A multi-tissue analysis identifies HLA complex group 9 gene methylation differences in bipolar disorder

Article (PDF Available)inMolecular Psychiatry 17(7):728-40 · June 2011with60 Reads
DOI: 10.1038/mp.2011.64 · Source: PubMed
Abstract
Epigenetic studies of DNA and histone modifications represent a new and important activity in molecular investigations of human disease. Our previous epigenome-wide scan identified numerous DNA methylation differences in post-mortem brain samples from individuals affected with major psychosis. In this article, we present the results of fine mapping DNA methylation differences at the human leukocyte antigen (HLA) complex group 9 gene (HCG9) in bipolar disorder (BPD). Sodium bisulfite conversion coupled with pyrosequencing was used to interrogate 28 CpGs spanning ∼700 bp region of HCG9 in 1402 DNA samples from post-mortem brains, peripheral blood cells and germline (sperm) of bipolar disease patients and controls. The analysis of nearly 40 000 CpGs revealed complex relationships between DNA methylation and age, medication as well as DNA sequence variation (rs1128306). Two brain tissue cohorts exhibited lower DNA methylation in bipolar disease patients compared with controls at an extended HCG9 region (P=0.026). Logistic regression modeling of BPD as a function of rs1128306 genotype, age and DNA methylation uncovered an independent effect of DNA methylation in white blood cells (odds ratio (OR)=1.08, P=0.0077) and the overall sample (OR=1.24, P=0.0011). Receiver operating characteristic curve A prime statistics estimated a 69-72% probability of correct BPD prediction from a case vs control pool. Finally, sperm DNA demonstrated a significant association (P=0.018) with BPD at one of the regions demonstrating epigenetic changes in the post-mortem brain and peripheral blood samples. The consistent multi-tissue epigenetic differences at HCG9 argue for a causal association with BPD.
ORIGINAL ARTICLE
A multi-tissue analysis identifies HLA complex group 9
gene methylation differences in bipolar disorder
Z Kaminsky
1,2,8
, M Tochigi
1,3,8
, P Jia
1,8
, M Pal
1
, J Mill
1,4
, A Kwan
1
, I Ioshikhes
5
, JB Vincent
6
, JL Kennedy
6
,
J Strauss
6
, S Pai
1
, S-C Wang
1,7
and A Petronis
1
1
The Krembil Family Epigenetics Laboratory, Neuroscience Department, Centre for Addiction and Mental Health, Toronto, ON,
Canada;
2
Department of Psychiatry and Behavioral Sciences, Johns Hopkins University, Baltimore, MD, USA;
3
Department of
Neuropsychiatry, Graduate School of Medicine, University of Tokyo, Tokyo, Japan;
4
King’s College London, Social, Genetic
and Developmental Psychiatry Department, Institute of Psychiatry, London, UK;
5
Ottawa Institute of Systems Biology (OISB)
and Department of Biochemistry, Microbiology and Immunology (BMI), University of Ottawa, Ottawa, ON, Canada;
6
Neuroscience Department, Centre for Addiction and Mental Health, Toronto, ON, Canada and
7
Systems Biology and
Bioinformatics Institute, National Central University, Jhongli, Taiwan
Epigenetic studies of DNA and histone modifications represent a new and important activity in
molecular investigations of human disease. Our previous epigenome-wide scan identified
numerous DNA methylation differences in post-mortem brain samples from individuals
affected with major psychosis. In this article, we present the results of fine mapping DNA
methylation differences at the human leukocyte antigen (HLA) complex group 9 gene (HCG9)in
bipolar disorder (BPD). Sodium bisulfite conversion coupled with pyrosequencing was used to
interrogate 28 CpGs spanning B700 bp region of HCG9 in 1402 DNA samples from post-
mortem brains, peripheral blood cells and germline (sperm) of bipolar disease patients and
controls. The analysis of nearly 40 000 CpGs revealed complex relationships between DNA
methylation and age, medication as well as DNA sequence variation (rs1128306). Two brain
tissue cohorts exhibited lower DNA methylation in bipolar disease patients compared with
controls at an extended HCG9 region (P = 0.026). Logistic regression modeling of BPD as a
function of rs1128306 genotype, age and DNA methylation uncovered an independent effect of
DNA methylation in white blood cells (odds ratio (OR) = 1.08, P = 0.0077) and the overall sample
(OR = 1.24, P = 0.0011). Receiver operating characteristic curve A prime statistics estimated a
69–72% probability of correct BPD prediction from a case vs control pool. Finally, sperm DNA
demonstrated a significant association (P = 0.018) with BPD at one of the regions demonstrating
epigenetic changes in the post-mortem brain and peripheral blood samples. The consistent
multi-tissue epigenetic differences at HCG9 argue for a causal association with BPD.
Molecular Psychiatry (2012) 17, 728–740; doi:10.1038/mp.2011.64; published online 7 June 2011
Keywords: epigenetics; DNA methylation; pyrosequencing; bipolar disorder
Introduction
Complex non-Mendelian diseases exhibit numerous
features that are difficult to explain by traditional
mechanisms such as predisposing DNA sequence
variation and hazardous environmental factors. The
non-Mendelian irregularities include the discordance
of monozygotic twins, parent of origin effects, gender
dimorphism and relatively late age at onset.
1–3
Putative
epigenetic misregulation that manifests via changes in
modifications of DNA and histones can offer a more
comprehensive and cohesive explanation for the
various facets of complex disease.
4
Epigenetic factors
such as DNA methylation control various genetic and
genomic functions, including gene transcription.
5,6
DNA methylation is known to act via three mechan-
isms: critical site, general density of methylated sites
in the DNA and multiple elements, which combines
components of the first two models.
7
Critical site
effects relate to the limitation of transcription factor
binding affinity due to the methylation status of
specific CpGs within their associated binding sites.
8,9
The overall density of methylcytosine in a gene
regulatory region also contributes to gene activity with
a large number of genes exhibiting an inverse correla-
tion between the degree of methylation and the level of
gene expression.
5,6
Epigenetic factors are relatively
stably inherited across somatic cells; however, sto-
chastic events and environmental influences can result
Received 23 January 2011; revised 17 March 2011; accepted 20
April 2011; published online 7 June 2011
Correspondence: Dr A Petronis, The Krembil Family Epigenetics
Laboratory, Neuroscience Department, Centre for Addiction and
Mental Health, Toronto, ON, Canada M5T 1R8.
E-mail: arturas_petronis@camh.net
8
Joint first authorship.
Molecular Psychiatry (2012) 17, 728740
&
2012 Macmillan Publishers Limited All rights reserved 1359-4184/12
www.nature.com/mp
in a drift of these regulatory molecular signals over
time.
10,11
There is increasing evidence that some
epigenetic signals can be transmitted from one genera-
tion to another,
12
which contradicts the idea that all
epigenetic signals are wiped out during reprogram-
ming in germ cells and points to a secondary
molecular mechanism of heritability.
Over the last decade, it has been suggested that
epigenetic misregulation is consistent with many
epidemiological, clinical and molecular findings in
bipolar disorder (BPD).
1,13,14
Partial epigenetic
instability, or metastability, can account for both the
discordance of monozygotic twins and clinical
fluctuations over the course of BPD. The opposite
disease phenotypes, depressive and manic phases,
between which the patients can make a nearly
complete recovery, are consistent with dynamic
changes in epigenetics rather than alternations to
the stable DNA sequence. Sex differences exist in
BPD, such as the higher incidence of rapid cycling,
mixed states and cyclothymia in women and a higher
prevalence of early-onset BPD in men. These differ-
ences, as well as sex effects revealed in genetic
linkage and association studies of BPD, could be
rationalized by the known epigenetic effects of sex
hormones.
15
Parent of origin-dependent clinical and
molecular differences in BPD are likely to result from
genomic imprinting,
16
a differential epigenetic modi-
fication of genes depending upon their parental
origin.
17
Epigenetic misregulation of gene activity is
also consistent with the numerous neurochemical
differences detected in BPD such as altered densities/
levels, of receptors, enzymes, peptides and neurome-
diators.
18
Increasing evidence for ‘soft’ inheritance
(that is, partial epigenetic stability in meiosis)
12
offers
a non-DNA sequence-based explanation for heritabi-
lity of BPD. It is telling that valproic acid, a mood
stabilizer and one of the most common medications to
treat BPD, is a histone deacetylase inhibitor
19
and can
also change DNA methylation.
20
Recent studies
detected that lithium, another mood stabilizer, modu-
lates histone acetylation and DNA methylation.
21,22
BPD may result from a series of unfavorable
epigenetic events beginning with a primary epige-
netic defect, or pre-epimutation that hypothetically
occurs in the germline during epigenetic reprogram-
ming
23–25
and is inherited. Pre-epimutations are
further altered during embryogenesis, childhood and
adolescence by the multiple effects of tissue differ-
entiation, stochastic factors, hormones and perhaps,
external environmental factors.
23,26,27
Decades may
elapse before epigenetic misregulation reaches the
level of an epimutation, a defect of sufficient severity
that the resultant cellular effects manifest clinically as
BPD. The severity of an epimutation may fluctuate, as
evidenced by clinical remission and relapse.
