebees in 12 trials, 6 with each odour presented in a pseudo-random order. The previously rewarded odour was not associated with a reward anymore (A−) while the previously unrewarded odour became rewarded (B+). To start from the same level of learning, in the first trial of the reversal phase we presented the odour A−, and kept for analysis only those bumblebees that extended their proboscis (Table 1). We analysed the performance of bumblebees in each phase separately by attributing them an acquisition score (sum of all trials where the bee responded to CS+ but not to CS−) for each phase.

We assessed the infection status of bumblebees that finished the tests by PCR using 218MITOC primers (Martín-Hernández et al., 2007). These primers are 100% specific for N. ceranae and have a 0% error rate (false negatives) for reactions containing DNA equivalent to 2000 spores (Martín-Hernández et al., 2007). For each bumblebee, the entire gut was extracted, homogenized in sterile dH₂O and vortexed with 2 mm glass beads (Labbox Labware). Genomic DNA was extracted using proteinase K (20 mg ml⁻¹; Euromedex) and 1 mmol l⁻¹ of Tris-EDTA buffer (pH 8). A sample with N. ceranae spores was included in each round of extraction as a positive control. PCR was performed with the Taq Polymerase Direct Loading Buffer (5 U μl⁻¹; MP Biomedicals) following the manufacturer's instructions. We used a final volume of 25 μl with 0.4 μmol l⁻¹ of each pair of primers (Martín-Hernández et al., 2007), 200 μmol l⁻¹ of dNTPs (Jena Biosciences), 0.48 μg μl⁻¹ of BSA (Sigma) and 2.5 μl of DNA sample. PCR reactions were carried out in a S1000™ Thermal Cycler (Bio-Rad). Thermal conditions were 94°C for 2 min, 35 cycles of 94°C for 30 s, 61.8°C for 45 s and 72°C for 2 min, and a final step of 72°C for 7 min. The length of the PCR products (i.e. 218 bp) was checked by electrophoresis with a 1.2% agarose gel stained with SYBR Safe DNA Stain (Edvotek, Washington, DC, USA). Positive and negative PCR controls were run in parallel. Based on the PCR results, we classified bumblebees into three different infection statuses: control, Nosema-exposed negative (NE−) or Nosema-exposed positive (NE+). Bumblebees that were not exposed to the parasite but nevertheless showed a positive result in PCR were excluded from the analysis (i.e. 6.26%; 23 out of 367 control bumblebees). These positives may be due to the fact that despite our precautions before starting the experiments (PCR screening of about 10% of the workers in the colonies, use of UV-treated pollens), it is possible that the commercial colonies we used were not fully free of parasites. Additionally, we cannot completely exclude low levels of cross-contamination between treatments during manipulations.

All analyses were conducted in RStudio (version 4.1.0). We evaluated the effects of parasite exposure and infection on learning curves, and gustatory, acquisition, learning and memory scores. The proportion of responses to the different sucrose concentrations was analysed in a generalized linear mixed model (GLMM) (package lme4; Bates et al., 2015), with infection status and concentration as fixed factors. Learning curves of absolute and reversal learning experiments were analysed with a binomial GLMM with infection status, trial and the interaction between them as fixed factors. Whenever the interaction was not significant it was removed from the model. The responses to rewarded and unrewarded odours during reversal learning were analysed separately. To determine the ability of bumblebees to learn to differentiate a rewarded from an unrewarded odour (i.e. differential phase) and the opposite (i.e. reversal phase) in the reversal learning experiment, we studied the interaction of infection status, trial and reward in a binomial GLMM, followed by a Tukey pairwise comparison applying the function lsmeans (package emmeans). All these models included colony and bee identity as random factors. Gustatory scores for sucrose responsiveness, and acquisition scores during learning experiments were analysed with a linear model (LM). These models included bee infection status as a fixed factor and colony of origin as a random factor. We performed Tukey post hoc pairwise comparisons (package multcomp; Hothorn et al., 2008) to assess the relationship between the three bee infection statuses. The effect of infection status on the ability of bumblebees to generate specific memory (i.e. response to CS+ only) or no memory (i.e. response to any odour) was compared with a Chi-square test. Raw data are available from Zenodo (doi:10.5281/zenodo.4376362).

