Reversible switching between epigenetic states in honeybee behavioral subcastes

Abstract

In honeybee societies, distinct caste phenotypes are created from the same genotype, suggesting a role for epigenetics in deriving these behaviorally different phenotypes. We found no differences in DNA methylation between irreversible worker and queen castes, but substantial differences between nurses and forager subcastes. Reverting foragers back to nurses reestablished methylation levels for a majority of genes and provides, to the best of our knowledge, the first evidence in any organism of reversible epigenetic changes associated with behavior.

Full text

© 2012 Nature America, Inc. All rights reserved.
nature neuroscience ADVANCE ONLINE PUBLICATION 1
Brief communications
Epigenetic changes are thought to underlie lineage-specific differ-
entiation, as the pattern of gene expression is stably changed, but
the DNA sequence remains the same. Recently, the epigenome of a
specific differentiation pathway was mapped, defining hundreds of
differentially methylated regions (DMRs) that define lineage commit-
ment in mouse hematopoietic progenitors. DNA methylation appears
to be critical in that system for lineage specificity, as lymphoid cells
show greater global DNA methylation than myeloid cells1. However,
the roles of the epigenome in global changes in organismal remod-
eling or in behavior have not been defined2.
The honeybee Apis mellifera is an ideal model organism for such
studies3, as it organizes social structures from distinct individual forms
that can emerge from one genome. A female embryo may become a
queen by receiving a diet of royal jelly and commit her life to egg-
laying (germline) or become a sterile helper ‘worker’ (soma)4. Workers
follow a rich behavioral program of nursing and later undergo a tran-
sition to foraging that involves extensive gene expression changes in
the brain5. In contrast with queens, worker behavior is remarkably
flexible: age-matched workers can nurse or forage, and foragers may
revert to nursing tasks6.
To investigate the potential role of DNA methylation in defin-
ing honeybee caste phenotypes, we compared the methylomes of
sister queens versus workers and sister nurses versus foragers by
whole genome bisulfite sequencing (WGBS) and comprehensive
high-throughput array-based relative methylation (CHARM) anal-
ysis1. CHARM covers 85% and WGBS covers 92% of the CpGs in
the 270-Mb genome, both revealing sparse methylation throughout
the bee genome (Supplementary Fig. 1).
We first compared five biological replicates of queens and workers,
both collected within 4 h of adult emergence from the pupal stage
(Fig. 1a). Brain was selected because of its influence on behavior and,
unlike ovary, is similar in size between queens and workers. CHARM
analysis found no significant DMRs (at a false discovery rate (FDR)
of 5%, permutation test) between queens and workers. WGBS of the
same samples found no differences using single CpG t tests corrected
for multiple testing. In addition, we tested the top-ranked differences
by CHARM, albeit statistically insignificant (FDR cutoff of 5%), using
bisulfite pyrosequencing, an independent measure of DNA methyla-
tion at the single-base level, and found no caste-specific differences
(Supplementary Fig. 2a–c).
Given these negative results, we then compared subcastes of
workers. Initially, most workers are nurses that care for the queen and
larvae inside the hive. About 2–3 weeks later, the majority switches to
foraging and collect pollen, nectar and water outside5. Using CHARM,
we identified 155 DMRs that distinguished nurses from age-matched
foragers (Figs. 1b and 2a,b, Table 1, Supplementary Table 1 and
Supplementary Fig. 3a–e). Approximately 70% of DMRs overlapped
exons (Table 1 and Supplementary Table 1), similar to previous find-
ings7,8. The genes associated with the 155 nurse-to-forager DMRs
appeared to be enriched for gene regulation and development through
transcriptional control and chromatin remodeling. Many histone modi-
fication writers, including LOC412350 (a histone deacetylase similar
to Hdac3), JIL-1 (a histone phosphotransferase) and LOC411070
(a histone H3 methyltransferase9), increased in methylation during
the nurse to forager transition. In addition, DEAD-box helicase genes
Iswi and spn-E have chromatin remodeling capacity and are involved
in morphogenesis10. Iswi in particular is involved in dendrite morpho-
genesis11 and may contribute to noted changes to the nurse brain
before foraging5.
