RESEARCH ARTICLE SUMMARY
Global Epigenomic Reconfiguration
During Mammalian Brain
Development
READ THE FULL ARTICLE ONLINE
http://dx.doi.org/10.1126/science.1237905
Cite this article as R. Lister et al.,
Science 341, 1237905 (2013).
DOI: 10.1126/science.1237905
Ryan Lister,* Eran A. Mukamel, Joseph R. Nery, Mark Urich, Clare A. Puddifoot, Nicholas D.
Johnson, Jacinta Lucero, Yun Huang, Andrew J. Dwork, Matthew D. Schultz, Miao Yu, Julian
Tonti-Filippini, Holger Heyn, Shijun Hu, Joseph C. Wu, Anjana Rao, Manel Esteller, Chuan He,
Fatemeh G. Haghighi, Terrence J. Sejnowski, M. Margarita Behrens,* Joseph R. Ecker*
FIGURES IN THE FULL ARTICLE
Fig. 1. Methylcytosine in mammalian
frontal cortex is developmentally dynamic
and abundant in CG and CH contexts.
Introduction: Several lines of evidence point to a key role for dynamic epigenetic changes during
brain development, maturation, and learning. DNA methylation (mC) is a stable covalent modification that persists in post-mitotic cells throughout their lifetime, defining their cellular identity.
However, the methylation status at each of the ~1 billion cytosines in the genome is potentially an
information-rich and flexible substrate for epigenetic modification that can be altered by cellular
activity. Indeed, changes in DNA methylation have been implicated in learning and memory, as well
as in age-related cognitive decline. However, little is known about the cell type–specific patterning
of DNA methylation and its dynamics during mammalian brain development.
Methods: We performed genome-wide single-base resolution profiling of the composition, patterning, cell specificity, and dynamics of DNA methylation in the frontal cortex of humans and
mice throughout their lifespan (MethylC-Seq). Furthermore, we generated base-resolution maps
of 5-hydroxymethylcytosine (hmC) in mammalian brains by TAB-Seq at key developmental stages,
accompanied by RNA-Seq transcriptional profiling.
Results: Extensive methylome reconfiguration occurs during development from fetal to young
adult. In this period, coincident with synaptogenesis, highly conserved non-CG methylation (mCH)
accumulates in neurons, but not glia, to become the dominant form of methylation in the human
neuronal genome. We uncovered surprisingly complex features of brain cell DNA methylation at
multiple scales, first by identifying intragenic methylation patterns in neurons and glia that distinguish genes with cell type–specific activity. Second, we report a novel mCH signature that identifies
genes escaping X-chromosome inactivation in neurons. Third, we find >100,000 developmentally
dynamic and cell type–specific differentially CG-methylated regions that are enriched at putative
regulatory regions of the genome. Finally, whole-genome detection of 5-hydroxymethylcytosine
(hmC) at single-base resolution revealed that this mark is present in fetal brain cells at locations
that lose CG methylation and become activated during development. CG-demethylation at these
hmC-poised loci depends on Tet2 activity.
Fig. 3. mCH is enriched in genes that escape
X inactivation.
Fig. 4. Cell type–specific and developmental
differences in mC between mouse neurons
and glia.
Fig. 5. hmCG is enriched within active
genomic regions in fetal and adult mouse
brain.
Fig. 6. Developmental and cell type–specific
differential mCG.
SUPPLEMENTARY MATERIALS
Materials and Methods
Figs. S1 to S12
Tables S1 to S5
References (63–78)
Human
Mouse
eye-opening
GC
CG
CG
% methylated
CG sites
...
Discussion: Whole-genome single-base resolution methylcytosine and hydroxymethylcytosine
maps revealed profound changes during frontal cortex development in humans and mice. These
results extend our knowledge of the unique role of DNA methylation in brain development and
function, and offer a new framework for testing the role
of the epigenome in healthy function and in pathologi100 birth
cal disruptions of neural circuits. Overall, brain cell DNA
methylation has unique features that are precisely conTA
served, yet dynamic and cell-type specific.
Fig. 2. mCH is positionally conserved and
is the dominant form of DNA methylation
in human neurons.
0
TA
GC
TA
CH
{CA,CC,CT}
mCH
38%
mCG
~50 year old
neurons
>6 week old
neurons
1.5
1
0.5
0
0
...
The DNA methylation landscape of human and mouse neurons is dynamically reconfigured through development.
Base-resolution analysis allowed identification of methylation in
the CG and CH context (H = A, C, or T). Unlike other differentiated
cell types, neurons accumulate substantial mCH during the early
years of life, coinciding with the period of synaptogenesis and
brain maturation.
% methylated
CH sites
AT
mCG mCH
53%
50
10
20
55
Age (years)
65 0
10
75
Age (weeks)
85
The list of author affiliations is available in the full article online.
*Corresponding author. E-mail: ryan.lister@uwa.edu.au (R.L.); mbehrens@salk.edu (M.M.B.); ecker@salk.edu (J.R.E.)
www.sciencemag.org SCIENCE VOL 341 9 AUGUST 2013
Published by AAAS
629
RESEARCH ARTICLE
Global Epigenomic Reconfiguration
During Mammalian Brain Development
Ryan Lister,1,2*† Eran A. Mukamel,3* Joseph R. Nery,1 Mark Urich,1 Clare A. Puddifoot,3
Nicholas D. Johnson,3 Jacinta Lucero,3 Yun Huang,4 Andrew J. Dwork,5,6 Matthew D. Schultz,1,7
Miao Yu,8 Julian Tonti-Filippini,2 Holger Heyn,9 Shijun Hu,10 Joseph C. Wu,10 Anjana Rao,4
Manel Esteller,9,11 Chuan He,8 Fatemeh G. Haghighi,5 Terrence J. Sejnowski,3,12,13
M. Margarita Behrens,3† Joseph R. Ecker1,13†
DNA methylation is implicated in mammalian brain development and plasticity underlying
learning and memory. We report the genome-wide composition, patterning, cell specificity, and
dynamics of DNA methylation at single-base resolution in human and mouse frontal cortex
throughout their lifespan. Widespread methylome reconfiguration occurs during fetal to young
adult development, coincident with synaptogenesis. During this period, highly conserved
non-CG methylation (mCH) accumulates in neurons, but not glia, to become the dominant
form of methylation in the human neuronal genome. Moreover, we found an mCH signature
that identifies genes escaping X-chromosome inactivation. Last, whole-genome single-base
resolution 5-hydroxymethylcytosine (hmC) maps revealed that hmC marks fetal brain cell genomes
at putative regulatory regions that are CG-demethylated and activated in the adult brain and
that CG demethylation at these hmC-poised loci depends on Tet2 activity.
ynamic epigenetic changes have been observed during brain development, maturation, and learning (1–6). DNA methylation
(mC) is a stable covalent modification that persists in postmitotic cells throughout their lifetime,
defining their cellular identity. However, the methylation status at each of the ~1 billion cytosines in
the genome is potentially an information-rich and
flexible substrate for epigenetic modification that
can be altered by cellular activity (7, 8). Changes
in DNA methylation were implicated in learning and memory (9, 10), as well as in age-related
D
1
Genomic Analysis Laboratory, The Salk Institute for Biological
Studies, La Jolla, CA 92037, USA. 2Plant Energy Biology [Australian Research Council Center of Excellence (CoE)] and Computational Systems Biology (Western Australia CoE), School of
Chemistry and Biochemistry, The University of Western
Australia, Perth, WA 6009, Australia. 3Computational Neurobiology Laboratory, The Salk Institute for Biological Studies,
La Jolla, CA 92037, USA. 4La Jolla Institute for Allergy and
Immunology and Sanford Consortium for Regenerative Medicine, La Jolla, CA 92037, USA. 5Department of Psychiatry,
Columbia University and The New York State Psychiatric Institute, New York, NY 10032, USA. 6Department of Pathology
and Cell Biology, Columbia University, New York, NY 10032,
USA. 7Bioinformatics Program, University of California at San
Diego, La Jolla, CA 92093, USA. 8Department of Chemistry and
Institute for Biophysical Dynamics, The University of Chicago,
Chicago, IL 60637, USA. 9Cancer Epigenetics Group, Cancer
Epigenetics and Biology Program (PEBC), Bellvitge Biomedical
Research Institute (IDIBELL), L’Hospitalet de Llobregat, Barcelona 08907, Spain. 10Department of Medicine, Division of
Cardiology, Stanford University School of Medicine, Stanford,
CA 94305, USA. 11InstitucióCatalana de Recerca i Estudis Avançats
(ICREA), Barcelona, Catalonia, Spain. 12Division of Biological
Sciences, University of California at San Diego, La Jolla, CA
92037, USA. 13Howard Hughes Medical Institute, The Salk
Institute for Biological Studies, La Jolla, CA 92037, USA.
*These authors contributed equally to this work.
†Corresponding author. E-mail: ryan.lister@uwa.edu.au (R.L.);
mbehrens@salk.edu (M.M.B.); ecker@salk.edu (J.R.E.)
cognitive decline (11). Mice with a postnatal deletion of DNA methyltransferases Dnmt1 and
Dnmt3a in forebrain excitatory neurons, or with a
global deletion of methyl-CpG-binding protein
2 (MeCP2), show abnormal long-term neural plasticity and cognitive deficits (2, 12).
DNA methylation composition and dynamics
in the mammalian brain are highly distinct. A modification of mC catalyzed by the Tet family of mC
hydroxylase proteins, 5-hydroxymethylcytosine
(hmC), accumulates in the adult brain (13–15)
along with its more highly oxidized derivatives
5-formylcytosine and 5-carboxylcytosine. These
modifications of mC were implicated as intermediates in an active DNA demethylation pathway (16–19). In addition, methylation in the
non-CG context (mCH, where H = A, C, or T) is
also present in the adult mouse and human brains
(20, 21) but is rare or absent in other differentiated cell types (22, 23). Little is known about
cell type–specific patterning of DNA methylation
and its dynamics during mammalian brain development. Here, we provide integrated empirical
data and analysis of DNA methylation at singlebase resolution, across entire genomes, with celltype and developmental specificity. These results
extend our knowledge of the unique role of DNA
methylation in brain development and function
and offer a new framework for testing the role of
the epigenome in healthy function and in pathological disruptions of neural circuits.
Accumulation of Non-CG DNA Methylation
During Brain Development
To identify the composition and dynamics of transcription and methylation during mammalian
brain development, we performed transcriptome
profiling (mRNA-Seq) and whole-genome bi-
www.sciencemag.org
SCIENCE
VOL 341
sulfite sequencing [MethylC-Seq (24)] to comprehensively identify sites of cytosine DNA
methylation (mC and hmC) and mRNA abundance at single-base resolution throughout the
genomes of mouse and human frontal cortex
(table S1). DNA methylation in embryonic stem
(ES) cells occurs in both the CG (mCG) and
non-CG (mCH) contexts, but mCH is largely
lost upon cell differentiation (22, 23, 25, 26). We
found that although mCH levels are negligible
in fetal cortex, abundant mCH occurs in adult
frontal cortex (Fig. 1A). mCH has previously been
identified throughout the genome of the adult
mouse brain (20) and at several hundred genomic
positions in the human adult brain (21). Supporting previous studies, we found that mammalian brain mCH is typically depleted in expressed
genes, with genic mCH level inversely proportional to the abundance of the associated transcript (Fig. 1, A and B) (20). This pattern is the
opposite of that observed in ES cells (22) and
suggests that genic mCH in the brain may inhibit
transcription. The absence of mCH in fetal brain
suggests that this signature for gene repression
is added to the genome at a later developmental stage.
We performed MethylC-Seq on mouse and
human frontal cortex during early postnatal, juvenile, adolescent, and adult stages (Fig. 1C). CH
methylation level, defined as the fraction of all
base calls at CH genome reference positions that
were methylated (denoted mCH/CH), accumulates in mouse and human brain during early
postnatal development to a maximum of 1.3 to
1.5% genome-wide at the end of adolescence
before diminishing slightly during aging. mCH
increases most rapidly during the primary phase
of synaptogenesis in the developing postnatal
brain, from 2 to 4 weeks in mouse (27) and in the
first 2 years in humans (28), followed by slower
accumulation of mCH during later adolescence.
mCH accumulation initially parallels the increase
in synapse density within human middle frontal
gyrus (synaptogenesis lasts from birth to 5 years),
but it subsequently continues to increase during
the period of adolescent synaptic pruning, which
in humans occurs between 5 and 16 years of age
(Fig. 1C). Notably, the accumulation of mCH in
mice from 1 to 4 weeks after birth coincides with
a transient increase in abundance of the de novo
methyltransferase Dnmt3a mRNA and protein
(Fig. 1D). Analysis of the context of mCH sites
showed that it is mainly present in the CA context
(fig. S1, A to F), as previously reported for mCH
(20, 22, 23, 26).
Overall, genomes in the frontal cortex are
highly methylated. Whereas CG partially methylated domains (PMDs) account for about a third of
the genome of various differentiated human cells
(22, 25), human brain genomes have negligible
CG PMDs, resembling pluripotent cell methylomes
(25) (fig. S1, G and H). Given the high spatial
concordance of CG PMDs and nuclear laminaassociated domains reported previously (29), the
9 AUGUST 2013
1237905-1
ADRA2B
ASTL
DUSP2
C
20 kb
1.6
100%
0
100%
100%
0
100%
60
5 yr
mCH/CH
mCG/CG
synapses†
0.8
64 yr
12 yr
2 yr
40
2 wk
0.4
80
55 yr
22 mo
4 wk
1.2
25 yr
16 yr
6 wk
10 wk
25 yr
mRNA
H1
fetal
mCG 25 yr
H1
fetal
mCH 25 yr
H1
20
35 do
Fetal
mouse
mRNA
hmC
100%
D
100%
0
100%
100%
0
100%
mCH/CH
Human H1 ESC
0.01
0
1
15,000
0.55
Fetal
0
20,000
1
mCH hmCH
F
0
15,000
1
0.91 0.87
0
0.017
(hmCH)
1.3
4
Fetal
35 d
2 yr
5 yr
12 yr
16 yr
25 yr
64 yr
40
60
80
Fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
94.1
6
human
1.0
0.5
0
0.02
0.01
0
0
96.3
2.9
2
0.5
0
20
CH
0.052 (mCH) 0 (hmCH)
2.9
Adult (6 wk)
1.0
0.01
hmCG mCG
0.20
20
1.5
Age (days post-conception)
mRNA rank
(least abundant
most abundant)
CG
Dnmt1
Dnmt3a
Dnmt3l
Dnmt3b
Dnmt3a protein
40
Max
mCH/CH mCG/CG
0.01
Density (a.u.)
0.02
mCH/CH mCG/CG
mCH/CH
Human 25 yr cortex
mCH/CH
Mouse 10 wk cortex
0.02
0
0.02
E
00
0
mouse chr2: 126,895,000 - 127,217,000
B
,0
Age (days post conception)
100%
mRNA
mCH
fetal
10 wk
fetal
10 wk
fetal
10 wk
fetal
6 wk
0
Protein (norm. to actin)
mCG
20 kb
0
birth
20
C
100 680
00
Ncaph
Itpripl1
0
50
birth
human
Fetal
1 wk
0
Snrnp200
,0
Tmem127
Stard7
Ciao1
10
Astl
Dusp2
mRNA-Seq FPKM
Genes
W
Adra2b
20 CEN 40 Chr 12 (Mb) 80
100
120
mouse
1.0
0.5
0
0.02
0
0 CEN
20
40 Chr 12 (Mb)
80
100
120
100
G
% of cytosine basecalls (mouse)
Immunoglobulin
VH locus
Genes
mRNA
VOL 341
SCIENCE
115
116
117
118
H
119
Density
(a.u.), z
9 AUGUST 2013
114
0.5
0.4
atin y
,
om
Chr sibility
es
acc
Fig. 1. Methylcytosine in mammalian frontal cortex is developmentally
ChIP input
dynamic and abundant in CG and CH contexts. (A) Browser representation
DNaseI HS
of mC and mRNA transcript abundance in human and mouse frontal cortex and
1
mCG/CG
human ES cells. Chr2, chromosome 2. (B) mCH/CH within gene bodies exhibits
0
0.03
mCH/CH 0
opposite correlation with gene expression in ES cells (ESC) and brain. Contours
5hmC 4
show data point density, and red line shows smoothed mCH/CH as a function (CMS-IP/input)
0
of mRNA. a.u., arbitrary units. (C) Synaptic density (for mouse, per 100 mm2; for
Chr 12 (Mb) 113
human, per 100 mm3) and mC level in CG and CH contexts through development
in mouse and human frontal cortex. †Synaptic density quantitation from De Felipe et al. (27) and
Huttenlocher and Dabholkar (28). (D) DNA methyltransferase mRNA and protein abundance (mean T SEM)
in mouse frontal cortex through development. FPKM, fragments per kilobase of exon per million fragments
mapped. (E) Fraction of cytosine base calls with each modification in fetal and adult mouse frontal cortex.
(F) Cortex mC level in CG and CH contexts throughout mouse and human chromosome 12 in 100-kb
bins smoothed with ~1-Mb resolution. CEN, centrosome. (G) Transcript abundance, chromatin accessibility [8-week mouse cortex ChIP input and DNaseI hypersensitivity (HS) normalized read density], and
mC levels in 5-kb bins at the mouse immunoglobulin VH locus. (H) Density (z) plot of 10-week mouse
frontal cortex mCH level (x) versus 8-week mouse cortex ChIP-input normalized read density (y) for all
10-kb bins of the mouse genome.
1237905-2
% mCG/CG
C
synaptic density†
Genes
mCH
NCAPH
ITPRIPL1
CIAO1
STARD7
TMEM127
SNRNP200
LOC285033
Ag
ed
W
Ju
ve
Ad nile
ol
es
Ad ce
ul nt
t
mCG
human chr2: 96,109,000 - 96,442,000
% mCH/CH
A
C
hi
ld
Ad hoo
ol d
es
ce
nt
Ad
ul
t
RESEARCH ARTICLE
0.3
0.02
0.2
0.1
www.sciencemag.org
0.01
0
0
/CH
H
mC
(Mm
0
1
FC
,
wk)
x
RESEARCH ARTICLE
paucity of CG PMDs in these brain methylomes
could indicate that lamina-associated domains
are altered or much less frequent in the brain.
The adult mammalian brain contains the highest levels of hmC that have been observed (15),
accounting for about 40% of methylated CG sites
in cerebellar Purkinje cells (30). hmC accumulates during early postnatal brain development in
mice (31, 32), becoming enriched in highly expressed genes (33). Given the evidence that hmC
can be an intermediate in an active DNA demethylation pathway (16, 17), high-resolution analysis of the genomic distribution of hmC is needed
to understand its role in the control of DNA methylation dynamics through brain development.
Standard bisulfite-sequencing data does not distin-
chr5: 87,917,000 - 88,317,000
mCG
MEF2C
B
mCH
20 kb
0.29
0.43
2.09
3.43
0.38
2.12
0.48
2.13
3.44
1.22
3.43
0.58
4.11
4.84
0.39
4.12
4.29
1.44
3.29
4.03
glia
3.22
3.24
3.56
4.20
(10 wk)
Tissue
mouse
50 bp
NeuN+ R2
50 bp chr3: 36,078,990 - 36,079,557
1
0
CH sites
R1
R2
R3
1
ChrX
CG
CH
0.6
0.4
0.2
Fig. 2. mCH is positionally conserved and is the dominant form of DNA
methylation in human neurons. (A) Browser representation of mCG and
mCH in NeuN+ and NeuN– cells. Human NeuN+/NeuN– samples: R1, 53-yearold female; R2, 55-year-old male. Mouse NeuN+/NeuN– samples: R1, 7-week
males; R2, 6-week females; R3, 12-month females (not shown). (B) Percentage
of methylated base calls in each sequence context throughout the genome. (C)
Box and whisker plot of mCG and mCH level in neurons and glia at genomic
www.sciencemag.org
SCIENCE
vs
Ne
uN
+
vs
ES
Hu
ES
6
vs
ES
H1
Ne
uN
+
vs
Ne
uN
+
Ne
uN
+
glia
0.3
1
Pearson correlation
Ne
uN
+
0
vs
0
1
CG
CH
0.8
vs
2
NeuN+ R2
Autosomes
1
Ne
uN
+
mCH
4
R2
R3
R1
R2
R1
R3
NeuN+ NeuN–
NeuN+ NeuN–
1-r
0.7 0 R2 R3 R1 R2 R1 R3
Per−site correlation
of mC/C (normalized)
glia
0
F
1
Ne
uN
+
NeuN+
mCH
NeuN+ R1
chr2: 171,489,889 - 171,490,454
ES
R2
human
H1
R1
Dnmt3a binding sites
Random
NeuN–
NeuN–
NeuN+ R3
R2
mCH
40
R1 R2 R3 R1 R2 R3
(25 yr)
Tissue
R1
60
NeuN–
NeuN+
CH sites
NeuN+ R1
mCH R2
NeuN+ NeuN–
80
% mCG
5.12
mCH mCG
E
NeuN+ NeuN–
1-r
0.7 0 R1 R2 R1 R2
R1 R2 R3 R1 R2 R3
R1 R2 R1 R2
NeuN+ NeuN–
chr13: 83,518,000 - 83,911,000
D
NeuN+
R1 R2 R3 R1 R2 R3
0
100
% mCH
4
2
C
NeuN+
6
4.66
20 kb
Mef2c
mouse
0
mCH
8
Genes
NeuN+ R1
R2
NeuN– R1
R2
NeuN+ R1
R2
NeuN– R1
R2
Dnmt3a ChIP-chip
20
mCG
10
W
C
Ne
uN
+
mCH mCG
human
Genes
NeuN+ R1
R2
NeuN– R1
R2
NeuN+ R1
R2
NeuN– R1
R2
% mC/C
A
for false detection, we estimated that 0.017% of
cytosine base calls were hmCH genome-wide (99%
confidence interval: 0 to 0.059%), and significant
hmCH was detected at few individual sites (fig.
S2, A and B). The overwhelming presence of
hmC in the CG context (99.98%) in mouse adult
and fetal frontal cortex is consistent with recent
findings in human ES cells, where 99.89% of hmC
is in the CG context (34). hmC was present at
many highly methylated CG sites (fig. S2C). Therefore, although only a small fraction of all cytosines
throughout the genome are methylated (mCG =
2.9%, mCH = 1.3%, hmC = 0.87%), mCH and
hmC constitute major, and nonoverlapping, components of the methylated fraction of the genome in adult frontal cortex (mCG = 57.2%,
guish between methylated and hydroxymethylated
sites, so methylcytosines identified by MethylC-Seq
analysis represent the sum of these two contributions. Therefore, we used Tet-assisted bisulfite
sequencing [TAB-Seq (34)], a base-resolution
technique that distinguishes hmC from C and mC
genome-wide, to profile hmC in mouse fetal and
adult frontal cortex (Fig. 1A). Integration of the
genome-wide profiles of mC and hmC enabled a
detailed breakdown of the methylated subset of
the genome at these distinct developmental stages
(Fig. 1E). hmC constitutes 0.20% of total cytosine
base calls in fetal cortex and increases to 0.87% in
adult cortex. This modification appears to be restricted to the CG context, as also observed in
human and mouse ES cells (34); after correction
regions bound by Dnmt3a versus a random set. Whiskers indicate 1.5 times the
interquartile range. (D) mCH correlation between NeuN+ and NeuN– cells in
mouse and human, measured in 10-kb bins. (E) Browser representation of mCH
sites in neurons. Scatter plots (right) show consistent mCH/CH at all single sites
in a 20-kb window overlapping the example region (left). (F) Correlation
analysis of methylation state at single sites between neurons and ES cells in
human and mouse. Correlation values are normalized by a simulation (62).