Recently, we completed the first DNA methylome
study in major psychosis using the 12K CpG island
microarray.
34
The study revealed a number of DNA
methylation differences associated with BPD and
schizophrenia (SCZ). This article reports data from
our continuing effort to validate and finely map the
identified epigenetic differences in the microarray
project. One of the genes identified as differentially
methylated in major psychosis was the human
leukocyte antigen (HLA) complex group 9 gene
(HCG9; Figure 1). In the microarray experiments,
the human CpG island microarray probe
UHNhscpg0007403 showed significantly lower DNA
methylation in major psychosis patients relative to
controls (fold change 0.42, false discovery rate
(FDR) P = 0.03). HCG9 is located on chromosome
6p21.33 and is of particular interest as recent
genome-wide association studies (GWAS) report an
association between genetic variants at chromosome
6p21.3-22 with SCZ and BPD.
28–30
HCG9 is a member
of the PERB11 gene family, also known as MIC, many
variants of which (such as MICA and MICB) function
as ligands for natural killer cell receptors.
31
Three
splice variants for HCG9 have been identified so far
(Figure 1). While it has been suggested based on
experimental evidence that HCG9 may be related to
stress response and cell death,
32
its function is
currently unknown.
The goals of our analysis of the HCG9 gene were
fivefold. Our primary objective was to validate the
microarray findings in the Stanley Medical Research
Institute (SMRI) post-mortem brain cohort, the sample
which was used in our 12K CpG island microarray-
based DNA methylome scan. Second, we attempted to
replicate our primary findings in an independent
post-mortem brain sample. In the event of validation,
our third objective was to characterize HCG9 methy-
lation differences in the context of the available
demographic and clinical information (for example,
age-dependent dynamics of DNA methylation).
Fourth, we were interested if pathological HCG9
methylation differences can be detected in peripheral
white blood cells (WBCs) and germline, tissues that
are not directly involved in psychiatric disease but
that are easily accessible and can be used in clinical
applications. Finally, our fifth objective was to
address the unknown function of the HCG9 gene in
order to better understand how it may fit into
the etiopathogenesis of BPD and other psychiatric
diseases.
Cytosine modifications were finely mapped using
the sodium bisulfite conversion approach coupled
with pyrosequencing. In total, we interrogated 1402
DNA samples across 28 CpG positions of HCG9.
Materials and methods
Samples
Post-mortem brain tissue of prefrontal, parietal and
occipital cortices and corpus callosum from indivi-
duals with BPD (N = 34), SCZ ( N = 35) and matched
controls (Con) (N = 35) was provided unblinded by the
SMRI brain-array collection. A brain prefrontal cortex
replication cohort from the McLean (McL) Hospital at
the National Brain Databank (N = 34 BPD, 30 SCZ and
50 Con) was tested along with a peripheral white
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
729
Molecular Psychiatry
blood cell (WBC 1) sample provided by Dr James
Kennedy (N = 100 BPD, 50 SCZ and 100 Con), a second
WBC sample (WBC 2) from GlaxoSmithKline, London,
UK (N = 270 BPD and 282 Con), and sperm samples
collected by our laboratory at the Centre for Addiction
and Mental Health (N = 29 BPD, 14 SCZ and 30 Con).
Diagnoses had been made according to the DSM-IV
(Diagnostic and Statistical Manual of Mental Disorders,
Fourth Edition). A summary of the demographic
information of samples used in this study is shown
in Table 1. Detailed information of the SMRI cohort
was described elsewhere.
33
Extraction of all DNA was
performed with a standard phenol–chloroform
extraction method. The project has been approved
by the Ethics Committee of the Centre for Addiction
and Mental Health, Toronto.
Bisulfite treatment was performed by use of a
standard protocol.
34
In brief, B500 ng of genomic
DNA was denatured in 0.3
M NaOH for 20 min
at 42 1C. After the addition of freshly prepared
saturated sodium metabisulfite (Sigma Aldrich,
St Louis, MO, USA) and 10 m
M hydroquinone (Sigma
Aldrich) solution, samples were subjected to a 5-h
incubation at 55 1C under exclusion of light. The
samples were then purified and desulfonated with
0.3
M NaOH by using a Montage PCR96 96-well
filtration plate (Millipore, Billerica, MA, USA).
Pyrosequencing
Four primer sets were designed to amplify the exon 1
and adjacent intron 1 region of the HCG9 gene
(Supplementary Table S1). Within these regions, six
pyrosequencing assays were designed. In total, 28
CpGs in HCG9 were covered. PCR amplifications
were performed with a standard PCR protocol in 50 ml
volume reactions containing 2 ml of sodium bisulfite-
treated DNA, 0.15–0.5 m
M primers, and master mix
containing JumpStart or Pla tinum Taq DNA polymerase
(Sigma Aldrich and Invitrogen, Carlsbad, CA, USA,
respectively). After agarose gel electrophoresis to
ensure successful amplification and specificity, PCR
amplicons were processed for pyrosequencing analysis
according to the manufacturer’s standard protocol
(Qiagen, Germantown, MD, USA). To validate the
assay method, various proportions (0, 25, 50, 75
and 100%) of mixtures of CpGenome Universal
Methylated and Unmethylated DNA (Millipore) were
processed with the same protocol. Before proceeding
with the clinical samples, it was confirmed that there
are linear correlations between stepwise increments
of methylated DNA and the measured methylation by
pyrosequencing analysis in all CpG sites.
Genotyping
Genotyping of rs1128306 was performed by pyrose-
quencing. Selected genotypes were confirmed by
resequencing in selected samples including the SMRI
brain, McL brain and WBC 2 samples. PCR was
performed using 10 ng of genomic DNA, 0.5 m
M
primers listed in Supplementary Table S1 and Taq
DNA polymerase (New England BioLabs, Ipswich,
MA, USA). PCR products were purified using
Figure 1 HCG9 investigated region. A diagram of the HCG9 gene located on Chr 6p21.33 (a) and magnified region
(b) subjected to fine mapping of methylated cytosines. Representations are to scale based on the human genome build
(NCBI 36) hg18. The position of the University Health Network (UHN) microarray probe (UHNhscpg0007403) that revealed
DNA methylation differences in major psychiatric disease in our microarray analysis is shown in light blue. Reference
sequence genes are depicted in dark blue while mRNA splice variants 1–3 for HCG9 as determined by AceView are in pink.
Relative gene expression probes from the Affymetrix HGU133A microarray relating to this gene are depicted in (a). Assays 1–
6in(b) represent the regions subjected to sodium bisulfite modification coupled with pyrosequencing. CpG dinucleotides
subjected to DNA methylation analysis are represented by numbered red lollipops while un-interrogated CpGs and those
omitted due to overlap with a database of single nucleotide polymorphisms (dbSNP) citation are shown as upside-down blue
lollipops.
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
730
Molecular Psychiatry
the QIAquick PCR purification kit (Qiagen) and
processed for sequencing by the Center for Applied
Genomics. Sequence alignment was performed by
using ClustalW2.
Data analysis
Normality for the tested distributions was evaluated
using an Anderson–Darling test. As a small percentage
of CpG positions displayed non-Gaussian distribu-
tions, we evaluated significance using non-parametric
tests. Mean values and s.e.m are reported in the text.
Pp0.05 was deemed significant. The sliding window
algorithm used as described in the text. As each test
is not independent, we corrected for multiple test-
ing using the Benjamini–Hochberg–Yekutieli-based
method for calculating the FDR. The effect of
potential confounding factors including brain pH,
post-mortem interval and gender was assessed
through non-parametric correlation. Correction for
age and rs1128306 genotype was performed by taking
the residuals of a linear model of DNA methylation as
a function of age and rs1128306 genotype.
Results
HCG9 methylation analysis in the brain of BPD patients
and controls
We investigated the regions surrounding the 12K CpG
island microarray probe of HCG9: the first exon and a
GC-rich region located 3
0
of the first exon that extends
into the first intron. After treatment with sodium
bisulfite, each genomic DNA sample was subjected to
six pyrosequencing assays (Figure 1b). To maximize
our chances of finding functionally relevant DNA
methylation differences, we evaluated the mean DNA
methylation density per individual for all adjacent
CpG positions and for all possible window sizes of
CpG 1–28 resulting in 406 independent tests. For each
of the 406 tested windows, a non-parametric Wilcoxon
rank sum test was performed on these mean HCG9
methylation values between the BPD and control
group; similar analysis was done to compare the SCZ
and control group. All tests were subjected to
Benjamini–Hochberg–Yekutieli-based FDR correction
for multiple testing, which controls the FDR under
dependency assumptions. This method was first used
to evaluate DNA methylation status in the SMRI brain
cohort (Con N = 35, SCZ N = 35, BPD N = 34; see
Table 1 for details) in attempts to validate the
microarray data. In SCZ studies testing the prefrontal
cortex of post-mortem brains, most results were
non-significant (a single CpG window at CpGs
13–15 that displayed higher methylation in SCZ
as compared with controls; SCZ = 17.4
±
2.67,
Con = 10.4
±
1.34, P = 0.031). SCZ samples in WBC 2
and germline were also tested with few significant
results; and therefore, we decided to focus our efforts
solely on BPD.