To determine whether the DMRs that we observed during the nurse
to forager transition are linked to phenotype, and not simply the
result of the transition, we induced the reversion of foragers back to
nurses using a strategy of hive trickery. To initiate reversion, we set up
foragers to return to a hive in which only queen and larvae are present
(Fig. 1b). The foragers will then segregate into reverted nurses that
pick up caregiver tasks and continuing foragers that do not change
behavior6. Reversion separates changes caused by nervous system
development, maturation and foraging experience that are shared
between reverted nurses and foragers, but not nurses, from changes
robustly linked to current behavior that are shared between reverted
nurses and nurses, but not foragers.
Reversible switching between
epigenetic states in honeybee
behavioral subcastes
Brian R Herb1,2,8, Florian Wolschin3,4,8, Kasper D Hansen1,5,
Martin J Aryee1,6, Ben Langmead5, Rafael Irizarry5,
Gro V Amdam3,4 & Andrew P Feinberg1,2,5,7
In honeybee societies, distinct caste phenotypes are created from
the same genotype, suggesting a role for epigenetics in deriving
these behaviorally different phenotypes. We found no differences
in DNA methylation between irreversible worker and queen
castes, but substantial differences between nurses and forager
subcastes. Reverting foragers back to nurses reestablished
methylation levels for a majority of genes and provides, to the
best of our knowledge, the first evidence in any organism of
reversible epigenetic changes associated with behavior.
1Center for Epigenetics, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA. 2Department of Medicine, Johns Hopkins University School of
Medicine, Baltimore, Maryland, USA. 3Department of Chemistry, Biotechnology and Food Science, Norwegian University of Life Sciences, Aas, Norway. 4School of
Life Sciences, Arizona State University, Tempe, Arizona, USA. 5Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland,
USA. 6Department of Oncology, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA. 7Department of Molecular Biology and Genetics,
Johns Hopkins University School of Medicine, Baltimore, Maryland, USA. 8These authors contributed equally to this work. Correspondence should be addressed to
G.V.A. (gro.amdam@asu.edu) or A.P.F. (afeinberg@jhu.edu).
Received 2 July; accepted 20 August; published online 16 September 2012; doi:10.1038/nn.3218

© 2012 Nature America, Inc. All rights reserved.
2 ADVANCE ONLINE PUBLICATION nature neuroscience
Brief communications
With CHARM, we found 107 DMRs for the forager-to–reverted
nurse transition. The genes associated with these CHARM DMRs
appeared to be enriched in transcription factors and DEAD-box
helicases, as seen in the nurse-to-forager CHARM DMRs (Fig. 2a,b,
Supplementar y Table 2 and Supplementary Fig. 4). Of these
107 CHARM DMRs, 57 overlapped with CHARM DMRs associated
with the nurse-to-forager transition, a markedly close concordance
(P < 2.2 × 10−16 by Fisher’s test, P < 10−3 based on 1,000 permuta-
tions; Supplementary Fig. 5). This subset of epigenetically reversible
genes showed enrichment for development, ATP binding and nuclear
pore formation (Supplementary Table 2). These genes include the
ortholog to kismet, LOC726524, which regulates developmental
genes such as hedgehog and affects learning and axon migration in
Drosophila12,13, and might explain observed differences in learning14
between nurses and foragers. In addition, DEAD-box helicase genes
LOC725306 (ref. 15) and LOC726524 both have roles in transcription,
whereas LOC411989 is involved in translation16.
To independently validate this result, we replicated the reversion
experiment and created six new pools of six brains for both foragers and
reverted nurses. We performed WGBS on these 12 samples and found
that 45 of 57 reversion DMRs showed the same direction of change in
methylation between CHARM and WGBS (Supplementary Fig. 6).