VOL 341
9 AUGUST 2013
1237905-3
RESEARCH ARTICLE
mCH = 25.6%, hmC = 17.2%). These data suggest that the steady-state population of hmC in
the adult brain is not an intermediate stage in the
demethylation of mCH. However, these steadystate measurements do not preclude the possibility that hmCH could be rapidly turned over after
conversion from mCH, leading to negligible detected hmCH despite Tet-mediated demethylation
at CH sites.
Protection of Inaccessible Genomic Regions
from de Novo Methylation
mCH accumulates in parallel across most of the
genome (Fig. 1F). However, we found numerous
(36 in human, 34 in mouse) noncentromeric,
megabase-sized regions that do not accumulate
mCH. These regions, which we termed mCH
deserts, are enriched for large gene clusters that
encode proteins involved in immunity and recep-
A
tors required for sensory neuron function (table
S2). One mCH desert spans the immunoglobulin
VH locus, which encodes variable domains of the
immunoglobulin heavy chain that rearrange in B
lymphocytes. The VH locus is transcriptionally
quiescent in the frontal cortex of 10-week-old mice,
and the chromatin state is highly inaccessible, as
inferred from deoxyribonuclease I (DNaseI) hypersensitivity profiling (35) and chromatin immunoprecipitation (ChIP) input sequence read density
data (36, 37) (Fig. 1G). In contrast, mCG is not
depleted in mCH deserts.
Genome-wide detection of hmC by cytosine
5-methylenesulphonate immunoprecipitation
(CMS-IP) (38, 39) revealed that hmC is also
strongly depleted in the VH locus. mCH deserts
are observed at other loci in the genome, including olfactory receptor gene clusters that form
heterochromatic aggregates required for mono-
allelic receptor expression in olfactory sensory
neurons (40, 41). Genome-wide comparison of
mCH/CH with chromatin accessibility, as inferred from ChIP input read density (36, 37),
for all 10-kb windows of the mouse genome
revealed two discrete groups of genomic regions
(Fig. 1H). Low-accessibility regions tend to contain minimal mCH, whereas more-accessible regions
of the genome show a proportional relationship
between genome accessibility and mCH levels.
Thus, although mCG is unaffected in these regions, lower chromatin accessibility appears to
be highly inhibitory to deposition of mCH and
hmC, potentially via inaccessibility to de novo
methyltransferases and Tet mC hydroxylases. Furthermore, this indicates that accumulation of mCH
and hmC during mammalian brain development
occurs via processes that are at least partly independent from methylation at CG dinucleotides.
human
chrX: 46,840,000 - 47,047,000
Genes
NDUFB11
UBA1
10 kb
USP11
CDK16
RBM10
25 yr
R1
NeuN+ mCH
R2
R1
NeuN – mCH
R2
mRNA
mouse
Genes
chrX: 148,510,000 - 148,852,000
Kdm5c
Gm15266
Iqsec2
20 kb
Tspy12
Gpr173
mRNA 10 wk
R1
NeuN+ mCH R2
R3
R1
NeuN – mCH R2
R3
0.4
0
-0.4
D
1
ChrX−escapee discriminability
1
0.5
0
0.05
0
0 1-2 3-4 5-6 7-8 9
Inactivated
Escaped
Intragenic mCH/CH
Female NeuN+
Intragenic mCH/CH
Female − Male
0
0.1
0.1
0.05
0
0.05
Male NeuN+
Fig. 3. mCH is enriched in genes that escape X inactivation. (A) Browser
representation showing mCH-hypermethylated female human and mouse genes
that escape X inactivation (shaded genes). (B) Box and whisker plots of gender
differences in promoter mCG and intragenic mCH in inactivated and escapee
genes on human chrX. (C) Scatter plot of gender differences in mCG and mCH
9 AUGUST 2013
CG (AUC=0.78)
CH (0.75)
Both (0.88)
0
0
Bi−allelic expression [0−9]
1237905-4
0.5
Chr2 (1,472)
X-inact.* (253)
X-escapee* (37)
X-escapee1§ (33)
X-escapee† (7)
Correct detection rate
C
Promoter mCG/CG
Female NeuN+
Promoter mCG/CG
Female − Male
B
VOL 341
0.1
0
1
False detection rate
in human chrX genes. Reported X inactivated and escapee genes: *Carrel and
Willard (49); §Sharp et al. (50); †predicted escapee genes, and autosomal
(Chr2) genes are indicated. (D) Discriminability analysis of genes that escape
female X inactivation using mC data, showing correct versus false detection
rate mapped for all possible mC/C thresholds.
SCIENCE
www.sciencemag.org
RESEARCH ARTICLE
A
mCG/CG
10 wk
fetal
fetal
-100 kb
1
2 wk
more, our data show that in human neurons mCH
is the dominant form of methylation in the genome: It is more abundant than mCG and occurs
in 5% of CH and 10% of CA sites (Fig. 2B and
fig. S1, A, B, and H). Of the total methylated
fraction of adult human neuronal genomes,
mCH accounts for ~53%, whereas mCG constitutes ~47%.
Although sparse in glia, mCH enrichment occurs within genes that are CH-hypomethylated in
neurons, such as Mef2c (Fig. 2A), a transcriptional activator that plays critical roles in learning and memory, neuronal differentiation (42),
hmCG/CG
10 wk NeuN+ NeuN–
6 wk
+100 kb
B
mCH/CH
fetal
-100 kb
2 wk
10 wk NeuN+ NeuN–
+100 kb
5
1
2
4
3
Genes
6
7
8
n.s.
0.8
-100 kb
1.5
mCH/CH
Median
mRNA−Seq FPKM
+100 kb
1
10
1
0.1
10
1
0.1
3’
5’
www.sciencemag.org
gene
3’
SCIENCE
1
10
100
mCG/CG
Median
mRNA−Seq FPKM
mCH/CH
+100 kb
10
1
0.1
10
1
0.1
10
1
0.1
5’
gene
Age (wks postconception)
VOL 341
10
Enrichment (q<0.05)
Consitutively
Downlow
regulated
0.1
gene
0.1
2
Normalized mCH/CH
-100 kb
10
5’
0
Normalized hmCG/CG
Upregulated Neuronal
tissue
mCG/CG
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
R1
R2
R3
glia
R1
R2
R3
0.5
Normalized mCG/CG
mRNA−Seq FPKM
C
1.2
Astrocyte
0.1 1 10
Constitutively high
9
10
25,260
Ne
Ne
uN
uN
–
+
Fig. 4. Cell type–specific
and developmental differences in mC between
mouse neurons and glia.
(A) Heat-map representation of 25,260 mouse
genes organized in gene
sets identified by k-means
clustering using normalized genic mCG and mCH
levels in adult developmental and NeuN+ and
NeuN– samples. Left-hand
plot shows mRNA abundance. (B) Enrichment or
depletion of each cluster
for developmental and celltype specific gene sets.
n.s., not significant (FET,
FDR < 0.05). (C) mCG and
mCH throughout gene body
and flanking 100 kb for
indicated gene sets. Transcript abundance (mRNASeq FPKM) over mouse
development is shown for
the same gene sets. Color
scales are as in (A).
position and patterning of mCG and mCH in
neurons and glia (Fig. 2A). Whereas differential
mCG between neurons and glia was restricted to
localized regions, neurons were globally enriched
for mCH compared with glia. Indeed, we discovered that the level of mCH in glia is similar to
that of fetal and early postnatal cortical tissue,
whereas adult neurons have the greatest frequency of mCH that has been observed in mammalian
cells. This indicates that the rapid developmental
increase in mammalian brain mCH that coincides
with the period of synaptogenesis is primarily
due to mCH accumulation in neurons. Further-
Constitutively high
Neuronal
Upregulated
Astrocytic
Downregulated
Constitutively low
Cell Type–Specific DNA Methylation Patterns
in Neurons and Glia
The diversity of neuronal and glial cells in the
frontal cortex raises the question of which features of DNA methylation are found in specific
cell types. We isolated populations of nuclei by
fluorescence-activated cell sorting that were highly enriched for neurons (NeuN+) or glia (NeuN–)
from human and mouse adult frontal cortex tissue. An additional glial population was isolated
from mice expressing enhanced green fluorescent
protein (eGFP) under the S100b promoter.
MethylC-Seq revealed differences in the com-
3’
5’
gene
3’
1
10
100
Age (wks postconception)
9 AUGUST 2013
1237905-5
RESEARCH ARTICLE
B
0.2
0.1
0.05
0.15
0.1
0.05
2kb
age
nic
Ad
ult
DH
S
Fet
al D
HS
Ad
ult
enh
Fet
al e
nh
S±
hmCH
Intr
TE
Is
CG
S±
TS
es
Ch
rX
tos
om
es
Ch
rX
Au
om
tos
Au
hmCG
2kb
0
0
-100 kb
C
0.15
hmCG/CG
hmC/C (6 wk)
A
mCH Position Is Highly Conserved
We examined whether the position of DNA methylation is stochastic or precisely controlled at
different genomic scales. The level of mCH in
10-kb windows throughout the genome was highly reproducible between independent samples of
the same cell type, with lower, but substantial,
correlation between cell types (Fig. 2D). Closer
inspection revealed consistency between the methylation level at individual mCH sites in neurons
from different individuals in both mice and hu-
Fetal
mans (Fig. 2E). At single-base resolution (fig. S4),
perfect correlation between individuals would not
be observed even if the true methylation level
were identical at each site because of the stochastic effect of a finite number of sequenced reads.
To correct for this, we normalized the observed
correlation by that of simulated data sets with the
same coverage per site as each of our experimental samples but with identical methylation
levels (Fig. 2F and fig. S4). To assess statistical
significance, we used a permutation test, which
compared the data correlation with the correlation after randomly shuffling the relative positions of CH sites in each sample (fig. S4). This
revealed that autosomal CG and CH sites have
nearly identical methylation levels in neuronal
populations isolated from different individuals of
the same species. Observed differences could be
explained by stochastic sampling rather than true
individual variation. Unexpectedly, normalized persite correlation is higher for mCH than mCG
between neuronal populations isolated from the
frontal cortices of different human individuals,
and mouse neuronal mCG and mCH per-site
+100 kb
Fetal
6 wk
Constitutively
high
Fetal
6 wk
Neuronal
Fetal
6 wk
Upregulated
Fetal
6 wk
Astrocyte
Fetal
6 wk
Downregulated
Fetal
6 wk
Constitutively
low
6 wk
0.8
1.2
Normalized
hmCG/CG
D
mCG/CG
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
mRNA-Seq
Fetal
-100 kb
6 wk
0
0.1
mCH/CH
hmCG/CG
Fetal
6 wk
+100 kb
0.2
Intragenic
hmCG/CG
Fetal
-100 kb
6 wk
+100 kb
1
Genes
Fig. 5. hmCG is enriched within active
genomic regions in fetal and adult mouse
brain. (A) hmCG level
in 6-week mouse frontal
cortex for autosomes and
ChrX. (B) Median hmCG
level within genomic features (error bars 32nd
to 68th percentile). enh,
enhancer. (C) Median normalized hmCG throughout
gene body and flanking
100 kb for indicated gene
sets. Bars show absolute
hmCG/CGlevelswithingene
bodies for each class. (D)
mC and hmC throughout
gene body and flanking
100 kb for each. Transcript abundance (mRNASeq FPKM) during mouse
development is also shown
(left).
display mCH hypermethylation in both neuronal
and glial populations. Consistent with CH methylation requiring Dnmt3a, glial hyper-mCH genes
frequently intersect areas of the genome bound
by Dnmt3a in mouse postnatal neural stem cells
(46). Dnmt3a-binding regions are greatly enriched
for mCH, particularly in glia, whereas mCG is
not enriched in Dnmt3a-binding regions in glia
and is depleted in neurons (Fig. 2C). Thus, there
is an association with Dnmt3a binding sites specific to mCH and not mCG, suggesting partial
independence between these two marks.
flank normalized hmCG/CG
synaptic plasticity (43), and regulation of synapse
number and function (44). Genome-wide surveys
identified 174 mouse genes in which glia were
hypermethylated relative to neurons in the CH
context (table S3). Unbiased gene ontology analysis revealed that these glial hyper-mCH genes
are highly enriched for roles in neuronal and synaptic development and function (table S3). These
genes also overlapped significantly with a set of
461 genes expressed at higher levels in neurons
than in astrocytes (13-fold higher overlap than
chance, P < 10−30, Fisher exact test, FET) (45)
and 233 developmentally up-regulated genes
(7.5-fold, P < 10−7, FET). These genes show
hypomethylation of CG and CH in neurons and
hypermethylation of CH in glia (fig. S3A), consistent with a potential role of mCH in transcriptional repression of neuronal genes in the glial
genome. Furthermore, genes associated with
oligodendrocyte or epithelial function accumulate mCH through development (fig. S3B), with
oligodendrocyte up-regulated genes showing intragenic mCH hypermethylation in neurons and
hypomethylation in glia, whereas epithelial genes
16,077
0.8
0.1
1
10
mRNA−Seq FPKM
1237905-6
9 AUGUST 2013
1.2
flank normalized
mCG/CG
VOL 341
SCIENCE
0.8
1.2
flank normalized
hmCG/CG
www.sciencemag.org
0.5
1.5
flank normalized
mCH/CH
RESEARCH ARTICLE
correlations are equivalent. Per-site correlation between two human ES cell lines (H1 and HUES6)
is also high (>0.8) for both mCG and mCH.
The high interindividual correlation of mCH
at the kilobase and single-site scales indicates that
methylation of CH positions, particularly in mammalian neurons, is a highly controlled process. It
is not consistent with a stochastic event that takes
place at any available CH position in a particular genomic region that accumulates mCH. Comparison of mCH between human and mouse
neurons at conserved exonic CH positions revealed a low but significant interspecies correlation (Fig. 2F; P < 0.005, shuffle test), possibly
indicating conservation of the cellular processes
that precisely target or restrict mCH at these positions. Last, per-site mCG and mCH correlation
between human ES cells and neurons is significantly lower, likely because of differences in the
processes governing methylation of particular genomic features in the distinct cell types, for example, enrichment and depletion of mCH in highly
transcribed genes in ES cells and neurons, respectively (Fig. 1, A and B).
The precise conservation of mCH position
may be partly caused by the physical configuration of DNA within nucleosomes. Consistent
with this, neuronal mCH patterns contain robust
periodic components at the scale of nucleosome
spacing [~170 base pairs (bp), fig. S5A] and the
DNA helix coil length (~10.5 bp, fig. S5B). Such
periodic components may arise from sequencedependent constraints on mCH position, which
would be the same in every neuronal cell. Alternatively, epigenetic heterogeneity within the
population of NeuN+ nuclei in our sample may
lead to stronger correlation for CH sites located
on the same physical chromosome, compared with
the correlation between the same locations on chromosomes from different cells. To test this, we measured the cross-correlation within individual reads,
revealing a contribution of within-chromosome
correlation to the periodic methylation pattern
(fig. S5C).
Gender-Specific DNA Methylation Patterns
on the X Chromosome
Interindividual correlation of mCG and mCH on
chromosome X (ChrX) is frequently lower than
on autosomes (Fig. 2F), prompting a closer analysis of ChrX mC patterns. ChrX mCG and mCH
levels were generally lower in females compared
with males, presumably because of the effect of
ChrX inactivation (fig. S5, F and G) (47, 48).
However, a subset of genes in both humans and
mice have significantly greater intragenic mCH
levels in females compared with males (Fig. 3A).
Inspection of these genes revealed that most were
previously found to escape inactivation in human
females (X-escapees), displaying biallelic expression (49) and a reduction in promoter mCG hypermethylation, a DNA methylation signature of
inactivated alleles (50). Quantification of human
gender differences in neuronal DNA methylation
for ChrX genes previously characterized as show-
ing biallelic expression (49) revealed that females
have reduced promoter mCG and a large increase
in intragenic mCH but not intragenic mCG (Fig.
3, B and C, and fig. S5, D and E). The sequence
composition of mCH is very similar in the whole
genome, within autosomal gene bodies, and within X-chromosome inactivated and escapee gene
bodies (fig. S5H). Analysis of gender-specific
methylation in additional human cell types revealed that female promoter mCG hypomethylation is observed at X-escapee genes in glia and
human embryonic stem cells (fig. S6). Intragenic
mCH hypermethylation of X-escapees was also
observed in female glia, albeit to a lesser extent
than in neurons, but was not present in ES cells.
Thus, X-escapee mCH hypermethylation may be
a feature that is specific to neural cell types. Although both promoter CG hypomethylation and
intragenic CH hypermethylation provide significant information for discriminating X-escapees
[Fig. 3D, discriminability index (area under the
curve, AUC) = 0.75 and 0.78, respectively], combining both mCG and mCH measurements boosts
discriminability (AUC = 0.88). By using this
intragenic mCH hypermethylation signature, we
identified seven new putative X-escapee genes
(table S4). On the basis of these data, we hypothesize that intragenic CH hypermethylation in
neurons may play a compensatory role in genes
that fail to acquire repressive CG hypermethylation in the promoter region, restoring equal gene
expression between male and female cells (51).
Distinct Genic DNA Methylation States
Demarcate Functionally Relevant Gene Clusters
DNA methylation within promoter regions and
in gene bodies is implicated in regulation of gene
expression (22, 52), suggesting that the precisely
conserved, cell type–specific DNA methylation
patterns may be related to specific neuronal and
glial cellular processes. We therefore used an unbiased approach to classify patterns of mCG and
mCH within each annotated gene body and in
flanking regions extending 100 kb up- or downstream. After normalizing the methylation pattern
around each autosomal gene by the local baseline
mCG or mCH level in each adult neuronal or
glial sample, we combined these features into a
large data matrix containing 4200 individual DNA
methylation measurements for each gene [seven
samples, two contexts (CG and CH), 300 1-kb
bins within and around each gene]. Using principal component (PC) analysis, we extracted five
methylation features (PCs) that together account
for 46% of the total data set variance (fig. S7A).
Gene sets with specific neuronal or astrocytic
expression, as well as ChrX genes, segregate within
PC space (fig. S6B). We then used k-means clustering to classify all genes into 15 clusters on the
basis of their mCG and mCH patterns (Fig. 4A
and fig. S8). Several dominant patterns of DNA
methylation and transcript abundance and dynamics between developmental and cellular states are
evident. A cluster of genes that progressively loses
gene-body mCG and mCH through development
www.sciencemag.org
SCIENCE
VOL 341
contains constitutively highly expressed genes that
are strongly enriched for neuronal function and
depleted for astrocyte-specific roles (Fig. 4A, box
1). These genes show intragenic mCG enrichment in glia and depletion in neurons (box 2),
indicating that glial gene body mCG resembles
that of the neural precursor cells that predominate
the fetal brain. This indicates that the loss of
mCG in brain tissue during development is due
to CG hypomethylation in mature neurons. These
constitutively highly expressed genes enriched for
neuronal function also show extensive intragenic
mCH hypomethylation in neurons in contrast to
glia (box 3), and they are enriched for hmCG (box 4)
as previously described (32, 33). Genes that are
not as highly transcribed, but that are associated
with neuronal function and are developmentally
up-regulated, also show intragenic mCG and mCH
hypomethylation in neurons but not glia (box 5).
For these gene sets, mCG and mCH enrichment
or depletion is precisely localized to transcribed
regions, suggesting that this modification of genic
mC is tightly coupled to transcription. Notably,
the bodies of constitutively high genes that are
not enriched for neuronal function (box 6) do not
show marked fetal/glial mCG enrichment or neuronal mCH depletion, indicating that this differential methylation is specific for genes enriched
for neuronal function and not simply an association with particular levels of transcriptional activity. Genes associated with astrocyte function
show an opposite pattern to genes associated with
neuronal function: a progressive increase in intragenic mCG and mCH in frontal cortex tissue over
development, neuronal mCG and mCH hypermethylation, and glial mCG and mCH hypomethylation (Fig. 4A, boxes 7 to 9). Last, genes with
constitutively low expression do not show developmental or cell type–specific DNA methylation
patterns (box 10), demonstrating that dynamic DNA
methylation in genes is highly associated with
differential transcriptional activity in mammalian
brain development and neural cell specialization.
Each of the gene clusters identified in our
unbiased analysis was significantly enriched or
depleted for cell type–specific function [neuronal
or astrocytic genes (45)] or particular expression
patterns (constitutively high or low expression,
developmentally up- or down-regulated) (Fig. 4B).
Profiling the median mCG and mCH of genes
within each of these categories allows direct comparison of developmental and cell type–specific
DNA methylation in mouse (Fig. 4C) and human
(fig. S3C). This analysis recapitulates many of the
conclusions of the unbiased clustering (Fig. 4A).
The inverse relationship observed between
genic mCH level and transcriptional activity is
consistent with a model whereby intragenic accumulation of mCH impedes transcriptional activity.