We tested BPD in multiple brain regions in two
samples: the SMRI (multiple regions: prefrontal
Table 1 Summary of the demographic variables of samples used in this study
N Age (years) Gender (M:F) PMI (h) AAO (years)
Postmortem brains
SMRI
BPD 34 45.4
±
10.7 16:18 37.9
±
18.6 25.3
±
9.23
SCZ 35 42.6
±
8.47 26:9 31.4
±
15.5 21.3
±
6.07
CON 35 44.2
±
7.58 26:9 29.4
±
12.9
McL
BPD 34 61.4
±
18.9 14:20 22.2
±
7.19
SCZ 30 59.8
±
13.3 20:10 22.8
±
6.30
CON 50 58.4
±
15.6 30:20 20.9
±
4.82
Peripheral white blood cells
Toronto
BPD 100 42.1
±
10 34:66 20.7 þ 0.77
SCZ 50 43.3
±
11.1 25:25 NA
CON 100 41.7
±
10.9 34:66
GSK
BPD 270 44.5
±
12.6 121:149 20.3 þ 0.55
CON 282 42.7
±
0.81 136:146
Germline (sperm)
BPD 29 40.2
±
11.6 20.8 þ 1.71
SCZ 14 37.4
±
11.2 NA
CON 30 37.7
±
10.3
Abbreviations: BPD, bipolar disorder; CON, control; GSK, GlaxoSmithKline; McL, brain samples from the McLean Brain
Tissue bank including the prefrontal cortex; PMI, post-mortem interval in hours; SCZ, schizophrenia; SMRI, brain samples
from the Stanley Medical Research Institute including prefrontal cortex, parietal cortex, occipital cortex and corpus
callosum.
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
731
Molecular Psychiatry
cortex, occipital cortex and corpus callosum) and
McL (prefrontal cortex). No significant differences
were observed in parietal, occipital or corpus callo-
sum brain regions. Prefrontal cortex in both the
SMRI and McL cohorts revealed the largest number
of significant P values (although none survived
correction for multiple testing) but the identified
CpG regions were discrepant between the two cohorts
(Supplementary Table S2). We observed generally
a lower degree of methylation in BPD of SMRI sample,
but higher in the McL sample. For example in SMRI
CpG 3 (BPD = 35.6
±
5.22, Con 50.4
±
4.48, P = 0.018)
but in McL brain sample (BPD = 37.1
±
1.36,
Con 31.2
±
1.72, P = 0.011). To understand the reasons
for the discrepancies, we investigated the effects
of age, medication and DNA sequence differences
on HCG9 methylation in the two brain collections.
Analysis of the discrepancies between the SMRI and
McL brain cohorts
HCG9 methylation increases with age in controls.
The
McL cohort represents significantly older individuals
than the SMRI cohort (SMRI = 44.7.
±
1.12 years,
McL = 60.1
±
1.70 years; t-test, P = 2.4 10
11
). For
control individuals in each cohort, we correlated the
age at death with the average DNA methylation in
each of the 406 CpG window combinations. In both
cohorts, we detected an increase in DNA methylation
with age. This phenomenon appears to be fairly
generalized over all 28 CpGs in the SMRI cohort
(Spearman’s r = 0.30, P = 0.037; Supplementary Table
S4). McL control DNA methylation was positively
correlated over CpGs 10–28 (Spearman’s r = 0.30,
P = 0.037). In a combined analysis of both brain
cohorts, CpGs ranging from 3 to 28 displayed
significant positive correlations with age (Figure 2a).
In the McL cohort, we plotted the average DNA
methylation difference in the most age discrepant
groups (30–40 vs 70–80 years) for all CpG windows
(Figure 2b). DNA methylation in younger BPD
patients was less than controls, resembling the
SMRI results. Together, these results corroborate the
hypothesis that age differences are at least partially
responsible for the discrepant initial findings
between the two brain cohorts.
Effect of medication on HCG9 methylation. In the
SMRI cohort (medication information for the McL
sample was not available), correlation of lifetime
antipsychotic measures (defined as fluphena-
zine equivalents in milligrams) with average age
corrected CpG methylation across the 28 represented
CpGs revealed a marginally significant association at
a single CpG 14 (Spearmans’ r = 0.34, P = 0.05), which
became insignificant after correction for multiple
testing. These results suggest that antipsychotic
medications do not affect DNA methylation at HCG9.
Single-nucleotide polymorphism rs1128306 is asso-
ciated with DNA methylation differences.
The third
factor that may have contributed to discrepancies
between the SMRI and McL cohorts is related to the
effect of DNA sequence variation on DNA methylation
status. Using sequencing data obtained for the HCG9
region of the SMRI data set, we evaluated 12 single-
nucleotide polymorphisms (SNPs) in the region for
evidence of allele-specific methylation. We detected no
DNA methylation difference between alleles of SNPs
rs17180353, rs2071568, rs373472, rs58031868,
rs6903753, rs690402, rs9260832 and rs9278524, while
SNPs rs400488, rs422640, rs9278523 and rs1128306
demonstrated significant association with mean DNA
methylation (Kruskal-Wallis w
2
= 13.26, Bonferroni
P = 0.0024 in all cases). These four SNPs were in
perfect linkage disequilibrium, and so only rs1128306
(G/A) was used as a marker of genetic polymorphism in
the region affecting DNA methylation, as it could
be genotyped using pyrosequencing assay 3 in the
remaining samples. We genotyped the SNP (rs1128306;
G/A) located between CpGs 8 and 9 (Figure 1) and
investigated it for association with HCG9 methylation
at each CpG position. DNA methylation between CpGs
5 and 9 was consistently lower in GG compared with
GA plus AA genotypes (numbers of AA homozygotes
were too small to investigate separately and therefore
such were included in the GA group) reaching
the maximal difference at CpG 5 (Wilcoxon rank
sum test, GG minus GA = 16.8
±
2.3, P =1.7 10
9
)
(Supplementary Figure S1; Supplementary Table S5).
Similar analyses of DNA samples extracted from WBC
and germline showed similar rs1128306 effects and
will be discussed in the sections below.
As the rs1128306 allele A in heterozygotes GA is
associated with a higher density of methylated
cytosines, we tested to see if the SMRI and McL
cohorts had differing case and control allele frequen-
cies for this allele. The BPD group in the SMRI cohort
has less allele A possessing individuals as compared
with that of the McL cohort (allele A frequency: SMRI
Con = 0.4, SMRI BPD = 0.16, McL Con = 0.4, McL
BPD = 0.26, w
2
= 31.2, P = 0.075). The higher propor-
tion of BPD cases containing the allele A in the McL
disease group could be another reason for the
observed higher DNA methylation in BPD patients
in this cohort.
Combined analysis of SMRI and McL brain
cohorts.
We eliminated the effects of age and the
rs1128306 genotype through linear model trans-
formation and performed non-parametric Wilcoxon
rank sum tests on the 406 windows of the overall
brain sample consisting of both SMRI and McL
cohorts. This analysis was equivalent to evaluating
significance with a linear model with age and
genotype as covariates (with the exception that it
was a more conservative non-parametric test). Despite
the differences previously identified, now these two
brain cohorts displayed numerous consistent results
in the exon 1 and exon 1 3
0
UTR region ranging
between CpGs 1 and 10 (BPD = 2.0
±
1.22,
Con 1.68
±
0.86, P = 0.026) with CpG 5 displaying
the largest effect (BPD = 2.3
±
1.52, Con = 1.86
±
1.13,
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
732
Molecular Psychiatry
P = 0.05) (Supplementary Table S3). Negative mean
DNA methylation values are the result of data
correction.
HCG9 methylation studies in WBCs
In addition to the brain studies, we investigated the
HCG9 methylation profiles in peripheral WBCs (total
N = 370 BPD patients and N = 382 controls) using the
same approach. Before disease analysis, we evaluated
the effect of age on DNA methylation in controls from
the WBC cohort. Similar to the brain studies, DNA
methylation increased with age over a majority of
CpGs in the WBC cohort with significant correlations
observed in windows between CpGs 8 and 28 (Spear-
man’s r = 0.3, P = 0.00012). Another similarity with
the brain findings was that GG homozygotes at
rs1128306 displayed significantly lower methylation
compared with GA heterozygotes in the control
population (Supplementary Figure S1; Supplemen-
tary Table S5). After correction for age and rs1128306
genotype, we evaluated DNA methylation associa-
tions with BPD. CpGs 5–8 displayed lower methyla-
tion in BPD (BPD = 0.39
±
0.14, Con = 0.35
±
0.21,
P = 0.0006) (Supplementary Figure S2). All of these
identified differences are located in the same region
and same direction as those identified in the brain
studies.
