DNA methylation changes elicited by social stimuli in the brains of worker honey bees



Social environments are notoriously multifactorial, yet studies in rodents have suggested that single variables such as maternal care can in fact be disentangled and correlated with specific DNA methylation changes. This study assesses whether non-detrimental social environmental variation in a highly plastic social insect is correlated with epigenomic modifications at the DNA methylation level. Honey bee workers perform tasks such as nursing and foraging in response to the social environment in the hive, in an age-linked but not age-dependent manner. In this study, the methylation levels of 83 cytosine–phosphate–guanosine dinucleotides over eight genomic regions were compared between the brains of age-matched bees performing nursing or foraging tasks. The results reveal more changes correlated with task than with chronological age, and also hive-associated methylation at some sites. One methylation site from a gene encoding Protein Kinase C binding protein 1 was consistently more methylated in foragers than nurses, which is suggested to lead to production of task-specific protein isoforms via alternative splicing. This study illustrates the ability of the neural epigenome to dynamically respond to complex social stimuli.

Changes at the chromatin level to DNA methylation and histone modifications are often associated with exposure to toxicants (Bollati & Baccarelli 2010), but these marks also respond to less physically affronting stimuli such as the social environment (Szyf et al. 2008). A pivotal demonstration of this phenomenon is that low levels of maternal care in rats increase DNA methylation in the promoter of the glucocorticoid receptor gene, with long-lasting effects (Weaver et al. 2004). Non-human primates too show long-lasting behavioural deficits when deprived of maternal care, and in rodents communal nesting improves social skills in adulthood (Branchi 2009). Early social environments have also been shown to leave specific marks on the epigenome in humans (McGowan et al. 2000). Social behaviour-related gene networks show some conservation between phyla suggesting these effects may occur outside mammalian model systems (Sokolowski 2010).

The honey bee provides an excellent opportunity to examine the interaction between the social environment and DNA methylation (Maleszka 2008; Miklos & Maleszka 2011). First, the honey bee has a conserved DNA methylation enzymology and methylates selected cytosines in a manner similar to that found in mammals (Lyko & Maleszka 2011). Second, honey bees are known for their complex repertoire of social behaviour, and workers follow a plastic but age-linked progression from in-hive tasks such as nursing and processing nectar, to foraging (Lindauer 1953). Task transitions during the in-hive phase are more subtle than the transition to foraging, consequently research has focussed on comparing in-hive workers to foragers.

The honey bee brain is especially plastic in the transition from nurses to foragers: the volume of the mushroom body neuropile increases with foraging experience (Maleszka et al. 2009; Withers et al. 1993). But importantly, the brain begins changing in preparation for foraging long before foraging actually takes place (Farris et al. 2001), indicating that at least some of the differences between in-hive bees and foragers in fact occur in anticipation of foraging. The honey bee brain also exhibits neuroanatomical plasticity in response to extreme social stimuli: solitary isolation and the presence of dead bees retard mushroom body growth, but this effect can be countered by social environment (Maleszka et al. 2009).

The progression to foraging also involves changes in the expression of hundreds of genes (Kucharski & Maleszka 2002a; Sen Sarma et al. 2007; Whitfield et al. 2003, 2006), some of which have been examined in more detail (Ament et al. 2008; Kucharski & Maleszka 2002b; Wolschin & Amdam 2007; Yamazaki et al. 2006). Many of these molecular changes occur around days 7–8 of the adult maturation and coincide with the completion of anatomical changes in the brain (Guez et al. 2001, 2003; Whitfield et al. 2006).

Cognitive age is correlated more with task than chronological age in the honey bee (Behrends & Scheiner 2010; Ray & Ferneyhough 1999; Rueppell et al. 2007; Whitfield et al. 2003), providing a unique model in which chronological age can be neatly disentangled from cognitive age. The experimental construction of single cohort honey bee colonies with bees of a narrow age range exploits task plasticity by forcing workers to start foraging precociously and keep nursing longer (Huang & Robinson 1996).

The primary focus of this study is to assess whether altered DNA methylation in the brain correlates with task status, as this could be one mechanism regulating task-specific gene expression. While detrimental stimuli are known to alter DNA methylation and histone acetylation, the research presented here addresses whether a naturally occurring non-detrimental social stimulus can also produce changes to these marks.

