European honey bees, Apis mellifera, of the standard commercial strain available in New South Wales, Australia were used for all experiments. Adult bees were collected on emergence from brood cells, placed in mesh cages (20 × 16 × 3 cm) with ad libitum access to honey (80 bees per cage) and housed in an incubator at 34°C for 6 days prior to learning experiments. Cage rearing offers greater control of bees' age and experience it differs fundamentally from life in the hive. This can be problematic for some experiments, but as it does not significantly affect brain development (Maleszka et al., 2009) or ability to retain olfactory memories (Arenas and Farina, 2008), we decided it was the best approach for our experiments. Behavioral experiments 1–4 were conducted at The Australian National University, Canberra, while remaining experiments were conducted at Macquarie University, Sydney.

The treatments used for all experiments consisted of either 3 μg of freebase cocaine (cocaine) dissolved in 1 μL dimethylformamide (DMF) or 1 μL DMF on its own (control). All chemicals were supplied by Sigma-Aldrich (St. Louis, MO, USA). The treatments were given topically by placing 1 μL of the solution onto the dorsal thorax of bees with a microcapillary pipette. Care was taken to prevent treatments from spreading to wing joints or across the wings. DMF rapidly penetrates bee cuticle and can conduct compounds into the haemolymph of the bees' open circulatory system, from where small compounds can access the brain and nervous system (Barron et al., 2007; Okada et al., 2015). This administration method has previously been shown to be effective for delivering cocaine to honey bees (Barron et al., 2009; Søvik et al., 2013, 2016; Scheiner et al., 2014).

At 6 days of age, bees were harnessed for proboscis extension response (PER) conditioning (Bitterman et al., 1983). The thorax and abdomen of bees were lightly restrained in 8 mm diameter metal tubes by a thin piece of tape placed behind the neck so the head was kept in place, but antennae and proboscis were free to move (Maleszka et al., 2000; Si, 2004; Lockett et al., 2014). Each bee was fed 2 drops (approx. 30 μL) of 1.5 M sucrose, and left overnight. On the following morning, bees were trained in either a differential (experiment 1–5), or absolute (experiment 6) conditioning paradigm. For differential conditioning bees were trained to distinguish two odors (limonene and natural vanilla), one paired to reward and the other to punishment. For absolute conditioning only a single odor associated with reward was used.

Reward training involved touching a droplet of 2 M sucrose solution to the antennae followed by offering sucrose to the proboscis. Punishment consisted of touching saturated NaCl solution to the antennae, which is strongly aversive to bees (Maleszka et al., 2000; de Brito Sanchez et al., 2005; Lockett et al., 2010, 2014). Presentation of sucrose to bees results in proboscis extension, and following paired presentation of odor and sucrose bees learn to extend their proboscis to an odor that is predictive of sucrose delivery. Following training with the aversive salt solution the proboscis is actively withheld (Smith et al., 1991). For acquisition training odors were presented for 3 s on their own, and for 2 s simultaneously with the reward/punishment. For extinction training odors were presented on their own for 5 s.

For both absolute and differential conditioning bees were given 3 learning trials, each separated by 6 min (acquisition training). In the evening bees were fed with 2 drops of 1.5 M sucrose and left overnight. On the following morning, 24 h after conditioning, bees were tested by presenting training odors alone, and whether or not bees extended their proboscis to the odors was recorded (acquisition test). For the differential conditioning, we immediately discarded all bees that were deemed non-learners after the acquisition test (i.e., responded to the odor associated with the punishment, failed to respond to the odor associated with the reward, or both). Because of this, only bees that had had the correct response (proboscis extension to the rewarding odor only) were included in the extinction training when using the differential training protocol, whereas all bees were included in extinction training for the absolute conditioning. Therefore, the training curves for extinction training in experiments with differential conditioning start with all bees responding, while those using absolute conditioning start with the same proportion of responses as seen in the acquisition test.

Testing was immediately followed by an additional four presentations of odors alone, each separated by 2 min (extinction training). Five hours later bees were tested again (odor presentation alone) and proboscis extension recorded (extinction test). This method follows the conditioning paradigm used by Maleszka et al. (2000) and training schedule of Lockett et al. (2010). For molecular experiments, only bees that responded correctly during the acquisition test was used for experiments that used extinction training. Analyzed bees were drawn randomly from a cohort of bees that had gone through the exact same training or drug treatment protocol.