Alternatively, the process of transcription could
interfere with mCH de novo methylation or induce active mCH demethylation, although these
are not consistent with the DNMT3A-dependent
intragenic mCH in human embryonic stem cells
that is positively correlated with gene expression
9 AUGUST 2013
1237905-7
mCG/CG
0
0
0
Fe
t
eu A al >
d
N
N + ult Ad
eu h > ul
y
N N− pe Fe t
e
r t
N uN hyp m al
eu + o C
N hy m G
− p C
hy o G
pe mC
r G
Al mC
lD G
M
R
s
N
fetal
H1 ES
fetal
adult
G
3x10-7
DNaseI HS
4
H3K4me1
2
1.5
2.5
W
C
3
H3K27ac
2
1.5
1
1.0
1.0
mCG / CG
0.5
0
0
0.01
0.02
mCH / CH
0.01
0
0.2
0
0.2
hmCG / CG
0.1
0
0
1.5
1.2
hmCG / CG
(norm)
-2
0
+2
Distance (kb)
1.0
0.8
-2
0
+2
Distance (kb)
111,738,000 - 111,743,200
chr9
20
CG-DMRs:
Adult > Fetal (10,870)
CG-DMRs:
Fetal > Adult (20,483)
15
19.7%
10
5
0
3.6%
0.13%
0.06%
d
sed
sed
sed
ase
ea
ea
ea
cre/ecr-/ncr -/ncr -/i
i
e
d
d
G 2
G t2
G 2
/CG et2
G/C Tet G/C Tet
G/C Te
mC in mC in
mC in mCG in T
1237905-8
9 AUGUST 2013
VOL 341
Fetal > Adult
I
Fraction of DMRs
% of CG−DMRs
(FDR=0.05)
H
W
C
mCG
=100%
1.0
hmC
Ne
u
N
–
0.1
hmC fetal > 6 wk
DMRs 6 wk > fetal
adult
hmCG = 17%
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
R1
R2
R3
R1
R2
R3
glia
fetal
6 wk
hmC
1x10-7
2.0
fetal
Adult > Fetal
Control: WT vs. WT
WT vs.Tet2 –/–
0.08
0.04
0
−0.2
0
0.2
−0.2
0
Difference in mCG/CG vs. WT
SCIENCE
www.sciencemag.org
0.2
61
CG-DMRs
Fetal > Adult (73)
adult
1x10-7
N+
Ne
u
CG-DMRs
Fetal > Adult (73)
NeuN+ hyper mCG
NeuN− hyper mCG
NeuN+ hypo mCG
NeuN− hypo mCG
11
2
0.1 10 10
24
Adult unique
enhancers (20)
mC
H3K27ac
31
Adult DHS (147)
7
0
47
3x10-7
fetal
adult
fetal
adult
fetal
adult
H3K4me1
tissue
mCH
28
10
CG-DMRs
Adult > Fetal
Fetal > Adult
F
DNaseI
8
57
Fetal unique
enhancers (19)
hmCG = 13%
0.2
2
CG-DMRs
Adult > Fetal (35)
19
CG-DMRs
10
Adult > Fetal 14
(36)
10
NeuN+ NeuN-
E
Fetal DHS (103)
hmCG = 0%
0.4
0.2
not sig.
(p>0.001)
60
mCG = 11%
0.4
30
mCG = 75%
0.6
0.25
hmCG = 31%
0.8
0.6
1
mCG = 92%
0.8
mCG
D
fetal
35 do
2 yr
5 yr
12 yr
16 yr
25 yr
64 yr
1
4
% of CG−DMRs
per region
tissue
1
16
0
NeuN−
42,059
hypo-mCG
R1
R2
R3
fetal
6 wk
NeuN- hmC
C
CGIs
Promoter (TSS ±2kb)
TES ±2kb
Intragenic
Adult DHS
Fetal DHS
Adult DHS (unique)
Fetal DHS (unique)
Adult enhancers
Fetal enhancers
NeuN+ NeuN−
48,745 38,649
CG-DMRs
NeuN−
10,787
NeuN+
58,445
NeuN+
B
1
NeuN+
138,346
hyper-mCG
0.25
NeuN−
16,139
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
R1
R2
R3
glia
tissue
Median mCG/CG
0
Fold enrichment/depletion
hmCG/CG
1
NeuN+
57,464
CG differentially methylated regions (CG-DMRs)
hypo-mCG
hyper-mCG
0
R1
R2
R1
R2
mCG/CG
A
mC
Fig. 6. Developmental
and cell type–specific
differential mCG. (A)
Heat map of absolute
mCG level in CG-DMRs
identified between neurons and glia and over
development in mouse
(left) and human (right).
(B) Fraction of all CGDMRs located in distinct
genomic features in mouse.
(C) Enrichment or depletion of distinct cell type–
specific anddevelopmental
CG-DMR sets within genomic features from (B).
(D and E) Intersection of
developmentally dynamic CG-DMRs and (D) DNaseI
hypersensitive sites or (E)
enhancers in mouse brain
in thousands. (F) Browser
representation of mouse
developmentally dynamic CG-DMRs and quantification of local enrichment
of chromatin modifications, genome accessibility, and mC. (G) TAB-Seq
reads showingfetal-specific
hmCG in the Fetal>Adult
CG-DMR in mouse. (H)
Proportion of mouse developmental CG-DMRs
where mCG/CG is significantly increased or
decreased in Tet2 knockout mice. (I) Distribution
of mCG level difference
between wild-type (WT)
and Tet2 mutant at mouse
CG-DMRs. Significantly
different DMRs are indicated by coloration.
n=
7
n= 2,7
3 06
n= 5,3
5 91
n= 7,4
16 64
n= ,1
5 3
n= 8,4 9
4
n= 10, 5
25 787
0,
93
2
RESEARCH ARTICLE
RESEARCH ARTICLE
(22, 23). Overall, glial mCG and mCH patterns
closely resemble those of the fetal and the early
postnatal brain, indicating that DNA methylation
in early mammalian brain developmental stages
may be a default state that largely persists through
to maturity in glial cells, whereas neuronal differentiation and maturation involve extensive reconfiguration of the DNA methylome that is
highly associated with cell type–specific changes
in transcriptional activity.
hmCG Is Enriched Within Active Genomic
Regions in Fetal and Adult Mouse Brain
Our base-resolution analysis of hmC using TABSeq revealed that intragenic and global hmCG
levels are largely equivalent between chromosomes,
whereas hmCG/CG is 22% lower on the male
ChrX, consistent with previous reports from enrichment based detection of hmC (32, 33) (Fig. 5A).
Analysis of hmCG levels in different genomic
regions revealed that, although adult hmCG/CG
is similar across transcriptional end sites and intragenic, DNaseI-hypersensitive (DHS), and enhancer
regions, the fetal frontal cortex shows a relative
enrichment of hmCG in DHS regions and enhancers, in particular enhancer regions that are
unique to the fetal developmental stage (Fig. 5B).
The inverse pattern can be observed for adult
mCG levels, which are lower in DHS regions and
enhancers (fig. S9, A and B), suggesting that regions of relatively high hmCG levels in the fetal
brain show relatively low mCG levels in the adult
brain. Analysis of intragenic hmCG enrichment
relative to flanking genomic regions, for cell type–
specific or developmentally dynamic gene sets
(Fig. 5C), showed that neuronal and astrocyte
gene bodies that are highly enriched with hmCG
in adult are also highly enriched at the fetal stage.
Thus, despite lower absolute levels of intragenic hmCG in the fetal stage, the adult patterns
of hmCG enrichment at these cell type–specific
genes are already forming in utero. Constitutively
lowly expressed genes show intragenic depletion
of hmCG, in contrast to constitutively highly
transcribed genes, which show localized enrichment of hmCG throughout part or all of the gene
body. Developmentally down-regulated genes show
enrichment of hmCG in the fetal frontal cortex but
not in adults, indicating that reduced transcription
is accompanied by a loss of hmCG enrichment.
Overall, transcriptional activity is associated with
intragenic hmCG enrichment, as reported (33),
with in utero establishment of adult hmCG patterns for cell type–specific genes and loss of hmC
enrichment associated with developmentally coupled transcriptional down-regulation.
Measurement of mC and hmC in all genes in
fetal and adult mouse frontal cortex indicated that
both mCG and hmCG are depleted at promoters
and in gene bodies of lowly expressed genes,
whereas hmCG is enriched throughout the gene
bodies of more highly transcribed genes (Fig. 5D).
The most highly expressed genes in the adult
frontal cortex show intragenic mCG hypomethylation (Figs. 4 and 5D) but still retain high intra-
genic hmCG. Ranking all genes by transcript
abundance, it is evident that the highest mean
intragenic hmCG levels, which occur in the most
highly transcribed genes, correspond to hmCG/
CG ~ 0.25 and mCG/CG ~ 0.5 (fig. S10D). Frontal
cortex development is accompanied by increased
enrichment of hmCG at intragenic regions that
are already hyper-hydroxymethylated at the fetal
stage (Fig. 5D and fig. S10), demonstrating that
adult patterns of genic hmC are already evident in
the immature fetal brain.
CG Differentially Methylated Regions
Enriched in Regulatory Regions
Because differences in genic mCG were observed
over development and between neuronal and glial
cell populations (Fig. 4), we scanned the human and
mouse methylomes to comprehensively identify
CG differentially methylated regions (CG-DMRs)
throughout the genome. CG-DMRs were identified between fetal and adult frontal cortex, neurons and glia, and combined into four sets: neuronal
and glial hyper- and hypo-methylated CG-DMRs.
In total, 267,799 human and 142,835 mouse
CG-DMRs were identified (median lengths: for
mouse, 473 bp; human, 533 bp), revealing several predominant dynamics in mCG during brain
development and cellular specialization (Fig. 6A
and fig. S11). Neuronal CG-DMRs are the most
numerous in both mice and humans, because of
the very distinct mCG patterns that emerge during
neuronal differentiation and maturation. At these
sites, CG methylation in adult neurons is distinct
compared with those in glial and/or fetal and early postnatal development frontal cortex tissue samples. Neuronal hypermethylated CG-DMRs also
show mCH hypermethylation (fig. S11). In mouse,
mCG/CG within neuronal hypomethylated
CG-DMRs declines to a stable level by 1 week
after birth. In contrast, neuronal hypermethylated
CG-DMRs do not begin to change until 1 week
after birth, after which they accumulate mCG
until 2 to 4 weeks of age. These data indicate that
increases in neuronal mCG occur during synaptogenesis after most decreases in neuronal mCG
have already occurred. Furthermore, we found
that hydroxymethylation in the adult cortex is
highest in CG-DMRs that show neuronal hypermethylation and is depleted from CG-DMRs that
display neuronal hypomethylation (Fig. 6A). This
suggests that hmCG may be most abundant in
neurons, rather than glial cells, in the frontal cortex.
Analysis of the genomic features in which
CG-DMRs are located revealed that although half
are found within gene bodies, they are not common within promoters and transcriptional start
and end regions. Instead, they are disproportionately located at DHS regions and enhancers unique
to fetal or adult brain (Fig. 6B). Closer inspection
of the enrichment and depletion of these CG-DMRs
revealed that fetal enhancers and DHS sites unique
to the fetal brain are enriched for hypermethylation in adult brain but not in the fetal brain (Fig.
6C). In contrast, adult enhancers and unique adult
DHS sites are highly associated with CG hyper-
www.sciencemag.org
SCIENCE
VOL 341
methylation in fetal stages but are not associated
with hypermethylated CG-DMRs in the adult
brain and in neurons. Thus, developmentally dynamic enhancers and DHS sites in frontal cortex
have dynamic CG methylation that is depleted
where chromatin accessibility and regulatory element activity increase, consistent with a range of
human cell lines (53).
To characterize gene functions associated with
the CG-DMRs, we analyzed the association between proximal genes (transcriptional start site
within 5 kb of the DMR) and cell type–specific or
developmentally dynamic gene sets (fig. S12A).
We observed an inverse relationship between methylation and gene function. Genes associated with
neuronal function and up-regulation during development are enriched for promoter hypermethylation in glia and hypomethylation in neurons,
whereas genes down-regulated during brain development and those related to astrocyte function
are enriched for promoter hypermethylation in neurons and hypomethylation in glia. Genes that are
constitutively expressed at either high or low levels are not associated with promoter/transcription
start site CG-DMRs, indicating that dynamic CG
methylation is highly associated with changes in
transcriptional activity and cell type–specific transcriptional regulation.
Because the majority of all developmentally
dynamic CG-DMRs are associated with DHS
sites, we examined the directional relationships
between dynamic mCG and DNA accessibility
states over development (Fig. 6D). Notably, DHS
sites unique to fetal frontal cortex overlap with
28% of CG-DMRs that gain methylation through
development (Adult>Fetal). However, these sites
only overlap 7.3% of CG-DMRs that lose mCG
during development (Fetal>Adult). Similarly, DHS
sites unique to adult frontal cortex rarely overlap
Adult>Fetal CG-DMRs. A similar analysis of developmentally dynamic enhancers active in only
one of the developmental stages (37) (Fig. 6E)
showed that enhancer activation is associated
with mCG hypomethylation of the enhancer,
whereas enhancer inactivation is associated with
enhancer mCG hypermethylation.
This inverse relationship between genome accessibility and mCG level at putative functional
regions of the genome suggests that nuclear factors that bind the region and increase accessibility may cause localized reduction in mCG, as
previously reported for a small number of DNA
binding proteins (54). Alternatively, mCG hypermethylation may cause reduced genome accessibility by direct inhibition of DNA-protein
interactions or induction of chromatin compaction, with loss of mCG enabling increased chromatin accessibility and genome interaction with
DNA binding factors.
Discrete regions that show increased or decreased CG methylation through development are
associated with specific local chromatin modifications. We found that CG-methylated regions of
the fetal frontal cortex that become hypomethylated in the adult (Fig. 6F, Fetal>Adult) gain
9 AUGUST 2013
1237905-9
RESEARCH ARTICLE
localized histone modifications characteristic of
active enhancers (H3K4me1 and H3K27ac) and
increased DNaseI hypersensitivity in the adult. In
addition, these regions have reduced accumulation of mCH (Fig. 6F), consistent with an overall
decrease in mC linked to increased genome accessibility and enhancer activity. In contrast, genomic regions that gain mCG during development
(Adult>Fetal) lose localized enrichment of H3K4me1,
H3K27ac, and DNaseI hypersensitivity and show
increased mCH. These changes indicate inactivation
of these genomic regions through brain development and suggest that this inactivation is associated
with increased local mCG.
A Hydroxymethylation Signature of
Developmentally Activated Regions
Adult>Fetal CG-DMRs show broad low-level
hmCG enrichment flanking the CG-DMR and a
localized depletion of hmC at the center in both
fetal and adult genomes (Fig. 6F), and the absolute abundance of hmC is several-fold lower in
fetal compared with adult frontal cortex (Fig. 1E).
This suggests that although Adult>Fetal CG-DMRs
gain mCG through development, they tend to
be refractory to conversion to hmC, potentially
because of lower accessibility to the Tet hydroxylases. In contrast, Fetal>Adult CG-DMRs have
a local enrichment of both mCG and hmCG in
the fetal cortex that becomes a local depletion in
the adult. Two enrichment-based genome-wide
hmC profiling techniques, CMS-IP (38) and biotinglucosyl tagging (31), confirmed the localized
enrichment of hmC at Fetal>Adult CG-DMRs
(fig. S12B). The localized enrichment of hmC at
these inaccessible and quiescent genomic regions,
which lose mCG and hmCG later in development, indicates that they may be premodified with
hmCG in the fetal stage to create a dormant state
that is poised for subsequent demethylation and
activation at a later developmental stage. Closer
inspection of base-resolution hmC data revealed
that 4% of the hmCG bases that have significantly higher hmC levels in fetal compared with adult
[false discovery rate (FDR) 0.05] directly overlap
with Fetal>Adult CG-DMRs, far exceeding the
number expected by chance (0.5%). This indicates that despite lower global levels of hmC in
the fetal brain, developmentally demethylated CGDMRs are enriched for hmCG bases that are
more highly hydroxymethylated in fetal than in
adult brain (Fig. 6G). The localized enrichment
of mCG at these CG-DMRs in the fetal cortex indicates that CG-demethylation has not yet taken place.
If fetal hmCG is poised at dormant genomic
regions in order to facilitate active DNA demethylation at later developmental stages, then the Tet
hydroxylase enzymes that catalyze conversion of
mC to hmC should be necessary for mCG hypomethylation in the adult frontal cortex at these
regions. To test this, we performed MethylC-Seq
of genomic DNA from frontal cortex tissue of adult
Tet2−/− mice. Adult>Fetal CG-DMRs, which gain
mCG through development, are largely unaffected
in Tet2−/− compared with wild-type adult mice
1237905-10
(Fig. 6, H and I; 3.6% hypermethylated, FET,
FDR 0.05). By contrast, a substantial fraction of
Fetal>Adult CG-DMRs are hypermethylated in
Tet2−/− (19.7%) versus wild type. The mutant
shows a small but significant increase in mCG at
Fetal>Adult CG-DMRs (Fig. 6I and fig. S12C)
(Tet2−/−: 7.9% T 4.6%, P < 10−11, Wilcoxon
signed rank test). The partial effect of the mutation
on CG methylation is not unexpected given that
all three Tet genes are expressed in the brain
(fig. S9C) and may exhibit some functional redundancy. Additionally, a genome-wide search
identified 14,340 CG-DMRs hypermethylated in
Tet2−/− relative to wild type (6 weeks, 10 weeks,
and 22 months), >fourfold more numerous than
hypomethylated CG-DMRs (3099). This further
indicates a role for Tet2 in mediating mCG demethylation during brain development.
Discussion
The essential role of frontal cortex in behavior
and cognition requires the coordinated interaction, via electrical and chemical signaling, of multiple neuronal cell types and a diverse population
of glial cells. Individual brain cells have unique
roles within circuits that are defined by their location and pattern of connections as well as by
their molecular identity. The development and
maturation of the brain’s physical structure and
the refinement of the molecular identities of neurons and glial cells occur in parallel in a finely
orchestrated process that starts early during the
embryonic period and continues, in humans, well
into the third decade of life (55, 56). An early
postnatal burst of synaptogenesis is followed by
activity-dependent pruning of excess synapses
during adolescence (28, 57, 58). This process
forms the basis for experience-dependent plasticity and learning in children and young adults (59),
and its disruption leads to behavioral alterations
and neuropsychiatric disorders (60). During this
period, profound transcriptional changes lead to
the appearance of adult electrophysiological characteristics in neocortical neurons.
Our study suggests a key role of DNA methylation in brain development and function. First,
CH methylation accumulates significantly in neurons through early childhood and adolescence,
becoming the dominant form of DNA methylation in mature human neurons. This shows that
the period of synaptogenesis, during which the
neural circuit matures, is accompanied by a parallel process of large-scale reconfiguration of the
neuronal epigenome. Indeed, central nervous system deletion of Dnmt3a during late gestation induces motor deficits, and animals die prematurely
(61). However, mice with a postnatal deletion
restricted to the pyramidal cell population (complete recombination around 1 month old) do not
show overt behavioral or transcriptional alterations (2). Our data suggest that expression of
Dnmt3a specifically around the second postnatal
week may be critical for establishing a normal
brain DNA methylation profile and allowing
healthy brain development.
9 AUGUST 2013
VOL 341
SCIENCE
Second, the precise positioning of mCG and
mCH marks, which are conserved between individuals and across humans and mice, is consistent with a functional role. Whether this is the case,
or whether the conserved patterns are instead a
reflection of conserved nucleosome position or
chromatin structure, requires further investigation.
Third, the relationship between DNA methylation
patterns and the function of neuron- or astrocytespecific gene sets suggests a role for DNA methylation in distinguishing these two broad classes
of cortical cells. DNA methylation could therefore play a key role in sculpting more-specific
cellular identities. If this is the case, we expect
that purified subpopulations will reveal high specificity of methylation at specific sites for particular cell types. Thus, the observation that most
CH sites with nonzero methylation are methylated in ~20 to 25% of sampled cells (fig. S1H)
could be explained by the heterogeneity of these
brain circuits rather than by stochastic methylation within each cell. These conclusions obtained
from our genome-wide, base-resolution, cell type–
specific DNA methylomes for brain cells through
key stages of development are the first steps toward
unraveling the genetic program and experiencedependent epigenetic modifications leading to a
fully differentiated nervous system.
References and Notes
1. E. Borrelli, E. J. Nestler, C. D. Allis, P. Sassone-Corsi,
Decoding the epigenetic language of neuronal plasticity.
Neuron 60, 961–974 (2008). doi: 10.1016/
j.neuron.2008.10.012; pmid: 19109904
2. J. Feng et al., Dnmt1 and Dnmt3a maintain DNA
methylation and regulate synaptic function in adult
forebrain neurons. Nat. Neurosci. 13, 423–430 (2010).
doi: 10.1038/nn.2514; pmid: 20228804
3. C. A. Miller, J. D. Sweatt, Covalent modification of DNA
regulates memory formation. Neuron 53, 857–869
(2007). doi: 10.1016/j.neuron.2007.02.022;
pmid: 17359920
4. S. Numata et al., DNA methylation signatures in
development and aging of the human prefrontal cortex.
Am. J. Hum. Genet. 90, 260–272 (2012). doi: 10.1016/
j.ajhg.2011.12.020; pmid: 22305529
5. M. Suderman et al., Conserved epigenetic sensitivity to
early life experience in the rat and human hippocampus.
Proc. Natl. Acad. Sci. U.S.A. 109 (suppl. 2), 17266–17272
(2012). doi: 10.1073/pnas.1121260109; pmid: 23045659
6. K. D. Siegmund et al., DNA methylation in the human
cerebral cortex is dynamically regulated throughout the
life span and involves differentiated neurons. PLoS ONE
2, e895 (2007). doi: 10.1371/journal.pone.0000895;
pmid: 17878930
7. K. Martinowich et al., DNA methylation-related chromatin
remodeling in activity-dependent BDNF gene regulation.
Science 302, 890–893 (2003). doi: 10.1126/science.
1090842; pmid: 14593184
8. J. U. Guo et al., Neuronal activity modifies the DNA
methylation landscape in the adult brain. Nat. Neurosci.
14, 1345–1351 (2011). doi: 10.1038/nn.2900;
pmid: 21874013
9. I. B. Zovkic, M. C. Guzman-Karlsson, J. D. Sweatt,
Epigenetic regulation of memory formation and
maintenance. Learn. Mem. 20, 61–74 (2013).
doi: 10.1101/lm.026575.112; pmid: 23322554
10. J. M. Levenson et al., Evidence that DNA (cytosine-5)
methyltransferase regulates synaptic plasticity in the
hippocampus. J. Biol. Chem. 281, 15763–15773 (2006).
doi: 10.1074/jbc.M511767200; pmid: 16606618
11. A. M. M. Oliveira, T. J. Hemstedt, H. Bading, Rescue of
aging-associated decline in Dnmt3a2 expression restores
www.sciencemag.org
RESEARCH ARTICLE
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
cognitive abilities. Nat. Neurosci. 15, 1111–1113 (2012).
doi: 10.1038/nn.3151; pmid: 22751036
P. Moretti et al., Learning and memory and synaptic
plasticity are impaired in a mouse model of Rett
syndrome. J. Neurosci. 26, 319–327 (2006).
doi: 10.1523/JNEUROSCI.2623-05.2006; pmid: 16399702
N. W. Penn, R. Suwalski, C. O’Riley, K. Bojanowski,
R. Yura, The presence of 5-hydroxymethylcytosine in
animal deoxyribonucleic acid. Biochem. J. 126, 781–790
(1972). pmid: 4538516
N. W. Penn, Modification of brain deoxyribonucleic acid
base content with maturation in normal and
malnourished rats. Biochem. J. 155, 709–712 (1976).
pmid: 949331
D. Globisch et al., Tissue distribution of 5-hydroxymethylcytosine
and search for active demethylation intermediates. PLoS ONE
5, e15367 (2010). doi: 10.1371/journal.pone.0015367;
pmid: 21203455
J. U. Guo, Y. Su, C. Zhong, G.-L. Ming, H. Song,
Hydroxylation of 5-methylcytosine by TET1 promotes
active DNA demethylation in the adult brain. Cell 145,
423–434 (2011). doi: 10.1016/j.cell.2011.03.022;
pmid: 21496894
Y.-F. He et al., Tet-mediated formation of 5carboxylcytosine and its excision by TDG in mammalian
DNA. Science 333, 1303–1307 (2011). doi: 10.1126/
science.1210944; pmid: 21817016
C.-X. Song et al., Genome-wide profiling of 5formylcytosine reveals its roles in epigenetic priming. Cell
153, 678–691 (2013). doi: 10.1016/j.cell.2013.04.001;
pmid: 23602153
L. Shen et al., Genome-wide analysis reveals TET- and
TDG-dependent 5-methylcytosine oxidation dynamics. Cell
153, 692–706 (2013). doi: 10.1016/j.cell.2013.04.002;
pmid: 23602152
W. Xie et al., Base-resolution analyses of sequence and
parent-of-origin dependent DNA methylation in the
mouse genome. Cell 148, 816–831 (2012).
doi: 10.1016/j.cell.2011.12.035; pmid: 22341451
K. E. Varley et al., Dynamic DNA methylation across
diverse human cell lines and tissues. Genome Res. 23,
555–567 (2013). doi: 10.1101/gr.147942.112;
pmid: 23325432
R. Lister et al., Human DNA methylomes at base
resolution show widespread epigenomic differences. Nature
462, 315–322 (2009). doi: 10.1038/nature08514;
pmid: 19829295
M. J. Ziller et al., Genomic distribution and inter-sample
variation of non-CpG methylation across human cell types.
