Genome-Wide DNA Methylation Scan in Major Depressive Disorder

Abstract

While genome-wide association studies are ongoing to identify sequence variation influencing susceptibility to major depressive disorder (MDD), epigenetic marks, such as DNA methylation, which can be influenced by environment, might also play a role. Here we present the first genome-wide DNA methylation (DNAm) scan in MDD. We compared 39 postmortem frontal cortex MDD samples to 26 controls. DNA was hybridized to our Comprehensive High-throughput Arrays for Relative Methylation (CHARM) platform, covering 3.5 million CpGs. CHARM identified 224 candidate regions with DNAm differences >10%. These regions are highly enriched for neuronal growth and development genes. Ten of 17 regions for which validation was attempted showed true DNAm differences; the greatest were in PRIMA1, with 12–15% increased DNAm in MDD (p = 0.0002–0.0003), and a concomitant decrease in gene expression. These results must be considered pilot data, however, as we could only test replication in a small number of additional brain samples (n = 16), which showed no significant difference in PRIMA1. Because PRIMA1 anchors acetylcholinesterase in neuronal membranes, decreased expression could result in decreased enzyme function and increased cholinergic transmission, consistent with a role in MDD. We observed decreased immunoreactivity for acetylcholinesterase in MDD brain with increased PRIMA1 DNAm, non-significant at p = 0.08.
While we cannot draw firm conclusions about PRIMA1 DNAm in MDD, the involvement of neuronal development genes across the set showing differential methylation suggests a role for epigenetics in the illness. Further studies using limbic system brain regions might shed additional light on this role.

Full text

Genome-Wide DNA Methylation Scan in Major
Depressive Disorder
Sarven Sabunciyan
1.
, Martin J. Aryee
2,5.
, Rafael A. Irizarry
1,3,5
, Michael Rongione
9
, Maree J. Webster
7
,
Walter E. Kaufman
3,6
, Peter Murakami
3
, Andree Lessard
8
, Robert H. Yolken
1
, Andrew P. Feinberg
3,4
,
James B. Potash
9
*, GenRED Consortium
"
1Department of Pediatrics, Stanley Division of Developmental Neurovirology, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland, United
States of America, 2Department of Oncology, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland, United States of America, 3Epigenetics
Center, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland, United States of America, 4Division of Molecular Medicine, Department of
Medicine, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland, United States of America, 5Department of Biostatistics, Bloomberg School
of Public Health, Johns Hopkins University, Baltimore, Maryland, United States of America, 6Center for Genetic Disorders of Cognition and Behavior, Kennedy Krieger
Institute, Baltimore, Maryland, United States of America, 7Uniformed Services University of the Health Sciences, Bethesda, Maryland, United States of America, 8Maryland
Psychiatric Research Center, University of Maryland School of Medicine, Baltimore, Maryland, United States of America, 9Department of Psychiatry, University of Iowa,
Iowa City, Iowa, United States of America
Abstract
While genome-wide association studies are ongoing to identify sequence variation influencing susceptibility to major
depressive disorder (MDD), epigenetic marks, such as DNA methylation, which can be influenced by environment, might
also play a role. Here we present the first genome-wide DNA methylation (DNAm) scan in MDD. We compared 39
postmortem frontal cortex MDD samples to 26 controls. DNA was hybridized to our Comprehensive High-throughput
Arrays for Relative Methylation (CHARM) platform, covering 3.5 million CpGs. CHARM identified 224 candidate regions with
DNAm differences .10%. These regions are highly enriched for neuronal growth and development genes. Ten of 17 regions
for which validation was attempted showed true DNAm differences; the greatest were in PRIMA1, with 12–15% increased
DNAm in MDD (p = 0.0002–0.0003), and a concomitant decrease in gene expression. These results must be considered pilot
data, however, as we could only test replication in a small number of additional brain samples (n = 16), which showed no
significant difference in PRIMA1. Because PRIMA1 anchors acetylcholinesterase in neuronal membranes, decreased
expression could result in decreased enzyme function and increased cholinergic transmission, consistent with a role in MDD.
We observed decreased immunoreactivity for acetylcholinesterase in MDD brain with increased PRIMA1 DNAm, non-
significant at p = 0.08. While we cannot draw firm conclusions about PRIMA1 DNAm in MDD, the involvement of neuronal
development genes across the set showing differential methylation suggests a role for epigenetics in the illness. Further
studies using limbic system brain regions might shed additional light on this role.
Citation: Sabunciyan S, Aryee MJ, Irizarry RA, Rongione M, Webster MJ, et al. (2012) Genome-Wide DNA Methylation Scan in Major Depressive Disorder. PLoS
ONE 7(4): e34451. doi:10.1371/journal.pone.0034451
Editor: Tadafumi Kato, RIKEN Brain Science Institution, Japan
Received December 7, 2011; Accepted February 28, 2012; Published April 12, 2012
Copyright: ß2012 Sabunciyan et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by a grant from the NIMH (National Institute of Mental Health) to Dr. Potash (R01MH074131), by the NHGRI (National Human
Genome Research Institute) to Dr. Feinberg (2P50HG003233), and by the Margaret Price Investigatorship, and the Stanley Medical Research Institute (SMRI). The
funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: james-potash@uiowa.edu
.These authors contributed equally to this work.
"Membership of the GenRED Consortium is provided in the Acknowledgments.