We tested responsiveness to different sucrose concentrations in 63 bumblebees showing a consistent response (37 controls, 16 NE−, 10 NE+; Table 1). Bumblebees increased their response to sucrose solution with concentration (Fig. 2A; GLMM, estimate=0.098, s.e.=0.009, P<0.001) in the three infection statuses (GLMM, infection status: χ²=4.754, d.f.=2, P=0.09). Gustatory scores ranged from 1.32±1.22 (mean±s.e.m.) in controls to 1.43±0.24 in NE− and 1.80±0.41 in NE+ bumblebees, but did not differ between infection status (Fig. 2B; LM, F=1.077, P=0.347). Therefore, exposure to N. ceranae affected neither the sucrose sensitivity nor the feeding motivation of bumblebees.

We analysed absolute conditioning in 420 bumblebees (141 controls, 228 NE−, 51 NE+; Table 1). The proportion of bumblebees showing a conditioned response to CS+ increased with trials (GLMM; trial: χ²=58.432, d.f.=2, P<0.001), but this trend was lower for NE− (estimate=−1.056, s.e.=0.2950, P=0.00034) and NE+ (estimate=−1.388, s.e.=0.474, P=0.003) bumblebees than for controls (Fig. 3). Likewise, exposed bumblebees (either NE− or NE+) had significantly lower acquisition scores (Fig. 3B; LM, infection status: χ²=15.897, d.f.=2, P<0.001) than controls. Acquisition scores were similar for NE− and NE+ (Tukey P>0.05; Table S1).

We analysed STM and LTM formation of bumblebee responders (i.e. those that showed at least one conditioned response in trial 2 and/or trial 3: 41.3% for STM and 50% for LTM; Table 1). One hour after training, bumblebees either responded to the CS+ only (i.e. specific memory; 82.02±0.011%, mean±s.e.m.) or did not respond to any odour (i.e. no memory), with no bumblebees responding to NOd (i.e. neither generalizers nor inverters; Fig. 3C). The proportion of bumblebees with specific memory was similar in the three infection statuses (χ²=0.013, d.f.=2, P=0.993), indicating that N. ceranae did not affect STM (Fig. 3C). A much lower proportion of the bumblebees (33.03±6.35%, mean±s.e.m.) showed specific memory for the conditioned odour after 24 h (Fig. 3D). Two bumblebees generalized their responses (one control and one NE−), and only one NE− bumblebee inverted its response (i.e. response to NOd only). The proportion of bumblebees showing specific memory to the CS+ did not differ between infection statuses (Fig. 3D; χ²=2.349, d.f.=2, P=0.310), presumably because of the low number of NE+ individuals (9 bumblebees).

We analysed differential learning in 125 bumblebees (64 controls, 41 NE−, 20 NE+; Table 1). The proportion of bumblebees responding to A+ was affected by the interaction between the infection status and trial (GLMM; infection status×trial: χ²=7.987, d.f.=2, P=0.018). Responses increased with trials in control and NE− bumblebees (estimate=1.220, s.e.=0.1879, P<0.001), but not in NE+ (estimate=−0.732, s.e.=0.299, P=0.014). Control bumblebees discriminated the two odours (i.e. higher proportion of responses to A+ than to B−) at trial 2 (Tukey; z=3.674, P=0.044) and NE− bumblebees discriminated the two odours at trial 4 (Tukey: z=4.083, P=0.009), but NE+ bumblebees did not show this ability after 5 trials (Tukey: z=2.514, P=0.665) (Fig. 4A). Parasite exposure affected acquisition scores (GLMM, infection status, χ²=29.978, d.f.=2, P<0.001; Fig. 4B). NE− and NE+ bumblebees showed similar acquisition scores (Tukey test: P>0.05; Table S1) and these scores were significantly lower than those of controls (Tukey test: P<0.05; Table S1). Thus overall, exposure to N. ceranae reduced differential learning performance. Exposed bumblebees were slower (i.e. NE−) or unable (i.e. NE+) to solve the task.