This overlap of DMRs between replicated experiments was highly sig-
nificant (P = 3.3 × 10−6). Furthermore, the 45 WGBS-correlated genes
showed enrichment for ATP binding and nuclear pore formation
(Supplementary Table 3), consistent with our analysis of the 57 CHARM
Queen Worker
Fertilized
egg
0
20
40
60
80
100
Methylation
0 DMRs
FDR < 0.05
Royal
jelly
a
Chr 1.38: 150318 152818
Q
Q
0
20
40
60
80
100
0
20
40
60
80
100
0
20
40
60
80
100
Queen
+
1/2 brood
Methylation
Methylation
Methylation
Nurses
+
1/2 brood
Foragers
Nurse to forager
155 DMRs
Forager to nurse
107 DMRs
Reversion
57 DMRs
Reverted
nurses
b
Chr 5.21: 81668 82668
Chr 12.25: 50288 52288 Chr 3.21: 414942 525642
Figure 1 DNA methylation changes were found between nurses and
foragers, but not between queens and workers. (a) We compared
newly emerged queens and workers using CHARM (n = 5 per
phenotype) and found no statistically significant differences
(FDR cutoff of 5%). (b) DNA methylation changes during the
nurse-to-forager transition and changes back during the forager-
to-nurse transition (n = 3 per phenotype). We found 155 DMRs
associated with the nurse-to-forager transition, 107 DMRs associated
with the forager-to-nurse transition and 57 DMRs common to both
lists that exhibited a nurse-specific signature.
−40 −20 0
20
40
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
c
P = 0.00103
Difference in expression (log
2
, F – RN)
Difference in methylation CHARM (F – RN)
ba
CHARM
methylation
WGBS
methylation
t test
RNAseq
expression
Exon
junctions
CpG
density
Genes
Forager
RNA reads
Reverted
nurse
RNA reads
20
40
60
80
100
20
40
60
80
100
−4
0
4
−4
0
4
0
400
800
0
400
800
0
0.1
−
+
Chr 15.20: 220306 221095
LOC408546/CG7956
Chr 8.14: 190545 191453
LOC412115/Atg9
Forager
Reverted nurse
Nurse
20
40
60
80
100
20
40
60
80
100
−4
0
4
−4
0
4
0
400
1,000
0
400
1,000
0
0.1
−
+
CHARM
methylation
WGBS
methylation
t test
RNAseq
expression
Exon
junctions
CpG
density
Genes
Forager
RNA reads
Reverted
nurse
RNA reads
Figure 2 DNA methylation distinguishes nurses, foragers and reverted nurses. (a,b) Two examples of CHARM DMRs. Top, percent methylation for both
CHARM and WGBS data sets, with points representing individual samples and the smoothed lines representing the average for the phenotype. The
t test panel displays the top 1% differentially methylated CpGs by t test. The color of the point indicates which phenotype had greater methylation at
that CpG (n = 6 per phenotype). The RNAseq expression panel is a t statistic based on the number or reads detected in the annotated exons, with the
color indicating the higher expressed phenotype. The exon junctions panel is a t statistic based on the number of reads detected spanning the exon
junctions, as predicted by the TopHat program, with the color indicating the higher expressed phenotype. Switching between higher expressed nurse
and forager exon junctions is indicative of alternative splicing events. The RNA reads panels indicate the number of reads per phenotype as compiled
by TopHat program (n = 6 per phenotype). The bottom two panels show the CpG density and the relative position of the gene. (c) Plot of relative gene
expression comparing foragers (F) to reverted nurses (RN). We tested 26 genes associated with DMRs for expression differences by real-time PCR
(n = 12 per phenotype). The plot depicts the difference in average log2 expression versus average difference in methylation as determined by CHARM.
Correlation analysis results in a P value of 0.001.

© 2012 Nature America, Inc. All rights reserved.
nature neuroscience ADVANCE ONLINE PUBLICATION 3
Brief communications
reversion DMRs. These results provide evidence for a nurse-specific
methylome that needs to be reestablished during the reversion.
To determine the relevance of these reversible DMRs, we performed
transcriptome sequencing (RNAseq) on six pools each of foragers and
reverted nurses. We then used the TopHat program to analyze the
RNAseq data to predict the location of annotated and unannotated
exons and to determine the prevalence of exon skipping. This analysis
revealed that 22 of the 45 WGBS-correlated reversion DMRs colo-
calized with alternative splicing events (Fig. 2b and Supplementary
Fig. 7a–f). These data suggest a high incidence of alternative splicing
events in DMRs and strengthen the potential role of DNA methyl-
ation in regulating alternative splicing8. We also found a negative
correlation between gene expression and levels of DNA methylation
between foragers and reverted nurses for 26 genes by real-time PCR
(P = 0.00103; Fig. 2c and Supplementary Fig. 8a–f).