Introduction
Family studies show that siblings of probands with major
depressive disorder (MDD) have about a three-fold elevated risk of
illness, while the estimated heritability of MDD from twin studies
is about 37% [1]. The modest level of heritability suggests that the
DNA sequence does not fully explain the variability in suscepti-
bility to this illness. Indeed, genome-wide association studies have
not yet definitively identified variants implicated in MDD, though
some intriguing results have been reported [2].
There are at least two other major kinds of explanations for this
variation in susceptibility. One is that environmental factors such
as stressful life events play a significant role in triggering MDD [3],
and another is that epigenetic factors are involved. These may be
interdependent as the environment may cause epigenetic changes.
In an animal model of early-life stress characterized by reduced
maternal care, epigenetic changes, including increased DNA
methylation (DNAm), were seen in the promoter region of the
glucocorticoid-receptor gene, and these persisted into adulthood,
where they correlated with disruption of the hypothalamic-
pituitary-adrenal axis [4]. Analogously, DNA from postmortem
hippocampus obtained from suicide victims with a history of
childhood abuse, also showed increased DNAm in the human
version of the same gene [5].
Epigenetics, which has been frequently implicated in cancers
[6], has also been implicated in brain diseases, such as Rett
PLoS ONE | www.plosone.org 1 April 2012 | Volume 7 | Issue 4 | e34451

syndrome [7] and fragile X syndrome [8]. There is now ample
evidence that DNAm plays a critical role in brain development
and function. One study found that abnormally hypomethylated
CNS neurons were impaired functionally and were selected
against in postnatal development [9]. We have shown that DNAm
signatures distinguished three brain regions—cortex, cerebellum,
and pons [10]. A role for epigenetics in MDD and other
psychiatric disorders has been suggested based on factors such as
the lack of complete concordance in monozygotic twins, the onset
of illness in adolescence or adulthood rather than childhood, the
often episodic nature of the illnesses, and the apparent relationship
to environmental factors, including stress [11].
There are several examples of epigenetic variation in candidate
MDD genes and in DNA treated with medications used for MDD.
For example, early life adversity increased DNAm in Bdnf in rats
[12]. Valproate [13], used to treat bipolar depression, and
haloperidol [14], used for psychotic depression, as well as the
antidepressants imipramine [15], tranylcypromine [16], and
fluoxetine [17] have been shown to induce epigenetic changes in
rodent brain. Further, administration of a histone deacetylase
inhibitor, sodium butyrate, produces an antidepressant effect in an
animal model [18].
Despite the availability of an essentially complete genome
sequence for several years, understanding of the methylome has
progressed more slowly, largely due to limitations in technology
affecting sensitivity, specificity, throughput, quantitation, and cost
among the previously used detection methods. Microarray-based
methods can interrogate much larger numbers of CpGs than other
approaches. One study to date has reported on a genome-wide
DNAm study in psychiatric disorders, demonstrating differences in
the 4–9% range between DNA from bipolar disorder or
schizophrenia brain samples vs. controls [19]. This study used
the methylation-sensitive restriction enzymes HpaII and McrBC to
prepare DNA, which they hybridized to a 12,192 CpG-island
microarray.
We have similarly used a methylation-sensitive restriction
enzyme-based method focused on McrBC, though we have
implemented it on a microarray platform (CHARM), which is
not biased towards CpG islands, but rather has features chosen
agnostically based on high CpG density. We have shown that
CHARM robustly distinguishes tissue types based on differential
DNAm profiles, and can also discriminate between colon cancer
and normal colon tissues [20].
We have now used CHARM analysis to study genome-wide
DNAm variation in 39 MDD and 26 control brains. Here we
report results of this experiment and of follow-up pyrosequencing
experiments to attempt to validate the initial findings and to
correlate DNAm differences with gene expression. While these
data should be followed up on a much larger replication set, the
absence of large DNAm differences in the brains of MDD patients
is itself important in considering the epigenetic hypothesis. These
results suggest that if DNAm plays a role in MDD, the most
critical target may not be the frontal cortex, but other regions,
such as hippocampus and amygdala, key components of the limbic
system, in which epigenetic changes have been shown to influence
cognitive and behavioral phenotypes [21].
Materials and Methods
Ethics Statement
The Johns Hopkins University IRB approved all research
involving human participants. Subjects gave written informed
consent under the IRB-approved protocol.
Brain DNA
Postmortem frontal cortex brain tissue, Brodmann area 10,
from 39 individuals with MDD and 27 matched controls were
donated by the Stanley Medical Research Institute, in two batches.
The first sample set consisted of 12 psychotic depression cases, 12
non-psychotic depression cases and 12 age and sex matched
controls. A second set consisted of 15 non-psychotic depression
cases and 15 age and sex matched controls. To increase power, the
two samples were analyzed together. A structured interview-based
DSM-IV diagnosis was assigned to each sample independently by
two senior psychiatrists, based on available medical records and a
series of interviews conducted with the family [22]. For each brain,
the cerebrum was hemisected, and one half was fixed in formalin
while the other was cut into 1.5 cm thick coronal slices and frozen
in a mixture of isopentane and dry ice. Right and left brain
hemispheres were randomly alternated for formalin fixing or
freezing. Frozen tissues were used for the DNAm studies.