Information available in the WBC cohort gave us
the opportunity to evaluate the effect of mood
stabilizers on DNA methylation in the region. We
determined daily mood stabilizer dosage information
in the WBC cohort (mood stabilizers included lithium
and valproate) and correlated our 406 possible CpG
windows against these values. No significant correla-
tions were observed for lithium dose, while for
valproate, we obtained significant correlations rang-
ing from CpGs 2 to 9 (Spearman’s r = 0.13, P = 0.05)
and CpG 28 (Spearman’s r = 0.14, P = 0.043). The
direction of correlation suggests that increased mood
stabilizer dose increases DNA methylation toward
control levels. While the strength of the correlations
is not strong enough to conclude that mood stabilizers
help to alleviate HCG9 DNA methylation abnormalities
in BPD patients, these results suggest our disease-
specific associations are not the spurious result of
a class of medications commonly prescribed to BPD
patients.
To understand the BPD predictive value of DNA
methylation changes in peripheral WBCs, we
performed a logistic regression analysis, modeling
Figure 2 Age-related increase in HCG9 methylation. (a) Spearman’s correlation coefficients (r) for age at death and average
DNA methylation in control individuals from Stanley Medical Research Institute (SMRI) and McLean (McL) brain cohorts
combined (N = 84); each line shows the corresponding window of CpGs. Only windows with Pp0.05 are depicted.
(b) Average difference in DNA methylation density between younger (purple: 30–40 years) and older (pink: 70–80 years)
bipolar disorder (BPD) and control group individuals (McL cohort only). Data shown for all possible CpG window variants.
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
733
Molecular Psychiatry
the effect of DNA methylation on disease, controlling
for age and rs1128306 genotype. The mean density of
modified cytosines at CpGs 6–9 displayed the most
significant effect (odds ratio (OR) = 1.07, P = 0.0046).
Next, we evaluated the logistic regression model
based on age, rs1128306 genotype and DNA methyla-
tion at CpGs 5–8 for its ability to predict a diagnosis of
BPD in two ways. We initially generated the model
using WBC 2 (N = 270 BPD patients and 282 controls)
to predict diagnosis in WBC 1 (N = 100 BPD patients
and 100 controls) (see Samples; Figure 3a) and
subsequently randomly selected 90% of the combined
WBC sample to predict diagnosis based on the
remaining 10% consisting of 38 BPD patients and
37 controls (Figure 3b). We compared the predicted
probabilities with the true diagnoses over a range of
probability thresholds to generate a receiver operating
characteristic curve and generated area under the
receiver operating characteristic curve (A prime)
values of 0.69 and 0.72, respectively (Figure 3). This
value means that, given one BPD and one control
sample, these models have 69 and 72% probability,
respectively, of assigning a higher value to the BPD
case.
HCG9 methylation studies in the germline
In the germline samples, the epigenetic effects of the
rs1128306 genotype were consistent with the other
tissues (Supplementary Figure S1; Supplementary
Table S5). After correction for rs1128306 genotype
and age, a significantly lower DNA methylation
density was detected at CpG 5 (BPD = 0.68
±
0.44,
Con = 0.65
±
0.54, P = 0.028), which is also consistent
with brain and WBC findings. Interestingly, in the
germline, unlike the brain and WBC, older age was
associated with lower HCG9 methylation. For example,
raw Spearman’s correlation values were r = 0.65
(P = 0.021) for CpGs 4–20, r = 0.4 (P = 0.033) for
CpGs 1, r = 0.41 (P = 0.035) for CpGs 11–12 and
r = 0.39 (P = 0.036) for CpG 16. The opposite age
effects in germline and WBC suggest that germline
methylation d ifferences in BPD patients were genuine
and not artifacts of WBC contamination, which may
reach 5% of the total cell count in semen samples.
Also, GG and GA methylation differences in the
germline have a peak at CpGs 5–8 but a dip in WBC
(Supplementary Figure S1), which argues for the same.
HCG9 methylation across tissues of BPD patients and
controls
In the process of this project, we noticed that despite
variable density of modified cytosines at each specific
position HCG9 methylation profiles were similar
across the brain, WBC and germline samples, which
suggests that germline HCG9 epigenetic pattern was
partially retained in the somatic tissues. As expected,
the average HCG9 methylation profile across controls
in each tissue correlated with the germline (Figure 4a;
Supplementary Table S6).
We next investigated the relative risk to disease of
genotype- and age-independent DNA methylation
effects using the entire multi-tissue sample. We
evaluated the cumulative effects for brain, WBC and
germline samples using a random effects meta-
analysis to combine logistic regression model DNA
methylation ORs for each tissue across the 406 CpG
windows. Lower DNA methylation at CpGs 5–9
represented the most significant effect on disease risk
(summary OR = 1.24, P = 0.0011; Figure 4b). The
results of this analysis suggest that, independent
of age and the rs1128306 genotype, lower DNA
methylation at HCG9 at CpGs 5–9 is associated with
BPD across the brain, WBC and germline.
Figure 3 Bipolar disorder (BPD) predictive model char-
acteristics. Receiver operating characteristic curves gener-
ated for the prediction of the WBC 1 cohort and a randomly
selected 10% of the combined WBC cohorts using logistic
regression models generated using the mean DNA methyla-
tion density at CpGs 5–8, rs1128306 genotype and age in the
WBC 2 (a) cohorts and 90% of the WBC sample (b). WBC,
white blood cell.
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
734
Molecular Psychiatry
Putative pathological roles of HCG9 in BPD
As the pathological role of intronic methylation
differences at HCG9 in BPD is not straightforward,
we sought to elucidate any relationship between DNA
methylation differences and HCG9 steady-state
mRNA levels. We analyzed three gene expression
data sets (studies 2, 3 and 7) from the SMRI genomics
database that were performed on the Affymetrix
Figure 4 (a) HCG9 DNA methylation profiles by tissue. Mean DNA methylation for the control groups for each of the
28 CpGs for the brain (blue), white blood cell (WBC) (red) and germline (green) cohorts is plotted, demonstrating the
conserved DNA methylation patterns across tissues. (b) Combined sample analysis of bipolar disorder (BPD) risk by genomic
location. The odds ratio (OR) for BPD risk due solely to DNA methylation differences in BPD patients and controls is plotted
relative to the middle genomic position of each of the 406 tested CpG windows. The rs1128306 single-nucleotide
polymorphism (SNP) and age were modeled as covariates in a logistic regression model and ORs due to DNA methylation
were combined with a random effects meta-analysis for all tissues. The middle genomic positions of CpG windows where
DNA methylation significantly contributes to risk are plotted in red. Of interest, when the rs1128306 polymorphism alone
was evaluated in BPD evidence for association was observed for the combined brain sample and WBC sample; however,
allele G exerted an increased risk in the combined brain (OR = 2.45, P = 0.017) but protective effect in WBC (OR = 0.72,
P = 0.044, respectively). No association was observed in the germline sample or an analysis of all samples combined. (c) DNA
methylation vs HCG9 steady-state mRNA levels. Non-parametric Spearman correlations (r) between DNA methylation and
the ratio of HCG9 splice variants 3 and 1 (windows with Pp0.05) (black, Y axis left), and mean histone occupancy scores as
modeled by a sequence-based histone prediction program (blue, Y axis right).
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
735
Molecular Psychiatry
HGU133A gene expression arrays, which contain 11
probes covering HCG9 (Figure 1a). A limited number
of McL cohort individuals (N = 34) also contained
steady-state mRNA data from the same array platform,
and we performed HCG9 mRNA analyses on the
combined SMRI and McL brain data set. As there are
indications that intronic chromatin modifications
may be involved in alternative splicing, we were
interested to ascertain steady-state mRNA levels for
individual HCG9 splice variants (Figure 1a). In order
to do this, we subjected all data to background
correction and quantile normalization and took the
mean log
2
probe value for each of the 11 probes
spanning HCG9. Probes 10 and 11 (probe group 1)
uniquely bind mRNA from splice variant 3, while
probes 1–4 (probe group 2) bind mRNA from variants
1 and 3 (Figure 1a). After averaging these two probe
groups, we determined the levels of HCG9 splice
variant 1 by taking the residuals of a linear model
between probe groups 1 and 2, statistically subtract-
ing splice variant 3 from the combined 1 and 3
measures. A significantly higher log ratio of HCG9
mRNA splice variant 3/1 was detected in BPD
compared with controls (Wilcoxon rank sum test,
mean
±
s.e.m.: BPD = 1.45
±
2.17, Con = 1.77
±
15.6,
P = 0.034).
Since the maximal DNA methylation differences in
BPD were detected around CpG 5 in the two brain
cohorts, and CpGs 5–9 represented the most signifi-
cant effect on disease risk in the total sample of the
three tissues, we hypothesized that epigenetic modi-
fications may be contributing to splicing decisions in
this region which corresponds to the overlapping
splice variants 1 and 3. The log ratio of splice variant
3 over splice variant 1 was tested using the DNA
methylation sliding window. CpG 6 (that is, next to
CpG 5) displayed a significant negative correlation
with splice variant ratios (Spearman’s r = 0.23,
P = 0.029; Figure 4c) although did not survive correc-
tion for multiple testing. The largest correlation was
detected for CpG 25 (Spearman’s r = 0.44,
P = 5.33 10
5
, FDR P = 0.022; Figure 4c). Interest-
ingly, CpG 25 maps next to CpGs 23–24 which in the
SMRI brain cohort displayed the highest significance
for differential methylation between the BPD and
control groups (BPD = 10.7
±
3.28, Con 15.2
±
1.60,
P = 0.0021) (Supplementary Table S1). Other signifi-
cant correlations are shown in Supplementary
Table S7. Both CpG 6 and CpG 25 corresponded to
peak mean histone occupancy scores 7–14 bp down-
stream of CpG 6 and B60 bp upstream of CpG 25 as
predicted by a nucleosome prediction algorithm
35
(Figure 4c). Another algorithm
36
mapped tentative
nucleosome centers 20–50 bp upstream of CpG 6 and
ranging from B45 bp upstream to 17 bp downstream
of CpG 25, while a third program, FineStr 1.0,
37
mapped nucleosome centers B60 bp downstream of
CpG 6 and 2 bp downstream of CpG 25. As the amount
of DNA that coils around a single histone octamer
is B150 bp, CpGs 5–8 and 24–27 can be reason-
ably associated with a single nucleosome peak.