In vertebrates, increasing age correlates with genomic hypomethylation and gene-specific promoter hypermethylation (Issa 2003; Wilson et al. 1987) and numerous other chromatin level changes (Thompson et al. 2010). Age-related epigenomic changes are also involved in age-related cognitive decline in vertebrates (Lockett et al. 2010b; Peleg et al. 2010). In contrast, extreme chronological age produces few cognitive deficits in the honey bee (Behrends & Scheiner 2010), although foraging duration is a strong predictor of cognitive senescence (Wolschin et al. 2009). Utilizing the separability of chronological age from cognitive age in the honey bee, the interaction between these two different measures of ‘age’ and DNA methylation in the brain was examined.

In this study intragenic methylation was assessed at cytosine–phosphate–guanosine (CpG) sites within eight genomic regions (described in detail later), comparing methylation levels between age-matched workers (13 or 29 days old), performing different tasks (nursing or foraging). Many significant changes were found, illustrating the highly dynamic nature of the neural methylome. More changes were correlated with task than chronological age, in line with current knowledge of cognitive aging in the honey bee; and the presence of hive-specific methylation patterns could indicate a superorganism ‘identity’ or transcriptional level control of nestmate identification signals. It is concluded that the social environment of the honey bee interacts with the epigenome, and that DNA methylation changes to certain genes may mediate task-specific alternative splicing.


Construction of single cohort colonies

Two separate single cohort colonies (SCCs) were constructed, and the experiments performed over February to March 2009 at the Australian National University (Canberra, Australia). Approximately 2000 worker bees emerging overnight – and therefore of the same age – were collected, marked with a dot of paint to the dorsal thorax and a queen added (the same queen was used for both SCCs). The entrance to the SCC was closed with mesh until bees reached 5 days of age, to avoid robbing by other hives before the bees were strong enough to defend their hive.

Dissection, DNA extraction and bisulfite conversion

Samples were collected when bees were 13- and 29-day old. ‘Foragers’ were bees returning to the hive carrying pollen, and ‘nurses’ were those nursing brood where possible, or otherwise processing nectar (tasks typically performed by young ‘nurse’ bees; Winston 1987), identified as bees head-first in cells containing brood or honey respectively. The behavioural status of each bee was confirmed with a morphological analysis of the hypopharyngeal glands (Maleszka et al. 2009). For each sample the brains of 10 bees were pooled and their DNA extracted and bisulfite converted (as in Kucharski et al. 2008) twice consecutively and stored at – 80°C.

Cloning and sequencing of full length amplicons

The main advantage of cloning and standard sequencing of long amplicons (compared with very short Illumina reads) is that each amplicon represents a single methylation event. Therefore, both methylation levels of individual cytosines and patterns of methylation within selected, relatively large regions of the target genes can be analysed with no need for applying complex mapping, filtration and statistical algorithms taking into account factors such as the sequencing quality, the bisulfite conversion rate, the number of reads covering each methylated site, etc.

The eight genes with putative brain functions selected for this study were chosen on the basis of our previous results (Foret et al. 2009; Kucharski et al. 2008; Lyko et al. 2010). These genes have been shown to be dynamically methylated in brains and larvae of queens and workers suggesting that they might be useful illustrators of methylation changes in the context of task specialization (see Table 1 for details). Intragenic regions were amplified from bisulfite-converted DNA by nested polymerase chain reaction (PCR), with 5′EcoRI (GCAGAATTC) or HindIII (CGCAAGCTT) adapters included in the forward and reverse nested PCR primers respectively to facilitate cloning (Table S1). Cycling for both primary and nested reactions consisted of 5 min at 94°C then 40 cycles of 15 seconds at 94°C, 15 seconds at 50°C and 1 min at 72°C, then a final 5 min at 72°C. For some amplicons the annealing temperature in the nested PCR was optimized: dynactin: 55°C, makorin: 60°C and PKCbp1: 65°C. Nested PCR products were resolved in agarose gels and the target bands excised and purified (Qiagen MinElute Gel Extraction kit, Qiagen, Hilden, Germany), restriction digested with EcoRI and HindIII (Promega, Madison WI, USA) for 30 min at 37°C, and repurified (Qiagen MinElute PCR Purification kit). Amplicons from each sample were then ligated to pBlueScript KS+ (Stratagene, La Jolla, CA, USA) and transformed into XL-1 competent cells (Stratagene). Recombinant clones were selected using blue/white selection and grown in 2.5 ml cultures for 16–24 h, then their plasmid DNA extracted and purified (QiaPrep Spin Miniprep kit). For each sample 16 or more clones were sequenced (AGRF, Brisbane, Australia); a total of 1215 sequences were used in the analyses presented in this study. More detailed molecular protocols were published previously (Ashby et al. 2007; Kucharski & Maleszka 1998).