Our data suggest a strong link between reversible DNA methylation
and nurse-forager transition and reversion, but no relationship to
queen-worker segregation. These data stand in contrast with a study
comparing 2.5-week-old mated queens and 8-d-old foraging-capable
workers8, which is likely explained by the difference in timing of the
data, that is, newly emerged queens and workers. Although DNA
methylation may be involved in distinguishing queens from workers
during development3, our data clearly indicate that the queen and
worker brain methylomes are the same at the time of emergence,
despite differences in body morphology.
In summar y, we found substantial DNA methylation changes
that accompany phenotype switching in honeybee subcastes. Genes
associated with these DMRs can potentially influence global gene
expression patterns by altering chromatin structure or regulating
transcriptional machinery. Profound phenotype shifts between
nurses and foragers may be orchestrated by a subset of genes, which
are themselves regulated by DNA methylation. Key regulatory genes
may either be differentially expressed or differentially spliced, which
we correlate with changes in DNA methylation. For example, the
eIF-4a homolog LOC411989, which is critical for translation initia-
tion17, exhibits alternative splicing in an exon that codes for RNA
binding (Supplementary Fig. 7c). Different isoforms of eIF-4a may
bind to RNA with greater affinity, thereby globally affecting the rate
or regulation of translation. As this differentially spliced exon is in
a DMR, methylation might be used to remember which isoform to
express in nurses or foragers.
We found that DNA methylation was able to revert, concomitant
with experimental reversion of foragers back to nurses, which we dem-
onstrated in replicated experiments. This suggests a subcaste-specific
methylation signature that assists in forming subcaste phenotypes.
Although studies in rodents found methylation changes associated
with learning, these changes disappear over several hours and do not
establish a stable phenotype18,19. Similarly, nurturing can induce long
lasting methylation marks in rodents20. Our results are, to the best of
our knowledge, the first to show reversible DNA methylation corres-
ponding to a reversible behavioral phenotype in any species.
METHODS
Methods and any associated references are available in the online
version of the paper.
Accession codes. WGBS and RNA sequencing data: NCBI SRA,
SRA050798. CHARM data: NCBI GEO, GSE36650.
Note: Supplementary information is available in the online version of the paper.
ACKNOWLEDGMENTS
We thank E. Fennern, N. Baker, K. Flores and O. Kaftanoglu for assistance with colonies,
bees and brain dissections, and A. Doi for reviewing the manuscript. We thank
G. Klein for serving as scientific schadchen to A.P.F. and G.V.A. after hearing them
lecture on different occasions at I. Ernberg’s “What is Life” series at the Karolinska
Institute, without which this research would not have taken place. G.V.A. was funded by
the Research Council of Norway #191699 and the PEW Charitable Trust #2009-000068-
001. A.P.F. was funded by US National Institutes of Health grant 1DP1OD008324.
AUTHOR CONTRIBUTIONS
B.R.H. performed genome-scale, gene-specific DNA methylation analysis and
performed gene expression analysis. F.W. raised bees and manipulated hives
for reversion experiment, and collected bees and dissected brains. R.I., M.J.A.,
K.D.H., B.L. and B.R.H. performed statistical analysis. B.R.H. and M.J.A. generated
microarray data sets. B.R.H., B.L. and K.D.H. generated WGBS and RNAseq data
sets. A.P.F. and G.V.A. conceived, designed and oversaw the experiments. A.P.F.,
B.R.H. and G.V.A. wrote the paper with the assistance of K.D.H., M.J.A., B.L. and F.W.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Published online at http://www.nature.com/doifinder/10.1038/nn.3218.
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. Ji, H. et al. Nature 467, 338–342 (2010).
2. Wolf, C. & Linden, D.E. Genes Brain Behav. 11, 3–28 (2012).
3. Wang, Y. et al. Science 314, 645–647 (2006).
4. Winston, M.L. The Biology of the Honey Bee (Harvard University Press, 1987).
5. Whitfield, C.W. Science 302, 296–299 (2003).
6. Robinson, G.E., Page, R.E., Strambi, C. & Strambi, A. Ethology 90, 336–348 (1992).
7. Elango, N., Hunt, B.G., Goodisman, M.A. & Yi, S.V. Proc. Natl. Acad. Sci. USA
106, 11206–11211 (2009).
