Cocaine Directly Impairs Memory Extinction and Alters Brain DNA Methylation Dynamics in Honey Bees

Drug addiction is a chronic relapsing behavioral disorder. The high relapse rate has often been attributed to the perseverance of drug-associated memories due to high incentive salience of stimuli learnt under the influence of drugs. Drug addiction has also been interpreted as a memory disorder since drug associated memories are unusually enduring and some drugs, such as cocaine, interfere with neuroepigenetic machinery known to be involved in memory processing. Here we used the honey bee (an established invertebrate model for epigenomics and behavioral studies) to examine whether or not cocaine affects memory processing independently of its effect on incentive salience. Using the proboscis extension reflex training paradigm we found that cocaine strongly impairs consolidation of extinction memory. Based on correlation between the observed effect of cocaine on learning and expression of epigenetic processes, we propose that cocaine interferes with memory processing independently of incentive salience by directly altering DNA methylation dynamics. Our findings emphasize the impact of cocaine on memory systems, with relevance for understanding how cocaine can have such an enduring impact on behavior.


INTRODUCTION
Commonly abused drugs cause debilitating drug addiction in a small fraction of users (McLellan et al., 2000). Addiction is a chronic, relapsing condition marked by compulsive drug seeking and craving (Robinson and Berridge, 2003). Recovering addicts suffer high relapse rates due to persistent drug associated memories (Hser et al., 2001). This has led some authors to conclude that drug addiction is a disease of learning and memory (Hyman, 2005;Hyman et al., 2006). Most drugs of abuse are strongly reinforcing and have high incentive salience (Siegel, 2005;Robinson and Berridge, 2008): that is, motivation to seek out drugs or drug associated cues is strong (Robinson and Berridge, 1993). Consequently, stimuli associated with drug administration are readily learned, and memories of them are persistent (Uslaner et al., 2006). It has been argued that increased incentive salience is why drug associated memories are particularly difficult to extinguish (Stewart, 2000), resulting in frequent relapses (Weiss et al., 2001). In mammalian brains, many drugs of abuse alter neurotransmission in the dopaminergic midbrain pathway either by increasing release or inhibiting clearance of dopamine (Kuhar et al., 1991;Han and Gu, 2006), thereby increasing the incentive salience of a given stimuli (Berridge, 2007). Altered incentive salience is not the only way by which memory processing can be altered, however. Many drugs also interfere with memory processing directly (Mittenberg and Motta, 1993). Cocaine induces widespread changes in DNA methylation patterns (Anier et al., 2010). This is of particular interest, since both DNA methyltransferase enzymes (DNMT) and Ten-eleven translocation (TET) proteins, responsible for methylation and demethylation of DNA, respectively, are both vital for memory formation (Day et al., 2013;Alaghband et al., 2016;Kennedy and Sweatt, 2016). Previous studies have shown cocaine-associated memories to be correlated with changes in DNA methylation (Tian et al., 2012) and to be highly resistant to extinction (Di Ciano and Everitt, 2004). It is not yet known, however, if the enduring nature of cocaine-associated memory is due to cocaine directly affecting the DNA methylation machinery, or if it is an indirect consequence of altered incentive salience.
Honey bees have long been used to study mechanisms of associative learning (Menzel et al., 1974;Bitterman et al., 1983;Hammer and Menzel, 1995), and more recently as a valuable model system for elucidating the effects of pharmacological manipulations on learning and memory (Felsenberg et al., 2011;Maleszka, 2014). Bees have functional DNA methylation and demethylation systems (Wang et al., 2006;Lyko and Maleszka, 2011;Wojciechowski et al., 2014;Maleszka, 2016), that are involved in memory processing. Following olfactory conditioning altered methylation patterns can be seen across the entire honey bee genome (Li et al., 2017).
DNMT function is required for forming stimulus-specific olfactory memories (Biergans et al., 2012(Biergans et al., , 2016, potentially due to its activity in the antennal lobes (Biergans et al., 2017). Pharmacological inhibition of DNMTs has also been shown to interfere with consolidation extinction of appetitive memories (Lockett et al., 2010;Gong et al., 2016). DNMT1b and DNMT3 and Tet are all upregulated following olfactory conditioning (Biergans et al., 2015), but no direct function of TET proteins during learning has been demonstrated in bees so far.
Because cocaine results in similar behavioral and neurochemical responses in bees and mammals Søvik, 2013;Søvik et al., , 2014, it presents itself as a valuable system to explore the basic interactions between drugs of abuse, epigenomic modifications and behavior Maleszka, 2014Maleszka, , 2016. Here we investigated the effects of cocaine on acquisition, consolidation, and retrieval of memories in honey bees when drug delivery was dissociated from conditioning, and explored whether cocaine affected brain DNA methylation systems.

Animals
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  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.