Table 1. Selected Apis genes used in this study
BeeBase gene numberSymbolGeneral functionAlternative splicingNumber of CpGsNumber of methylated CpGs
  1. *Numbers of methylated CpGs refer to the entire gene and are based on genome-wide sequencing of DNAs from the brains of queens and workers (Lyko et al. 2010). Due to the paucity of sequencing reads, the number of methylated CpGs over the length of the fax could not be analysed with confidence. As such, the number of methylated CpGs given for fax is taken from this study and refers only to the analysed amplicon. See Tables S1 and S2 for more details.

GB12148 UPL Ubiquitin protein ligase E2: Brain development and adult synaptic plasticityAlternative 3′ end8536
GB12499 PKCbp1 Protein kinase C binding protein 1: Interacts with PKC and binds DNA via PWWP domainAlternative 3′ end5843
GB16176 Nadrin Rho GTPase-activating protein: Novel brain GTPase with multiple spliced variantsAlternative cassette exon and unspliced introns6046
GB17360 makorin Ubiquitin protein ligase E3 makorin 1: Brain development and adult synaptic plasticityAlternative exon splice sites2319
GB17380 fax Failed axon connections: Implicated in axonogenesisAlternative 5′ end12176*
GB17514 FAF FAS-associated factor 1 (Caspar): Multi-domain protein involved in signalling cascadesCassette exon with alternative splice site2317
GB18370 dynactin Dynactin subunit: Neuronal functions, axonal retrograde transportExon skipping1813
GB30092 NTFY Histone-like transcription factor: Several functions including axon guidanceAlternative 3′ end457

Sequence analysis and statistics

All clone sequences were analysed for the methylation status of each CpG using custom Microsoft Word® macros and each manually verified. The methylation status of each CpG was scored as 0 or 1. The resulting binary data were analysed statistically with spss Statistics 17.0 (IBM) or GenStat software (Version 12.0 ©2009 VSN International) using a generalized linear model of the binomial family, with task (nurse vs. forager), age (13 vs. 29 days) and hive (hive 1 vs. hive 2) used as factors to model the state of each CpG. For all analyses 0.05 was used as the threshold of statistical significance.


Task, age and hive correlate with DNA methylation

In total, this study surveyed the methylation levels of 83 CpGs across eight genes in the brains of age-matched honey bee workers performing nursing or foraging tasks. Comparisons between the average methylation levels in these samples in two separate hives revealed consistent and significant effects of age, task and hive on DNA methylation level, and also in some cases interactions between these factors. Out of the 83 CpGs surveyed, the methylation levels of 52 CpGs (63%) were correlated with at least one of the factors in this analysis (Fig. 1). Changes explained by a single factor can be interpreted most meaningfully, and in assessing the effects of task and age on methylation those changes not interacting with hive are more broadly biologically relevant. When broken down by gene, significant correlations especially with the primary factors (age, task and hive) are shown to be concentrated within certain regions (Table 2). dynactin and UPL methylation correlated more often with hive, whereas the methylation of many CpGs within Nadrin correlated with task. No CpGs within the examined region of makorin correlated significantly with task, age or hive (although there were correlations with factor interactions), and the correlations observed within FAF, fax, NTFY and PKCbp1 were distributed across the three factors examined in this study.

Figure 1.