We analysed reversal learning in 103 bumblebees that finished the differential phase, and that responded to A− in the first trial of the reversal phase (56 controls, 32 NE−, 15 NE+; Table 1). All bumblebees reduced their response to A− over trials (GLMM, trial: estimate=−1.187, s.e.=0.134; P<0.001) and increased their response to B+ (GLMM, trial; estimate=1.215, s.e.=0.124; P<0.001) (Fig. 4D). Infection status did not affect the response to A− (GLMM; infection status: χ²=9, d.f.=2, P=0.389). The proportion of bumblebees responding to B+ was not different between controls and NE− (estimate=−0.579, s.e.=0.348, P=0.096). However, it was significantly reduced in NE+ (estimate=−1.597, s.e.=0.450, P<0.001). Control bumblebees reversed their response to odours earlier, at trial 5 (i.e. higher proportion of bees responding to B+ than A−; Tukey: z=4.478, P=0.002), while NE− did so at trial 6 (Tukey: z=4.281, P=0.006), and NE+ never did so (Tukey; trial 6: z=2.103, P=0.960). This was also reflected in the acquisition scores, which were significantly different between bumblebees of different infection statuses (Fig. 4E; GLMM, infection status, χ²=6.783, P=0.033). NE+ bumblebees had lower acquisition scores than controls (Tukey test: P=0.053; Table S1) and NE− bumblebees (Tukey test: P=0.044; Table S1), suggesting they had a lower ability to reverse the task. Thus, overall, exposure to N. ceranae also impaired the reversal phase of reversal learning. We found no effect of N. ceranae exposure on either phase of reversal learning when bumblebees were tested at 2 days post-exposure, suggesting that stress due to parasite exposure or parasite infection requires a longer time to be established (see SupplementaryMaterials and Methods, Table S2 and Fig. S1).

Bees are exposed to a number of parasites that can affect cognitive ability supporting crucial behaviour (Koch et al., 2017; Schmid-Hempel, 2013). Previous studies exploring the effect of N. ceranae on absolute olfactory learning and memory in bees reported contrasting results, presumably because of differences in conditioning protocols and infection rates across studies (Bell et al., 2020; Charbonneau et al., 2016; Gage et al., 2018; Piiroinen and Goulson, 2016; Piiroinen et al., 2016). Here, we ran a suite of standard olfactory cognitive assays showing that feeding bumblebees spores of this parasite consistently impaired different types of olfactory learning but not memory 7 days after exposure.

Exposure to N. ceranae in food impaired the ability of bumblebees to associate an odour with a reward (absolute learning), discriminate two odours (differential learning) and learn an opposite association (reversal learning). These are fundamental cognitive operations a bee must display to efficiently forage on flowers (Giurfa and Sandoz, 2012). Our results agree with a previous study reporting reduced absolute learning in N. ceranae exposed bumblebees (Piiroinen and Goulson, 2016). We also found that N. ceranae did not affect sucrose responsiveness, contrary to observations in honey bees in which it was found to increase their hunger (Mayack and Naug, 2009). The parasite may thus not produce the same energetic stress observed previously in honey bees (Mayack and Naug, 2009), where N. ceranae seems to be a more specific parasite (van der Steen et al., 2022). However, further experiments are needed to confirm this as protocols in these previous studies differed slightly from ours. For instance, under our conditions we cannot discard the possibility that response to high sucrose concentrations was due to sensitization as no water was used between sucrose concentrations. In bees, sucrose perception through the antennae and olfactory learning require processing of olfactory information through the antennal lobes (Giurfa and Sandoz, 2012). While simple forms of associations can be acquired only with functional antennal lobes, other types of learning (i.e. reversal learning, configural learning) also require information processing in the mushroom bodies (Devaud et al., 2007, 2015; Boitard et al., 2015). In our experiments, the fact that all types of learning were impaired and that sucrose sensitivity was not suggests that N. ceranae did not specifically target the antennal lobes or the mushroom bodies. Rather, it probably impacted the learning processes in general.

By contrast, we found no evidence that N. ceranae influenced memory or sucrose sensitivity. During training, animals learn and form short-term memories that are later consolidated and transformed into stable long-term memories (Menzel and Müller, 1996) after protein synthesis (Menzel, 2001). In our experiments, N. ceranae impaired neither STM nor LTM.