PLoS Genet. 7, e1002389 (2011). doi: 10.1371/
journal.pgen.1002389; pmid: 22174693
R. Lister et al., Highly integrated single-base resolution
maps of the epigenome in Arabidopsis. Cell 133,
523–536 (2008). doi: 10.1016/j.cell.2008.03.029;
pmid: 18423832
R. Lister et al., Hotspots of aberrant epigenomic
reprogramming in human induced pluripotent stem cells.
Nature 471, 68–73 (2011). doi: 10.1038/nature09798;
pmid: 21289626
B. H. Ramsahoye et al., Non-CpG methylation is
prevalent in embryonic stem cells and may be mediated
by DNA methyltransferase 3a. Proc. Natl. Acad. Sci.
U.S.A. 97, 5237–5242 (2000). doi: 10.1073/
pnas.97.10.5237; pmid: 10805783
J. De Felipe, P. Marco, A. Fairén, E. G. Jones, Inhibitory
synaptogenesis in mouse somatosensory cortex. Cereb.
Cortex 7, 619–634 (1997). doi: 10.1093/cercor/7.7.619;
pmid: 9373018
P. R. Huttenlocher, A. S. Dabholkar, Regional differences
in synaptogenesis in human cerebral cortex. J. Comp.
Neurol. 387, 167–178 (1997). doi: 10.1002/(SICI)10969861(19971020)387:2<167::AID-CNE1>3.0.CO;2-Z;
pmid: 9336221
B. P. Berman et al., Regions of focal DNA
hypermethylation and long-range hypomethylation in
colorectal cancer coincide with nuclear lamina-associated
domains. Nat. Genet. 44, 40–46 (2012). doi: 10.1038/ng.969;
pmid: 22120008
S. Kriaucionis, N. Heintz, The nuclear DNA base
5-hydroxymethylcytosine is present in Purkinje neurons
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
and the brain. Science 324, 929–930 (2009);
10.1126/science.1169786. doi: 10.1126/science.
1169786; pmid: 19372393
C.-X. Song et al., Selective chemical labeling reveals the
genome-wide distribution of 5-hydroxymethylcytosine.
Nat. Biotechnol. 29, 68–72 (2011). doi: 10.1038/
nbt.1732; pmid: 21151123
K. E. Szulwach et al., 5-hmC-mediated epigenetic
dynamics during postnatal neurodevelopment and aging.
Nat. Neurosci. 14, 1607–1616 (2011). doi: 10.1038/
nn.2959; pmid: 22037496
M. Mellén, P. Ayata, S. Dewell, S. Kriaucionis, N. Heintz,
MeCP2 binds to 5hmC enriched within active genes and
accessible chromatin in the nervous system. Cell 151,
1417–1430 (2012). doi: 10.1016/j.cell.2012.11.022;
pmid: 23260135
M. Yu et al., Base-resolution analysis of 5-hydroxymethylcytosine
in the mammalian genome. Cell 149, 1368–1380 (2012).
doi: 10.1016/j.cell.2012.04.027; pmid: 22608086
J. A. Stamatoyannopoulos et al., An encyclopedia of
mouse DNA elements (Mouse ENCODE). Genome Biol.
13, 418 (2012). doi: 10.1186/gb-2012-13-8-418;
pmid: 22889292
R. K. Auerbach et al., Mapping accessible chromatin
regions using Sono-Seq. Proc. Natl. Acad. Sci. U.S.A.
106, 14926–14931 (2009). doi: 10.1073/
pnas.0905443106; pmid: 19706456
Y. Shen et al., A map of the cis-regulatory sequences in
the mouse genome. Nature 488, 116–120 (2012).
doi: 10.1038/nature11243; pmid: 22763441
W. A. Pastor et al., Genome-wide mapping of
5-hydroxymethylcytosine in embryonic stem cells.
Nature 473, 394–397 (2011). doi: 10.1038/nature10102;
pmid: 21552279
Y. Huang, W. A. Pastor, J. A. Zepeda-Martínez, A. Rao,
The anti-CMS technique for genome-wide mapping of
5-hydroxymethylcytosine. Nat. Protoc. 7, 1897–1908
(2012). doi: 10.1038/nprot.2012.103; pmid: 23018193
A. Magklara et al., An epigenetic signature for
monoallelic olfactory receptor expression. Cell 145,
555–570 (2011). doi: 10.1016/j.cell.2011.03.040;
pmid: 21529909
E. J. Clowney et al., Nuclear aggregation of olfactory
receptor genes governs their monogenic expression. Cell
151, 724–737 (2012). doi: 10.1016/j.cell.2012.09.043;
pmid: 23141535
H. Li et al., Transcription factor MEF2C influences
neural stem/progenitor cell differentiation and
maturation in vivo. Proc. Natl. Acad. Sci. U.S.A. 105,
9397–9402 (2008). doi: 10.1073/pnas.0802876105;
pmid: 18599437
M. W. Akhtar et al., In vivo analysis of MEF2 transcription
factors in synapse regulation and neuronal survival.
PLoS ONE 7, e34863 (2012). doi: 10.1371/journal.
pone.0034863; pmid: 22496871
A. C. Barbosa et al., MEF2C, a transcription factor that
facilitates learning and memory by negative regulation of
synapse numbers and function. Proc. Natl. Acad.
Sci. U.S.A. 105, 9391–9396 (2008). doi: 10.1073/
pnas.0802679105; pmid: 18599438
J. D. Cahoy et al., A transcriptome database for astrocytes,
neurons, and oligodendrocytes: a new resource for
understanding brain development and function. J. Neurosci.
28, 264–278 (2008). doi: 10.1523/JNEUROSCI.
4178-07.2008; pmid: 18171944
H. Wu et al., Dnmt3a-dependent nonpromoter DNA
methylation facilitates transcription of neurogenic genes.
Science 329, 444–448 (2010). doi: 10.1126/science.
1190485; pmid: 20651149
T. Mohandas, R. S. Sparkes, L. J. Shapiro, Reactivation of
an inactive human X chromosome: Evidence for X
inactivation by DNA methylation. Science 211, 393–396
(1981). doi: 10.1126/science.6164095; pmid: 6164095
K. Plath, S. Mlynarczyk-Evans, D. A. Nusinow, B. Panning,
Xist RNA and the mechanism of X chromosome
inactivation. Annu. Rev. Genet. 36, 233–278 (2002).
doi: 10.1146/annurev.genet.36.042902.092433;
pmid: 12429693
L. Carrel, H. F. Willard, X-inactivation profile reveals
extensive variability in X-linked gene expression in
www.sciencemag.org
SCIENCE
VOL 341
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
females. Nature 434, 400–404 (2005). doi: 10.1038/
nature03479; pmid: 15772666
A. J. Sharp et al., DNA methylation profiles of human
active and inactive X chromosomes. Genome Res. 21,
1592–1600 (2011). doi: 10.1101/gr.112680.110;
pmid: 21862626
C. M. Johnston et al., Large-scale population study
of human cell lines indicates that dosage
compensation is virtually complete. PLoS Genet.
4, e9 (2008). doi: 10.1371/journal.pgen.0040009;
pmid: 18208332
A. K. Maunakea et al., Conserved role of intragenic DNA
methylation in regulating alternative promoters. Nature
466, 253–257 (2010). doi: 10.1038/nature09165;
pmid: 20613842
R. E. Thurman et al., The accessible chromatin landscape
of the human genome. Nature 489, 75–82 (2012).
doi: 10.1038/nature11232; pmid: 22955617
M. B. Stadler et al., DNA-binding factors shape the
mouse methylome at distal regulatory regions. Nature
480, 490–495 (2011). pmid: 22170606
Z. Petanjek et al., Extraordinary neoteny of
synaptic spines in the human prefrontal cortex.
Proc. Natl. Acad. Sci. U.S.A. 108, 13281–13286
(2011). doi: 10.1073/pnas.1105108108
pmid: 21788513
B. Kolb et al., Experience and the developing prefrontal
cortex. Proc. Natl. Acad. Sci. U.S.A. 109
(suppl. 2), 17186–17193 (2012). doi: 10.1073/
pnas.1121251109; pmid: 23045653
H. T. Chugani, A critical period of brain development:
Studies of cerebral glucose utilization with PET.
Prev. Med. 27, 184–188 (1998). doi: 10.1006/
pmed.1998.0274; pmid: 9578992
P. Levitt, Structural and functional maturation of the
developing primate brain. J. Pediatr. 143 (suppl.),
S35–S45 (2003). doi: 10.1067/S0022-3476(03)00400-1;
pmid: 14597912
M. V. Johnston, Plasticity in the developing brain:
implications for rehabilitation. Dev. Disabil. Res. Rev. 15,
94–101 (2009). doi: 10.1002/ddrr.64; pmid: 19489084
P. J. Uhlhaas, W. Singer, The development of neural
synchrony and large-scale cortical networks during
adolescence: Relevance for the pathophysiology of
schizophrenia and neurodevelopmental hypothesis.
Schizophr. Bull. 37, 514–523 (2011). doi: 10.1093/
schbul/sbr034; pmid: 21505118
S. Nguyen, K. Meletis, D. Fu, S. Jhaveri, R. Jaenisch,
Ablation of de novo DNA methyltransferase
Dnmt3a in the nervous system leads to
neuromuscular defects and shortened lifespan.
Dev. Dyn. 236, 1663–1676 (2007). doi: 10.1002/
dvdy.21176; pmid: 17477386
Materials and methods are available as supplementary
materials on Science Online.
Acknowledgments: We thank W. F. Loomis for critical
reading of this manuscript; J. Chambers for technical
assistance with animal breeding; D. Chambers for technical
assistance with cell sorting; M. Lutz and T. Berggren for
provision of the HUES6 cells; F. Yue, G. Hon, and B. Ren
for providing mapped mouse ChIP-Seq data and assistance
with TAB-Seq; and A. Nimmerjahn for providing the
S100b-eGFP mice. Human brain tissue was obtained from
the National Institute of Child Health and Human Development
Brain and Tissue Bank for Developmental Disorders at the
University of Maryland, Baltimore, Maryland, the IDIBELL
Biobank, which is part of the BrainNet Europe Bank funded by
the European Commission (LSHM-CT-2004-503039). We thank
the Stamatoyannopoulos laboratory (University of Washington)
and the Mouse ENCODE Consortium for generating and
providing access to the DNaseI hypersensitivity data sets (35).
This work was supported by a National Institutes of Mental
Health grant to M.M.B. and J.R.E. (MH094670); the Howard
Hughes Medical Institute to T.J.S. and J.R.E.; the Gordon and
Betty Moore Foundation (GMBF3034) to J.R.E.; NIH grant
HG006827 to C.H.; NIH RO1 grants AI44432, HD065812, and
CA151535; grant RM-01729 from the California Institute of
Regenerative Medicine; and Translational Research grant
9 AUGUST 2013
1237905-11
RESEARCH ARTICLE
TRP 6187-12 from the Leukemia and Lymphoma Society
(to A.R.). R.L. was supported by an Australian Research Council
Future Fellowship. Support was also provided by the Government
of Western Australia through funding for the Western
Australia CoE for Computational Systems Biology. E.A.M. was
supported by the Center for Theoretical Biological Physics,
University of California San Diego, and NIH (National Institute
of Neurological Diseases and Stroke grant K99NS080911).
W.A.P. was supported by a predoctoral graduate research
1237905-12
fellowship from the NSF. Analyzed data sets can be accessed
at http://neomorph.salk.edu/brain_methylomes. Sequence
data can be downloaded from National Center for
Biotechnology Information GEO (GSE47966). Tet2 mutant
mice are available under a material transfer agreement from
the La Jolla Institute for Allergy and Immunology. Correspondence
and requests for materials should be addressed to R.L.
(ryan.lister@uwa.edu.au), M.M.B. (mbehrens@salk.edu), and
J.R.E. (ecker@salk.edu).
9 AUGUST 2013
VOL 341
SCIENCE
Supplementary Materials
www.sciencemag.org/content/341/6146/1237905/suppl/DC1
Materials and Methods
Figs. S1 to S12
Tables S1 to S5
References (63–78)
15 March 2013; accepted 7 June 2013
10.1126/science.1237905
www.sciencemag.org
www.sciencemag.org/cgi/content/full/science.1237905/DC1
Supplementary Materials for
Global Epigenomic Reconfiguration During Mammalian Brain
Development
Ryan Lister,* Eran A. Mukamel, Joseph R. Nery, Mark Urich, Clare A. Puddifoot,
Nicholas D. Johnson, Jacinta Lucero, Yun Huang, Andrew J. Dwork, Matthew D.
Schultz, Miao Yu, Julian Tonti-Filippini, Holger Heyn, Shijun Hu, Joseph C. Wu, Anjana
Rao, Manel Esteller, Chuan He, Fatemeh G. Haghighi, Terrence J. Sejnowski, M.
Margarita Behrens,* Joseph R. Ecker*
*Corresponding author. E-mail: ryan.lister@uwa.edu.au (R.L.); mbehrens@salk.edu (M.M.B.);
ecker@salk.edu (J.R.E.)
Published 4 July 2013 on Science Express
DOI: 10.1126/science.1237905
This PDF file includes:
Materials and Methods
Figs. S1 to S12
Table S1 to S5 captions
References
Other Supplementary Material for this manuscript includes the following:
available at www.sciencemag.org/cgi/content/full/science.1237905/DC1
Tables S1 to S5
Materials and Methods
Animals
The mouse lines C57Bl/6 and S100b-eGFP (B6;D2-Tg(S100B-EGFP)1Wjt/J), both from
Jackson Laboratories, ME were bred and maintained in our animal facility in 12 h
light/dark cycles with food ad libitum. To produce the S100b-eGFP animals used for
FACS, animals were crossed to C57BL/6 to produce heterozygotes for eGFP expression.
Tet2-/- mice were produced by targeted disruption of the Tet2 gene as described
previously (63). Animals were weaned in groups of 3-4 per cage, and used at the
postnatal time-points indicated. Three animals obtained from separate litters were
processed together for DNA and RNA isolations. All protocols were approved by the
Salk Institute's Institutional Animal Care and Use Committee (IACUC).
Human samples
Human brain tissue was obtained from the NICHD Brain and Tissue Bank for
Developmental disorders at the University of Maryland, Baltimore, MD, the IDIBELL
Biobank (Barcelona, Spain), which is part of the BrainNet Europe Bank (Munich,
Germany) funded by the European Commission (LSHM-CT-2004-503039). Procedures
were approved by the Institutional Review Board of the Salk Institute.
Tissue production
Animals were anesthetized with isoflurane and decapitated. Brains were quickly
extracted and placed in dissection media (64). The frontal cortex was obtained by slicing
the adult brains coronally at Bregma 1 mm, and dissecting the frontal cortical tissue,
being careful to avoid contamination with olfactory or putamen regions, under a
dissection microscope. Meninges were then dissected out from the frontal cortex, which
was rapidly frozen on dry ice or minced and processed for nuclei isolation and FACS.
Brain dissections in younger animals occurred at a similar coordinate (i.e., when the genu
of the corpus callosum unites between the two hemispheres). For the embryonic cortex,
the front third of the cortical plate was dissected under a microscope.
Protein isolation and Western blot
Tissue dissected as above was disrupted in RIPA buffer (Hepes 50 mM, pH 8.0; 1% NP40; 0.7% Deoxycholate; 0.5 M LiCl; complete protease inhibitors) with the addition 0.5%
SDS, by incubation on ice for 30 min followed by centrifugation at 15,000 x g to separate
nucleic acids. Protein was determined using a commercial kit (BCA, Pierce, Rockford,
IL)). Fifty micrograms of protein for each timepoint were separated in 7.5% SDS-PAGE
gels and transferred onto nitrocellulose membranes. Dnmt3a was detected by incubation
with anti-Dnmt3a antibody (1:300, Imgenex Cat# IMG-268A, San Diego, CA) followed
by HRP conjugated secondary antibodies and chemiluminescence (Cell Signaling,
Danvers, MA) in four independent experiments.
Nuclei isolation
2
Mouse: Frontal cortices from four animals for each time-point were processed as
described (65) with the following modification: isolated nuclei from 4 animals were
incubated for 1 h at 4°C in 1 ml of PBS + 2% horse serum containing a 1:1000 dilution of
AlexaFluor488 conjugated anti-NeuN monoclonal antibody (Millipore Cat# MAB377X,
Temecula, CA) before FACS.
Human: production of nuclei followed a similar procedure as for mice. Freshly frozen
samples of middle frontal gyrus were obtained from the NICHD Brain and Tissue Bank
for Developmental disorders at the University of Maryland. Sample demographics are
described in Table S5.
Dissociation of frontal cortex tissue for FACS
Frontal cortices from S100b-eGFP mice at 7-8 week old, obtained as above were minced
and placed in oxygen bubbled dissection media containing 1 mg/ml freshly prepared
pronase. After incubation at 37°C for 30 min, the tissue was washed thoroughly with the
same media, resuspended in 5 ml PBS containing 1 % horse serum and gently triturated
using a 5 ml pipette. The solution was then filtered through a 70 µm nylon mesh and
subjected to FACS.
FACS
Cells (S100b-eGFP) or nuclei (NeuN) were sorted using a FacsVantage SE DiVacell
sorter (BD Biosciences, San Jose, CA) using an 80 µm tip. Singlet cells were gated based
on forward scatter pulse height and pulse width characteristics. eGFP positive cells or
AlexaFluor488 labeled nuclei were discriminated from autofluorescence by plotting
green fluorescence versus orange fluorescence, and bright expressors were chosen for
sorting. Sorted cells or nuclei were centrifuged and pellets rapidly frozen in dry ice until
processed for nucleic acid isolation.
MethylC-Seq
MethylC-Seq library generation, read mapping, processing, and analysis were performed
as described previously (25), aligning reads to the mouse mm9 and human hg18 reference
genomes, except a previous filter that excluded reads containing >3 cytosine bases in the
CH context was not applied in this study. Library amplification was performed with
either PfuTurboCx Hotstart DNA polymerase (Agilent Technologies, Santa Clara, CA) or
KAPA HiFi HotStart Uracil+ DNA polymerase (Kapa Biosystems, Woburn, MA).
mRNA-Seq
mRNA-Seq libraries were generated from total RNA with polyA+ selection of mRNA
using the TruSeq RNA Sample Prep Kit v2 (Illumina, San Diego, CA). Strand-specific
libraries were constructed using a dUTP methodology as described previously (66), while
non-strand-specific libraries were constructed with the TruSeq RNA Sample prep Kit vs
as per manufacturer’s instructions. TopHat2 and Cufflinks2 packages (67, 68) were used
to map sequence reads to the mouse mm9 and human hg18 reference genomes and
quantitate differential gene expression.
Enrichment-based detection of 5-hydroxymethylcytosine (hmC)
3
Cytosine 5-methylenesulphonate immunoprecipitation (CMS-IP): mouse E13 fetal cortex
or 10 wk frontal cortex genomic DNA fragments (100 - 200 bp) were first end repaired,
3’ adenylated, ligated to methylated Illumina adaptor oligonucleotides and bisulfite
converted as per the MethylC-Seq protocol (25). Bisulfite conversion of hmC to cytosine
5-methylenesulphonate (CMS) was followed by immunoprecipitation of gDNA
fragments containing CMS with a specific antiserum to CMS. Bisulfite converted gDNA
was first denatured for 10 minutes at 95°C (0.4 M NaOH, 10 mM EDTA), neutralized by
addition of cold 2 M Ammonium Acetate pH 7.0, incubated with anti-CMS antiserum in
1x IP buffer (10 mM sodium phosphate pH 7.0, 140 mM NaCl, 0.05% Triton X-100) for
2 h to overnight at 4°C, and then precipitated with Protein G beads. Precipitated DNA
was washed 3 times with 1x IP buffer and eluted with Proteinase K, then purified with
Phenol Chloroform. Following immunoprecipitation, sequencing libraries were amplified
by 8 cycles of PCR and DNA sequencing was performed using a HiSeq 2000 Genome
Analyzer (Illumina) as described previously (38).
Biotin-glucosyl tagging and enrichment of hmC (biotin-gmC): the Hydroxymethyl
Collector Kit (Active Motif, Carlsbad, CA) was used to selectively tag hmC bases within
mouse E13 fetal cortex or 6 wk frontal cortex genomic DNA fragments (100 - 200 bp)
with glucose and biotin moieties to form biotin-N3-5-gmC (biotin-gmC), as described
previously (31). Genomic DNA fragments containing biotin-gmC were then enriched and
purified by high-affinity capture using streptavidin magnetic beads. Enriched DNA
containing biotin-gmC was used to generate Illumina DNA sequencing libraries with the
TruSeq DNA Sample Prep Kit v2 and sequenced using a HiSeq 2000 Genome Analyzer
(Illumina).
TAB-Seq
Performed as described in Yu et al. 2012 (34). Conversion and protection control
sequences, read filtering, and quantitation of hmC protection and mC non-conversion are
detailed below (Correction for non-conversion and protection).
Data Analysis Methods
Bisulfite non-conversion rate estimation for MethylC-Seq
To estimate the bisulfite non-conversion frequency, the frequency of all cytosine
basecalls at reference cytosine positions in the lambda genome (unmethylated spike in
control) was normalized by the total number of base-calls at reference positions in the
lambda genome. This was performed for all cytosines as well as specific sequence
contexts (CG, CH, CA, CC, CT, see Table S1).