Drug Treatments
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 Søvik et al., , 2016Scheiner et al., 2014).

Training Protocols
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., 2010Lockett et al., , 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.

DNA Methylation Enzyme Activity Quantification
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:

Specimen Preparation for Quantification of Cocaine in Honey Bee Brains by Liquid Chromatography-Mass Spectrometry/Mass Spectrometry
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.

LC-MS/MS
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 R 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 H 2 O 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.

Statistical Tests
To determine if the cocaine treatment had any effect on learning a χ 2 -test was conducted for each of the two tests (acquisition and extinction). For acquisition and extinction curves a χ 2 -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 ttests, with effect sizes given as Cohen's d. All statistical analysis were conducted in R 3.4.1 (R Development Core Team, 2017).

Experimental Order
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 Weakly Inhibits Acquisition and Recall of Memory
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: χ 2 = 8.8245, p = 0.0030, n = 236, φ = 0.1933, Figure 2A) and after acquisition training (Experiment 2: χ 2 = 3.9503, p = 0.0469, n = 234, φ = 0.1299, Figure 2B), and 2 h before the recall of acquisition memory (Experiment 3: χ 2 = 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).

Cocaine Strongly Inhibits Consolidation of Extinction
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: χ 2 = 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: χ 2 = 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: χ 2 = 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.

The Effects of Cocaine on TET Is Context Dependent
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 Frontiers in Physiology | www.frontiersin.org FIGURE 1 | Schematic of training schedules and treatment regimens. Acquisition and extinction trials are presented in yellow and blue, respectively. The darker shades represent test of the same type of memory. The black blocks signify when drug treatment was given, while red blocks show when animals were sacrificed. Note that time is not presented as a linear scale in diagram (h, hours; m, minutes). Dashed lines denote absolute conditioning was used instead of differential.
FIGURE 5 | Experiment 6: results for bees trained in a single odor conditioning PER paradigm. Cocaine treatment was given 1 h after extinction training.

Retention of Cocaine Post-treatment
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).