Formalin-fixed, paraffin-embedded sections were employed for
the immunohistochemical analysis of acetylcholinesterase (AChE).
All frozen tissue was stored at 270uC. DNA was extracted using
the MasterPure DNA Purification kit (Epicentre Biotechnologies).
A replication sample set was provided by the Maryland Psychiatric
Research Center. This consisted of post-mortem BA10 samples
from 16 subjects with MDD and 13 controls. These samples were
age, sex, and race matched.
Lymphoblastoid cell line DNA
Cases of MDD (N = 30) were selected from the Genetics of
Recurrent Early Onset Depression (GenRED) study. Clinical
methods have been described elsewhere [23]. MDD cases had two
or more episodes of DSM-IV MDD with onset before age 31.
Subjects gave written informed consent under IRB-approved
protocols. European-American controls (N = 30) selected from the
NIMH Genetics Initiative repository had no MDD. DNA for
GenRED cases and MGS controls was provided from EBV-
transformed lymphoblastoid cell lines by the NIMH Center for
Collaborative Genetics Studies. A replication set of 90 MDD cases
and 90 controls were also run.
CHARM platform
The CHARM assay was performed as described previously
[20]. Briefly, 10 mg of DNA were sheared in 100 ml using a
Hydroshear device (Genomic Solution) to 1.6 kb–3 kb. Sheared
DNA was then divided into two fractions. One fraction was
digested overnight at 37uC with the methyl-sensitive enzyme
McrBC (NEB). Following digestion cut and uncut fractions from
the same sample were electrophoresed in adjacent wells of a 1%
agarose gel. Areas corresponding to the 1.6 kb–3 kb regions were
excised and purified using Qiagen Spin Gel Purification columns.
The gel-purified DNA was quantified on a spectrophotometer and
30 ng of DNA from each fraction was amplified using a
GenomePlex Whole Genome Amplification Kit (SIGMA). The
amplified DNA was then isolated with a Qiagen PCR Purification
column, then quantified using a spectrophotometer. The untreat-
ed, total DNA fraction was labeled with Cy3 and the methyl-
depleted DNA fraction was labeled with Cy5 and hybridized onto
the custom NimbleGen 2.1 M feature CHARM microarray
(design previously described [24]).
Pyrosequencing
1mg of genomic DNA was bisulfite treated using the Epitect kit
(Qiagen). CpG unbiased primers were designed to PCR amplify
92 CpG sites in 17 genes. Nested PCR was performed. Amplicons
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 2 April 2012 | Volume 7 | Issue 4 | e34451

DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 3 April 2012 | Volume 7 | Issue 4 | e34451

were analyzed on a PSQ HS 96 pyrosequencer (Biotage), and
CpG sites were quantified, from 0% to 100% methylation, using
Pyro Q-CpG software [25].
Real-time gene expression
RNA was extracted from frontal cortex using the RNAeasy Kit
(Qiagen). MonsterScript 1st – Strand cDNA Synthesis Kit
(Epicentre) was used to generate cDNA for subsequent quantita-
tive real-time PCR. Negative RT samples were used to ensure the
absence of contamination. All reactions were carried out in
triplicate using 16TaqMan master mix (Applied Biosystems), 16
TaqMan probe for each gene, and 10 ng of template in a volume
of 20 mL. Real-time reactions were performed on an Applied
Biosystems 7900HT Real-Time PCR System. Each set of
triplicates was checked to ensure that the threshold cycle (Ct)
values were all within 1 Ct of each other. The delta-delta-Ct
method was used to determine sample quantity.
Microarray data preprocessing
Hybridization quality was assessed by comparing the untreated
fraction signal intensity for each genomic probe to that of
background (anti-genomic) probes, with the expectation that the
genomic probes should register significantly higher signals. Poor
hybridization was indicated by genomic probe signal levels not
being significantly higher than background probe levels. Using this
metric eight arrays were identified as having failed hybridization
and discarded.
Detection of differentially methylated regions (DMRs)
Normalized methylation log-ratios were smoothed using a
weighted sliding window as previously described [24]. For each
probe, the average log-ratio and standard deviation were
computed for cases and controls allowing a Z-score to be
calculated for each probe. Under the assumption that most
regions are not differentially methylated, the median absolute
deviation of t-scores across all probes was used to determine the
standard deviation of the null distribution. Contiguous regions of
$6 smoothed Z-scores with p,0.005 were identified as candidate
DMRs. For these regions, a Bayesian model was used to convert
log ratios of intensities to estimated percent methylation [26]. P-
values were assigned by comparing the DMR areas to a null
distribution generated by permuting sample labels.
Gene Ontology analysis
We sought to determine whether our nominally significant
differentially methylated regions were in or near genes that
clustered together functionally. We determined the nearest gene
for each differential DNAm region and thus created a list of genes
with differential DNAm. We then asked whether this gene list was
enriched for GO Biological Process categories [27] using the NIH
DAVID tool [28]. We calculated an expected number of genes we
would see from our data set in each category under the null
hypothesis and compared that with the observed number to obtain
a p-value using the Fisher exact test. To better determine the
statistical significance of these results we further calculated a False
Discovery Rate using the Benjamini-Hochberg method [29].