8. Lyko, F. et al. PLoS Biol. 8, e1000506 (2010).
9. Suganuma, T. et al. Nat. Struct. Mol. Biol. 15, 364–372 (2008).
10. Tsukiyama, T., Daniel, C., Tamkun, J. & Wu, C. Cell 83, 1021–1026 (1995).
11. Barak, O. et al. EMBO J. 22, 6089–6100 (2003).
12. Terriente-Félix, A., Molnar, C., Gómez-Skarmeta, J.L. & de Celis, J.F. Dev. Biol.
350, 382–392 (2011).
13. Melicharek, D.J., Ramirez, L.C., Singh, S., Rhea, T. & Marenda, D.R. Hum. Mol.
Genet. 19, 4253–4264 (2010).
14. Amdam, G.V. Aging Cell 10, 18–27 (2011).
15. Goldman-Levi, R., Miller, C., Bogoch, J. & Zak, N.B. Nucleic Acids Res. 24,
3121–3128 (1996).
16. Linder, P. Nucleic Acids Res. 34, 4168–4180 (2006).
17. Parsyan, A. et al. Nat. Rev. Mol. Cell Biol. 12, 235–245 (2011).
18. Guo, J.U. et al. Nat. Neurosci. 14, 1345–1351 (2011).
19. Miller, C.A. & Sweatt, J.D. Neuron 53, 857–869 (2007).
20. Weaver, I.C. et al. Nat. Neurosci. 7, 847–854 (2004).
Table 1 Summary of DMRs
DMR type Number of DMRs
Percentage of DMRs
overlapping exons GO category examples Example genes
Nurse-to-forager transition 155 72.2% Helicase activity, chromatin
remodeling, neuron development
alpha-Cat, Stat92E, Dhc64C, alpha-Spec,
hts, spn-E
Forager-to-nurse transition 107 72.9% Nucleoside binding, helicase
activity, cytoskeleton organization
Hel89B, Hsc70Cb, CG7177, Upf1, CG2017,
eIF-4a, kis, Dhc64C
Overlap 57 73.7% Nuclear import, cell differentiation,
ATP binding
Ranbp21, Fs(2)Ket, Mtor eIF-4a, kis, poe,
bur, BicD

© 2012 Nature America, Inc. All rights reserved.
nature neuroscience doi:10.1038/nn.3218
ONLINE METHODS
Bee preparation. For each replicate, two colonies were initially prepared for nurse
and forager rearing. They each consisted of 6,000–7,000 newly emerged (0–24 h
old) workers born to five sister queens of our standard research stock and a
wild-type queen of Californian commercial origin. All individuals were paint-
marked (Testors) on the thorax; after a solid foraging pattern was established,
foragers were paint-marked on the abdomen for the purpose of tracking their
life history. The behavioral reversion was carried out essentially as described
before21. This reversion resulted in two forager-derived colonies and two nurse-
derived colonies. Young brood of Californian commercial origin was provided to
the colonies as incentive for reversal from foraging to nursing behavior. For the
replicate used in the CHARM and bisulfite pyrosequencing assays, bees of three
groups were collected 12–14 d after the reversion: continuous foragers (forager
bees that had not reverted to nursing behavior), reverted nurse bees (forager bees
that exhibited nurse-like behavior (head in brood cells, sluggish response to a
flight challenge) in the brood area) and continuous nurses (bees that had never
been observed foraging and exhibited at least one nurse-like behavioral trait at
the time of collection). For the replicate used in the WGBS and RNAseq assays,
continuous foragers and reverted nurse bees were collected for the validation of
the CHARM results.
Queens and workers were derived from eggs produced by a single drone–
inseminated honeybee queen that belonged to a standard research stock with
restricted genetic background22 and were left to develop into 2-d-old larvae in
the hive. Subsequently, larvae at this crucial point in the divergent development
of queen and worker traits were either allowed to develop into worker pupae
or, for queen pupae development, they were manually grafted into queen cells
and raised as previously described23. On the day before emergence (the day of
sample collection), queen and worker pupae were transferred into an incubator
at 33 °C, 65–70% relative humidity.
For all bees, individuals from both experiments were collected directly into
2 ml of 80% ice-cold ethanol and stored at 4 °C until the central brains were
dissected (<48 h). Following dissections, central brains were immediately trans-
ferred into liquid N2 and stored at −80 °C until further use.