It has recently been questioned whether bumblebees are natural hosts of N. ceranae based on the lack of evidence of parasitic forms inside host cells (Gisder et al., 2020). Several studies have nevertheless reported N. ceranae in wild bumblebees at low (e.g. 4.76%; Sinpoo et al., 2018) and high prevalence (e.g. 72%; Arbulo et al., 2015). Whether or not bumblebees are suitable hosts for N. ceranae replication, our results imply they are impacted by an acute exposure to the parasite. Such exposure may be extremely frequent in nature because of the high prevalence of N. ceranae in honey bees (Runckel et al., 2011) that contaminate flowers with spores through physical contact or in their faeces (Graystock et al., 2015). Our protocol of parasite exposure significantly increased the infection rate of bumblebees to 28% in comparison to previous studies (Piiroinen and Goulson, 2016; Piiroinen et al., 2016), which allowed the evaluation of cognitive traits in bumblebees in the three infection statuses. Bumblebees that tested positive to N. ceranae showed a tendency for lower cognitive performance than those that were exposed but tested negative. They reached the lowest learning during absolute conditioning and did not discriminate odours, suggesting that infection may interfere with some aspects of cognition. However further experiments are needed to tackle this question, as the lower performance of exposed bumblebees positive to N. ceranae could also be related to a worse health status and therefore higher susceptibility to become infected.

Through which mechanism does the parasite impair learning? Ingestion of N. ceranae spores exerts a stress that can reduce cognition. Nosema ceranae is known to alter the immune system of bees; for example, by modulating the expression of antimicrobial peptides (Antúnez et al., 2009; Botías et al., 2021; Sinpoo et al., 2018). Stimulation of the immune system with non-pathogenic elicitors, such as lipopolysaccharides (LPS), was shown to reduce learning ability in honey bees, which were less able to associate an odour with a reward (Mallon et al., 2003), and bumblebees, which showed a lower performance in odour (Mobley and Gegear, 2018) and colour differential learning tasks (Mobley and Gegear, 2018). It is thus possible that the observed effects of N. ceranae exposure on bumblebee cognition were caused by activation of the immune response. Exposed positive bumblebees showed lower learning performance than exposed negative bumblebees during differential and reversal learning tasks, suggesting a further effect of parasite infection, rather than just exposure. In honey bees, infection with N. ceranae was shown to downregulate the expression of genes in the brain (Doublet et al., 2016), some of which are linked to olfaction (Badaoui et al., 2017; Doublet et al., 2016), potentially leading to changes in behaviour and cognition. Whether N. ceranae downregulates gene expression in the brain of bumblebees needs to be addressed. Interestingly, none of these effects were observed at 2 days post-exposure (see Supplementary Materials and Methods, Table S2 and Fig. S1). So far it is unknown whether N. ceranae triggers the immune response at this time. In honey bees, the earliest effects have been observed after 3 days (Chaimanee et al., 2012).

As olfactory learning is essential for foraging, this sublethal effect of N. ceranae exposure on bumblebee cognition can compromise colony foraging success, as well as chemical communication between bees, ultimately leading to colony collapse. Parasite load in the field can range from a few to thousands spores (Meana et al., 2010). Here, we used a substantially higher spore load of N. ceranae to infect commercially reared bumblebees than the actual infection rates found in wild bumblebees (e.g. 6800 spores per individual; Graystock et al., 2013). Commercial and wild bumblebee colonies exhibit physiological and behavioural differences as a result of different selective pressures (Velthuis and van Doorn, 2006), and may, therefore, show different susceptibility to parasites. Therefore, further studies are needed to analyse the effects of different concentrations of N. ceranae spores and their possible interactions with other stressors in the field in order to assess their real impact on wild pollinators. Beyond bees, these effects may also have broader fundamental consequences for plants and parasites. From the plant perspective, an impaired flower constancy by pollinators may increase pollen transfer between incompatible flowers of different species, and therefore reduce pollination efficiency. From the parasite perspective, foraging errors due to impaired learning by bees may decrease they tendency for flower constancy. This would be beneficial for the parasite as it may favour its spread across flower species and thus possibly increase the range of hosts.

We thank Philipp Heeb, Cristian Pasquaretta and Maria Eugenia Villar Damiani for constructive discussions, Charline Schartier for helping running some experiments, and Martin Giurfa for lending the automated PER conditioning setup.