Analysis of pyrosequencing data
For each of the 17 most differentially methylated regions, we
assessed pyrosequencing data based on primers designed across
the most CpG dense part of the region implicated by CHARM. A
linear regression model was used to assess the statistical
significance of the effect of case-control status on DNAm. These
were then corrected at two levels of stringency: 1) taking the best p-
value for each gene and correcting for 17 tests (the number of
regions tested); and 2) taking all p-values and correcting for 92 tests
(the number of CpGs tested). We then tested DNAm levels at all
CpGs against a number of additional sample variables including:
pH, postmorterm interval, age, sex, side of brain assayed, smoking
at time of death, and lifetime alcohol use, using a univariate
regression model. Resulting p-values were corrected for the
number of tests performed (92). For PRIMA1 each of these was
added as a covariate into a regression equation with case status as
the primary independent variable and DNAm as the dependent
variable.
Immunohistochemistry
Ten micron-thick paraffin sections from four subjects with
MDD and five controls were processed for AChE immunostain-
ing. Sections were incubated with a rabbit polyclonal antibody
targeting a signature epitope of an AChE precursor recombinant
protein, particularly suitable for tissue immunohistochemistry
(HPA019704; Sigma, St. Louis, MO), at a 1:25 dilution and
subsequently processed by a modification of the avidin-biotin-
peroxidase method as we have previously described [30]. Several
AChE immunostaining parameters were measured semi-quanti-
tatively in a blind fashion, using Likert scale scores (0–4) as
reported [30]: overall intensity of staining, degree of reticular
neuropil staining, and density of perikaryal neurite clusters. Scores
were compared by the Mann-Whitney-U test. In addition, we
performed qualitative evaluations of neuronal perikaryal and
nuclear staining.
Figure 1. Examples of CHARM results for two of the regions showing greatest DNAm differences between MDD cases and controls.
The plots show percent methylation versus genomic location with each point representing the methylation level of an individual sample for a given
probe. The curve represents averaged smoothed percent methylation values. The locations of CpG dinucleotides are indicated with black tickmarks
on the X-axis. CpG density was calculated across the region using a standard density estimator and is represented by the smoothed black line. The
location of the CpG island is denoted on the X-axis as an orange line. Gene annotation is indicated, showing LASS2 in (a) and PRIMA1 in (b). The thin
outer grey line represents the transcript, while the thin inner lines represent a coding region. Filled in grey boxes represent exons.
doi:10.1371/journal.pone.0034451.g001
Table 1. Stanley Medical Research Institute MDD and control brain samples.
N Age M F PMI (hr) pH % suicide L R
Control 27 48.2610.5 23 4 26.5615.5 6.660.5 0 12 15
MDD 39 44.6610.6 28 11 44.5632.8 6.660.5 53.8 21 18
doi:10.1371/journal.pone.0034451.t001
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 4 April 2012 | Volume 7 | Issue 4 | e34451

Results
Characteristics of postmortem brain samples are provided in
Table 1. Of these 66 samples, 58 were used in our analyses. Data
for eight were removed because of inadequate quality of array
hybridization. CHARM analysis identified 438 nominally signif-
icant candidate DMRs between MDD and controls (Table S1). Of
these, 224 DMRs showed differences .10%; the largest difference
was 22%. Figure 1 shows examples of two regions with the greatest
DNAm differences. We note that their magnitude was modest
compared to another disease vs. control CHARM experiment in
which we observed colon cancer vs. normal colon DNAm
differences of up to 52%. Nonetheless, their magnitude was not
unexpected given the results of a comparable study of psychiatric
brain samples with DNAm differences in the single digits [19]. We
calculated a false discovery rate (FDR) for each DMR to account
for multiple testing. None of the DNAm differences reached the
threshold for statistical significance (q-value,0.1) after correcting
for multiple testing. However, we sought to further characterize
the results with additional exploratory analyses.
We assessed the DNAm differences between MDD and controls
using the Biological Processes categories of the Gene Ontology
database [27]. The set of overrepresented categories includes
many processes related to neurogenesis and central nervous system
development (Table S2). These categories are intriguing given the
neurotrophic model of MDD that posits a critical role for deficits
in neuronal growth in the etiopathogenesis of the illness [31].
We attempted validation for 17 DMRs chosen because they
were among those showing the greatest DNAm differences
between MDD and controls, and were in or near genes. Within
these regions bisulfite pyrosequencing was conducted across 92
CpG dinucleotides. We observed nominally significant DNAm
differences in 10 of the regions. The four regions with the strongest
results, those in or near the genes LASS2,CPSF3,ZNF263, and
PRIMA1 (Figure 2), remained statistically significant after
correcting for 17 tests. The greatest DNAm difference for each
gene was 4, 8, 8, and 15 percent, respectively, with the MDD
samples being the more highly methylated for each of the four.
When we corrected for 92 CpGs tested, only four consecutive
CpGs in PRIMA1, with 12–15% increased DNAm in MDD,
remained significant (p = 0.00019–0.00028).
For all of the 17 regions tested, we tested the impact of
additional demographic, clinical, and biologic variables on DNAm
(Table S3). After correction for 17 regions tested, DNAm was not
predicted by pH, post-mortem interval, age, sex, side of brain,
smoking, psychotic status, or alcohol use. For PRIMA1, two
variables predicted DNAm for CpG-2 at a nominal level of
significance: increased age was associated with decreased DNAm
(p = 0.04), as was lower pH (p = 0.03) (Table 2). When these two
variables were included as covariates in a regression the
relationship between MDD and DNAm remained significant
(p = 0.008–0.02). We further examined whether medication usage
might account for the increased DNAm at PRIMA1 in MDD
samples by focusing on the subset of seven samples that were
medication free. DNAm for these were 3–6% greater than for the
remaining 32 MDD samples (p = 0.15), suggesting that medication
was not responsible for the difference between MDD and controls.