Effects of task, age and hive on DNA methylation. The number of methylated CpGs at which DNA methylation is significantly correlated with age (blue), task (red) and hive (yellow) and interactions of these factors are shown in a Venn diagram (P < 0.05, generalized linear binomial modelling). Numbers sum to more than 83 because the methylation levels of some CpGs were correlated with more than one factor or factor interaction.

Table 2. Effects of task, age and hive on DNA methylation, by gene
GeneFactor or factor interaction
HiveAgeTaskH × AH × TA × TH × A × T
  1. The number of CpGs in each gene which significantly correlate with the factors and factor interactions included in this analysis, within the amplified region from each gene, are shown below (P < 0.05, generalized linear binomial model). Correlation values were calculated using GenStat software. The complete data set from which these summaries are derived is described in Table S3, and includes amplified regions from dynactin (panel A), FAF (panel B), fax (panel C), makorin (panel D), Nadrin (panel E), NTFY (panel F), PKCbp1 (panel G) and UPL (panel H). P-values for correlation of the methylation of each CpG with each factor and factor interaction are displayed in each panel, alongside graphs illustrating the average methylation level at each CpG in each sample. Test statistics and degrees of freedom associated with these P-values are given in Table S4.

Number of CpGs significantly correlated
dynactin 5014022
FAF 1122101
fax 2120100
makorin 0001132
Nadrin 3063441
NTFY 1201000
PKCbp1 2221241
UPL 5311030

Methylation of PKCbp1 CpG #4 is strongly correlated with task

Significant correlations between methylation level and task were observed at many CpGs within the analysed amplicon of PKCbp1, the strongest of which was CpG #4 where methylation levels are significantly higher in the brains of foragers than nurses (Fig. 2). Other CpGs whose methylation levels correlated significantly with task alone are dynactin CpG #3, FAF CpG #3, fax CpG #1 and Nadrin CpG #13.

Figure 2.

Correlation between DNA methylation level and task within PKCbp1. CpG #4 within the examined region of PKCbp1 is more methylated in the brains of foragers than nurses, at both 13 days (d) and 29 d of age and in both hives (generalised linear binomial modelling: χ2 = 16.606, df = 1, p < 0.001). No other factors or factor interactions are significantly correlated with the methylation level of this CpG (Table S3 panel G).

Methylation of PKCbp1 CpGs #1, #3 and #6 correlates positively with cis CpG #4 methylation

By comparing the methylation state of PKCbp1 CpG #4 with the methylation states of other CpGs within each clone, correlation of methylation along the same strand of DNA (i.e. cis correlation) could be examined. This is one advantage of using a cloning and sequencing approach. PKCbp1 clones from all samples were pooled and modelled with a generalized linear model of the binomial family, using CpG #4 methylation status as a factor. The methylation states of CpGs #1, #6 and especially #3 correlated significantly with that of CpG #4 on the same DNA molecule (Table 3). Correlations between the methylation states of all PKCbp1 CpGs in cis are given in Table 4. It can be seen that the cis correlations of the methylation states of CpGs #1, #3 and #6 with the methylation state of CpG #4 are all positive.

Table 3. Statistical significance of correlation of cis methylation states of PKCbp1 CpGs with CpG #4 methylation state
  PKCbp1 CpG
  1. P-values are obtained from χ2 values, which are described as approximate due to their non-normal error distribution. Correlation values were calculated using GenStat software.

Table 4. Correlations of cis methylation states of PKCbp1 CpGsThumbnail image of


Task and age correlate with DNA methylation

Many of the CpGs examined in this study show variation in the level of DNA methylation, and in many cases this variation correlates significantly with the age, task or hive of those bees. For some genes many CpGs correlate with one of these factors (Table 2). While these findings could be interpreted as each representative of DNA methylation changes across the whole length of the gene, it is important to note that many CpGs also exist outside the amplified regions (e.g. PKCbp1, Fig. S1, supporting information). The use of a cloning and sequencing approach is restricted in this sense, and provides a sample of each gene rather than an absolute representation. Effects could well differ over the length of the gene, especially if DNA methylation control of alternative splicing is focussed locally around splice sites (Lyko et al. 2010). However, our previous analyses of dynactin show that individual CpG sites reveal greater differences between the castes than those illustrated by the average methylation levels across the entire fragment (Kucharski et al. 2008). This finding suggests that certain CpG sites might be preferentially methylated to produce specific gene products and could be used as better illustrators of methylation dynamics than entire genes.