Identification of CG Differentially Methylated Regions (CG-DMRs)
To identify DMRs, we used a two-step process. The first step involved performing a root
mean square test, as outlined previously (69), on each individual CG. For this test, we
constructed a contingency table where the rows indicated a particular sample and the
columns indicated the number of reads that supported a methylated cytosine or an
unmethylated cytosine at this position in a given sample. The p-values were simulated
4
using 5000 permutations. For each permutation, a new contingency table was generated
by randomly assigning reads to cells with a probability equal to the product of the row
marginal and column marginal divided by the total number of reads squared. To speed up
this process, if a p-value returned 50 permutations with a statistic greater than or equal to
the original test statistic, we stopped running permutations, thus using adaptive
permutation testing. To determine a p-value cutoff that would control the false discovery
rate (FDR) at our desired rate (see details for each DMR analysis, below), we used the
procedure outlined previously (70). Briefly, this method first generates a histogram of the
p-values and calculates the expected number of p-values to fall in a particular bin under
the null. This expected count is computed by multiplying the width of the bin by the
current estimate for the number of true null hypotheses (m0), which is initialized to the
number of tests performed. It then looks for the first bin where the expected number of pvalues is greater than or equal to the observed value, starting from the most significant
bin and working its way towards the least significant. The differences between the
expected and observed counts in all the bins up to this point are summed, and a new
estimate of m0 is generated by subtracting this sum from the current total number of tests.
This procedure was iterated until convergence, which we defined as a change in the m0
estimate less than or equal to the FDR. With this m0 estimate, we were able to estimate
the FDR of a given p-value by multiplying the p-value by the m0 estimate (the expected
number of positives at that cutoff under the null hypothesis) and dividing that product by
the total number of significant tests we detected at that p-value cutoff. We chose the
largest p-value cutoff that still satisfied our FDR requirement. Once this p-value cutoff
was chosen, for mCG-DMR analysis, significant sites were combined into blocks if they
were within 500 bases of one another and had methylation changes in the same direction,
for example if at both sites sample A was hypermethylated and sample B was
hypomethylated. Furthermore, blocks that contained fewer than 4 differentially
methylated sites were discarded. Finally, CG-DMR blocks were filtered based on a
requirement for a minimum number of samples to all show the same significant
differential methylation patterns. These sample comparison details for the CG-DMR sets
described in this study are described below.
- Human fetal vs adult frontal cortex CG-DMRs were identified (FDR = 0.05) where
fetal frontal cortex was differentially methylated compared to both 16 yr and 25 yr frontal
cortex samples.
- Human NeuN+/NeuN- vs fetal frontal cortex CG-DMRs were identified (FDR =
0.05) between NeuN+/NeuN- samples and fetal frontal cortex, requiring both human
NeuN+/NeuN- samples (53 yr female R1, 55 yr male R2) to be differentially methylated
compared to fetal.
- Human NeuN+ vs NeuN- CG-DMRs were identified (FDR = 0.05) between NeuN+
and NeuN- samples, requiring both human individuals (53 yr female, 55 yr male) to be
differentially methylated between cell types.
- Mouse fetal vs adult frontal cortex CG-DMRs were identified (FDR = 0.01) where
fetal frontal cortex was differentially methylated compared to both 6 wk and 10 wk
frontal cortex samples.
- Mouse NeuN+/NeuN- vs fetal frontal cortex CG-DMRs were identified (FDR =
0.01) between NeuN+/NeuN- samples and fetal frontal cortex, requiring !2 of 3 mouse
5
NeuN+/NeuN- samples (7 wk male R1, 6 wk female R2, 12 mo female R3) to be
differentially methylated compared to fetal.
- Mouse NeuN+ vs NeuN- CG-DMRs were identified (FDR = 0.01) between NeuN+
and NeuN- samples, requiring !2 of 3 mouse samples (7 wk male R1, 6 wk female R2, 12
mo female R3) to be differentially methylated between cell types.
- Mouse Tet2-/- vs wild-type CG-DMRs were identified (FDR = 0.01) where Tet2-/was differentially methylated compared to ! 2 of the 6 wk, 10 wk and 22 mo wild-type
adult frontal cortex samples.
- Mouse fetal hmCG vs 6 wk hmCG CG-DMRs were identified (FDR = 0.05) as
above except individual significantly differentially hydroxymethylated bases were
identified and no block joining was performed.
DNaseI hypersensitivity
Data generated by the Mouse ENCODE Consortium (35) was mapped to the mouse mm9
reference genome with Bowtie (71) and peaks of read enrichment were identified using
MACS (72).
ChIP-Seq and enhancers
Mouse E14.5 embryonic brain predicted enhancer regions, adult cortex predicted
enhancer regions, and sequence reads mapped to mm9 for H3K4me1, H3K27ac and input
were from Shen et al. 2012 (37). Coordinates of predicted enhancer regions that
overlapped with DNaseI hypersensitive regions were adjusted to center the enhancers on
the midpoint of DNaseI hypersensitive sites.
Dnmt3a binding sites
Liftover was used to convert the coordinates of Dnmt3a binding sites identified in mouse
postnatal neuronal stem cells (46) to the mm9 reference.
IP detection of hmC
CMS-IP reads were mapped to the mouse mm9 genome as described previously for
MethylC-Seq sequencing data (25). Biotin-gmC reads were mapped to the mouse mm9
genome using Bowtie2 (71).
Estimation of methylation level (mC/C) with correction for non-conversion and
protection
The majority of our analyses of DNA methylation are based on the level (mC/C,
mCG/CG, or mCH/CH), which is an estimate of the fraction of cytosines in the
sequenced population which are methylated. To estimate mC/C, we computed the
fraction of all MethylC-Seq base-calls at cytosine reference positions that were cytosine
(protected from bisulfite conversion). We then corrected these estimates for known
experimental biases. Both the whole-genome bisulfite sequencing and TAB-Seq
techniques used in this study are subject to a low error rate, r, due to the failure of the
chemical conversion of unmethylated cytosine to uracil. Such errors leads to false
positive detections of mC. In addition, TAB-Seq suffers from a small rate of error due to
non-protection, s, which is the probability that a hydroxymethylated cytosine is converted
6
by the bisulfite reaction and leads to a false negative (missed call). These rates were
calibrated as described previously (34), summarized as follows.
TAB-Seq cytosine non-conversion and hmC protection rates: M.SssI-converted
pUC19 DNA (NEB), fully methylated (mC) at all CG cytosines, was spiked into the
mouse genomic DNA sample at 0.5% (w/w) prior to library preparation (34). The
frequency of C basecalls at pUC19 CG reference positions was calculated from reads
uniquely mapped to the pUC19 reference, following removal of clonal reads and reads
containing >3 cytosine basecalls in the CH context (34). The cytosine non-conversion
rate represents the aggregate frequency of failed mC conversion by Tet1 and failure of
bisulfite conversion.
hmC non-protection rate: The 38-48 kb region of the unmethylated lambda phage
genome (Promega, Madison, WI) was amplified using ZymoTaq DNA polymerase in 5
separate, nonoverlapping 2 kb PCR amplicons in the presence of dATP, dGTP, dTTP and
d5hmCTP (Zymo Research), as described previously (34). PCR products purified by
agarose-gel electrophoresis were spiked into the mouse genomic DNA sample at 0.5%
(w/w) prior to library preparation. The frequency of C basecalls at CG reference positions
(protection frequency) was calculated from all reads uniquely mapped to the 38-48 kb
region of the lambda genome, only considering CG reference positions where the closest
neighboring reference C is at least 4 bases away (e.g. hmCGNNhmC), in order to reflect
the subset of bases in the similar hmC context as mammalian genomes (34). Finally,
given the commercial source of 5hmdCTP used to synthesize the control contains ~5%
dCTP, which does not get protected, the protection frequency was increased by 5%.
Thus, the non-protection frequency (1 - protection frequency) represents the rate at which
hmC bases failed to be glucosylated and protected from Tet1 conversion and subsequent
bisulfite conversion.
We corrected the observed number of converted and protected reads (a, c
respectively) to take these rates into account. If the true rate of methylation or
hydroxymethylation at a given site is p, the probability of observing (c, a) is
Binom[c|a+c,q], where ! ! ! ! ! ! ! ! ! ! ! is the probability of a read being
!!!
sequenced as cytosine. The maximum likelihood estimate of p is therefore ! !
!
!!!!!!!!!
!!!!!
. When ! ! ! we set ! ! !! In all of our samples, !! ! ! !. For some analyses,
we therefore used an approximation valid for low non-conversion rate, ! ! ! ! ! !
!
! !.
!!!!!
!!!
Supplementary Text
Figure-specific data analysis details:
Fig. 1A, 2A, 2E and Fig. 3A (browser representations)
The AnnoJ browser (24) shows tracks of hg18 and mm9 UCSC gene annotations,
mRNA-Seq read density, and DNA methylation on the Watson and Crick strands (W, C).
DNA methylation at each CG or CH site is represented as a vertical tick (green or blue,
respectively) whose height corresponds to the fraction of all reads mapped to that site
7
which were protected from bisulfite conversion (mC/C, range from 0 to 1). Regions
which are homologous in human and mouse are indicated by light-blue shading.
Fig. 1B
CH methylation level (mCH/CH) for each gene as a function of the rank of mRNA
transcript abundance (mRNA-Seq fragments per kilobase of mapped read (FPKM)). We
analyzed all genes which are expressed (FPKM > 0.01), whose length > 1 kb, mCH/CH <
0.03, and we omitted genes with no MethylC-Seq coverage for CG or CH reference
positions. Note: mCG/CG and mCH/CH are defined as the total number of mC basecalls
in each context, normalized by the total number of basecalls in the same context (both
methylated and unmethylated), corrected by subtraction of context specific nonconversion rates (see above).
Fig. 1E
We calculated the fraction of all cytosine basecalls in each context, considering only sites
covered by at least 10 reads in the mouse MethylC-Seq and TAB-Seq samples (fetal and
6wk). The proportion of basecalls was corrected for each sample’s non-conversion rate as
described above.
Fig. 1F
For all 100 kb contiguous windows of mouse and human chromosome 12, mCG/CG and
mCH/CH were calculated, with subtraction of mCG or mCH non-conversion values.
Fig. 1G
For all 5 kb contiguous windows of the mouse genome in chr12:112,250,000 119,500,000, the following mouse data was summarised:
- mRNA-Seq (10 wk frontal cortex): fraction of all reads located in window.
- ChIP-input reads (8 wk cortex): fraction of all reads located in window.
- DnaseI HS (8 wk cortex): fraction of all reads located in window.
- mCG/CG and mCH/CH: fraction of basecalls in window that are methylated in
each context.
- hmC CMS-IP (10 wk frontal cortex): [fraction of all CMS-IP reads located in
window] / [fraction of all CMS-input reads located in window].
- Note: mRNA, ChIP input and DNaseI HS tracks use arbitrary units proportional to
read density.
Fig. 1H
Chromatin accessibility vs CH methylation level (mCH/CH). For all 10 kb contiguous
windows of the mouse genome, we calculated mCH/CH level from 10 wk mouse frontal
cortex MethylC-Seq and estimated chromatin accessibility as ChIP-Seq input read
density (reads per 10 kb per million total reads) for 8 wk mouse cortex (37). We omitted
any 10 kb regions where ChIP-seq input read density = 0 or mCH/CH = 0. Density of
bins is shown for mCH/CH " 0.02 and ChIP-seq input read density " 0.5.
Fig. 2B: as per Fig. 1E. Only sites that were covered by at least one read in all samples
were included, and correction for non-conversion was applied.
8
Fig. 2C
For each mouse NeuN+, NeuN- and glia sample, mCG/CG and mCH/CH were calculated
for all Dnmt3a binding regions (46) (n = 67,050) and a set of 100,000 regions randomly
distributed throughout the genome, each the size of the mean length of Dnmt3a binding
regions (555 bp). Data is presented in a box and whisker plot with outliers not shown.
Fig. 2D
Heatmaps show Pearson correlation of mCH/CH within bins of size 10kb throughout the
human and mouse genome. The median total coverage in each bin (i.e., the number of
CH sites times the number of reads overlapping that site) was >24,000 for all samples.
This provides confidence that the estimated mCH/CH was not severely affected by
stochastic sampling of reads. Finally, we applied hierarchical clustering to all the data
sets with complete linkage and distance given by one minus the sample correlation.
Fig. 2F and S4
We developed statistical procedures to analyze the base-resolution, or per-site, correlation
of methylation level between samples within the same species (intra-species) and across
species (inter-species comparison of mouse and human). For this analysis, we focused on
samples with substantial mCH, including NeuN+ in mouse (R1, R2, R3) and human (R1,
R2), as well as human stem cell lines (H1, HUES6).
Intra-species correlation: To reduce the effect of stochastic sampling, we set a minimum
coverage threshold of 10 reads per site. Any site which had fewer than 10 reads in any of
the samples was excluded. To account for slight differences in the bisulfite nonconversion between samples, we corrected the methylation level (mC/C) using the
procedure described above. We then calculated the Pearson correlation of the corrected
mC/C across all single sites between each pair of samples. These values defined the
correlation values for “individual replicates,” shown as open circles in Fig. S4. In some
cases more than two samples were available for comparison, so we computed a group
average correlation coefficient by summing each of the pairwise covariances of each
independent pair of samples, then normalizing by the appropriate sum of variances. For
the three mouse NeuN+ replicates, let the three replicates be m1, m2 and m3. The
average correlation, which determines the bar height in Fig. 2F and S4, is then:
!!"# ! !!!"!!" ! !!"!!" ! !!"!!" !!!!!"!!" ! !!"!!" ! !!"!!" !
Inter-species correlation: To compare methylation patterns between mouse and human,
we first identified homologous CH sites. We restricted analysis to CH sites of high
coverage in all three mouse NeuN+ samples (!20 reads). For each of these sites, we used
the UCSC liftOver utility to determine whether a unique homologue site exists in the
human (73). We then filtered these homologous sites to retain only those which were CH
sites in both species, and which had a minimum coverage in all human NeuN+ or ESC
samples (!10 reads).
We identified 66,489 autosomal and 3,323 ChrX sites for comparison of human and
mouse NeuN+; for comparison of mouse NeuN+ with human H1 ESC we used slightly
9
more (88,771 autosomal, 5,120 ChrX). The correlation coefficients for individual
pairwise comparisons of replicates were then calculated as for the intra-species
correlations (above).
For the human-mouse NeuN+ comparison, we combined 6 total comparisons
(human replicates r1={h1,h2} vs. mouse replicates r2={m1,m2,m3}) to form an average
correlation coefficient as follows:
!!"# !
!!
!!
!! !!!!!!
!! !!!!!!
!"
!" !!"!!"
This procedure was also used to compute an average correlation for comparison of
the two embryonic stem cell lines with the human and mouse NeuN+ replicates.
To assess the magnitude and significance of the correlation values we observed, we
carried out two control analyses for each of the above correlation measurements.
Simulation of full correlation to assess correlation magnitude: Because our estimate of
methylation level at each site is based on the observed frequency of methylated reads
(mC/C), this measurement suffers from stochastic noise arising from the random
sampling of reads from the population of DNA fragments. Such noise lowers the
correlation coefficient we observe between each pair of samples. To correct for this, we
performed a simulation to determine how much the correlation coefficient is reduced by
sampling noise. We did this by simulating pairs of samples with identical methylation
patterns, but independent sampling noise. At each site, we combined the number of
observed methylated and unmethylated reads from two experimental samples
(!! ! !! ! !! ! !! , respectively) and determined the total empirical frequency of methylated
reads: ! ! !!! ! !! !!!!! ! !! ! !! ! !! !! We then simulated random samples from a
binomial distribution with probability of success f and total trials !!! ! !! ! or !!! ! !! !,
respectively. Finally, we computed the correlation coefficient using these simulated
samples. Fig. S4 shows that these simulated correlation coefficients were ~0.8 for most
comparisons. Since the true correlation of the simulated samples, which would be
observed in the limit of very high coverage, is 1, in Fig. 2F we normalized the observed
correlation of the original data by the simulated correlation. Fig. S4 shows the correlation
of the data, without normalization
Shuffle control to assess correlation significance: A second control analysis was
conducted to compare the observed correlation with what would be expected in case the
true correlation was 0. The finite size of the data set leads to a non-zero observed
correlation, particularly for inter-species comparisons that cannot take advantage of large
populations of homologous sites. In addition, long length-scale correlations related to
genomic regions such as CG-islands may induce correlation that is not specific to a
particular site (Fig. S5A). For each comparison of a pair of samples, we performed a set
of 200 shuffle controls in which the methylation level at a given site in sample 1 was
compared with the methylation level at a randomly chosen site in sample 2, constrained
to be <1 Mb distant. After shuffling sites, we computed the correlation coefficient exactly
as for the unshuffled data.
10
Fig. 3: Analysis of gender-specific methylation patterns on X-chromosome
We compared the brain DNA methylation patterns in our human female and male NeuN+
samples with two published surveys of human X-linked genes. The first of these studies
measured expression from the active and inactive ChrX alleles using a human/rat somatic
cell hybrid assay (49). The authors of that study assigned each surveyed gene a score in
the range 0-9 corresponding to the number of individual hybrid lines in which expression
from the inactivated ChrX was detected. A score of 0 thus corresponds to complete
inactivation, whereas 9 corresponds to complete escape from X-inactivation. We
remapped these genes to the reference used in this study (hg18) using the UCSC liftOver
utility (73). We also compared our data with a second survey that used an
immunoprecipitation based method (MeDIP) to profile DNA methylation from peripheral
blood in normal human females (karyotype 46, XX) as well as Turner syndrome patients
(45, X) (50). That study reported that inactivated genes feature hypermethylation of
promoter region CG islands, whereas escapee genes show low levels of promoter
methylation. These two studies using complementary techniques each provide a list of
putative X-inactivated and X-escapee genes, which we used to compare with the mCG
and mCH patterns in our neuronal samples. To avoid noise and bias, we excluded short
genes (<2 kb in length) as well as any gene with low coverage (<50% of the mean read
density for the entire ChrX).
Fig. 3B and S4D
For each of the genes assayed in (49), we examined the total mCG/CG within the
promoter region, defined to be a 2 kb region centered on the TSS, and the total mCG/CG
or mCH/CH within the gene body. The box plot shows the difference between female and
male methylation level for genes ranked according to the X-inactivation status index. For
each box, the central line is the group median, the box edges are the 25th and 75th
percentiles, and the whiskers extend to the maximum and minimum values (including all
“outliers”).
Fig. 3C and S5E
Scatter plots of human female vs. male NeuN+ mCG/CG within the promoter region
(TSS±1kb) of each gene, or intragenic mCG/CG or mCH/CH.
Fig. 3D
Receiver operating characteristic (ROC) analysis was used to assess how well DNA
methylation patterns allow discrimination of X-escapee genes. For this analysis, we used
only those genes whose expression was measured by (49). We defined escapee genes as
those which were expressed from both the active and inactivated ChrX in all of the cell
lines tested (score = 9). Genes with score 0-8 were defined as non-escapee. We tested
discriminability using the female-male difference in promoter region mCG/CG, which
was previously shown to correlate with X-inactivation status (50), as well as the
difference in neuronal intragenic mCH/CH. In addition, we tested a linear combination of
four different features (female-male promoter region mCG/CG and mCH/CH, femalemale intragenic mCG/CH and mCH/CH) with coefficients determined by linear
regression against the X-inactivation score (49). For each of these three measures of
11
DNA methylation, we created an ROC curve by plotting the fraction of escapee genes
that are correctly discriminated, as well as the fraction of false detections, at each value
of a discrimination threshold. The area under the ROC curve (AUC) is a statistical
measure of discriminability, which ranges from 0.5 when little or no discrimination
information is present to 1 for perfect discriminability.
Fig. 4A, S7: Unbiased clustering of gene sets based on DNA methylation profiles
To identify sets of genes that share similar DNA methylation patterns in an unbiased
fashion, we developed a two-stage procedure that first represents all annotated genes
within a low dimensional feature space, and then groups nearby genes in this space to
create gene clusters.
The starting point is a large data array with one entry for each gene in each of 7
samples (6 wk, 7 wk and 12 mo NeuN+ and NeuN-, and glia (S100b+)) and in each
sequence context (CG and CH). For each entry, we profiled the methylation level
(mCG/CG and mCH/CH) in bins of size 1 kb starting 100 kb upstream of the TSS and
ending 100 kb downstream of the transcription end site (TES). To compare genes with
different lengths, we divided each gene body into 10 non-overlapping bins of equal size
extending from the TSS to the TES. We then linearly interpolated the gene-body mC/C
data at 100 evenly spaced bins within the gene body in order to give roughly equal weight
to the gene-body and flanking methylation data. The absolute mCG/CG and mCH/CH
level in each sample for each gene was normalized by the median value in the distal
flanking region (50-100 kb upstream of TSS or downstream of TES). Normalized mC/C
values were then log-transformed. Combining all of the normalized methylation data
points, we obtained a matrix of 4,200 features for each of 25,260 genes. Any bins with
missing data due to insufficient coverage in one of the samples (5.9% of the total) were
replaced with the median value of the entire data set. We performed singular value
decomposition on this data matrix to identify the linear combinations of methylation
features that account for the largest fraction of the total data variance [principal
components (PCs), Fig. S7]. We retained the top 5 PCs as a low-dimensional
representation of robust genomic methylation features, accounting for 46% of the total
data variance. Although the remaining PCs contain a great deal of meaningful data which
may be useful for more refined analysis of DNA methylation features, our analysis
focused on this reduced space.
Next, we used k-means clustering to estimate gene sets with highly similar withinset methylation patterns. We chose to extract k=15 clusters to capture a diverse range of
methylation features, while still allowing visualization and statistical enrichment analysis
of functional association for each gene set. We repeated the clustering procedure 10 times
using random initialization of the cluster centers, choosing as the final estimate the run
with the smallest within-cluster sum of distances from each point to the cluster centroid.
To display the heatmaps of mRNA-Seq and mC/C patterns for each of 25,260 genes, we
smoothed and downsampled the genes 40-fold to allow representation of genome-scale
features.
Fig. 4B: Enrichment analysis of functional gene categories
To relate DNA methylation profiles with cell-type specific gene functions and with
specific patterns of developmental regulation, we defined six mouse gene sets, as follows:
12
1. Neuronal genes (461) which were reported to have >3-fold enrichment in mouse
neurons (45). To increase the specificity of this list, we also required that genes were
homologous to rat genes which are highly expressed in neuronal somata (74).
2. Astrocytic genes (2,618), reported to have >20-fold enriched expression in astrocytes
(45).
3. Constitutively high genes (60) which did not appear in the neuronal or astrocytic gene
set; whose transcripts were highly abundant (mRNA-Seq FPKM ! 10) in brain tissue
samples from seven developmental stages (fetal, 1, 2, 4, 6 and 10 wk, and 22 mo); and
which had "20% variation between the maximum and minimum abundance in these
samples.