Figure 2. Results of bisulfite pyrosequencing for validation of
CHARM in brain samples. Regions in or near four genes showed
differences that remained statistically significant after correction for
having tested 17 genes: (a) LASS2, (b) CPSF3, (c) ZNF263, (d) PRIMA1. The
grey bars represent values from control brain sample DNA, while the
black bars represent those from MDD brain samples. The Y-axis is
percent DNA methylation, while the X-axis shows distance along the
chromosome for each CpG dinucleotide assayed. One asterisk indicates
a difference between MDD and control of p,0.05. Two asterisks
indicates p,0.0029 (a correction for 17 regions tested). Three asterisks
indicates p,0.00054 (a correction for 92 CpGs tested).
doi:10.1371/journal.pone.0034451.g002
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 5 April 2012 | Volume 7 | Issue 4 | e34451

To assess the potential functional impact of increased PRIMA1
DNAm in MDD, we tested mRNA levels of the gene in the same
brain samples that were used for the DNAm experiments. Levels
were altered in the MDD brain samples in the expected direction,
being decreased 53% (p = 0.047).
Because of the potential clinical value of blood-derived
biomarkers, we sought to determine whether PRIMA1 DNAm
Table 2. PRIMA1 DNAm by diagnosis, and by covariate status
a.
MDD vs. control Covariates (p-values)
CpG
Control %
DNAm MDD % DNAm Dx (p-value) PMI Brain pH Side of Brain Age Sex Smoking Alcohol
1 40.3 50.9 0.0062 0.66 0.062 0.99 0.053 0.57 0.30 0.74
2 43.4 58.7 0.00027 0.50 0.031 0.90 0.044 0.88 0.51 0.71
3 51.9 67.2 0.00028 0.76 0.052 0.72 0.063 0.64 0.58 0.55
4 57.3 71.4 0.00026 0.76 0.052 0.80 0.083 0.85 0.52 0.60
5 64.4 76.7 0.00019 0.91 0.083 0.96 0.072 0.95 0.70 0.46
6 93.5 95.6 0.050 0.06 0.034 0.77 0.010 0.50 0.36 0.89
a
Dx = diagnosis; PMI = postmortem interval; DNAm = DNA methylation; p-values,0.05 are italicized for clarity.
doi:10.1371/journal.pone.0034451.t002
Figure 3. Immunohistochemical pattern of AChE in frontal cortex. (A) In controls there is diffuse and intense pattern of immunoreactivity
involving mainly the neuropil. (B) In MDD subjects, though variable, immunostaining was reduced. Both 2006. (C) In controls, there is virtually no
perikaryal staining. (D) The latter contrasts with the pattern observed in some areas in MDD subjects, in which groups of pyramidal neurons display
intense perikaryal staining, suggesting redistribution of the enzyme to the cell body. The red circles highlight examples. Both 6406.
doi:10.1371/journal.pone.0034451.g003
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 6 April 2012 | Volume 7 | Issue 4 | e34451

differences could be detected between subjects from our GenRED
study as compared to normal controls collected for genetic studies.
We used DNA from these subjects’ lymphoblastoid cell lines and
saw results similar to those in brain. DNAm was increased in
MDD subjects as compared to controls (by 7–10%, p = 0.0006–
0.01) for three of the four PRIMA1 CpGs (Figure S1).
We attempted to replicate both the brain and the blood results
using independent sample sets. In 16 MDD postmortem brain
samples and 13 controls, we failed to detect a significant difference
in DNAm at any of the four previously implicated PRIMA1 CpGs.
DNAm levels were virtually identical between groups (Supporting
Information S1). The biggest difference was a 4.4% decrease in
methylation for the cases at the third CpG (p = 0.16). Similarly
DNAm in CpGs in LASS2,CPSF3,ZNF263 did not differ
significantly between groups (Supporting Information S1). When
we examined lymphoblastoid cell line DNA from an additional 90
MDD cases and 90 controls, we could not replicate the DNAm
differences in PRIMA1 observed in the prior sample set
(Supporting Information S1).
Using immunohistochemistry, we investigated whether MDD
subjects with high DNAm and low expression for PRIMA1 would
show reduced immunoreactivity for AChE as compared to
controls with the opposite pattern. Such a result would be
consistent with the changes we observed in PRIMA1 DNAm
influencing cholinergic transmission. In a semi-quantitative
comparison between frontal cortex tissues from four MDD
subjects and five controls, we found that overall AChE staining
intensity was reduced in the MDD subjects on average 42%,
however, this difference did not reach statistical significance
(p = 0.08). We also observed that subjects with MDD had a larger
number of superficial pyramidal neuron perikaryal staining,
despite overall reduction in neuropil immunoreactivity, suggesting
redistribution of AChE towards the cell bodies (Figure 3).