CHARM DNA methylation analysis. Genomic DNA was isolated from brains
using the Masterpure kit from Epicentre. Brains from the nurse/forager study
were pooled in groups of seven for each biological replicate, three replicates per
phenotype. Brains from the queen/worker study were pooled in groups of eight
for each biological replicate, five replicates per phenotype. Genome-wide methyl-
ation was assessed by CHARM24,25 and performed as previously described1.
A custom-designed Nimblegen 2.1 million feature array was designed for the
honeybee genome, which covers approximately 200 Mb of non-repeat genomic
sequence and includes ~8.7 million CpG sites covering ~85% of the ~10.2 million
CpG sites in the genome.
DMRs for the nurse/forager study were determined by a cutoff of at least 10%
methylation difference24, and DMRs with an average methylation close to the
baseline of 20% were eliminated. DMRs from pairwise comparisons were com-
bined to determine the relative methylation of all three phenotypes. Methylation
differences between continuous nurses and both reverted nurses and continuous
foragers were determined, and these differences were plotted. Clustering analy-
sis identified three classes of DMRs (Supplementary Fig. 9). GO analysis26 of
each class was performed by first determining the closest Apis mellifera gene to
each DMR. The orthologous gene in Drosphila melanogaster was found for each
Apis mellifera gene, and this new gene list was used for GO analysis using the
web tools available on the DAVID bioinformatics database at http://david.abcc.
ncifcrf.gov/. DMRs for the queen/worker study were determined by calculating
a FDR score for each potential DMR.
Bisulfite pyrosequencing. Approximately 400 ng of pooled genomic DNA
that was also used for the CHARM analysis was bisulfite converted using the
Zymo DNA-Methylation Gold kit. We used nested PCR to amplify regions
of interest in DMRs and quantified the level of methylation using the Biotage
PSQ HS96 pyrosequencer. The percent methylation for every CpG in our target
region was calculated using the Q-CpG methylation software (Biotage).
Control DNA was prepared using the Repli-G kit (Qiagen) from genomic
DNA. Repli-G amplified DNA served as a 0% methylated control. The 100%
methylated control was created by treating the Repli-G amplified DNA with SssI
methyltransferase (NEB), which methylates every CpG site. The 25%, 50% and 75%
methylated controls were created by mixing 0% and 100% controls. All controls
were bisulfite treated with same Zymo kit as test samples. Primer sequences are
listed in Supplementary Table 4.
Real-time quantitative RT-PCR. RNA was extracted from single brains by first
lysing cells in Chaos buffer (4.5 M guanidinium thiocyanate, 2% N-lauroylsar-
cosine (wt/vol), 50 mM EDTA, 25 mM Tris-HCl, 0.1 M β-mercaptoethanol)
followed by phenol, chloroform, then purified with Qiagen RNeasy columns.
cDNA was synthesized using Quantitect Reverse Transcription Kit (Qiagen) and
1 ng of cDNA was used for each real-time PCR reaction. Fast Sybr green (Applied
Biosystems) was used for real-time PCR reaction and quantified by 7900HT
(Applied Biosystems). Primer sequences are listed in Supplementary Table 5.
WGBS. Libraries of queen/worker and reverted nurse/forager samples were
created using TruSeq DNA library preparation kits (Illumina) with some
modification to the standard protocol. Genomic DNA samples were prepared
by homogenizing pools of whole brains. For queen/worker samples, five pools
of eight brains per pool were prepared. For reverted nurse/forager samples, six
pools of six brains per pool were prepared. Genomic DNA was extracted from
homogenized brains using either Masterpure kit from Epicentre (queen/worker),
or lysing cells in Chaos buffer (4.5 M guanidinium thiocyanate, 2% N-lauroylsar-
cosine, 50 mM EDTA, 25 mM Tris-HCl, 0.1M β-mercaptoethanol) and purified
with Qiagen DNeasy columns (reverted nurse/forager). For all samples, genomic
DNA was sheared to an average size of 350 bp using a Covaris sonicator with the
following settings: duty cycle = 10%, intensity = 5.0, bursts per second = 200,
duration = 60 s. Blunt ends were created on the DNA fragments using a unique
protocol to eliminate the introduction of non-genomic cytosines into the frag-
ments, which would be falsely interpreted as unmethylated cytosines during
subsequent analysis. To achieve this, we only used A, G and T nucleotides with
a mixture of the enzymes T4 DNA polymerase, Klenow DNA polymerase and
T4 PNK(NEB) to perform end repair of fragments. Illumina adapters were then
ligated to the fragments after the addition of a single A, per TruSeq protocol.