4. Constitutively low genes (5,278) which did not appear in the neuronal or astrocytic
gene set; whose transcript abundance remained below a threshold (FPKM " 0.5); and
which had "20% variation between the maximum and minimum abundance in these
samples.
5. Upregulated genes (233) which are either neuronal or astrocytic genes and which have
>10-fold increased expression at 2 wk compared with fetal.
6. Down-regulated genes (207) which are either neuronal or astrocytic genes and which
have >5-fold lower expression at 2 wk compared with fetal.
For each functional gene set, we computed the overlap with each of the 15 gene
clusters identified by k-means (Fig. 4A). We compared this number with the expectation
for random gene sets with the same size to compute a fold-enrichment value, and we
assessed statistical significance using Fisher’s exact test. We applied a BenjaminiHochberg procedure (75) to control the false discovery rate < 0.05.
Fig. 4C
For each gene set, we computed the median of the flank-normalized mC/C profiles in
each developmental and cell-type specific sample. The mRNA-Seq expression data
(right-hand column) show the median expression level over all genes in each set.
Fig. 5A
Total hmCG/CG and hmCH/CH are shown for autosomes and ChrX in 6 wk cortex.
Methylation values have been corrected for the biases introduced by non-conversion and
non-protection (see above).
Fig. 5B
We profiled the total hmCG/CG within the following regions: CG islands (UCSC),
promoter regions (TSS±2kb), gene ends (TES±2kb), gene bodies (TSS-TES), DHS (35)
identified in adult or fetal brain, DHS identified exclusively in adult or fetal, and
enhancers identified in adult or fetal brain (37). The plot shows the median over all the
individual regions of each type, and the error bars indicate the 32-68th percentile range
corresponding to ~1 standard deviation.
Fig. 5C
Heatmaps show the median profile of flank-normalized hmCG/CG around six categories
of genes. Methods are the same as Fig. 4C.
13
Fig. 5D
Heatmaps showing the profile of mC/C and hmC/C for each of 16,077 genes expressed in
adult mouse cortex (mRNA-Seq FPKM > 0.1 in 6 wk tissue). Genes are ranked by their
expression in adult cortex and the flank-normalized mC/C or hmC/C pattern around each
gene is displayed, as described in the methods for Fig. 4A (above).
Fig. 6A
Starting with six distinct categories of CG-DMRs for each species (described above), we
excluded those on the X and Y chromosomes and organized the remaining DMRs into 4
largely non-redundant sets comprising most of the specific CG-DMR types. These were
defined as follows:
NeuN+ hyper-mCG: This set includes CG-DMRs where mCG/CG is larger in NeuN+
compared with NeuN– and Fetal; larger than Fetal only; or larger than NeuN– only
NeuN– hyper-mCG: This set includes CG-DMRs where mCG/CG is larger in NeuN–
compared with NeuN+, and those in which mCG/CG is larger in NeuN– than in fetal.
NeuN+ hypo-mCG: This set includes CG-DMRs where mCG/CG is smaller in NeuN+
compared with NeuN– and Fetal; or smaller compared with Fetal only.
NeuN– hypo-mCG: This set includes CG-DMRs where mCG/CG is smaller in NeuN–
compared with NeuN+ and Fetal, or Fetal only.
We also created two additional DMR sets (heatmaps not shown):
Fetal hyper-mCG: This set includes CG-DMRs where mCG/CG is larger in fetal
compared with adult NeuN+ and/or NeuN–.
Adult hyper-mCG: This set includes CG-DMRs where mCG/CG is larger in adult NeuN+
and/or NeuN– compared with fetal.
To display heatmaps of the methylation level in each CG-DMR, we first sorted the
DMRs within each set by the difference between NeuN+ and NeuN– mCG/CG, averaged
over replicates. We then smoothed and down-sampled the ordered DMR methylation
patterns by 10 DMRs. The line plots below each heatmap show the median methylation
level for each CG-DMR category, while the shaded regions show the 32nd-68th
percentiles.
Fig. 6B
To assess the genomic distribution of CG-DMRs, we computed the fraction of all CGDMRs (the union of the six sets listed above) that overlap the genomic regions defined
above (see Fig. 5B and associated methods).
Fig. 6C
We compared the number of CG-DMRs of each type overlapping each region, with the
expected number that would be observed if both sets were randomly distributed. Suppose
14
there are N CG-DMRs and R regions, and let the total genomic length of the CG-DMRs
be !! and the total genomic length of the regions be !! . Then the expectation value of
the number of overlapping regions is !! ! !!!! ! !!! !!!, where G is the total length
of the genome. In the limit !! ! !! ! !, the number of observed overlapping regions will
be Poisson distributed with mean !! . To assess the significance of enrichment or
depletion, we therefore compared the actual number of observed overlaps with the
Poisson cumulative distribution function. No multiple-comparison correction was applied
to this analysis.
Fig. 6D,E
For each category of developmental CG-DMRs (Fetal > Adult or Adult > Fetal), we
calculated the fraction that overlap a fetal or adult DHS site or a fetal- or adult-specific
enhancer. The Venn diagrams represent these overlaps, and the relative sizes and
positions of the circles are chosen to approximate the true proportions as closely as
possible. Exact proportionality of the Venn diagrams with the true overlaps is sometimes
not geometrically feasible.
Fig. 6F
For each CG-DMR, data was analyzed in contiguous 100 bp bins from 2.5 kb upstream to
2.5 kb downstream of the central position of the CG-DMR. H3K4me1, H3K27ac and
CMS-IP read density profiles were normalized to their respective input control samples.
The DNaseI HS read density profile uses arbitrary units proportional to read density.
Fig. 6H
To test the hypothesis that Tet2 is involved in active demethylation during development
of brain cells, we analyzed differences between mCG/CG in adult wild-type mice (6 wk)
and Tet2-/- knockout mice (63). At each developmental CG-DMR, we counted the total
methylated and unmethylated reads within a window of size 1 kb centered on the
midpoint of the DMR. We then used Fisher’s exact test to assess the significance of
differences between the 6 wk wild-type sample and each of the other four samples. We
applied a Benjamini-Krieger-Yekutieli procedure to control FDR<0.05 (MATLAB
procedure fdr_bky by David M. Groppe) (76). The bars in Fig. 6H show the proportion of
developmental CG-DMRs that had higher or lower methylation in the Tet2-/- mice
compared to the 6 wk wild-type.
Fig. 6I
Histograms show the distribution of the difference in mCG/CG within 1 kb windows
centered on each developmental CG-DMR between adult WT (6 wk) and two groups:
control (4 wk, 10 wk and 22 mo WT), as well as the Tet2-/- knockout. Filled bars show
the magnitude of differences at CG-DMRs that were significantly altered (see statistical
procedure above, Fig. 6H).
Fig. S1A-D
The fraction of all basecalls methylated in each context was calculated as follows (where
CN=CC, CA, CT or CG):
15
Total mCG basecalls (corrected for CG BS non-conv.): mCGbasecalls-corr = #mCGbasecalls
- #CGbasecalls* mCGBS-non-conv
Total mCA basecalls (corrected for CA BS non-conv.):
mCAbasecalls-corr = #mCAbasecalls
- #CAbasecalls* mCABS-non-conv
Total mCC basecalls (corrected for CC BS non-conv.): mCCbasecalls-corr = #mCCbasecalls #CCbasecalls* mCCBS-non-conv
Total mCT basecalls (corrected for CT BS non-conv.):
mCTbasecalls-corr = #mCTbasecalls #CTbasecalls* mCTBS-non-conv
Fraction of all C basecalls that are mCG: mCGfraction = mCGbasecalls-corr / #CNbasecalls
Fraction of all C basecalls that are mCA: mCAfraction = mCAbasecalls-corr / #CNbasecalls
Fraction of all C basecalls that are mCC: mCCfraction = mCCbasecalls-corr / #CNbasecalls
Fraction of all C basecalls that are mCT: mCTfraction = mCTbasecalls-corr / #CNbasecalls
Fraction of all C basecalls that are mCH: mCHfraction = mCAfraction + mCCfraction +
mCTfraction
Fig. S1G: Partially methylated domains were identified as described previously (22).
Fig. S2A,B
Overall, before correcting for the rate of false positives introduced by bisulfite nonconversion and failure of Tet1 oxidation, we measured 0.5262% non-converted cytosines
in TAB-Seq reads in the CH context in adult (6wk) mouse cortex. Our empirical control
measurement of non-conversion in unmethylated pUC19 DNA in the CH context yielded
a rate of 0.5245%, with a 99% confidence interval spanning 0.467 - 0.587% (based on fit
of a binomial distribution using the Clopper-Pearson method; MATLAB function
“binofit”). We thus estimate the global hmCH/CH to be 0.017%, with a 99% confidence
interval 0 - 0.059%.
To test whether individual CH sites contain statistically significant hmCH, we
developed a statistical model based on the null hypothesis that CH sites are composed of
1.4% mCH, 98.6% CH, and 0% hmCH. The non-conversion rate for unmethylated CH
sites was assumed to be 0.52%, whereas the sum of bisulfite non-conversion and Tet1
non-oxidation for mC sites was set at 2.08% based on measurement of pUC19 DNA
(Table S1). Our model assumes that the number of non-converted reads will follow a
mixture of two binomial distributions with these two rates. Based on this model we
calculated the p-value associated with the reads observed at each CH site across the
genome. We detect only 50 significant sites [FDR=0.01, Benjamini-Hochberg-Yekutieli
procedure (76)]. These sites were marginally significant, with hmCH/CH ~ 0.5.
Fig. S2C
Our base-resolution measurement of hmC (TAB-Seq) allowed us to examine the
relationship between the hydroxymethylation level (hmCG/CG) and the total methylation
level (mC/C as measured in bisulfite sequencing, which is the sum of the frequency of
hydroxymethylated and methylated sites). We restricted analysis to sites covered by at
least 10 reads in the TAB-Seq data set. We sorted each of these sites into ten equally
spaced bins according to the total level of methylation observed in bisulfite sequencing
(mC/C) of the same sample. Within each bin, we plot the proportion of sites that have
statistically significant 5hmC/C (uncorrected p<0.01, Fisher’s exact test). Significance
16
was assessed by comparing the number of hmC reads with the cumulative binomial
distribution with mean parameter given by the bisulfite non-conversion rate.
Fig. S3A and Table S2
For each of the three adult mouse brain cell-type specific methylomes (R1, R2, R3) we
used Fisher’s exact test to assess whether intragenic mCH/CH is larger in glia (NeuN–)
vs. neurons (NeuN+) for each gene. A final list of 174 glial hyper-mCH genes was
obtained by requiring a significant effect (FDR=0.05, Benjamini-Hochberg) in all three
comparisons.
Fig. S3B
We profiled mC/C around genes reported to be enriched in oligodendrocytes (45) and in
brain epithelial cells (78). The top oligodendrocyte genes were those with at least 20-fold
enrichment (45).
Fig. S3C
We repeated our analysis of methylation patterns around functional gene categories in
human brain samples by mapping each mouse gene to a human ortholog using the geneoriented ortholog database (GOOD) (77).
Fig. S8: Methods as in Fig. 4.
17
A
% mCC/CC
B
2
0
4
% mCT/CT 2
% mCC/CC
0
80
6
10
1
0
0
2.5
6
2.0
5
0
2
2
1
1
0
0
2
2
1
1
8
Position
CAT
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
NeuN+ 7wk
NeuN– 7wk
NeuN+ 6wk
NeuN– 6wk
NeuN+ 12mo
NeuN– 12mo
glia S100b+
Tet2-/-
CTT
0.6
mCG
0.5
0.4
0.3
0.2
0.1
0
6
20
4
2
0
0 2 CAA
CAC
CCT
CCA
CCC
CCG
10
for
0
12x10
8
Fraction of mCG sites
0
0
Fraction of mCH sites
1
0.1
n f IMR9
ibr
ob 0
la
H1 st
iPS ES
C- C
1
fet 9.11
al
25 brain
yr
bra
in
1
0.2
H
es
ki
2
Position
CTG
2
0 2 0.3
G
Total PMD length (bp)
Hypermethylated
All
Unmethylated
cytosines mC/C < 0.01
0
CTA
CTT
CAA
CAT
CAG
CTC
0
CAC
% mCHN/CHN
0
0.1
Hs NeuN+
2
1
0.2
F
3
0
D
0.3
E
4
0.5
CAG
0
1.0
CTC
0
fetal
35 do
2 yr
5 yr
12 yr
16 yr
25 yr
NeuN+ 53 yr
NeuN– 53 yr
55 yr
NeuN+ 55 yr
NeuN– 55 yr
64 yr
H1 ES
iPCS-19.11
IMR90 fibroblast
foreskin fibroblast
2
fetal
35 do
2 yr
5 yr
12 yr
16 yr
25 yr
NeuN+ 53 yr
NeuN– 53 yr
55 yr
NeuN+ 55 yr
NeuN– 55 yr
64 yr
H1 ES
iPCS-19.11
IMR90 fibroblast
foreskin fibroblast
1
% mC/C
4
1.5
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
NeuN+ 7wk
NeuN– 7wk
NeuN+ 6wk
NeuN– 6wk
NeuN+ 12mo
NeuN– 12mo
glia S100b+
Tet2-/-
2
6
Mouse NeuN+ 7wk
% mCHN/CHN
% mC/C
3
% mCH/CH
8
4
Human NeuN+ 53yr
% mCH/CH
5
20
CCT
0
2
CCA
0
3
CCG
2
40
CTA
4
4
60
CCC
6
CTG
30
1
0
5
8
% mCA/CA
60
% mCT/CT
% mCG/CG
% mCA/CA
% mCG/CG
10
Conservation (bits)
0
12
90
C
1
40
60
80 100
mCH
0.4
0.3
0.2
0.1
0
20
40
60
80 100
Methylation level
Hs 25yr frontal cortex
Hs 53yr NeuN+
Hs 55yr NeuN+
!"#$%&'$%
!(")#$%&'()&$*+,-.,/01$'*$23$45675/$)&&$016'85/,$()8,0)&&8"$0)6,9'+5:,;$(1$8,-.,/0,$0'/6,<6"$*'+$,)07$7.2)/$
)/;$2'.8,$8)2=&,>$!*"+#$?+,-.,/01$'*$23$5/$/,.+'/)&$8)2=&,8$45675/$,)07$3@A$0'/6,<6>$!,#$@.2)/$/,.+'/$
'+$!!#$'+$9&5)&$0,&&$8,-.,/0,$0'/8,+B)65'/$)6$7597&1C2,671&)6,;$)/;$./2,671&)6,;$3@$='8565'/8>$D;,'9+)28$87'4$
8,-.,/0,$ 0'/8,+B)65'/$ )/;$ 0'2='8565'/$ )6$ ='8565'/8$ *&)/E5/9$ 67,$ 2'86$ 7597&1$ 2,671&)6,;$ 3@$ 856,8$ !6'=$ +'4#"$
./2,671&)6,;$856,8$!25;;&,#"$'+$)&&$3@$='8565'/8$!('66'2#>$!-#$F'6)&$&,/967$'*$=)+65)&&1$2,671&)6,;$;'2)5/8$5/$
7.2)/$9,/'2,8>$!.#$G+'='+65'/$'*$2,671&)6,;$()8,8$5/$3%$'+$3@$0'/6,<6$67)6$)+,$2,671&)6,;$)6$)$95B,/$&,B,&$
5/$7.2)/$0'+6,<$)/;$/,.+'/$8)2=&,8>$$
HI
A
S8&FRQWURO
0
3ïYDOXH
10
'DWD
1XOOH[SHFWDWLRn
S %RQIHUURQL)
)'5 %+
)'5 %+<
ï
10
ï
10
ï
10
2
10
3
10
C
ZN7$%ï6HT
0
3ïYDOXH
10
ï
10
1
50 10
108
1XPEHURI&+VLWHV
1
(p<0.01, uncorrected)
B
1
10
)UDFWLRQRIVLWHVZLWK
VLJQLILFDQWKP&&
0
10
Fetal
6 wk
0.5
0
0
0.5
P&*&*
1
!"#$%&'%
!"#"$#%&'( &)( *+,-&.+/"#*+0$+#&1%'"( %'( #*"( 23( $&'#".#( 4+( 56789":;( <',( #*"( -"0<#%&'1*%=( 4"#>""'( /2?( <',(
*/2?(0"@"0A(B1%'C(1":D"'$%'C(&)(=B2EF($&'#-&0(!G6;(>"($<0%4-<#",(#*"(-<#"(&)()<01"(=&1%#%@"(*/2(4<1"$<001(
,D"(#&('&'8$&'@"-1%&'(&)()D00+(D'/"#*+0<#",(23(1%#"1(H-23IJAJJKLMN(<1(>"00(<1(#*"('&'8$&'@"-1%&'(<',O&-('&'8
&.%,<#%&'(&)()D00+(/"#*+0<#",(H/2M(2?(1%#"1(H-2?IJAJLJPMA(5*"1"(#>&(=<-</"#"-1(,")%'"(#*"(0%Q"0%*&&,()D'$#%&'(
)&-( #*"( ,<#<( D',"-( #*"( 'D00( *+=&#*"1%1A( H(M( 6'<0+1%1( &)( #*"( =B2EF( $&'#-&0( ,<#<( >%#*( /D0#%=0"8$&/=<-%1&'(
$&--"$#%&'( H7&')"--&'%;( 7"'R</%'%83&$*4"-C( H73M( H!"M( &-( 7"'R</%'%83&$*4"-C8S"QD#%"0%( H73SM( H!#MM( 1*&>1(
'&()<01"(=&1%#%@"1A(H)M(5*"(1</"(<'<0+1%1(<==0%",(#&(<,D0#(/&D1"(56789":(,<#<(-"1D0#1(%'(TKJ(1%C'%)%$<'#(1%#"1(
#*-&DC*&D#( #*"( C"'&/"A( G&#"( #*<#( #*"( 73( =-&$",D-"( <11D/"1( 1#<#%1#%$<0( %',"="',"'$"( </&'C( <00( 1%#"1;( <',( %1(
#*"-")&-"( <( 0"11( $&'1"-@<#%@"( <==-&<$*A( H*M( 5*"( )-<$#%&'( &)( /2?( 1%#"1( #*<#( 1*&>( <'+( 1%C'%)%$<'#( */2( 1%C'<0(
H=UJAJE;(4%'&/%<0(".<$#(#"1#M()-&/(#*"(56789":(,<#<;(C%@"'(<(1="$%)%$(0"@"0(&)(/2?A(%(
EF
Glial hyper-mCH (174)
Ne
Ne
tissue
uN
uN
+
–
tissue
+100 kb
-100 kb
+100 kb
10
1
0.1
5!
gene
3!
5!
mCG/CG
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
R1
R2
R3
glia
R1
R2
R3
-100 kb
gene
3!
Birth 10 100
$JHZNVSRVWïFRQFHSWLRQ
median
P51$ï6HT)3.0
mCH/CH
+100 kb
-100 kb
+100 kb
10
1
0.1
10
Top oligodendrocytes
(42)
Oligodendrocytes (540)
B
-100 kb
median
P51$ï6HT)3.0
mCH/CH
Ne
Ne
uN
uN
+
–
A
mCG/CG
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
R1
R2
R3
glia
R1
R2
R3
1
0.1
Epithelial
(131)
10
1
0.1
5!
3!
5!
mCG/CG
-100 kb
gene
3!
Birth 10 100
$JHZNVSRVWïFRQFHSWLRQ
mCH/CH
+100 kb
-100 kb
median
P51$ï6HT)3.0
+100 kb
10
1
Neuronal
0
1
mCG/CG
0
0.06
mCH/CH
<0.1
10
1
<0.1
Upregulated
Constitutively high
Ne Ne
tissue
uN uN
– +
C
H1 ESC
fetal
35 do
2 yr
5 yr
12 yr
16 yr
25 yr
64 yr
R1
R2
R1
R2
gene
10
1
Astrocyte
<0.1
10
1
Downregulated
<0.1
0 1 2
Normalized mCH/CH
1
<0.1
Consitutively
low
0.8 1 1.2
Normalized mCG/CG
10
10
1
<0.1
5!
gene
3!
5!
gene
3!
Birth 10 100
$JH\HDUVSRVWïFRQFHSWLRQ
!"#$%&'%
!"#$% !"&% '()*+,(*+'% ,-.-% /*01% 2.0% 342.56.,% 788% 5/$% 3*)% !*+9-% ,-.-9% '(2'% 9(*:% ;(<% ,4624% !"&%
(1=-)!-'(142'6*.%2.0%;)<%*46,*0-.0)*>1'-%2.0%-=6'(-4624%,-.-9?%@)2.9>)6='%2/+.02.>-%;ABCDE-F%GHIJ<%*K-)%
0-K-4*=!-.'%69%9(*:.%3*)%'(-%92!-%,-.-%9-'9?%;*<%!"#$%!"&%'()*+,(*+'%,-.-%/*01%2.0%342.56.,%788%5/$%3*)%
6.06>2'-0%,-.-%9-'9%6.%(+!2.?%@)2.9>)6='%2/+.02.>-%;ABCDE-F%GHIJ<%*K-)%0-K-4*=!-.'%69%9(*:.%3*)%'(-%92!-%
,-.-%9-'9?%
L8
CG
$XWRVRPHV
CH
CG
&KU;
CH
1
0.8
0.6
0.4
0.2
Ne
uN
+
vs
H1
ES
Ne
uN
+
vs
ES
Hu
ES
6
vs
ES
H1
Ne
uN
+
vs
Ne
uN
+
Ne
uN
+
vs
Ne
uN
+
ņ
!N
eu
N+
Ņ
!N
eu
N+
0
vs
3HUïVLWHFRUUHODWLRQP&&
,QGLYLGXDOUHSOLFDWHV
6KXIIOHG1RFRUUHODWLRQ
6LPXODWHG)XOOFRUUHODWLRQ
!"#$%&'%
"#$%&'(!)'%%#*$+,'(!)'#--,),#(+!'-!.#+/0*$+,'(!&+$+#!$+!&,(1*#!&,+#&!2#+3##(!(#4%'(&!$(5!67!)#**&!,(!/4.$(!$(5!
.'4&#8!9$+$!$%#!+/#!&$.#!$&!,(!:,18!;:<!24+!&/'3(!3,+/!('!('%.$*,=$+,'(!20!+/#!&,.4*$+#5!-4**!)'%%#*$+,'(!>$*4#8!
+/
7/4--*#5! 5$+$! $%#! %#?%#&#(+#5! 20! +/#! .#5,$(! '>#%! ;@@! %$(5'.! &/4--*#&! $(5! #%%'%! 2$%&! &/'3! +/#! ABCA !
?#%)#(+,*#&8!