To assess the activity of DNA methylation enzymes in individual bee brains, honey bees were chilled to −20°C for 3 min before the central brain (excluding optic lobes and gnathal ganglia) was removed in freshly prepared PBS solution. This procedure was performed 1 h after cocaine delivery. Nuclear proteins were extracted using the EpiQuik^TM Nuclear extraction kit (Epigentek Group Inc., Farmingdale, NY) and DNA methylation enzyme activity was then measured using an EpiQuik^TM DNMT Activity/Inhibition Assay Ultra Kit (Epigentek Group Inc., Farmingdale, NY), according to the manufacturer instructions. Protein concentration was determined with a Bradford assay (Bio-Rad Laboratories Inc., Hercules, CA), and DNA methylation enzyme activity was estimated as follows: DNA methylation enzyme activity=1,000×Sample OD-Blank ODProtein amount×Incubation time

To examine transcription levels of DNA methyltransferase 3 (AmDNMT3, GenBank gene ID: 410798) and TET dioxygenase (AmTET, GenBank gene ID: 412879 412878) following cocaine treatment and extinction training, whole bees were frozen in liquid nitrogen 1 h after treatment (2 h after extinction training). Heads were partially lyophilized prior to the removal of central brains. Dissected brains were stored at −80°C until RNA extraction with the PureLink® RNA Mini Kit (Ambion®, Life Technologies, Grand Island, NY) following supplied guidelines. Each biological replicate consisted of RNA extracted from a single brain. For each experiment, six replicates were performed for each treatment group. Total RNA was reverse transcribed using SuperScript® III First-Strand Synthesis SuperMix kit (Invitrogen™, Life Technologies, Grand Island, NY). Transcription levels were assessed by quantitative real time polymerase chain reactions (qRT-PCR). We performed 10 μL reactions using SsoAdvanced™ Universal SYBR® Green Supermix (Bio-Rad Laboratories Inc., Hercules, CA). Primers used for AmDNMT3 were: forward: 5′-GAACTCGTTGAAGCAAGGCA-3′; reverse: 5′-AACGTTTGCACGCTCCAAGA-3′, and for AmTET: forward 5′-GTCAGTGAGATCAGAGGAGC-3′; reverse 5′-TGGTGCAAGGCTGAGGTACA-3′. The housekeeping genes AmUGT (uridine 5′-diphospho-glucuronosyltransferase) and AmeIF.S8 (eukaryotic initiation factor S8) were used as controls (AmUGT, GenBank gene ID: 412198, forwards primer: 5′-CGTTGATGCTGATCAGGTTG-3′; reverse primer: 5′-CGTCGAAATCGCTTCAAGTC-3′; AmeIF.S8, GenBank gene ID: 551184, forwards primer: 5′-TGAGTGTCTGCTATGGATTG CAA-3′; reverse primer: 5′-TCGCGGCTCGTGGTAAA-3′). All primers crossed an intron junction in order to avoid potential problems with contamination from genomic DNA, and have previously been used successfully (Foret et al., 2012; Wojciechowski et al., 2014).

Bees were treated with cocaine in the same manner as for behavioral experiments and placed in an incubator for 15, 30 min, 1, 4, or 24 h, after which central brains were extracted and stored at −80°C until analysis. For analysis brains were resuspended in 60 μL pH 6.0 phosphate buffer and lysed by sonication. Cellular debris was collected by centrifugation at 12,000 g for 10 min at 4°C and the supernatant was removed for analysis. A separate set of untreated brains were processed as above (n = 3–6 for each concentration point) and spiked with concentrations of cocaine ranging from 1 to 0.1 ng/mL. These served as a standard for quantification.

Liquid chromatography-mass spectrometry/mass spectrometry (LC-MS/MS) experiments were performed with an AB SCIEX QTRAP 5500 (SCIEX, Framingham, MA, USA) triple quadrupole linear ion trap mass spectrometer. The LC system used was an Agilent 1100 (Agilent, Santa Clara, CA, USA). Analyst® TF software (version 1.6.2) was used for acquisition and quantitation. Chromatographic separation was achieved using a Zorbex SB-C18 column (150 × 0.5 mm) (Agilent). Elution was performed isocratically with 35% methanol/65% (0.1% formic acid) in H₂O at a flow rate of 30 μL/min for 8 min total run time. Ten microliter injection volume was used.

MS data was collected in positive ion mode. For quantification and validation of cocaine, the following transitions were monitored, as previously described by Shakleya and Huestis (2009) m/z 304 to 182 (quantification) and m/z 304 to 82 (validation). Peak areas were measured for quantification of each sample.

To determine if the cocaine treatment had any effect on learning a χ²-test was conducted for each of the two tests (acquisition and extinction). For acquisition and extinction curves a χ²-test was used for each point in the curve, adjusted using Bonferroni's correction to account for multiple testing. Effect sizes were estimated using Pearson's φ. Mann-Whitney U-test was used to compare DNA methylation enzyme activity between cocaine and control treated bees. The effect size was estimated using rank biserial correlations (r). Transcript levels were compared by t-tests, with effect sizes given as Cohen's d. All statistical analysis were conducted in R 3.4.1 (R Development Core Team, 2017).

In order to examine the effects of cocaine on learning, independently of its effects on reward perception, we treated bees with cocaine at various time points before and after acquisition and extinction training. Since we found that cocaine most strongly inhibited consolidation of extinction memory (see below), we tried testing for acquisition memory at the same time point to ensure the effect of cocaine was specific to consolidation of extinction memory. In all of the experiments so far, we used a differential conditioning paradigm. In this paradigm bees are exposed to both aversive and appetitive conditioning. To ensure that the effect of cocaine on consolidation on extinction memory was not the result of an interaction between the processing of these two distinct kinds of memories we repeated the whole conditioning procedure, treating bees 1 h after extinction training, with an absolute conditioning paradigm.