Libraries were then size selected by cutting a 400–500-bp fragment from an aga-
rose gel (BioRad-Certified Low Range Ultra Agarose, NEB 100-bp DNA Ladder,
Invitrogen-SYBR Gold nucleic acid gel stain) and purified using Qiagen MinElute
Gel Extraction Kit. The purified libraries were then bisulfite converted and puri-
fied using Zymo EZ DNA Methylation Gold. Libraries were then amplified using
a mixture of Uracil insensitive polymerases; Denville Choice Taq and Agilent
Turbo Pfu. Queen/worker samples were amplified for 12 cycles and reverted
nurse/forager for 15 cycles using the TruSeq PCR conditions.
RNA sequencing. RNA samples for the six pools of six brains of reverted nurse
and foragers were derived from the same lysate that was used to create the
reverted nurse and forager WGBS libraries. RNAseq libraries were created using
the Illumina TruSeq RNA sample prep kit with no modifications to the standard
protocol. This kit enriches for mRNA by using beads bound with a poly-T oligo
to bind to the poly-A tails of mRNA.
Data analysis of queen and worker WGBS. We ran the Bsmooth27 bisulfite
alignment pipeline (version 0.4.5-beta) on the 100 × 100 nucleotide paired-end
HiSeq 2000 sequencing reads obtained for each queen and worker pool. We used
Bsmooth’s Bowtie 2–based alignment pipeline, which employs a version of the
unbiased and efficient in silico bisulfite conversion approach introduced previ-
ously28. We used Bowtie 2 version 2.0.0-beta5 (ref. 29). We aligned to a reference
index consisting of the Baylor Human Genome Sequencing Center A. mellifera
assembly version 4.0, the A. mellifera mitochondrial sequence and the lambda
phage genome. Supplementary Table 6 summarizes alignment results.
We then used Bsmooth to extract read-level measurements. One read-level
measurement corresponds to one instance in which an aligned read overlapped
a CpG in the reference genome. The measurement records the genomic posi-
tion of the CpG, the allele observed in the read, its base quality, the alignment’s
mapping quality and other related measures.
Using Bsmooth, we filtered read-level measurements in three ways. First, we
removed read-level measurements from alignments with mapping quality less
than 10. Second, we remove read-level measurements in which the allele in the
alignment was neither C nor T. Third, we removed read-level measurements

© 2012 Nature America, Inc. All rights reserved.
nature neuroscience
doi:10.1038/nn.3218
from sequencing cycles that we deemed unreliable after visually inspecting the
M-bias plot. That is, we plotted the fraction of methylated read-level measure-
ments versus sequencing cycle. Ideally, this plot should be flat and horizontal,
indicating no strong relationship between sequencing cycle and fraction of
methylated read-level measurements. In practice, we found peaks and troughs
at the extremes of both mates. We filtered out measurements from the affected
cycles. In the case of this data set, we filtered out read-level measurements from
the 5′-most eight nucleotides of mate 1, the 3′-most four nucleotides of mate 1
and the 5′-most eight nucleotides of mate 2.
After filtering, we used Bsmooth to sort read-level measurements by genome
coordinate and compile them into a table summarizing methylation measure-
ments at each CpG in the reference genome. We used the summarized methyl-
ation measurements over the lambda genome, which we assume is entirely
unmethylated, to estimate the bisulfite conversion rate. Supplementary Table 7
summarizes the read-level measurements obtained, how they were filtered and
the estimated bisulfite conversion rates. After filtering we only included evidence
with a quality score greater than or equal to 20 for a particular CpG.
We then used Bsmooth to smooth the data and determine the correlation in
methylation between the WGBS data and CHARM data. Given that ~85% of
CpGs that have greater than 25% methylation are located in genes, we compared
the methylation levels between WGBS and CHARM by segmenting genes into
1,000-bp windows and found the average smoothed methylation value in each
window. Each window was required to contain at least four CHARM probes and
eight CpGs. This analysis was performed for each queen and worker sample,
and correlation values range from 0.691 to 0.807, mean = 0.755 (Supplementary
Fig. 1a,b). Methylation profiles were compared to determine the reproducibility
of detecting regions of methylation (Supplementary Fig. 1c–f).