In terms of single factor correlations, across all genes task and hive correlate with the methylation levels of the most CpGs (19 and 18 CpGs, respectively), although age also correlates significantly with the methylation levels of 7 CpGs (Fig. 1).

Considering only those methylation changes not interacting with hive, more CpGs exhibited methylation level changes correlated with task than with age (Fig. 1). This is in line with the accepted notion that cognitive age in the honey bee is more correlated with task than with chronological age (Withers et al. 1993). In wasps too, the neural transcriptome is associated with brood provisioning behaviour (Toth et al. 2007).

Although the impacts of task and hive are stronger, the methylation levels of some CpGs do change with chronological age (Fig. 1), indicating that bees are not immune to the effects of time. Methylation of certain CpGs within Nadrin decreases in the MB Kenyon cells with extreme old age, (93 days bees, compared to 7 days) (Maleszka et al., unpublished data). Among the nine CpGs significantly correlated with age in this study, seven were more methylated in older bees. In vertebrates, the trend with increasing age is a combination of genomic hypomethylation and gene-specific promoter hypermethylation (Christensen et al. 2009; Issa 2003; Wilson et al. 1987). While the data presented here are only a small sample of the methylome, if representative they indicate that some regions increase in methylation with age, comparable to the gene-specific hypermethylation observed in vertebrates.

The methylation levels of 16 CpGs were significantly associated with the task × age interaction, demonstrating that the effects of task on DNA methylation levels can be different at different ages. Interestingly, miRNAs in the honey bee brain too cluster in groups upregulated either in young nurses or in old foragers, rather than by task irrespective of age (Behura & Whitfield 2010). Together with the results presented here this suggests younger vs. older bees performing the same task do differ to some extent. Non-coding RNAs such as miRNAs are thought to interact complexes coordinating chromatin level processes such as DNA methylation and histone acetylation to provide the necessary sequence specificity (Mattick et al. 2009), suggesting interactions between miRNAs and methyl-CpG-associated complexes could regulate age polyethism.

Overall, the changes in DNA methylation levels observed in the brain with task and age were subtle, supporting the ‘systems-level’ notion that transcriptional changes in task progression and aging are regulated by tweaking immense gene networks (Behura & Whitfield 2010; Kucharski & Maleszka 2002a; Sen Sarma et al. 2007; Whitfield et al. 2003).

Hive-specific methylomes

The methylation levels of many of the CpGs surveyed in this study correlated with the hive-of-origin, suggesting hive-specific DNA methylation profiles shared by nursing and foraging members of each colony. Honey bee colonies can be regarded as superorganisms analogous to individuals of other species, and indeed similar strategies are used by both vertebrate individuals and honey bee colonies for many facets of life (Seeley 1989; Wheeler 1928). It could be that honey bee colonies accumulate specific DNA methylation patterns with time, in much the same way as do human individuals (Bjornsson et al. 2008; Fraga et al. 2005). The data presented here support the honey bee superorganism metaphor extending to the molecular level. Colony-specific methylation cultures may be reflected genome wide, although nestmate identification constitutes an especially intriguing possible use for epigenomic signatures. Guard bees quickly and easily distinguish their nest-mates using cuticular hydrocarbons: workers share their colony's distinct identifying blend (Howard & Blomquist 2005), and transcriptional differences resulting from differential DNA methylation could form the molecular basis of this phenomenon. Hive-specific epigenetic cultures are unlikely to reflect distinct worker genotypes, as workers in each hive were a heterogeneous mix of emerging workers sourced from several hives.

These results also reveal some hive × age and hive × task interactions, indicating that age and task are significantly correlated with DNA methylation levels but that this can differ between hives. Task-associated gene expression changes too vary greatly between studies, leading the authors of one meta-analysis to conclude that ‘The imperfect overlap of results from the individual- and meta-study analyses corroborates reports that the association between the expression of numerous genes and behavioural maturation is highly sensitive to other genetic or environmental factors’ (Adams et al. 2008), a hypothesis supported by the DNA methylation data presented here.