Cocaine treatment applied 1 h before training (Figure 1) does not alter the rate of acquisition or extinction of learning of an appetitive memory as evidenced by comparison of acquisition and extinction learning curves for cocaine treated and control bees for any of the experiments (Figures 2, 3). There were, however, differences in the recall of acquisition memory for bees treated with cocaine 1 h before (Experiment 1: χ² = 8.8245, p = 0.0030, n = 236, ϕ = 0.1933, Figure 2A) and after acquisition training (Experiment 2: χ² = 3.9503, p = 0.0469, n = 234, ϕ = 0.1299, Figure 2B), and 2 h before the recall of acquisition memory (Experiment 3: χ² = 12.043, p = 0.0005, n = 139, ϕ = 0.2943, Figure 2C). In other words, bees that had been treated with cocaine prior to the acquisition test (Experiments 1–3), regardless of time point (before conditioning, after conditioning, or before recall) performed worse in the recall of acquisition test than control bees (Figure 2). There was no difference in response rate to the aversive memory for any of the treatment timepoints (Figure S1).

When we tested the recall of extinction conditioning, however, it was only when cocaine was administered 1 h post-extinction training that there was a difference between cocaine and control groups (Experiment 4: χ² = 16.7965, p < 0.00001, n = 66, ϕ = 0.5044, Figure 3D). Here, cocaine treatment impaired consolidation of extinction conditioning resulting in a stronger response to the training odor during the extinction test (Figure 3). Because the extinction test was 5 h after conditioning (as opposed to the 24 h gap between acquisition training and testing) we could not be sure if the effect seen was specific to the extinction paradigm or a general performance change 4 h after cocaine treatment. We therefore treated bees with cocaine 1 h after acquisitioning training and tested them 4 h later. We did not detect any difference between treatment and control treated bees in response to odor paired with sucrose (Experiment 5: χ² = 0.5489, p = 0.4588, n = 101, Figure 4) or NaCl (Figure S2). This suggests the effect seen in Figure 3 is specific to consolidation of extinction memory.

We further examined the robustness of this phenomenon by training bees in an absolute conditioning paradigm with a rewarded odor only. When cocaine treatment was given 1 h after training the same effect was seen (Experiment 6: χ² = 21.2706, p < 0.000001, n = 84, ϕ = 0.5032, Figure 5). Thus, we conclude that cocaine has a strong inhibitory effect on consolidation of extinction memory.

DNA methylation enzyme activity was increased in honey bee brains 1 h following cocaine administration (Experiment 7: Mann-Whitney U = 29, p = 0.01276, r = 0.4994, Figure 6A). Cocaine treatment did not affect transcription levels of AmDNMT3, which is believed to be responsible for de novo methylation in honey bees (Wang et al., 2006), in bees 1 h after cocaine treatment (Experiment 7: AmUGT: t₁₀ = 1.3439, p = 0.1940; AmeIF.S8: t₁₀ = 0.0001, p = 0.9999; Figure 6B). We next examined the effects of cocaine treatment and associative learning on levels of the honey bee homolog of TET dioxygenase, which has been shown to be responsible for removal of methylation marks in honey bees (Wojciechowski et al., 2014). Cocaine treatment caused a significant decrease in transcript levels of AmTET (Experiment 7: AmUGT: t₁₀ = −5.0172, p < 0.0001, d = −1.5866; Ame.IF.S8: t₁₀ = −5.3780, p < 0.0001, d = −1.7007; Figure 6C).

The effects of cocaine on AmTET, but not AmDNMT3, expression levels varied with the learning experience of the bees. Levels of AmDNMT3 mRNA were not affected by cocaine in bees that had gone through extinction conditioning (Experiment 8: AmUGT: t₁₀ = 0.8532, p = 0.4135; AmeIF.S8: t₁₀ = −0.6977, p = 0.4927; Figure 7A). Cocaine treatment did not alter AmDNMT3 levels in bees that had not gone through extinction training either (Experiment 9: AmUGT: t₁₀ = 0.9113, p = 0.3836; AmeIF.S8: t₁₀ = 1.8837, p = 0.0729; Figure 7B). In contrast, levels of AmTET were significantly reduced following cocaine treatment after extinction conditioning (Experiment 8: AmUGT: t₁₀ = −3.6832, p = 0.0013, d = −1.1105; AmeIF.S8: t₁₀ = −7.8450, p < 0.0001, d = −2.3653; Figure 7C). Intriguingly, when bees that were treated with cocaine after acquisition training AmTET levels increased significantly (Experiment 9: AmUGT: t₁₀ = 14.1652, p < 0.0001, d = 4.2710; AmeIF.S8: t₁₀ = 6.9097, p < 0.0001, d = 2.0834; Figure 7D). Taken together, these results suggest that the effects of cocaine on AmTET levels were dependent on the learning experience of the bees.

The total amounts of cocaine present in brains peaked 30 min after treatment and gradually declined to almost zero over a 4 h period (Figure 8).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.