To assess whether there was any difference between queens and workers, we
carried out the following analysis. First, we only analyzed CpGs with a cover-
age across the ten samples of greater than 10. For each of these CpGs, we per-
formed a t test for difference in methylation mean between the two groups. The
t test allows us to measure biological variability. The P values were corrected for
multiple testing using the Benjamini-Horchberg procedure and no CpGs were
significant at a 0.05 false discovery rate.
Data analysis of forager and reverted nurse WGBS. The same methods were
used to analyze the forager and reverted nurse sequencing data as were used
to analyze the queen and worker sequencing data (see above). Supplementary
Table 8 summarizes alignment results and Supplementary Table 9 summarizes
the read-level measurements obtained, how they were filtered and the estimated
bisulfite conversion rates.
To determine whether the differences in methylation between foragers
and reverted nurses that were discovered using CHARM also exist in WGBS
data set, we performed t tests on individual CpGs within 500 bp of CHARM
reversion DMRs. The average difference of the top three CpGs ranked by
significance in the DMR was calculated and compared with the average
CHARM methylation in the DMR. These results are presented as a scatter
plot in Supplementar y Figure 6.
To asses the significance of the overlap in direction of change between
CHARM and WGBS, we used the following test. Assuming that there is no cor-
relation between CHARM and WGBS there should be a 50% chance that the
direction of change is the same. Thus, we calculate the probability of observing
a more extreme statistic by P(X ≥ 45) + P(X ≤ 11) with X being binomially dis-
tributed with 57 trials and 0.5 chance of success.
Data analysis of forager and reverted nurse RNA sequencing. We used
TopHat30 v1.3.3 to align the 100 × 100 nucleotide paired-end HiSeq 2000
sequencing reads obtained for each nurse and reverted forager pool. We aligned
to a reference index consisting of the Baylor Human Genome Sequencing Center
A. mellifera assembly version 4.0 and the A. mellifera mitochondrial sequence.
Supplementary Table 10 summarizes alignment results.
TopHat alignments were overlapped with annotated exons obtained from
NCBI to form a table of overlap counts per gene and per sample. An alignment
was said to overlap an exon if there was any reference position covered both by
the exon and by the alignment. A spliced alignment that spanned an exon with-
out either mate overlapping the exon did not count as overlapping the exon.
Junctions and junction counts emitted by TopHat were combined to form a
table of counts per junction and per sample. Two junctions that were not identi-
cal, but where their boundaries differed by no more than five nucleotides on
one or both sides, were considered to be identical. This was necessary because
there is some variability in where exactly TopHat will place junction boundaries.
Junction counts were used to determine differentially expressed exon junctions
between foragers and reverted nurses. We define alternative splice events as the
presence of at least two distinct exon junctions that have opposite expression
in the same gene or DMR. To determine the frequency of alternative splicing
events in DMRs, we expanded the DMR 500 bp on each side and checked for
the presence of exon junctions that were more expressed in foragers colocalizing
with exon junctions that were more expressed in reverted nurses. Examples are
presented in Figure 2b and Supplementary Figure 7.
21. Amdam, G.V. et al. Exp. Gerontol. 40, 939–947 (2005).
22. Page, R.E. & Fondrk, M.K. Behav. Ecol. Sociobiol. 36, 135–144 (1995).
23. Laidlaw, H.H. & Page, R.E. Queen Rearing and Bee Breeding (Wicwas Press,
1997).
24. Aryee, M.J. et al. Biostatistics 12, 197–210 (2011).
25. Irizarry, R.A. et al. Genome Res. 18, 780–790 (2008).
26. Huang, D.W., Sherman, B.T. & Lempicki, R.A. Nat. Protoc. 4, 44–57 (2009).
27. Hansen, K.D. et al. Nat. Genet. 43, 768–775 (2011).
28. Lister, R. et al. Nature 462, 315–322 (2009).
29. Langmead, B. & Salzberg, S.L. Nat. Methods 9, 357–359 (2012).
30. Trapnell, C., Pachter, L. & Salzberg, S.L. Bioinformatics 25, 1105–1111 (2009).