Consistently with our initial observation of lower
methylation at HCG9 in BPD patients, these data
suggest that as HCG9 DNA methylation decreases in
BPD, there is an increase in the ratio of splice variant
3 to splice variant 1 in the brain.
HCG9 has three potential splice variants, generating
77, 37 and 121 amino-acid (aa) proteins (Figure 1).
Using Protein BLAST, we determined the 37 and 77
aa peptides lack conserved domains and do not share
homology with any proteins in the non-redundant
protein database. The 121 aa peptide also lacked
conserved domains, however, did produce a single hit
in a known locus with the exported acetyl esterase
(YP_001901494) in Xanthomonas campestris. These
class of enzymes have been implicated in regulating
B-cell immunological tolerance, and the sialic acid
acetylesterase was recently associated with autoim-
mune disorders including rheumatoid arthritis and
type 1 diabetes.
38
We next attempted to elucidate the function of
HCG9 by performing gene ontology (GO) analysis
on those genes whose expression from the previously
mentioned SMRI data sets is most tightly correlated
with that of HCG9. For the BPD and control
individuals in the SMRI studies 2, 3 and 7 performed
on the HGU133A gene expression platform, we
correlated the expression of all probes against that
of HCG9 using Spearman method followed by FDR
correction. We selected only those genes significantly
correlated with HCG9 in all three studies using
an FDR P < 0.05 and subsequently performed GO
analysis. GO categories associated with genes in this
selected group were tested against their frequency
of occurrence on the Affymetrix HGU133A array.
Significance was assessed using the binomial test in R
and corrected for multiple testing using the Bonferroni
method. Significant GO categories (Bonferroni
P < 0.01) appear in Supplementary Table S8. The
results suggest that when expression of HCG9
changes, so does that of genes involved in signaling
and immune system-related functions such as inflam-
mation and regulation of B cell-based immunological
tolerance, among others.
Discussion
Microarray-based DNA methylation profiling pro-
vides an unbiased approach for the identification of
epigenetic differences that may have an etiological
role in human disease. Our earlier study of B1%
of the methylome revealed a number of candidate
regions displaying altered DNA methylation signa-
tures associated with major psychosis.
39
In this
project, we explored methylation patterns of an HLA
gene, HCG9,inpost-mortem brain tissues, WBC, and
the germline, which all consistently revealed lower
methylation of HCG9 in BPD patients compared with
the controls.
HCG9 methylation patterns in the brain are
complex and depend on age as well as DNA sequence
variation. Several reports implicated age as a
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
736
Molecular Psychiatry
contributing factor to DNA methylation change.
40–42
HCG9 methylation increases in the aging brain and
seems to be approaching the normal state observed in
controls, which is consistent with the common
observation that clinical symptoms of BPD in aging
individuals become less severe.
43
Alternatively, DNA
methylation levels decreased with age in the germline
sample, toward the risk epigenotype. This finding
suggests that with increasing paternal age, the prob-
ability of contributing low DNA methylation sperm
upon fertilization increases and may have relevance
to epidemiological observations of higher incidence
of BPD among children of older fathers.
44–46
DNA methylation differences associated with
genetic polymorphisms have been documented and
are known as allele-specific methylation patterns.
39,47
The rs1128306 SNP at HCG9 also contributes to DNA
methylation differences across individuals. More
specifically, allele A exhibited a higher density of
methylated cytosines in B65 bp in each direction
of CpG 8 in comparison with the alternative allele G.
This was observed across all three investigated
tissues. While we identified associations of this SNP
with BPD between the combined brain sample and
WBC sample, alone, the sample sizes of these
individual cohorts are likely too small to identify
true SNP associations with BPD and may explain why
allele G appeared to exert a risk inducing and
protective effect in the respective cohorts. These
results highlight the possibility that DNA polymorph-
isms associated with disease, such as those identified
in GWAS studies, may be acting synergistically
with epigenetic misregulation.
48,49
The relative con-
tribution of DNA sequence variation vs epigenetic
variation to the disease phenotype may vary signifi-
cantly. The extreme case is when only epigenetic
differences have a pathological role while SNPs are
only the reporters of such epimutations. It is also
interesting to note that the G/A polymorphism
(flanking nucleotides are C and A) may have origi-
nated from a C-T mutation in the complementary
strand (T[C/T]G). The dinucleotide CpG is a ‘hotspot’
for mutation in the human genome, most likely due to
high frequency of spontaneous deamination of
5-methylcytosine to thymidine.
50
The ancestral allele
of rs1128306 SNP is unknown but based on a
primarily Caucasian population, allele G(C) is three
times as common as allele A(T) (0.75 and 0.25,
respectively), and the later allele exhibits higher
methylation compared with the former, consistent
with the epigenetic origin of rs1128306 SNP. There-
fore, in addition to a direct pathogenic effect, DNA
methylation can contribute to human genetic disease
indirectly, that is, via increased mutation rates.
The observation of a DNA sequence–epigenetic
association is particularly interesting in the light of
recent findings suggesting that 6p21.3-6p22 (HCG9
maps to NCBI 36; chr6:30050871-30054156) has been
implicated by several large GWAS in major psycho-
sis.
28–30
The rs1128306 SNP was a strong marker for a
portion of the epigenetic variation in the region.
Unfortunately, no linkage disequilibrium data exists
in HapMap for this SNP; however, this and possibly
other SNPs could be linked to the markers tested by
the GWAS, suggesting that a portion of the disease
association signal coming out of the major histocom-
patibility complex (MHC) could represent this and
possibly other epigenetically misregulated regions.
Of particular interest and added value were the two
non-brain tissues, WBC and sperm, which ‘mirrored’
most of the brain findings and provided additional
confidence that the HCG9 methylation changes in the
post-mortem brains of BPD patients are genuine. The
identification of an epigenetic difference in BPD
detectable in WBCs holds promise for the identification
of predictive biomarkers for the disease. Although area
under the receiver operating characteristic curves
demonstrated only a mildly predictive value for disease,
given that HCG9 was selected from an interrogation of
only B1% of the epigenome,
39
the HCG9 epigenetic
biomarker for BPD stands as a proof of principle that the
epigenomic studies of multiple tissues may result in
clinical applications. Identification of biomarkers
should be far more successful if the candidate genes
or regions are selected from the most significant changes
observed in epigenome-wide analyses.
In epigenetic studies, cellular heterogeneity is a
potential confounder. Peripheral WBC DNA comes
from a wide variety of blood cells, of which the
proportion of cells retaining the inherited or acquired
pre-epimutation may be quite small. For example, the
pre-epimutation may be lost in differentiation from
a precursor stem cell to some mature cell types and
not others. Similarly, in the brain tissues, HCG9
methylation differences could be resultant from
having only a particular subset of neurons or glial
cells affected and ultimately conferring the observed
epigenetic difference. The epigenetic noise caused
by cellular heterogeneity could be affecting our
predictive model of BPD diagnosis. One of the next
steps of HCG9 methylation studies in BPD or other
psychiatric disease could be dedicated to investiga-
tion of particular WBC subtypes or homogenous brain
cell populations. Cellular heterogeneity issues call for
the development and optimization of technologies
capable of isolating and investigating small numbers
of cells, even to the level of the single cell to
adequately address these issues.
One major caveat of epigenetic studies of complex
disease is that the etiological significance of identified
DNA methylation changes is not immediately evident
as epigenetic patterns in the diseased tissue may be
influenced by the disease state, treatment or other
events related to the pathological process. We sought
to evaluate the cause vs effect relationship between the
identified DNA methylation differences and disease at
the HCG9 gene by analyzing non-brain tissues, WBC
and germline. The lower HCG9 methylation density in
peripheral WBCs mapping to the same region of HCG9
nominated from the combined brain cohort argues
(although does not prove) for an etiological role of
HCG9 methylation differences in BPD. The causal link
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
737
Molecular Psychiatry
between HCG9 epimutation and BPD is further
supported by our analysis of sperm DNA, which also
revealed HCG9 methylation differences (for example,
CpG 5) in BPD patients. In general, DNA methylation
is known to be subject to reprogramming during both
gametogenesis and after fertilization in humans and
mice; however, examples exist where epigenetic
alterations in the parental generation are passed to
the offspring.