DISCUSSION
Our experiments show a far stronger effect of cocaine on consolidation of extinction memory (Figures 3D, 5) than consolidation of acquisition memory (Figures 2B, 4), as evidenced by the difference in effect size between these experiments. Cocaine treatment affects performance in the recall of acquisition memory when it was applied before testing ( Figure 2C). Treatment prior to acquisition training does not affect the rate of acquisition, but still impair memory formation (Figure 2A). Cocaine did not affect naïve odor responses as no difference was seen between treatment groups on the first exposure to odors in any of our experiments where cocaine was administered prior to acquisition training.
The lack of any effect of cocaine on rate of learning (demonstrated by the overlapping acquisition and extinction curves of the two experimental groups, Figures 2-4) was expected, as we intentionally chose time points for cocaine administration that would not cause cocaine to interfere with perception of stimuli during acquisition and extinction conditioning. This design allowed us to examine any direct effects of cocaine on learning and memory, rather than the effects of changed incentive salience for the conditioned stimuli. The strongest effect of cocaine, seen in this context, was an impairment of the consolidation of extinction memory (Figures 3D, 5). This suggests that part of the reason why cocaine-associated memories are so hard to extinguish, could be that in addition to increasing the incentive salience of stimuli (Uslaner et al., 2006), cocaine also actively inhibits consolidation of extinction. A likely mechanistic explanation for this phenomenon could be that cocaine interferes with the epigenetic mechanisms of memory consolidation (Day and Sweatt, 2010;Robison and Nestler, 2011).
There is ample evidence from the mammalian literature that chronic cocaine administration interferes with DNA methylation dynamics (Robison and Nestler, 2011). In mice levels of DNMT3a, but not DNMT1 and DNMT3b, increase following chronic cocaine exposure (LaPlant et al., 2010). This increase persist for at least 28 days after the end of drug treatments (LaPlant et al., 2010). The altered levels of DNMT3a have been shown to affect the DNA methylation patterns in the brains of mice following cocaine exposure (Anier et al., 2010). Further, Feng et al. (2015) has shown that levels of TET are also decrease after chronic cocaine exposure. The effects on both DNMT3a and TET have been shown to be localized to particular brain regions and affect the methylation and demethylation of specific genomic regions (LaPlant et al., 2010;Feng et al., 2015). Thus, it is becoming increasingly clear that changes to brain DNA methylation dynamics play a role in the behavioral outcomes associated with cocaine abuse.
In this study we show that in bees a single cocaine treatment can alter DNA methylation enzyme activity levels ( Figure 6A) and transcription of AmTET (Figures 6C, 7C,D). So far the results for AmTET largely mimics what is seen in mammalian systems, however, unlike studies in mice, we failed to detect any change in AmDNMT3 levels in both naïve and trained bees. The most likely explanation for this difference is that mice were treated chronically over the course of 28 days, while the bees in our study received a single cocaine treatment. It is of course possible that AmDNMT3 levels are affected in bees following chronic exposure.
Another possibility is that AmDNMT3 is not a primary de novo methyltransferase in honey bees (Wedd and Maleszka, 2016). Recent studies suggest that the classical roles of DNMT1 and DNMT3s in establishing methylation patterns need to be redefined to include the evident de novo activity of DNMT1 and DNMT3s' involvement in maintenance (Jeltsch and Jurkowska, 2014). The variation of DNMTs across invertebrates is also suggestive of diverse roles for these enzymes. In the honey bee, the DNA methylation toolkit consists of two copies of DNMT1 and one copy of DNMT3, but many insects lack DNMT3 and still methylate their genome implying that DNMT1 provides de novo activity in these organisms (Wedd and Maleszka, 2016).
Perhaps the most surprising result is the context dependency of the effects of cocaine on AmTET levels. We can only speculate why levels were shown to increase in animals following acquisition conditioning, while they were decreased in naïve animals and animals that had gone through extinction training, but it is a clear demonstration that when considering how drugs of abuse might interact with the DNA methylation/demethylation machinery it is important to remember that what the animal is actively doing or exposed to can play an important role. A further issue that must be taken seriously is where in the brain the enzymes responsible, for methylation and demethylation, act during memory formation and where cocaine is exerting its effects. In this study we used a systemic cocaine treatment and all analysis of gene expression or enzyme activity occur at the whole-brain level, we are thus not able to conclude anything clearly about this question. Biergans et al. (2017) beautifully demonstrated how DNMT activity in the antennal lobes mediates odor specificity during learning. Another interesting to note is that many of the methylation related genes shown to have FIGURE 7 | The effects of cocaine and extinction training on levels of DNA methylation enzymes 1 h after treatment. (A) There was no significant difference in AmDNMT3 mRNA levels following cocaine or control treatments given after extinction training (AmUGT: t 10 = 0.8532, p = 0.4135; AmeIF.S8: t 10 = −0.6977, p = 0.4927). (B) Similarly, levels were not affected by cocaine in bees that did not go through extinction training (AmUGT: t 10 = 0.9113, p = 0.3836; AmeIF.S8: t 10 = 1.8837, p = 0.0729). (C) There was a significant reduction in levels of AmTET mRNA in bees treated with cocaine following extinction training (AmUGT: t 10 = −3.6832, p = 0.0013, d = −1.1105; AmeIF.S8: t 10 = −7.8450, p < 0.0001, d = −2.3653). (D) Intriguingly, bees that only went through acquisition training, prior to being treated with cocaine showed the opposite effect (AmUGT: t 10 = 14.1652, p < 0.0001, d = 4.2710; AmeIF.S8: t 10 = 6.9097, p < 0.0001, d = 2.0834). *Denotes statistically significant differences.
FIGURE 8 | Amounts of cocaine in the honey bee brain after topical treatment. Levels of cocaine increased sharply 30 min after treatment, thereafter levels gradually decreased, until they were almost absent at 4 h. altered expression patterns following olfactory conditioning are highly expressed in the honey bee mushroom bodies (Biergans et al., 2015). We therefore think that the antennal lobe and mushroom bodies are key structures for future investigations.
The finding that cocaine interferes with consolidation of extinction learning, potentially by changing levels of AmTET, has implications for the application of extinctionbased therapies. While this method has worked well when attempting to extinguish fearful memories (Schiller et al., 2010), it has been less successful for treating addiction (Conklin and Tiffany, 2002), as marked by very high rates of relapse in recovering addicts (McLellan et al., 2000;Hser et al., 2001). If cocaine uniquely interferes with mechanisms involved in consolidation of extinction memory, it could potentially mean that extinction therapies are severely compromised in recovering addicts who still occasionally use cocaine.

AUTHOR CONTRIBUTIONS
ES, AB, and RM conceived the study. ES, PD, WK, PH, EB, and JP performed the experiments. ES analyzed the data. ES, AB, RM, PD, WK, PH, EB, and JP wrote the manuscript.

ACKNOWLEDGMENTS
This work was supported by Australian Research Council grant DP0986021 awarded to AB and RM. JP was supported by an iMQRES scholarships awarded by Macquarie University and a DAAD Doktorandenstipendium awarded by the German Academic Exchange service. We thank David Maltby and Benjamin Crossett of the Mass spectrometry core facility at the Charles Perkins Centre at the University of Sydney for use of facilities and assistance with LC/MS. We thank Katherine Berthon for experimental assistance, Dr. Dorothea Eisenhardt for discussions about experimental design, and Dr. Alexis Hill for comments on the manuscript. The authors declare no competing financial interests.