Discussion
We report here on the first genome-wide DNA methylation
comparison between MDD and control brain. Although the
magnitude of DNAm differences we observed was relatively small
and did not survive correction for multiple testing, the DMRs
identified were in or near genes enriched for roles in neuronal
growth and development, suggesting that the differences picked up
by our CHARM experiment, despite being relatively small, might
be biologically meaningful. Our validation experiment showed the
greatest differences in PRIMA1, with 12–15% increased DNAm in
MDD. Consistent with this result, PRIMA1 expression was
decreased in MDD brain samples. The DNAm changes in the
brain were also reflected in DNA from an initial set of
lymphoblastoid cell lines, with MDD cases again showing greater
DNAm than controls. However, we were unable to replicate
PRIMA1 DNAm differences in additional sample sets of brain and
lymphoblastoid cell lines. Further, although we observed de-
creased immunoreactivity for AChE in MDD tissues that had
increased PRIMA1 DNAm, this change did not reach statistical
significance. Therefore, we cannot draw firm conclusions about a
potential role for PRIMA1 DNAm in MDD.
PRIMA1 is of substantial biological interest in MDD because of
its relationship to cholinergic neurotransmission. The gene
encodes a protein that both guides the transport of acetylcholin-
esterase to neuronal membranes [32] and anchors it there [33].
When PRIMA1 is knocked down by antisense cDNA [33] or
knocked out [32], there is a decrease in localization of AChE at the
neuronal membrane, or of AChE activity, respectively. AChE
hydrolyzes acetylcholine, thus less of its activity means more
cholinergic transmission. Janowsky and colleagues proposed that
increased cholinergic transmission is a central mechanism in
depression, noting that reserpine, which can cause depression, is
cholinomimetic, and the tricyclic antidepressants are anticholin-
ergic [34]. Additional evidence in support of this hypothesis
includes the induction of depressive symptoms by the administra-
tion of physostigmine, a more specific cholinometic agent [35],
and the alleviation of such symptoms by the use of more specific
anticholinergic medications such as scopolamine [36], a musca-
rinic acetylcholine receptor antagonist, and mecamylamine, a
nicotinic acetycholine receptor antagonist [37]. Intriguingly, stress,
which plays a key role in MDD etiology, has been shown to
influence cholinergic gene expression in mouse brain [38].
Compared to DNAm differences seen in prior studies using the
CHARM platform to compare tissue or cell types, or colon cancer
vs. normal colon, the magnitude of those seen in our study was
modest. This is, perhaps, not surprising given the findings of the
only other genome-wide DNAm studies in psychiatric illness, that
of Mill et al [19] and Dempster et al [39], which similarly found
small, though statistically significant, differences between cases and
controls. It is likely because the magnitude of our DNAm
differences hovered around the limit of resolution of CHARM
that a number of our candidate DMRs did not validate. Since
completing this experiment, we have developed improvements to
CHARM that increase its signal-to-noise ratio. In addition, the
next generation of CHARM includes coverage of a greater
number of CpGs, augmenting beyond the ,20% of all CpGs that
were initially on the array. In the current experiment we employed
a conservative statistical threshold to guard against false positives.
It is possible that a relaxed threshold might have captured more
signals reflecting true biological differences between depression
and controls.
Our failure to detect a robustly replicating signal makes it hard
to draw firm conclusions about a role for DNAm in the frontal
cortex of subjects with MDD. It is possible that larger
etiopathologically relevant DNAm changes might exist in other
brain regions known to be involved in MDD, such as the limbic
regions anterior cingulate cortex [40], amygdala, and hippocam-
pus [41]. We have previously shown brain region-specific variation
in DNAm [10]. Further, disease-related DNAm variation might be
restricted to particular cell types, such as neurons only or even,
more narrowly, subtypes of neurons, such as pyramidal cells.
However, there may be a substantial portion of DMRs that are not
cell type- or tissue-specific. We note that these generalized MDD
DMRs might be the most valuable as they both shed light on
etiopathogenesis, and also potentially provide biomarkers that can
be studied in living patients. Blood-based DMRs would also allow
for much larger numbers of samples to be assayed and for
correlation on a large scale with genotype.
Supporting Information
Figure S1 Results of bisulfite pyrosequencing of six
PRIMA1 CpGs in lymphablastoid cell line samples. The
grey bars represent values from control sample DNA, while the
black bars represent those from MMD samples. The Y-axis is
percent DNA methylation, while the X-axis shows each CpG
arrayed along the chromosome. Asterisks indicate a difference
between MDD and control of p,0.01.
(TIFF)
Table S1 The results of the primary experiment using
CHARM to compare postmortem brain samples be-
tween MDD cases and controls.
(DOC)
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 7 April 2012 | Volume 7 | Issue 4 | e34451

Table S2 The result of taking all genes in or near
nominally significant differentially methylated regions
and examining their representation in Gene Ontology
Categories.
(DOC)
Table S3 Bisulfite pyrosequencing was used to experi-
mentally validate some of the regions that showed
differential methylation between MDD and controls by
CHARM analysis. This table shows those that were nominally
validated. P-values for regression of pyrosequencing methylation at
individual CpGs (rows) on 6 covariates (columns). The last column
shows the F-statistic p-value for the multiple regression of
methylation on all 6 covariates.
(DOC)
Supporting Information S1 Supplementary Tables S4,
S5, S6 show results of replication attempts in postmor-
tem brain and in lymphoblastoid cell lines.