51–53
The results of germline studies allow
us to infer the possibility that HCG9 epimutation may
represent one of the heritable epigenetic risk factors in
BPD. It is important to keep this finding in perspective,
however, as the epigenetic difference was identified in
the sperm of men affected with BPD and not the
fathers of BPD patients. We can only suggest that the
identified HCG9 methylation profiles are to some
extent meiotically stable and thus may have the
potential to survive successive epigenetic resetting
and be passed to the next generation. The mechanisms
of such a passage of epigenetic information are beyond
the scope of this study but may involve a communica-
tion between inherited DNA methylation and histone
protein modifications, non-coding RNAs or interaction
with DNA sequence variation in the region.
Functional implications of the detected epigenetic
difference at CpGs 5–9, just downstream of the first
coding exon, on HCG9 activity and its role in BPD
require further study. Although traditionally the
regulatory function of DNA methylation is most
characteristically recognized in its role in the gene
promoter,
54–57
in the case of HCG9 and BPD, the
disease-related mechanism may be different. One
possibility is that DNA methylation at CpGs 5–9
may be contributing to alternative splicing scenarios
of the HCG9 mRNA transcript. According to AceView,
HCG9 has three mRNA splice variants, only two of
which (77 and 121 aa) may be encoding protein
products; however, it remains possible that one or
both of these splice variants represent non-coding
transcripts. There is a growing body of evidence that
chromatin conformation and histone modifications
help direct aspects of co-transcriptional splicing.
58
In
particular, reduction of histone 3 lysine 4 tri-methyla-
tion (H3K4me3) in conjunction with CHD1 chromatin-
remodeling ATPase knockdown was found to alter
splicing efficiency,
59
while increased levels of histone
3 lysine 9 (H3K9) acetylation triggered by neuronal
depolarization were associated with exon skipping.
60
In our experiments, we found correlations between
DNA methylation and HCG9 splice variant ratios at
peaks of predicted histone occupancy, suggesting that
our DNA methylation levels may be a reflection of
associated histone modification status and that
epigenetic alterations in these positions are important
for alternative splicing mechanisms. The decreased
HCG9 DNA methylation in BPD patients may account
for the increased ratio of splice variant 3 to splice
variant 1 in the brain of affected individuals. The
observed increased ratio of the 121 aa over 77 aa
protein in BPD may represent a gain of deleterious
neural inflammatory function or a loss of protective
anti-inflammatory function in BPD and may contri-
bute to disease risk.
Using GO analysis, we observed that HCG9 is co-
expressed with genes involved in inflammation while
Protein BLAST identified homology of the 121 aa
peptide with a class of genes negatively associated
with inflammatory response.
61
CNS inflammation has
been hypothesized as a contributing factor for BPD,
62
while pro-inflammatory biomarkers have been found
to be predictive of acute episodes in BPD.
63
Further-
more, antipsychotic and antidepressant medications
have been known for their ability to reduce inflam-
mation.
64–68
A downregulation of HCG9 was observed
in a recent transcriptome analysis after treatment with
pro-inflammatory factors, which is consistent with
the interpretation that one of the splice variants of
HCG9 confers an anti-inflammatory signal.
32
Loss of
HCG9 protective function as a result of alternative
splicing could account for the deleterious effects of
the epimutation and its association with disease.
Recently, one study identified a change in HCG9
expression levels with the addition of a histone
deacetylase inhibitor suberoylanilide hydroxamic
acid and taxanes to human breast cancer cells,
69
further corroborating an epigenetic component to
HCG9 expression and suggesting that epigenetic
modifiers may have therapeutic benefit in BPD.
Another important activity in the further character-
ization of the role of HCG9 in BPD is identification of
the role of 5-hydroxymethylcytosine, which is known
to be present at high levels in the brain.
70,71
It has been
recently shown that 5-hydroxymethylcytosine, like
5-methylcytosine, does not undergo C-to-T transitions
after bisulfite treatment, indicating that these two
modified cytosine species are indistinguishable by
the bisulfite sequencing technique.
72
Therefore, all
our estimates of methylated cytosines in HCG9,infact,
represent both types of cytosine modification. Recently,
some new approaches have been developed to differ-
entiate the two types of cytosine modifications
73–75
and
these can be used in future studies to address the roles
of 5-hydroxymethylcytosine in health and disease.
The results of this study highlight the importance of
controlling for confounding influences in brain
tissue-based studies of complex disease. We identi-
fied significant DNA methylation associations with
age and genetic influences that, if ignored, have the
potential to result in false-positive or false-negative
findings. Additionally, we used the analytical strategy
that favors replication across different tissues rather
than direct statistical evidence for isolated analysis of
each tissue. By using a sliding window-based method
and performing statistical tests on all possible
adjacent CpG combinations, we identified overlaps
across tissues and replication cohorts that we would
have otherwise missed. This approach may become
exemplary in the primary studies of DNA modifica-
tion differences when neither the number of cytosines
involved, nor the multiple ways of their interaction is
known. This study highlights the need for methodo-
logical principles for epigenomic and epigenetic
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
738
Molecular Psychiatry
studies of complex disease that are quite different from
the ones used in DNA sequence-based approaches.
Note added in proof: While the manuscript was
reviewed and processed, we performed a pilot
analysis of 5-hydroxymethylation (5-hmC) at CpG 6
which happens to be at the target site for MspI
restriction enzyme. Glucosylation of the 5-hmC at
CCGG renders such targets resistant to MspI diges-
tion, and the differential enrichment of MspI-target
amplicons with and without glucosylation estimates
5-hmC. In this experiment, a lower level of 5-hmC
was identified in a subset of BPD prefrontal cortex
samples compared to controls, suggesting that a
portion of the identified DNA methylation differences
in this study may be related to 5-hmC and warranting
further follow-up analyses.
Conflict of interest
The authors declare no conflict of interest.
Acknowledgments
We would like to thank the Stanley Medical Research
Institute (SMRI) and Dr Francine Benes of the McLean
Hospital Brain Bank for the provided brain tissues.
ZK and JM were CIHR fellows. MT was supported by
the Japan Society for the Promotion of Science
Postdoctoral Fellowships for Research Abroad. AP is
Senior Fellow, Ontario Mental Health Foundation,
and Tapscott Chair in Schizophrenia Studies, Uni-
versity of Toronto. This work was supported by the
Canadian Institutes for Health and Research (186007)
and the National Institutes of Health (MH074127;
MH088413) to AP.
Web Resources
AceView, http://www.ncbi.nlm.nih.gov/IEB/Research/
Acembly/
Clustal W2, http://www.ebi.ac.uk/Tools/clustalw2/
index.html
FineStr, http://www.cs.bgu.ac.il/~nucleom/
Nucleosome Prediction Algorithm, http://genie.weiz-
mann.ac.il/software/nucleo_prediction.html/
Online Mendelian Inheritance in Man (OMIM),
http://www.ncbi.nlm.nih.gov/Omim/
Stanley Medical Research Institute Genomics Data-
base, http://www.stanleygenomics.org/
References
1 Petronis A. Human morbid genetics revisited: relevance of
epigenetics. Trends Genet 2001; 17: 142–146.
2 Robertson KD. DNA methylation and human disease. Nat Rev
Genet 2005; 6: 597–610.
3 Feinberg AP. Phenotypic plasticity and the epigenetics of human
disease. Nature 2007; 447: 433–440.
4 Petronis A. Epigenetics as a unifying principle in the aetiology of
complex traits and diseases. Nature 2010; 465: 721–727.
5 Yeivin A, Razin A. Gene methylation patterns and expression. In:
Jost J, Saluz, H (eds). DNA Methylation: Molecular Biology and
Biological Significance. Birkhauser Verlag: Basel, 1993 pp. 523–568.
6 Holliday R, Ho T, Paulin R. Gene silencing in mammalian cells. In:
VEA Russo RM, Riggs AD (eds). Ep igenetic Mechanisms of Gene
Regulation. Cold Spring Harbor Laboratory Press: Cold Spring
Harbor, New York, 1996, pp 47–59.
7 Riggs AD, Xiong Z, Wang L, Lebon JM. Methylation dynamics,
epigenetic fidelity and X chromosome structure. Novartis Found
Symp 1998; 214: 214–225; discussion 225–32.
8 Ehrlich M, Ehrlich, K. Effect of DNA methylation and the binding
of vertebrate and plant proteins to DNA. In: Jost J, Saluz P (eds).
DNA Methylation: Molecular Biology and Biological Significance.
Birkhauser Verlag: Basel, Switzerland, 1993, pp 145–168.
9 Riggs A, Xiong Z, Wang L, Jm L. Methylation dynamics, epigenetic
fidelity and X chromosome structure. In: Wolffe, A (ed).
Epigenetics. John Wiley & Sons: Chistester, 1998, pp 214–227.
10 Fraga MF, Ballestar E, Paz MF, Ropero S, Setien F, Ballestar ML et
al. Epigenetic differences arise during the lifetime of monozygotic
twins. Proc Natl Acad Sci USA 2005; 102: 10604–10609.