;D
D
Same strand
Opposite strand
0.11
Intragenic mCG/CG
)HPDOHï0DOH
mCH/CH
A
0.1
0.09
0.08
ï
ï
ï
ï
0
200
400
600
0.2
0
-0.2
0
1-2 3-4 5-6 7-8
9
Inactivated
Escaped
%LïDOOHOLFH[SUHVVLRQ>ï@
800
Lag (bp)
C
0.1
0.09
E
Within-read
Across reads
Shuffled
0.2
mCH/CH
mCH/CH
0.11
0.1
0.08
50
0
100
25
Lag (bp)
50
Chr2 (1472)
X-inact.* (253)
X-escapee* (37)
X-escapee§ (33)
X-escapee† (7)
0.2
0
75
0.5
0.5
0.5
0.2
0.1
0
1
0
0DOHP&*&*
0.1
0.2
0DOHP&+&+
Chromosome X
1
0.5
0
0
0.5
H
1
0.3
0.2
0.1
0
0
0.1
0.2
0DOHP&+&+
0DOHP&*&*
1
100 10,000
Number of 1kb bins
Human female
53 yr NeuN+
1
0DOH1HX1
G
Female mCG/CG
Female mCH/CH
Female mCG/CG
1
0
0.4
Lag (bp)
Autosomes
0
0.6
Female mCH/CH
F
0.8
0
0
0
Intragenic mCG/CG
1
Female NeuN+
B
1 10 100 1,000
Number of 1kb bins
Human male
55 yr NeuN+
Fraction of mCH
1
0.5
mCA
mCT
mCC
W
ho
le
ge
&
C nom
KU
hr
e
;
2
g
ï
,Q ene
D
&
s
FW
KU
; LYD
WH
ï
(V G
FD
SH
H
W
ho
le
ge
&
C nom
KU
hr
e
;
2
g
ï
,Q ene
D
&
s
FW
KU
; LYD
WH
ï
(V G
FD
SH
H
0
!"#$%&'%
!()*"#$%&'()'*'+,#(-#./0#12)#./3#'2#45.12#6%567#*%889#!:;#,&#-%.18%"<#!+"#='+4'2>&%1)#15+(*(&&%81+'(2#(-#
./3?/3# !&%)"# '9# %2412*%)# &%81+'@%# +(# +4%# *(&&%81+'(2# (-# 95..%)# .%+4,81+'(2# A1++%&29# !B85%"<# !,"# C(D# 12)#
E4'9F%&# A8(+9# (-# G%2)%&>9A%*'-'*# )'--%&%2*%9# '2# '2+&1G%2'*# ./0# '2# '21*+'@1+%)# 12)# %9*1A%%# G%2%9# (2# 45.12#
*4&H<#!-"#I*1++%&A8(+#(-#G%2)%&#)'--%&%2*%9#'2#'2+&1G%2'*#./0#'2#45.12#*4&H#G%2%9<#J%A(&+%)#*4&H#'21*+'@1+%)#
M
O
12)#%9*1A%%#G%2%9#!K#/1&&%8#!"#$%&L# #I41&A#!"#$%&"N#A&%)'*+%)#%9*1A%%#G%2%9#! "N#12)#15+(9(.18#!/4&P"#G%2%9#1&%#
'2)'*1+%)<#!!"#./0#12)#./3#E'+4'2#Q#FB#B'29#'9#4'G48,#*(29'9+%2+#B%+E%%2#.18%#12)#-%.18%#45.129<#!."#/4&H#
./0#12)#./3#1&%#&%)5*%)#'2#-%.18%#*(.A1&%)#E'+4#.18%<#!/"#/4&H#'21*+'@1+%)#G%2%9#'2#-%.18%#2%5&(29#41@%#
1#98'G4+8,#8(E%&#-&1*+'(2#(-#./R#)'25*8%(+')%9<#
PP
Human adult NeuN+
(55 yr male; 53 yr female)
Female
1
0.5
C
Human ESC
(H1 male; HUES6 female)
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
1
0
0
0.5
0
1
0.05
0.05
0.04
0.04
0.03
0.03
0.02
0.02
0.01
0.01
0.5
1
Chr2 (1472)
X-inact.* (253)
X-escapee* (37)
X-escapee§ (33)
X-escapee† (7)
0.1
0.05
0
Female
Human adult NeuN–
(55 yr male; 53 yr female)
0
0
0
0
0
Intragenic mCG/CG
B
0.5
0
Female
Intragenic mCH/CH
Promoter mCG/CG
A
0.05
0.1
0
0.02
0
0.04
1
1
1
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0
0.5
Male
1
0.04
0
0
0
0.02
0
0.5
Male
1
0
0.5
1
Male
!"#$%&'%
!"#$"%&'(")*+*),-"./012.*3#,(2.."%#',2)%3'',)"11,.0("'4,521",6'4,+"-21",(%3-3."%,-7!87!,9.3(,%3:;<,="#*),
-7>87>,9-*$$1",%3:;,2#$,="#*),-7!87!,9?3..3-,%3:;,*#,/@-2#,2$@1.,+%3#.21,)3%."A,#"@%3#',9(;<,=1*2,9);<,
2#$,"-?%03#*),'."-,)"11,1*#"',9*;4,B*++"%"#.,'0-?31',)3%%"'(3#$,.3,7/%C,="#"',%"(3%."$,.3,?",*#2).*62."$,3%,.3,
F
H
"')2(",*#2).*62.*3#,9D,72%%"1,!"#$%&E, ,G/2%(,!"#$%4;<,(%"$*)."$,"')2("",="#"',9 <,./*','.@$0;<,2#$,2@.3'3-21,97/%I;,
="#"'4,J"@%3#',2#$,=1*2<,?@.,#3.,"-?%03#*),'."-,)"11'<,2%",/0("%-"./012."$,'(")*+*)2110,*#,+"-21",)/%C&"')2("",
="#"'4,
IK
A
Fraction of total data variance
1
0.8
0.6
0.4
0.2
0
100
101
102
103
Principal component (PC) rank
PC1
B
0
Max
Density of genes (a.u.)
PC2
PC2
ChrX genes (1175)
Neuronal genes (61)
Astrocytic genes (50)
PC3
PC3
PC4
!"#$%&'%
!"#$%#&'() %*+&*$,$-) .!/0) '$'(12#2) *3) +/) &'--,"$2) #$) 456478) +*92,) :,$,2;) .(0) /9+9('-#<,) 3"'%-#*$) *3) ='-')
<'"#'$%,) ,>&('#$,=) ?1) ,'%@) !/;) A@,) 3#"2-) 5) !/2) B,",) ",-'#$,=) 3*") %(92-,") '$'(12#2;) .)0) C%'--,") &(*-2) *3) -@,)
&"*D,%-#*$) *3) +/) &'--,"$) *3) ,'%@) :,$,) *$-*) -@,) 2&'%,) =,3#$,=) ?1) -B*) *3) -@,) &"#$%#&'() %*+&*$,$-2) .!/20;)
/*(*",=) &*#$-2) 2@*B) %@"*+*2*+,) E6) $,9"*$'() '$=) '2-"*%1-#%) :,$,26) #((92-"'-#$:) -@,) 2,:",:'-#*$) *3) :,$,) +/)
&'--,"$2)'%%*"=#$:)-*)39$%-#*$)'$=)",:9('-#*$;)
4F
mCG/CG
10 wk
fetal
fetal
-100 kb
1
2 wk
hmCG/CG
10 wk NeuN+ NeuN–
6 wk
+100 kb
mCH/CH
fetal
-100 kb
2 wk
B
10 wk NeuN+ NeuN–
+100 kb
Constitutively high
Neuronal
Upregulated
Astrocytic
Downregulated
Constitutively low
A
"
!
#
*
(
Genes
)
&
25,260
%
$
!'
0.1 1 10
P51$ï6HT)3.0
n.s.
0
1
mCG/CG
0
0.25
hmCG/CG
0
0.04
mCH/CH
0.1
10
Enrichment (q<0.05)
!"#$%&'%
!(")#$%&'$()*)$+,-.$/012$3$),'$4&-56$&','$50*&$)6$)74-89*'$!966-,.)80:'(#$4;)8'$+-,$.<=<2$
>?
A
mCG/CG
0
1
hmCG/CG
0
0.2
mCH/CH hmCH/CH
0
1 0
0.2
CGIs
Intragenic
Adult enh
Fetal enh
Fe
ta
7
wk 6 w l
Ne k
uN+
Fe
ta
7
wk 6 w l
Ne k
uN+
CGIs
Intragenic
Adult enh
Fetal enh
B
(h)mCH/CH
0.5
0.005
Fetal enh
Adult enh
Intragenic
CGIs
Fetal enh
Adult enh
Intragenic
1
2
3
53 yr
10
0
16 yr
10
1
5 yr
10
fetal
6 wk
10 wk
C
Fetal hmC
6 wk hmC
6 wk mC
0.01
0
CGIs
0
0.015
(h)mCG/CG
1
!"#$%&'%
!("#$%&'(#)*+,*+-#)*.,*.-#/)*+,*+#'01#/)*.,*.#20#3'4/#530%)24#6352%07#8'(93:#/';3#<330#4%6634&31#
=%6# &/3# :')>(3?:>342=24# 6'&3# %=# <2:9(=2&3# 0%0?4%0;36:2%07# !)"# @312'0# /)*# '01# )*# (3;3(# 20# 12==3630&# 530%)3#
&/
=3'&963:#!366%6#<'6:#AB?CD #>36430&2(3"7#/)*#1'&'#'63#19>(24'&31#=6%)#E257#FG7#!*"#$6'0:462>&#'<901'043#196205#
13;3(%>)30&#=%6#&/3#&/633#!"##=')2(H#5303:#)3':9631#<H#)IJK?L3M7#
BC
A
B
C
-100 kb
+100 kb
-100 kb
+100 kb
Gene
20,184
1
1
10 100 1000
0
P51$ï6HT
FPKM
% mCG/CG
0
25
% hmCG/CG
0
3
0.8
% mCH/CH
E
0.3
0.2
0.1
0
1.2
Normalized
mCG/CG
0.8
1.2
Normalized
hmCG/CG
0.8
1.2
Normalized
mCH/CH
0.3
Genic hmCG/CG
Genic hmCG/CG
D
1
100
0.2
10
0.1
1
Number of genes
0.1
0
0
0.5
Genic mCG/CG
(hmCG/CG subtracted)
1
0
0.02
Genic mCH/CH
0.04
!"#$%&'(%
!"#$%&'()*&+,-"%.""(,/0,$(1,*/0,$(1,2"(","3+4"))&'(5,6)7,84$()94&+%,$-:(1$(9",;'4,"$9*,';,<=>?@A,2"("),&(,
B, .C, /':)", ;4'(%$#, 9'4%"3, .&%*, "3+4"))&'(, #"D"#, EFGHI=5?5, 6*7, J-)'#:%", #"D"#, ';, /"%*K#$%&'(, 6/0LM0L>,
/0NM0N7,$(1,*K14'3K/"%*K#$%&'(,6*/0LM0L7,.&%*&(,"$9*,2"(",$(1,&(,;#$(C&(2,?==,C-,4"2&'()5,6+7,/0,$(1,
*/0,#"D"#,('4/$#&O"1,-K,%*",;#$(C&(2,4"2&'(5,6,>-7,8'%$#,&(%4$2"(&9,/0M0,&),9'/+$4"1,.&%*,*/0LM0L5,
<P
Median mCG/CG
1
0.6
0.4
0.2
0
NeuN- hmC
tissue
0.08
0.06
0.04
0.02
0
NeuN+ hyper mCG
NeuN+ hypo mCG
0.4
0.2
0
NeuN- hmC
tissue
0.8
1
0.6
NeuN+ hyper mCG
NeuN+ hypo mCG
R1
R2
NeuN+ NeuN-
NeuN+
138346
hyper-mCG
0
tissue
NeuN+ NeuN-
H1 ES
38649
1
R1
R2
NeuN+
48745
CG-DMRs
mCG/CG
R1
R2
fetal
35 do
2 yr
5 yr
12 yr
16 yr
25 yr
64 yr
42059
hypo-mCG
NeuN+
138346
hyper-mCG
NeuN+
57464
hyper-mCG
0
H1 ES
38649
CG-DMRs
10787
mCH/CH, hmCH/CH
R1
R2
fetal
35 do
2 yr
5 yr
12 yr
16 yr
25 yr
64 yr
NeuN+
48745
NeuN+
58445
CG-DMRs
0.07
42059
hypo-mCG
16139
hypo-mCG
0
R1
R2
R3
fetal
6 wk
NeuN+
glia
R1
R2
R3
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
R1
R2
R3
fetal
6 wk
NeuN+
57464
hyper-mCG
0.25
Median mCH/CH
10787
hmCG/CG
Median mCG/CG
NeuN+
glia
NeuN+
58445
CG-DMRs
R1
R2
R3
16139
hypo-mCG
mCG/CG
Median mCH/CH
0.8
1 0
fetal
1 wk
2 wk
4 wk
6 wk
10 wk
22 mo
tissue
0
mCH/CH
0.09
0.08
0.06
0.04
0.02
0
!"#$%&''
!"#$%#&'( )*( %+,( #-.( %+!( #$( +,/012'( 3."-$3*3".( 4"$5""-( -"67)-'8( 9:3#8( #-.( *7)-$#:( ;)7$"<( $=7)69=(
.">":)&%"-$(3-(%)6'"(#-.(=6%#-?(
@A
A
Gene set
B
Promoter (TSS±5kb)
CG-DMRs
CG-DMRs
fetal
Neuronal
adult
3
hmC
3
(CMS-IP)
Astrocyte
2
2
Downregulated
14x10-8
Consitutively low
10x10-8
(biotin-gluc)
10
1HX1ï
NeuN+
1HX1ï
hmC
NeuN+
n.s.
0.25
1
4
Fold enrichment
fetal
adult
4
Upregulated
Adult > Fetal
Fetal > Adult
Constitutively high
6x10
hyper- hypomCG mCG
8
6x10-8
-8
-2 -1 0 +1 +2
-2 -1 0 +1 +2
Distance (kb)
Distance (kb)
1
0
1
0
1
WT (6 wk)
0
1
1
0
1
WT (6 wk)
4 x 10-3
2
Tet2-/-
WT (4 wk,10 wk,22 mo)
Fetal
Fetal > Adult
1
WT (6 wk)
1
WT (6 wk)
WT (6 wk)
0
1
Tet2-/-
1
0
Fraction of DMRs
(p<0.01)
0
WT (4 wk,10 wk,22 mo)
1
Fetal
Adult > Fetal
C
1
WT (6 wk)
!"#$%&'(%
!"#$%&'(")*+(,-()$."/.0/1234567/$"/8)/)&(/)#8"7%#$,)$."8-/7)8#)/7$)(/.0/%(--3)9,(/7,(%$0$%/8"+/+(:(-.,'(")8--9/
+9"8'$%/ ;("(7/ $"/ '.<7(=/ >)?/ @<8")$0$%8)$."/ .0/ -.%8-/ ("#$%&'(")/ .0/ &'1/ 8)/ +(:(-.,'(")8--9/ +9"8'$%/ 123
4567/A9/&'1/("#$%&'(")/'()&.+.-.;$(7=/>*?/4("7$)9/'8,7/7&.B$";/8--/+(:(-.,'(")8-/1234567/>-(0)3&8"+/
,-.)7?C/.#/."-9/)&.7(/B&$%&/&8:(/7$;"$0$%8")-9/'.#(/.#/-(77/'12*12/%.',8#(+/B$)&/8/B$-+3)9,(/8+<-)/A#8$"/>D/
%&%
BE?=/ F&(/ !"#$ / '<)8")/ '$%(/ &8:(/ $"%#(87(+/ '12*12/ %.',8#(+/ B$)&/ GFC/ ,8#)$%<-8#-9/ 0.#/ H()8-IJ+<-)/ 123
4567C/$"+$%8)$";/F()3+(,("+(")/+('()&9-8)$."/8)/)&(7(/7$)(7/+<#$";/+(:(-.,'(")=/
KL
Table S1 (separate file)
Sample information and sequencing metrics.
Table S2 (separate file)
mCH deserts in human and mouse and the genes located within these domains.
Table S3 (separate file)
Glial mCH-hypermethylated genes.
Table S4 (separate file)
Gender-specific methylation on human X chromosome.
Table S5 (separate file)
Sample demographics.
30
References and Notes
1. E. Borrelli, E. J. Nestler, C. D. Allis, P. Sassone-Corsi, Decoding the epigenetic
language of neuronal plasticity. Neuron 60, 961–974 (2008).
doi:10.1016/j.neuron.2008.10.012 Medline
2. J. Feng, Y. Zhou, S. L. Campbell, T. Le, E. Li, J. D. Sweatt, A. J. Silva, G. Fan, Dnmt1
and Dnmt3a maintain DNA methylation and regulate synaptic function in adult
forebrain neurons. Nat. Neurosci. 13, 423–430 (2010). doi:10.1038/nn.2514
Medline
3. C. A. Miller, J. D. Sweatt, Covalent modification of DNA regulates memory
formation. Neuron 53, 857–869 (2007). doi:10.1016/j.neuron.2007.02.022
Medline
4. S. Numata, T. Ye, T. M. Hyde, X. Guitart-Navarro, R. Tao, M. Wininger, C.
Colantuoni, D. R. Weinberger, J. E. Kleinman, B. K. Lipska, DNA methylation
signatures in development and aging of the human prefrontal cortex. Am. J. Hum.
Genet. 90, 260–272 (2012). doi:10.1016/j.ajhg.2011.12.020 Medline
5. M. Suderman, P. O. McGowan, A. Sasaki, T. C. Huang, M. T. Hallett, M. J. Meaney,
G. Turecki, M. Szyf, Conserved epigenetic sensitivity to early life experience in
the rat and human hippocampus. Proc. Natl. Acad. Sci. U.S.A. 109, (Suppl 2),
17266–17272 (2012). doi:10.1073/pnas.1121260109 Medline
6. K. D. Siegmund, C. M. Connor, M. Campan, T. I. Long, D. J. Weisenberger, D.
Biniszkiewicz, R. Jaenisch, P. W. Laird, S. Akbarian, DNA methylation in the
human cerebral cortex is dynamically regulated throughout the life span and
involves differentiated neurons. PLoS ONE 2, e895 (2007).
doi:10.1371/journal.pone.0000895 Medline
7. K. Martinowich, D. Hattori, H. Wu, S. Fouse, F. He, Y. Hu, G. Fan, Y. E. Sun, DNA
methylation-related chromatin remodeling in activity-dependent BDNF gene
regulation. Science 302, 890–893 (2003). doi:10.1126/science.1090842 Medline
8. J. U. Guo, D. K. Ma, H. Mo, M. P. Ball, M. H. Jang, M. A. Bonaguidi, J. A. Balazer,
H. L. Eaves, B. Xie, E. Ford, K. Zhang, G. L. Ming, Y. Gao, H. Song, Neuronal
activity modifies the DNA methylation landscape in the adult brain. Nat.
Neurosci. 14, 1345–1351 (2011). doi:10.1038/nn.2900 Medline
9. I. B. Zovkic, M. C. Guzman-Karlsson, J. D. Sweatt, Epigenetic regulation of memory
formation and maintenance. Learn. Mem. 20, 61–74 (2013).
doi:10.1101/lm.026575.112 Medline
10. J. M. Levenson, T. L. Roth, F. D. Lubin, C. A. Miller, I. C. Huang, P. Desai, L. M.
Malone, J. D. Sweatt, Evidence that DNA (cytosine-5) methyltransferase
regulates synaptic plasticity in the hippocampus. J. Biol. Chem. 281, 15763–
15773 (2006). doi:10.1074/jbc.M511767200 Medline
11. A. M. M. Oliveira, T. J. Hemstedt, H. Bading, Rescue of aging-associated decline in
Dnmt3a2 expression restores cognitive abilities. Nat. Neurosci. 15, 1111–1113
(2012). doi:10.1038/nn.3151 Medline
12. P. Moretti, J. M. Levenson, F. Battaglia, R. Atkinson, R. Teague, B. Antalffy, D.
Armstrong, O. Arancio, J. D. Sweatt, H. Y. Zoghbi, Learning and memory and
synaptic plasticity are impaired in a mouse model of Rett syndrome. J. Neurosci.
26, 319–327 (2006). doi:10.1523/JNEUROSCI.2623-05.2006 Medline
13. N. W. Penn, R. Suwalski, C. O’Riley, K. Bojanowski, R. Yura, The presence of 5hydroxymethylcytosine in animal deoxyribonucleic acid. Biochem. J. 126, 781–
790 (1972). Medline
14. N. W. Penn, Modification of brain deoxyribonucleic acid base content with
maturation in normal and malnourished rats. Biochem. J. 155, 709–712 (1976).
Medline
15. D. Globisch, M. Münzel, M. Müller, S. Michalakis, M. Wagner, S. Koch, T. Brückl,
M. Biel, T. Carell, Tissue distribution of 5-hydroxymethylcytosine and search for
active demethylation intermediates. PLoS ONE 5, e15367 (2010).
doi:10.1371/journal.pone.0015367 Medline
16. J. U. Guo, Y. Su, C. Zhong, G.-L. Ming, H. Song, Hydroxylation of 5-methylcytosine
by TET1 promotes active DNA demethylation in the adult brain. Cell 145, 423–
434 (2011). doi:10.1016/j.cell.2011.03.022 Medline
17. Y.-F. He, B. Z. Li, Z. Li, P. Liu, Y. Wang, Q. Tang, J. Ding, Y. Jia, Z. Chen, L. Li, Y.
Sun, X. Li, Q. Dai, C. X. Song, K. Zhang, C. He, G. L. Xu, Tet-mediated
formation of 5-carboxylcytosine and its excision by TDG in mammalian DNA.
Science 333, 1303–1307 (2011). doi:10.1126/science.1210944 Medline
18. C.-X. Song, K. E. Szulwach, Q. Dai, Y. Fu, S. Q. Mao, L. Lin, C. Street, Y. Li, M.
Poidevin, H. Wu, J. Gao, P. Liu, L. Li, G. L. Xu, P. Jin, C. He, Genome-wide
profiling of 5-formylcytosine reveals its roles in epigenetic priming. Cell 153,
678–691 (2013). doi:10.1016/j.cell.2013.04.001 Medline
19. L. Shen, H. Wu, D. Diep, S. Yamaguchi, A. C. D’Alessio, H. L. Fung, K. Zhang, Y.
Zhang, Genome-wide analysis reveals TET- and TDG-dependent 5methylcytosine oxidation dynamics. Cell 153, 692–706 (2013).
doi:10.1016/j.cell.2013.04.002 Medline
20. W. Xie, C. L. Barr, A. Kim, F. Yue, A. Y. Lee, J. Eubanks, E. L. Dempster, B. Ren,
Base-resolution analyses of sequence and parent-of-origin dependent DNA
methylation in the mouse genome. Cell 148, 816–831 (2012).
doi:10.1016/j.cell.2011.12.035 Medline
21. K. E. Varley, J. Gertz, K. M. Bowling, S. L. Parker, T. E. Reddy, F. Pauli-Behn, M.
K. Cross, B. A. Williams, J. A. Stamatoyannopoulos, G. E. Crawford, D. M.
Absher, B. J. Wold, R. M. Myers, Dynamic DNA methylation across diverse
human cell lines and tissues. Genome Res. 23, 555–567 (2013).
doi:10.1101/gr.147942.112 Medline
22. R. Lister, M. Pelizzola, R. H. Dowen, R. D. Hawkins, G. Hon, J. Tonti-Filippini, J. R.
Nery, L. Lee, Z. Ye, Q. M. Ngo, L. Edsall, J. Antosiewicz-Bourget, R. Stewart, V.