Task-associated methylation of PKCbp1 may control alternative splicing

Our analysis detected one particular CpG whose methylation level varies in extremely strong correlation with task status, PKCbp1 CpG #4 (Fig. 2). This region of the PKCbp1 gene was previously verified to be methylated in the honey bee (Foret et al. 2009), and CpG #4 immediately precedes an intron/exon boundary (Fig. S1). Because intragenic DNA methylation is thought to mediate alternative splicing in the honey bee brain (Lyko et al. 2010), it is hypothesized that this particular CpG could be controlling the production of alternatively spliced task-specific PKCbp1 isoforms. In a survey of 42 proteins in the brain, not one differed in abundance between nurses and foragers (Brockmann et al. 2009). This is a surprising finding given the dramatic neural changes associated with foraging behaviour, yet could be explained by task-associated alternative splicing producing different isoforms of neural proteins.

With the recent availability of next-generation transcriptome data for the honey bee, it is now possible to easily obtain vast numbers of short read cDNA sequences and examine them for alternatively spliced isoforms (www.ncbi.nlm.nih.gov/sra). Short reads homologous to PKCbp1 do in fact reveal alternative splicing near CpG#4, therefore, it is entirely feasible that the methylation status of CpG #4 could determine task-specific alternative splicing of PKCbp1.

Within promoters of vertebrate genes, groups of neighbouring CpGs often seem to change methylation levels en masse whereas single CpG methylation is regarded as a much more precise event (Weaver et al. 2004). In the results presented above, methylation of PKCbp1 CpG #4 but not the neighbouring CpGs is strongly correlated with task, suggesting a sequence-specific single CpG event. Interestingly, examination of individual PKCbp1 clones reveals significant positive correlation of the methylation status of CpGs #1, #3 and #6 with the cis methylation status of CpG #4 (Tables 3 and 4). That neighbouring CpGs are co-methylated with CpG #4 yet not in a manner correlated with task highlights the precision of the task-specific methylation observed at CpG #4. Single CpGs could well be enough to influence alternative splicing, as methyl-binding proteins are capable of binding to single methyl-CpGs (Nan et al. 1993), and evidence for alternative splicing in response to a single CpG has been reported previously (Klamt et al. 1998). Furthermore PKCbp1 CpG #4 immediately precedes an exon/intron boundary (Fig. S1).

PKCbp1 is believed to bind protein kinase C (PKC), which has dual roles in foraging and memory processing in the honey bee (Humphries et al. 2003). In addition to its ability to form a complex with PKC, PKCbp1 also has the DNA-binding domain PWWP found in proteins such as DNA methyltransferases that target distinct genomic regions. One possibility is that PKC/PKCbp1 complex is involved in sequence-specific interactions linked to the epigenomic machinery that regulates brain plasticity genes. A corollary of the link between PKC, foraging and memory is that differential DNA methylation-controlled PKCbp1 splicing, if verified to be associated with task, might also be associated with memory processing. Foraging is regarded as having greater memory requirements than in-hive tasks, consequently the many changes occurring in the transition to foraging likely equip these bees for the cognitive demands of foraging. DNA methylation as a whole is involved in memory processing in the honey bee (Lockett et al. 2010a), although which specific genes show altered DNA methylation levels is as yet unknown. The data presented here suggest PKCbp1 as a strong candidate.

The results presented herein show dynamic socially induced changes to neural DNA methylation levels, correlated with task and hive-of-origin, and to a lesser extent chronological age. These findings suggest the existence of epigenomic cultures in honey bee colonies as superorganisms, and support the control of senescence in honey bee workers via social stimuli rather than time. Our study provides initial evidence for the involvement of epigenomic modifications at the DNA methylation level in mediating behavioural plasticity of honey bee workers by generating context-dependent molecular flexibility in the brain. Our results support the view that the highly dynamic nature of this process orchestrates the complex interplay between social stimuli and the genome and ultimately determines workers' behavioural outcomes.


We thank Paul Helliwell for his expert beekeeping and help with setting up the SCCs. This work was supported by the Australian Research Council grant DP1092706 and National Health and Medical Research Council grant 585442 awarded to R. M.