11 Kaminsky ZA, Tang T, Wang SC, Ptak C, Oh GH, Wong AH et al.
DNA methylation profiles in monozygotic and d izygotic twins.
Nat Genet 2009; 41: 240–245.
12 Richards EJ. Inherited epigenetic variation—revisiting soft inheri-
tance. Nat Rev Genet 2006; 7: 395–401.
13 Petronis A. Epigenetics and bipolar disorder: new opportunities
and challenges. Am J Med Genet 2003; 123C: 65–75.
14 Petronis A. The genes for major psychosis: aberrant sequence or
regulation? Neuropsychopharmacology 2000; 23: 1–12.
15 Kaminsky Z, Wang SC, Petronis A. Complex disease, gender and
epigenetics. Ann Med 2006; 38: 530–544.
16 Hall JG. Genomic imprinting: review and relevance to human
diseases. Am J Hum Genet 1990; 46: 857–873.
17 Barlow DP. Gametic imprinting in mammals. Science 1995; 270:
1610–1613.
18 Newberg AR, Catapano LA, Zarate CA, Manji HK. Neurobiology of
bipolar disorder. Expert Rev Neurother 2008; 8: 93–110.
19 Guidotti A, Dong E, Kundakovic M, Satta R, Grayson DR, Costa E.
Characterization of the action of antipsychotic subtypes on
valproate-induced chromatin remodeling. Trends Pharmacol Sci
2009; 30: 55–60.
20 Dong E, Chen Y, Gavin DP, Grayson DR, Guidotti A. Valproate
induces DNA demethylation in nuclear extracts from adult mouse
brain. Epigenetics 2010; 5: 730–735.
21 Kwon B, Houpt TA. Phospho-acetylation of histone H3 in the
amygdala after acute lithium chloride. Brain Res 2010; 1333: 36–47.
22 Mccoll G, Killilea DW, Hubbard AE, Vantipalli MC, Melov S,
Lithgow GJ. Pharmacogenetic analysis of lithium-induced delayed
aging in Caenorhabditis elegans. J Biol Chem 2008; 283: 350–357.
23 Petronis A. The origin of schizophrenia: genetic thesis, epigenetic
antithesis, and resolving synthesis. Biol Psychiatry 2004; 55: 965–970.
24 Schumacher A, Petronis A. Epigenetics of complex diseases: from
general theory to laboratory experiments. Curr Top Microbiol
Immunol 2006; 310: 81–115.
25 Flanagan JM, Popendikyte V, Pozdniakovaite N, Sobolev M,
Assadzadeh A, Schumacher A et al. Intra- and interindividual
epigenetic variation in human germ cells. Am J Hum Genet 2006;
79: 67–84.
26 Sutherland JE, Costa M. Epigenetics and the environment. Ann NY
Acad Sci 2003; 983: 151–160.
27 Jaenisch, R, Bird, A. Epigenetic regulation of gene expression:
how the genome integrates intrinsic and environmental signals.
Nat Genet 2003; 33(Suppl): 245–54.
28 Shi J, Levinson DF, Duan J, Sanders AR, Zheng Y, Pe’er I et al.
Common variants on chromosome 6p22.1 are associated with
schizophrenia. Nature 2009; 460: 753–757.
29 Purcell SM, Wray NR, Stone JL, Visscher PM, O’donovan MC,
Sullivan PF et al. Common polygenic variation contributes to risk
of schizophrenia and bipolar disorder. Nature 2009; 460: 748–752.
30 Stefansson H, Ophoff RA, Steinberg S, Andreassen OA, Cichon S,
Rujescu D et al. Common variants conferring risk of schizophrenia.
Nature 2009; 460: 744–747.
31 Gaudieri, S, Longman-Jacobsen N, Tay GK, Dawkins RL. Sequence
analysis of the MHC class I region reveals the basis of the genomic
matching technique. Hum Immunol 2001; 62: 279–285.
32 Seidl, C, Port M, Apostolidis C, Bruchertseifer F, Schwaiger M,
Senekowitsch-Schmidtke R et al. Differential gene expression
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
739
Molecular Psychiatry
triggered by highly cytotoxic alpha-emitter-immunoconjugates in
gastric cancer cells. Invest New Drugs 2010; 28: 49–60.
33 Torrey EF, Webster M, Knable M, Johnston N, Yolken RH.
The Stanley foundation brain collection and neuropathology
consortium. Schizophr Res 2000; 44: 151–155.
34 Kaminsky Z, Petronis A. Methylation SNaPshot: a method for the
quantification of site-specific DNA methylation levels. Methods
Mol Biol 2009; 507: 241–255.
35 Kaplan N, Moore IK, Fondufe-Mittendorf Y, Gossett AJ, Tillo D,
Field Y et al The DNA-encoded nucleosome organization of a
eukaryotic genome. Nature 2009; 458: 362–366.
36 Mavrich TN, Ioshikhes IP, Venters BJ, Jiang C, Tomsho LP, Qi J
et al. A barrier nucleosome model for statistical positioning of
nucleosomes throughout the yeast genome. Genome Res 2008; 18:
1073–1083.
37 Gabdank I, Barash D, Trifonov EN. FineStr: a web server for single-
base-resolution nucleosome positioning. Bioinformatics 2010; 26:
845–846.
38 Surolia I, Pirnie SP, Chellappa V, Taylor KN, Cariappa A, Moya J
et al. Functionally defective germline variants of sialic acid
acetylesterase in autoimmunity. Nature 2010; 466: 243–247.
39 Mill J, Tang T, Kaminsky Z, Khare T, Yazdanpanah S, Bouchard L
et al. Epigenomic profiling reveals DNA-methylation changes
associated with major psychosis. Am J Hum Genet 2008; 82:
696–711.
40 Fuke C, Shimabukuro M, Petronis A, Sugimoto J, Oda T, Miura K
et al. Age related changes in 5-methylcytosine content in
human peripheral leukocytes and placentas: an HPLC-based
study. Ann Hum Genet 2004; 68(Part 3): 196–204.
41 Rakyan VK, Down TA, Maslau S, Andrew T, Yang TP, Beyan H
et al. Human aging-associated DNA hypermethylation occurs
preferentially at bivalent chromatin domains. Genome Res 2010;
20: 434–439.
42 Teschendorff AE, Menon U, Gentry-Maharaj A, Ramus SJ,
Weisenberger DJ, Shen H et al. Age-dependent DNA methylation
of genes that are suppressed in stem cells is a hallmark of cancer.
Genome Res 2010; 20: 440–446.
43 Depp CA, Jeste DV. Bipolar disorder in older adults: a critical
review. Bipolar Disord 2004; 6: 343–367.
44 Frans EM, Sandin S, Reichenberg A, Lichtenstein P, Langstrom N,
Hultman CM. Advancing paternal age and bipolar disorder.
Arch Gen Psychiatry 2008; 65: 1034–1040.
45 Menezes PR, Lewis G, Rasmussen F, Zammit S, Sipos A,
Harrison GL et al. Paternal and maternal ages at conception and
risk of bipolar affective disorder in their offspring.
Psychol Med
2010; 40: 477–485.
46 Crow TJ. Mutation and psychosis: a suggested explanation of
seasonality of birth. Psychol Med 1987; 17: 821–828.
47 Kerkel K, Spadola A, Yuan E, Kosek J, Jiang L, Hod E et al.
Genomic surveys by methylation-sensitive SNP analysis identify
sequence-dependent allele-specific DNA methylation. Nat Genet
2008; 40: 904–908.
48 Tycko, B. Allele-specific DNA methylation: beyond imprinting.
Hum Mol Genet 2010; 19(R2): R210–R220.
49 Meaburn, EL, Schalkwyk LC, Mill J. Allele-specific methylation in
the human genome Implications for genetic studies of complex
disease. Epigenetics 2010; 5: 578–582.
50 Cooper DN, Krawczak M. Cytosine methylation and the fate of
CpG dinucleotides in vertebrate genomes. Hum Genet 1989; 83:
181–188.
51 Morgan HD, Sutherland HG, Martin DI, Whitelaw E. Epigenetic
inheritance at the agouti locus in the mouse. Nat Genet 1999; 23:
314–318.
52 Oates NA, Van Vliet J, Duffy DL, Kroes HY, Martin NG, Boomsma
DI et al. Increased DNA methylation at the AXIN1 gene in a
monozygotic twin from a pair discordant for a caudal duplication
anomaly. Am J Hum Genet 2006; 79: 155–162.
53 Rakyan VK, Chong S, Champ ME, Cuthbert PC, Morgan HD,
Luu KV et al. Transgenerational inheritance of epigenetic states at
the murine Axin(Fu) allele occurs after maternal and paternal
transmission. Proc Natl Acad Sci USA 2003; 100: 2538–2543.
54 Dulac C. Brain function and chromatin plasticity. Nature 2010;
465: 728–735.
55 Ehrlich M, Wang RY. 5-Methylcytosine in eukaryotic DNA.
Science 1981; 212: 1350–1357.
56 Brown DD. Gene expression in eukaryotes. Science 1981; 211:
667–674.
57 Razin A, Riggs AD. DNA methylation and gene function. Science
1980; 210: 604–610.