Ruotti, A. H. Millar, J. A. Thomson, B. Ren, J. R. Ecker, Human DNA
methylomes at base resolution show widespread epigenomic differences. Nature
462, 315–322 (2009). doi:10.1038/nature08514 Medline
23. M. J. Ziller, F. Müller, J. Liao, Y. Zhang, H. Gu, C. Bock, P. Boyle, C. B. Epstein, B.
E. Bernstein, T. Lengauer, A. Gnirke, A. Meissner, Genomic distribution and
inter-sample variation of non-CpG methylation across human cell types. PLoS
Genet. 7, e1002389 (2011). doi:10.1371/journal.pgen.1002389 Medline
24. R. Lister, R. C. O’Malley, J. Tonti-Filippini, B. D. Gregory, C. C. Berry, A. H.
Millar, J. R. Ecker, Highly integrated single-base resolution maps of the
epigenome in Arabidopsis. Cell 133, 523–536 (2008).
doi:10.1016/j.cell.2008.03.029 Medline
25. R. Lister, M. Pelizzola, Y. S. Kida, R. D. Hawkins, J. R. Nery, G. Hon, J.
Antosiewicz-Bourget, R. O’Malley, R. Castanon, S. Klugman, M. Downes, R.
Yu, R. Stewart, B. Ren, J. A. Thomson, R. M. Evans, J. R. Ecker, Hotspots of
aberrant epigenomic reprogramming in human induced pluripotent stem cells.
Nature 471, 68–73 (2011). doi:10.1038/nature09798 Medline
26. B. H. Ramsahoye, D. Biniszkiewicz, F. Lyko, V. Clark, A. P. Bird, R. Jaenisch, NonCpG methylation is prevalent in embryonic stem cells and may be mediated by
DNA methyltransferase 3a. Proc. Natl. Acad. Sci. U.S.A. 97, 5237–5242 (2000).
doi:10.1073/pnas.97.10.5237 Medline
27. J. De Felipe, P. Marco, A. Fairén, E. G. Jones, Inhibitory synaptogenesis in mouse
somatosensory cortex. Cereb. Cortex 7, 619–634 (1997).
doi:10.1093/cercor/7.7.619 Medline
28. P. R. Huttenlocher, A. S. Dabholkar, Regional differences in synaptogenesis in
human cerebral cortex. J. Comp. Neurol. 387, 167–178 (1997).
doi:10.1002/(SICI)1096-9861(19971020)387:2<167::AID-CNE1>3.0.CO;2-Z
Medline
29. B. P. Berman, D. J. Weisenberger, J. F. Aman, T. Hinoue, Z. Ramjan, Y. Liu, H.
Noushmehr, C. P. Lange, C. M. van Dijk, R. A. Tollenaar, D. Van Den Berg, P.
W. Laird, Regions of focal DNA hypermethylation and long-range
hypomethylation in colorectal cancer coincide with nuclear lamina-associated
domains. Nat. Genet. 44, 40–46 (2012). doi:10.1038/ng.969 Medline
30. S. Kriaucionis, N. Heintz, The nuclear DNA base 5-hydroxymethylcytosine is present
in Purkinje neurons and the brain. Science 324, 929–930 (2009).
doi:10.1126/science.1169786 Medline
31. C.-X. Song, K. E. Szulwach, Y. Fu, Q. Dai, C. Yi, X. Li, Y. Li, C. H. Chen, W.
Zhang, X. Jian, J. Wang, L. Zhang, T. J. Looney, B. Zhang, L. A. Godley, L. M.
Hicks, B. T. Lahn, P. Jin, C. He, Selective chemical labeling reveals the genomewide distribution of 5-hydroxymethylcytosine. Nat. Biotechnol. 29, 68–72 (2011).
doi:10.1038/nbt.1732 Medline
32. K. E. Szulwach, X. Li, Y. Li, C. X. Song, H. Wu, Q. Dai, H. Irier, A. K. Upadhyay,
M. Gearing, A. I. Levey, A. Vasanthakumar, L. A. Godley, Q. Chang, X. Cheng,
C. He, P. Jin, 5-hmC-mediated epigenetic dynamics during postnatal
neurodevelopment and aging. Nat. Neurosci. 14, 1607–1616 (2011).
doi:10.1038/nn.2959 Medline
33. M. Mellén, P. Ayata, S. Dewell, S. Kriaucionis, N. Heintz, MeCP2 binds to 5hmC
enriched within active genes and accessible chromatin in the nervous system. Cell
151, 1417–1430 (2012). doi:10.1016/j.cell.2012.11.022 Medline
34. M. Yu, G. C. Hon, K. E. Szulwach, C. X. Song, L. Zhang, A. Kim, X. Li, Q. Dai, Y.
Shen, B. Park, J. H. Min, P. Jin, B. Ren, C. He, Base-resolution analysis of 5hydroxymethylcytosine in the mammalian genome. Cell 149, 1368–1380 (2012).
doi:10.1016/j.cell.2012.04.027 Medline
35. J. A. Stamatoyannopoulos, M. Snyder, R. Hardison, B. Ren, T. Gingeras, D. M.
Gilbert, M. Groudine, M. Bender, R. Kaul, T. Canfield, E. Giste, A. Johnson, M.
Zhang, G. Balasundaram, R. Byron, V. Roach, P. J. Sabo, R. Sandstrom, A. S.
Stehling, R. E. Thurman, S. M. Weissman, P. Cayting, M. Hariharan, J. Lian, Y.
Cheng, S. G. Landt, Z. Ma, B. J. Wold, J. Dekker, G. E. Crawford, C. A. Keller,
W. Wu, C. Morrissey, S. A. Kumar, T. Mishra, D. Jain, M. Byrska-Bishop, D.
Blankenberg, B. R. Lajoie, G. Jain, A. Sanyal, K.-B. Chen, O. Denas, J. Taylor,
G. A. Blobel, M. J. Weiss, M. Pimkin, W. Deng, G. K. Marinov, B. A. Williams,
K. I. Fisher-Aylor, G. Desalvo, A. Kiralusha, D. Trout, H. Amrhein, A.
Mortazavi, L. Edsall, D. McCleary, S. Kuan, Y. Shen, F. Yue, Z. Ye, C. A. Davis,
C. Zaleski, S. Jha, C. Xue, A. Dobin, W. Lin, M. Fastuca, H. Wang, R. Guigo, S.
Djebali, J. Lagarde, T. Ryba, T. Sasaki, V. S. Malladi, M. S. Cline, V. M. Kirkup,
K. Learned, K. R. Rosenbloom, W. J. Kent, E. A. Feingold, P. J. Good, M. Pazin,
R. F. Lowdon, L. B. Adams, An encyclopedia of mouse DNA elements (Mouse
ENCODE). Genome Biol. 13, 418 (2012). doi:10.1186/gb-2012-13-8-418
Medline
36. R. K. Auerbach, G. Euskirchen, J. Rozowsky, N. Lamarre-Vincent, Z. Moqtaderi, P.
Lefrançois, K. Struhl, M. Gerstein, M. Snyder, Mapping accessible chromatin
regions using Sono-Seq. Proc. Natl. Acad. Sci. U.S.A. 106, 14926–14931 (2009).
doi:10.1073/pnas.0905443106 Medline
37. Y. Shen, F. Yue, D. F. McCleary, Z. Ye, L. Edsall, S. Kuan, U. Wagner, J. Dixon, L.
Lee, V. V. Lobanenkov, B. Ren, A map of the cis-regulatory sequences in the
mouse genome. Nature 488, 116–120 (2012). doi:10.1038/nature11243 Medline
38. W. A. Pastor, U. J. Pape, Y. Huang, H. R. Henderson, R. Lister, M. Ko, E. M.
McLoughlin, Y. Brudno, S. Mahapatra, P. Kapranov, M. Tahiliani, G. Q. Daley,
X. S. Liu, J. R. Ecker, P. M. Milos, S. Agarwal, A. Rao, Genome-wide mapping
of 5-hydroxymethylcytosine in embryonic stem cells. Nature 473, 394–397
(2011). doi:10.1038/nature10102 Medline
39. Y. Huang, W. A. Pastor, J. A. Zepeda-Martínez, A. Rao, The anti-CMS technique for
genome-wide mapping of 5-hydroxymethylcytosine. Nat. Protoc. 7, 1897–1908
(2012). doi:10.1038/nprot.2012.103 Medline
40. A. Magklara, A. Yen, B. M. Colquitt, E. J. Clowney, W. Allen, E. MarkenscoffPapadimitriou, Z. A. Evans, P. Kheradpour, G. Mountoufaris, C. Carey, G.
Barnea, M. Kellis, S. Lomvardas, An epigenetic signature for monoallelic
olfactory receptor expression. Cell 145, 555–570 (2011).
doi:10.1016/j.cell.2011.03.040 Medline
41. E. J. Clowney, M. A. LeGros, C. P. Mosley, F. G. Clowney, E. C. MarkenskoffPapadimitriou, M. Myllys, G. Barnea, C. A. Larabell, S. Lomvardas, Nuclear
aggregation of olfactory receptor genes governs their monogenic expression. Cell
151, 724–737 (2012). doi:10.1016/j.cell.2012.09.043 Medline
42. H. Li, J. C. Radford, M. J. Ragusa, K. L. Shea, S. R. McKercher, J. D. Zaremba, W.
Soussou, Z. Nie, Y. J. Kang, N. Nakanishi, S. Okamoto, A. J. Roberts, J. J.
Schwarz, S. A. Lipton, Transcription factor MEF2C influences neural
stem/progenitor cell differentiation and maturation in vivo. Proc. Natl. Acad. Sci.
U.S.A. 105, 9397–9402 (2008). doi:10.1073/pnas.0802876105 Medline
43. M. W. Akhtar, M. S. Kim, M. Adachi, M. J. Morris, X. Qi, J. A. Richardson, R.
Bassel-Duby, E. N. Olson, E. T. Kavalali, L. M. Monteggia, In vivo analysis of
MEF2 transcription factors in synapse regulation and neuronal survival. PLoS
ONE 7, e34863 (2012). doi:10.1371/journal.pone.0034863 Medline
44. A. C. Barbosa, M. S. Kim, M. Ertunc, M. Adachi, E. D. Nelson, J. McAnally, J. A.
Richardson, E. T. Kavalali, L. M. Monteggia, R. Bassel-Duby, E. N. Olson,
MEF2C, a transcription factor that facilitates learning and memory by negative
regulation of synapse numbers and function. Proc. Natl. Acad. Sci. U.S.A. 105,
9391–9396 (2008). doi:10.1073/pnas.0802679105 Medline
45. J. D. Cahoy, B. Emery, A. Kaushal, L. C. Foo, J. L. Zamanian, K. S. Christopherson,
Y. Xing, J. L. Lubischer, P. A. Krieg, S. A. Krupenko, W. J. Thompson, B. A.
Barres, A transcriptome database for astrocytes, neurons, and oligodendrocytes: a
new resource for understanding brain development and function. J. Neurosci. 28,
264–278 (2008). doi:10.1523/JNEUROSCI.4178-07.2008 Medline
46. H. Wu, V. Coskun, J. Tao, W. Xie, W. Ge, K. Yoshikawa, E. Li, Y. Zhang, Y. E.
Sun, Dnmt3a-dependent nonpromoter DNA methylation facilitates transcription
of neurogenic genes. Science 329, 444–448 (2010). doi:10.1126/science.1190485
Medline
47. T. Mohandas, R. S. Sparkes, L. J. Shapiro, Reactivation of an inactive human X
chromosome: evidence for X inactivation by DNA methylation. Science 211,
393–396 (1981). doi:10.1126/science.6164095 Medline
48. K. Plath, S. Mlynarczyk-Evans, D. A. Nusinow, B. Panning, Xist RNA and the
mechanism of X chromosome inactivation. Annu. Rev. Genet. 36, 233–278
(2002). doi:10.1146/annurev.genet.36.042902.092433 Medline
49. L. Carrel, H. F. Willard, X-inactivation profile reveals extensive variability in Xlinked gene expression in females. Nature 434, 400–404 (2005).
doi:10.1038/nature03479 Medline
50. A. J. Sharp, E. Stathaki, E. Migliavacca, M. Brahmachary, S. B. Montgomery, Y.
Dupre, S. E. Antonarakis, DNA methylation profiles of human active and inactive
X chromosomes. Genome Res. 21, 1592–1600 (2011). doi:10.1101/gr.112680.110
Medline
51. C. M. Johnston, F. L. Lovell, D. A. Leongamornlert, B. E. Stranger, E. T.
Dermitzakis, M. T. Ross, Large-scale population study of human cell lines
indicates that dosage compensation is virtually complete. PLoS Genet. 4, e9
(2008). doi:10.1371/journal.pgen.0040009 Medline
52. A. K. Maunakea, R. P. Nagarajan, M. Bilenky, T. J. Ballinger, C. D’Souza, S. D.
Fouse, B. E. Johnson, C. Hong, C. Nielsen, Y. Zhao, G. Turecki, A. Delaney, R.
Varhol, N. Thiessen, K. Shchors, V. M. Heine, D. H. Rowitch, X. Xing, C. Fiore,
M. Schillebeeckx, S. J. Jones, D. Haussler, M. A. Marra, M. Hirst, T. Wang, J. F.
Costello, Conserved role of intragenic DNA methylation in regulating alternative
promoters. Nature 466, 253–257 (2010). doi:10.1038/nature09165 Medline
53. R. E. Thurman, E. Rynes, R. Humbert, J. Vierstra, M. T. Maurano, E. Haugen, N. C.
Sheffield, A. B. Stergachis, H. Wang, B. Vernot, K. Garg, S. John, R. Sandstrom,
D. Bates, L. Boatman, T. K. Canfield, M. Diegel, D. Dunn, A. K. Ebersol, T.
Frum, E. Giste, A. K. Johnson, E. M. Johnson, T. Kutyavin, B. Lajoie, B. K. Lee,
K. Lee, D. London, D. Lotakis, S. Neph, F. Neri, E. D. Nguyen, H. Qu, A. P.
Reynolds, V. Roach, A. Safi, M. E. Sanchez, A. Sanyal, A. Shafer, J. M. Simon,
L. Song, S. Vong, M. Weaver, Y. Yan, Z. Zhang, Z. Zhang, B. Lenhard, M.
Tewari, M. O. Dorschner, R. S. Hansen, P. A. Navas, G. Stamatoyannopoulos, V.
R. Iyer, J. D. Lieb, S. R. Sunyaev, J. M. Akey, P. J. Sabo, R. Kaul, T. S. Furey, J.
Dekker, G. E. Crawford, J. A. Stamatoyannopoulos, The accessible chromatin
landscape of the human genome. Nature 489, 75–82 (2012).
doi:10.1038/nature11232 Medline
54. M. B. Stadler, R. Murr, L. Burger, R. Ivanek, F. Lienert, A. Schöler, E. van
Nimwegen, C. Wirbelauer, E. J. Oakeley, D. Gaidatzis, V. K. Tiwari, D.
Schübeler, DNA-binding factors shape the mouse methylome at distal regulatory
regions. Nature 480, 490–495 (2011). Medline
55. Z. Petanjek, M. Judaš, G. Šimic, M. R. Rasin, H. B. Uylings, P. Rakic, I. Kostovic,
Extraordinary neoteny of synaptic spines in the human prefrontal cortex. Proc.
Natl. Acad. Sci. U.S.A. 108, 13281–13286 (2011). doi:10.1073/pnas.1105108108
Medline
56. B. Kolb, R. Mychasiuk, A. Muhammad, Y. Li, D. O. Frost, R. Gibb, Experience and
the developing prefrontal cortex. Proc. Natl. Acad. Sci. U.S.A. 109, (Suppl 2),
17186–17193 (2012). doi:10.1073/pnas.1121251109 Medline
57. H. T. Chugani, A critical period of brain development: studies of cerebral glucose
utilization with PET. Prev. Med. 27, 184–188 (1998).
doi:10.1006/pmed.1998.0274 Medline
58. P. Levitt, Structural and functional maturation of the developing primate brain. J.
Pediatr. 143, (Suppl), S35–S45 (2003). doi:10.1067/S0022-3476(03)00400-1
Medline
59. M. V. Johnston, Plasticity in the developing brain: implications for rehabilitation.
Dev. Disabil. Res. Rev. 15, 94–101 (2009). doi:10.1002/ddrr.64 Medline
60. P. J. Uhlhaas, W. Singer, The development of neural synchrony and large-scale
cortical networks during adolescence: relevance for the pathophysiology of
schizophrenia and neurodevelopmental hypothesis. Schizophr. Bull. 37, 514–523
(2011). doi:10.1093/schbul/sbr034 Medline
61. S. Nguyen, K. Meletis, D. Fu, S. Jhaveri, R. Jaenisch, Ablation of de novo DNA
methyltransferase Dnmt3a in the nervous system leads to neuromuscular defects
and shortened lifespan. Dev. Dyn. 236, 1663–1676 (2007).
doi:10.1002/dvdy.21176 Medline
62. Materials and methods are available as supplementary material on Science Online.
63. M. Ko, H. S. Bandukwala, J. An, E. D. Lamperti, E. C. Thompson, R. Hastie, A.
Tsangaratou, K. Rajewsky, S. B. Koralov, A. Rao, Ten-Eleven-Translocation 2
(TET2) negatively regulates homeostasis and differentiation of hematopoietic
stem cells in mice. Proc. Natl. Acad. Sci. U.S.A. 108, 14566–14571 (2011).
doi:10.1073/pnas.1112317108 Medline
64. J. W. Kinney, C. N. Davis, I. Tabarean, B. Conti, T. Bartfai, M. M. Behrens, A
specific role for NR2A-containing NMDA receptors in the maintenance of
parvalbumin and GAD67 immunoreactivity in cultured interneurons. J. Neurosci.
26, 1604–1615 (2006). doi:10.1523/JNEUROSCI.4722-05.2006 Medline
65. Y. Jiang, A. Matevossian, H.-S. Huang, J. Straubhaar, S. Akbarian, Isolation of
neuronal chromatin from brain tissue. BMC Neurosci. 9, 42 (2008).
doi:10.1186/1471-2202-9-42 Medline
66. S. Zhong, J. G. Joung, Y. Zheng, Y. R. Chen, B. Liu, Y. Shao, J. Z. Xiang, Z. Fei, J.
J. Giovannoni, High-throughput illumina strand-specific RNA sequencing library
preparation. Cold Spring Harb. Protoc. 2011, 940–949 (2011).
doi:10.1101/pdb.prot5652 Medline
67. C. Trapnell, D. G. Hendrickson, M. Sauvageau, L. Goff, J. L. Rinn, L. Pachter,
Differential analysis of gene regulation at transcript resolution with RNA-seq.
Nat. Biotechnol. 31, 46–53 (2013). doi:10.1038/nbt.2450 Medline
68. C. Trapnell, A. Roberts, L. Goff, G. Pertea, D. Kim, D. R. Kelley, H. Pimentel, S. L.
Salzberg, J. L. Rinn, L. Pachter, Differential gene and transcript expression
analysis of RNA-seq experiments with TopHat and Cufflinks. Nat. Protoc. 7,
562–578 (2012). doi:10.1038/nprot.2012.016 Medline
69. W. Perkins, M. Tygert, R. Ward, Computing the confidence levels for a root-meansquare test of goodness-of-fit. Appl. Math. Comput. 217, 9072–9084 (2011).
doi:10.1016/j.amc.2011.03.124
70. T. Bancroft, C. Du, D. Nettleton, Estimation of false discovery rate using sequential
permutation p-values. Biometrics 69, 1–7 (2013). doi:10.1111/j.15410420.2012.01825.x Medline
71. B. Langmead, C. Trapnell, M. Pop, S. L. Salzberg, Ultrafast and memory-efficient
alignment of short DNA sequences to the human genome. Genome Biol. 10, R25
(2009). doi:10.1186/gb-2009-10-3-r25 Medline
72. Y. Zhang, T. Liu, C. A. Meyer, J. Eeckhoute, D. S. Johnson, B. E. Bernstein, C.
Nusbaum, R. M. Myers, M. Brown, W. Li, X. S. Liu, Model-based analysis of
ChIP-Seq (MACS). Genome Biol. 9, R137 (2008). doi:10.1186/gb-2008-9-9-r137
Medline
73. A. S. Hinrichs, D. Karolchik, R. Baertsch, G. P. Barber, G. Bejerano, H. Clawson, M.
Diekhans, T. S. Furey, R. A. Harte, F. Hsu, J. Hillman-Jackson, R. M. Kuhn, J. S.
Pedersen, A. Pohl, B. J. Raney, K. R. Rosenbloom, A. Siepel, K. E. Smith, C. W.
Sugnet, A. Sultan-Qurraie, D. J. Thomas, H. Trumbower, R. J. Weber, M.
Weirauch, A. S. Zweig, D. Haussler, W. J. Kent, The UCSC Genome Browser
Database: update 2006. Nucleic Acids Res. 34, D590–D598 (2006).
doi:10.1093/nar/gkj144 Medline
74. I. J. Cajigas, G. Tushev, T. J. Will, S. tom Dieck, N. Fuerst, E. M. Schuman, The
local transcriptome in the synaptic neuropil revealed by deep sequencing and
high-resolution imaging. Neuron 74, 453–466 (2012).
doi:10.1016/j.neuron.2012.02.036 Medline
75. Y. Benjamini, Y. Hochberg, J. R. Stat. Soc., B 57, 289 (1995).
76. Y. Benjamini, D. Yekutieli, The control of the false discovery rate in multiple testing
under dependency. Ann. Stat. 29, 1165–1188 (2001).
doi:10.1214/aos/1013699998
77. M.-R. Ho, C.-H. Chen, W.-C. Lin, Gene-oriented ortholog database: a functional
comparison platform for orthologous loci. Database (Oxford) 2010, baq002
(2010). doi:10.1093/database/baq002 Medline
78. R. Daneman, L. Zhou, D. Agalliu, J. D. Cahoy, A. Kaushal, B. A. Barres, The mouse
blood-brain barrier transcriptome: a new resource for understanding the
development and function of brain endothelial cells. PLoS ONE 5, e13741 (2010).
doi:10.1371/journal.pone.0013741 Medline