(DOCX)
Acknowledgments
The authors would like to thank SMRI and Dr. Fuller Torrey for providing
postmortem MDD brain samples. GenRED consortium co-authors are:
James A. Knowles, Myrna M. Weissman, William Coryell, William A.
Scheftner, and Douglas F. Levinson. Drs. Potash and Feinberg made equal
contributions and should be considered co-senior authors. Some of this
material was presented at the World Congress of Psychiatric Genetics, San
Diego, November 4–8, 2009. Data and biomaterials were collected in six
projects that participated in the National Institute of Mental Health
(NIMH) Genetics of Recurrent Early-Onset Depression (GenRED) project.
From 1999–2003, the Principal Investigators and Co-Investigators were:
New York State Psychiatric Institute, New York, NY, Myrna M.
Weissman, Ph.D. and James K. Knowles, M.D., Ph.D.; University of
Pittsburgh, Pittsburgh, PA, George S. Zubenko, M.D., Ph.D. and Wendy
N. Zubenko, Ed.D., R.N., C.S.; Johns Hopkins University, Baltimore, J.
Raymond DePaulo, M.D., Melvin G. McInnis, M.D. and Dean
MacKinnon, M.D.; University of Pennsylvania, Philadelphia, PA, Douglas
F. Levinson, M.D. (GenRED coordinator), Madeleine M. Gladis, Ph.D.,
Kathleen Murphy-Eberenz, Ph.D. and Peter Holmans, Ph.D. (University
of Wales College of Medicine); University of Iowa, Iowa City, IW,
Raymond R. Crowe, M.D. and William H. Coryell, M.D.; Rush
University Medical Center, Chicago, IL, William A. Scheftner, M.D.
Rush-Presbyterian.
Author Contributions
Conceived and designed the experiments: APF JBP RHY. Performed the
experiments: SS MR MJW WEK AL. Analyzed the data: MJA RAI PM.
Contributed reagents/materials/analysis tools: GC. Wrote the paper: SS
MJA WEK APF JBP.
References
1. Sullivan PF, Neale MC, Kendler KS (2000) Genetic epidemiology of major
depression: Review and meta-analysis. Am J Psychiatry 157: 1552–1562.
2. Shyn SI, Shi J, Kraft JB, Potash JB, Knowles JA, et al. (2011) Novel loci for
major depression identified by genome-wide association study of sequenced
treatment alternatives to relieve depression and meta-analysis of three studies.
Mol Psychiatry 16: 202–15.
3. Paykel ES, Myers JK, Dienelt MN, Klerman GL, Lindenthal JJ, et al. (1969) Life
events and depression. A controlled study. Arch Gen Psychiatry 21: 753–760.
4. Weaver IC, Cervoni N, Champagne FA, D’Alessio AC, Sharma S, et al. (2004)
Epigenetic programming by maternal behavior. Nat Neurosci 7: 847–54.
5. McGowan PO, Sasaki A, D’Alessio AC, Dymov S, Labonte B, et al. (2009)
Epigenetic regulation of the glucocorticoid receptor in human brain associates
with childhood abuse. Nat Neurosci 12: 342–348.
6. Feinberg AP, Vogelstein B (1983) Hypomethylation distinguishes genes of some
human cancers from their normal counterparts. Nature 301: 89–92.
7. Amir RE, Van den Veyver IB, Wan M, Tran CQ, Francke U, et al. (1999) Rett
syndrome is caused by mutations in X-linked MECP2, encoding methyl-CpG-
binding protein 2. Nat Genet 23: 185–188.
8. Oberle I, Rousseau F, Heitz D, Kretz C, Devys D, et al. (1991) Instability of a
550-base pair DNA segment and abnormal methylation in fragile X syndrome.
Science 252: 1097–1102.
9. Fan G, Beard C, Chen RZ, Csankovszki G, Sun Y, et al. (2001) DNA
hypomethylation perturbs the function and survival of CNS neurons in postnatal
animals. J Neurosci 21: 788–797.
10. Ladd-Acosta C, Pevsner J, Sabunciyan S, Yolken RH, Webster MJ, et al. (2007)
DNA methylation signatures within the human brain. Am J Hum Genet 81:
1304–1315.
11. Mill J, Petronis A (2007) Molecular studies of major depressive disorder: The
epigenetic perspective. Mol Psychiatry 12: 799–814.
12. Roth TL, Lubin FD, Funk AJ, Sweatt JD (2009) Lasting epigeneti c influence of
early-life adversity on the BDNF gene. Biol Psychiatry 65: 760–769.
13. Milutinovic S, D’Alessio AC, Detich N, Szyf M: Valproate induces widespread
epigenetic reprogramming which involves demethylation of specific genes.
Carcinogenesis (2007) ; 28: 560–571.
14. Shimabukuro M, Jinno Y, Fuke C, Okazaki Y (2006) Haloperidol treatment
induces tissue- and sex-specific changes in DNA methylation: A control study
using rats. Behav Brain Funct 2: 37.
15. Tsankova NM, Berton O, Renthal W, Kumar A, Neve RL, et al. (2006)
Sustained hippocampal chromatin regulation in a mouse model of depression
and antidepressant action. Nat Neurosci 9: 519–525.
16. Lee MG, Wynder C, Schmidt DM, McCafferty DG, Shiekhattar R (2006)
Histone H3 lysine 4 demethylation is a target of nonselective antidepressive
medications. Chem Biol 13: 563–567.