58 Schwartz S, Ast G. Chromatin density and splicing destiny: on the
cross-talk between chromatin structure and splicing. EMBO J 2010;
5: 578–582.
59 Sims 3rd RJ, Millhouse S, Chen CF, Lewis BA, Erdjument-Bromage
H, Tempst P et al. Recognition of trimethylated histone H3 lysine 4
facilitates the recruitment of transcription postinitiation factors
and pre-mRNA splicing. Mol Cell 2007; 28: 665–676.
60 Schor IE, Rascovan N, Pelisch F, Allo M, Kornblihtt AR. Neuronal
cell depolarization induces intragenic chromatin modifications
affecting NCAM alternative splicing. Proc Natl Acad Sci USA
2009; 106: 4325–4330.
61 Hubbard RE, O’mahony MS, Calver BL, Woodhouse KW. Plasma
esterases and inflammation in ageing and frailty. Eur J Clin
Pharmacol 2008; 64: 895–900.
62 Matigian N, Windus L, Smith H, Filippich C, Pantelis C, Mcgrath J
et al. Expression profiling in monozygotic twins discordant for
bipolar disorder reveals dysregulation of the WNT signalling
pathway. Mol Psychiatry 2007; 12: 815–825.
63 Kapczinski F, Dal-Pizzol F, Teixeira AL, Magalhaes PV,
Kauer-Sant’anna M, Klamt F et al. Peripheral biomarkers and
illness activity in bipolar disorder. J Psychiatr Res 2010; 45:156161.
64 Laan W, Grobbee DE, Selten JP, Heijnen CJ, Kahn RS, Burger H.
Adjuvant aspirin therapy reduces symptoms of schizophrenia
spectrum disorders: results from a randomized, double-blind,
placebo-controlled trial. J Clin Psychiatry 2010; 71: 520–527.
65 Monji A, Kato T, Kanba S. Cytokines and schizophrenia: microglia
hypothesis of schizophrenia. Psychiatry Clin Neurosci 2009; 63 :
257–265.
66 Goldstein BI, Kemp DE, Soczynska JK, Mcintyre RS. Inflammation
and the phenomenology, pathophysiology, comorbidity, and
treatment of bipolar disorder: a systematic review of the literature.
J Clin Psychiatry 2009; 70: 1078–1090.
67 Sugino H, Futamura T, Mitsumoto Y, Maeda K, Marunaka Y.
Atypical antipsychotics suppress production of proinflammatory
cytokines and up-regulate interleukin-10 in lipopolysaccharide-
treated mice. Prog Neuropsychopharma col Biol Psychiatry 2009;
33: 303–307.
68 Muller N. Inflammation and the glutamate system in schizophre-
nia: implications for therapeutic targets and drug development.
Expert Opin Ther Targets 2008; 12: 1497–1507.
69 Chang H, Jeung HC, Jung JJ, Kim TS, Rha SY, Chung HC.
Identification of genes associated with chemosensitivity to
SAHA/taxane combination treatment in taxane-resistant breast
cancer cells. Breast Cancer Res Treat 2011; 125: 55–63.
70 Tahiliani M, Koh KP, Shen Y, Pastor WA, Bandukwala H, Brudno Y
et al. Conversion of 5-methylcytosine to 5-hydroxymethylcytosine
in mammalian DNA by MLL partner TET1. Science 2009; 324:
930–935.
71 Kriaucionis S, Heintz N. The nuclear DNA base 5-hydroxymethyl-
cytosine is present in Purkinje neurons and the brain. Science
2009; 324: 929–930.
72 Huang, Y, Pastor WA, Shen Y, Tahiliani M, Liu DR, Rao A. The
behaviour of 5-hydroxymethylcytosine in bisulfite sequencing.
PLoS One 2010; 5: e8888.
73 Zheng Y, Cohen-Karni D, Xu D, Chin HG, Wilson G, Pradhan S
et al. A unique family of Mrr-like modification-dependent
restriction endonucleases. Nucleic Acids Res 2010; 38: 5527–5534.
74 Szwagierczak A, Bultmann S, Schmidt CS, Spada F, Leonhardt H.
Sensitive enzymatic quantification of 5-hydroxymethylcytosine in
genomic DNA. Nucleic Acids Res 2010; 38: e181.
75 Jin SG, Kadam S, Pfeifer GP. Examination of the specificity of DNA
methylation profiling techniques towards 5-methylcytosine and
5-hydroxymethylcytosine. Nucleic Acids Res 2010; 38: e125.
Supplementary Information accompanies the paper on the Molecular Psychiatry website (http://www.nature.com/mp)
Analysis of HCG9 methylation differences in BPD
Z Kaminsky et al
740
Molecular Psychiatry
    • "Consistent methylation results across brain and blood tissues have been reported for particular promoter CpG islands in other studies not included in this review [17] . Notably, two of the included studies showed comparable methylation results for major histocompatibility complex 9 (HCG9) [46] and synapsin 3 (SYNIII) [35] in both post-mortem brain tissue as well as a blood-derived DNA. "
    [Show abstract] [Hide abstract] ABSTRACT: Background Increasing evidence suggests the involvement of epigenetic processes in the development of schizophrenia and bipolar disorder, and recent reviews have focused on findings in post-mortem brain tissue. A systematic review was conducted to synthesise and evaluate the quality of available evidence for epigenetic modifications (specifically DNA methylation) in peripheral blood and saliva samples of schizophrenia and bipolar disorder patients in comparison to healthy controls. Methods Original research articles using humans were identified using electronic databases. There were 33 included studies for which data were extracted and graded in duplicate on 22 items of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement, to assess methodological precision and quality of reporting. Results There were 15 genome-wide and 18 exclusive candidate gene loci investigations for DNA methylation studies. A number of common genes were identified as differentially methylated in schizophrenia/bipolar disorder, which were related to reelin, brain-derived neurotrophic factor, dopamine (including the catechol-O-methyltransferase gene), serotonin and glutamate, despite inconsistent findings of hyper-, hypo-, or lack of methylation at these and other loci. The mean STROBE score of 59 % suggested moderate quality of available evidence; however, wide methodological variability contributed to a lack of consistency in the way methylation levels were quantified, such that meta-analysis of the results was not possible. Conclusions Moderate quality of available evidence shows some convergence of differential methylation at some common genetic loci in schizophrenia and bipolar disorder, despite wide variation in methodology and reporting across studies. Improvement in the clarity of reporting clinical and other potential confounds would be useful in future studies of epigenetic processes in the context of exposure to environmental and other risk factors.
    Full-text · Article · Dec 2016
    • "The authors of the latter study have argued that the mildly predictive value found for this marker may be accounted for the cellular heterogeneity in the white blood cells and brain tissue DNA, which seems to be an issue in most of the DNA methylation studies. Interestingly, given that a hypomethylation of HCG9 was also found in the sperm of BD patients, it is possible to speculate that this methylation pattern is transmitted to the offspring and may account for the familial risk of BD shown by this population (Kaminsky et al., 2012). "
    [Show abstract] [Hide abstract] ABSTRACT: Bipolar disorder (BD) is a multifactorial illness thought to result from an interaction between genetic susceptibility and environmental stimuli. Epigenetic mechanisms, including DNA methylation, can modulate gene expression in response to the environment, and therefore might account for part of the heritability reported for BD. This paper aims to review evidence of the potential role of DNA methylation in the pathophysiology and treatment of BD. In , several studies suggest that alterations in DNA methylation may play an important role in the dysregulation of gene expression in BD, and some actually suggest their potential use as biomarkers to improve diagnosis, prognosis, and assessment of response to treatment. This is also supported by reports of alterations in the levels of DNA methyltransferases in patients and in the mechanism of action of classical mood stabilizers. In this sense, targeting specific alterations in DNA methylation represents exciting new treatment possibilities for BD, and the ‘plastic’ characteristic of DNA methylation accounts for a promising possibility of restoring environment-induced modifications in patients.
    Full-text · Article · Jun 2016
    • "The question of whether methylation measured in the blood is informative for organ-or tissue-specific diseases is still unsettled. Nevertheless, several studies suggest that blood is a useful surrogate tissue [39][40][41][42][43]. For example, Talens et al. (2010) compared epigenetic profiles of candidate loci in blood and buccal cells and found that DNA methylation was the same for more than half of the sites in both tissues. "
    [Show abstract] [Hide abstract] ABSTRACT: Current risk assessment practices and toxicity information are hard to utilize for assessing the health impact of combined or cumulative exposure to multiple chemical and non-chemical stressors encountered in the “real world” environment. Non-chemical stressors such as heat, radiation, noise, humidity, bacterial and viral agents, and social factors, like stress related to violence and socioeconomic position generally cannot be currently incorporated into the risk assessment paradigm. The Science and Decisions report released by the National Research Council (NRC) in 2009 emphasized the need to characterize the effects of multiple stressors, both chemical and non-chemical exposures. One impediment to developing information relating such non-chemical stressors to health effects and incorporating them into cumulative assessment has been the lack of analytical tools to easily and quantitatively monitor the cumulative exposure to combined effects of stressors over the life course.
    Full-text · Article · Apr 2016
Show more

Recommended publications

Discover more