17. Cassel S, Carouge D, Gensburger C, Anglard P, Burgun C, et al. (2006)
Fluoxetine and cocaine induce the epigenetic factors MeCP2 and MBD1 in
adult rat brain. Mol Pharmacol 70: 487–492.
18. Schroeder FA, Lin CL, Crusio WE, Akbarian S (2007) Antidepressant-like
effects of the histone deacetylase inhibitor, sodium butyrate, in the mouse. Biol
Psychiatry 62: 55–64.
19. Mill J, Tang T, Kaminsky Z, Khare T, Yazdanpanah S, et al. (2008)
Epigenomic profiling reveals DNA-methylation changes associated with major
psychosis. Am J Hum Genet 82: 696–711.
20. Irizarry RA, Ladd-Acosta C, Wen B, Wu Z, Montano C, et al. (2009) The
human colon cancer methylome shows similar hypo- and hypermethylation at
conserved tissue-specific CpG island shores. Nat Genet 41: 178–186.
21. Sweatt JD (2009) Experience-dependent epigenetic modifications in the central
nervous system. Biol Psychiatry 65: 191–197.
22. Torrey EF, Webster M, Knable M, Johnston N, Yolken RH (2000) The Stanley
Foundation brain collection and neuropathology consortium. Schizophr Res 44:
151–155.
23. Levinson DF, Zubenko GS, Crowe RR, DePaulo RJ, Scheftner WS, et al. (2003)
Genetics of recurrent early-onset depression (GenRED): Design and preliminary
clinical characteristics of a repository sample for genetic linkage studies.
Am J Med Genet 119B: 118–130.
24. Irizarry RA, Ladd-Acosta C, Carvalho B, Wu H, Brandenburg SA, et al. (2008)
Comprehensive high-throughput arrays for relative methylation (CHARM).
Genome Res 18: 780–790.
25. Tost J, Gut IG (2007) DNA methylation analysis by pyrosequencing. Nat Protoc
2: 2265–2275.
26. Aryee MJ, Wu Z, Ladd-Acosta C, Herb B, Feinberg AP, et al. (2011) Accurate
genome-scale percentage DNA methylation estimates from microarray data.
Biostatistics 12: 197–210.
27. Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, et al. (2004) The gene
ontology (GO) database and informatics resource. Nucleic Acids Res 32:
D258–61.
28. Huang da W, Sherman BT, Lempicki RA (2009) Systematic and integrative
analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4:
44–57.
29. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: A practical
and powerful approach to multiple testing. J Royal Stat Soc,
Series B (Methodological) 57: 289–300.
30. Kaufmann WE, MacDonald SM, Altamura CR (2000) Dendritic cytoskeletal
protein expression in mental retardation: An immunohistochemical study of the
neocortex in Rett syndrome. Cereb Cortex 10: 992–1004.
31. Duman RS, Heninger GR, Nestler EJ (1997) A molecular and cellular theory of
depression. Arch Gen Psychiatry 54: 597–606.
32. Dobbertin A, Hrabovska A, Dembele K, Camp S, Taylor P, et al. (2009)
Targeting of acetylcholinesterase in neurons in vivo: A dual processing function
for the proline-rich membrane anchor subunit and the attachment domain on
the catalytic subunit. J Neurosci 29: 4519–4530.
33. Perrier AL, Massoulie J, Krejci E (2002) PRiMA: The membrane anchor of
acetylcholinesterase in the brain. Neuron 33: 275–285.
34. Janowsky DS, el-Yousef MK, Davis JM, Sekerke HJ (1972) A cholinergic-
adrenergic hypothesis of mania and depression. Lancet 2: 632–635.
35. Risch SC, Kalin NH, Janowsky DS (1981) Cholinergic challenges in affective
illness: Behavioral and neuroendocrine correlates. J Clin Psychopharmacol 1:
186–192.
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 8 April 2012 | Volume 7 | Issue 4 | e34451

36. Furey ML, Drevets WC (2006) Antidepressant efficacy of the antimuscarinic
drug scopolamine: A randomized, placebo-controlled clinical trial. Arch Gen
Psychiatry 63: 1121–1129.
37. Bacher I, Wu B, Shytle DR, George TP (2009) Mecamylamine – a nicotinic
acetylcholine receptor antagonist with potential for the treatment of neuropsy-
chiatric disorders. Expert Opin Pharmacother 10: 2709–2721.
38. Kaufer D, Friedman A, Seidman S, Soreq H (199 8) Acute stress facilitates long-
lasting changes in cholinergic gene expression. Nature 393: 373–377.
39. Dempster EL, Pidsley R, Schalkwyk LC, Owens S, Georgiades A, et al. (2011)
Disease-associated epigenetic changes in monozygotic twins discordant for
schizophrenia and bipolar disorder. Hum Mol Genet. 20: 4786–4796.
40. Mayberg HS, Lozano AM, Voon V, McNeely HE, Seminowicz D, et al. (2005)
Deep brain stimulation for treatment-resistant depression. Neuron 45: 651–660.
41. Campbell S, Marriott M, Nahmias C, MacQueen GM (2004) Lower
hippocampal volume in patients suffering from depression: A meta-analysis.
Am J Psychiatry 161: 598–607.
DNA Methylation in Major Depressive Disorder
PLoS ONE | www.plosone.org 9 April 2012 | Volume 7 | Issue 4 | e34451