SHORT COMMUNICATION
Confirmation of region-specific patterns of gene expression
in the human brain
Carl Ernst & Adolfo Sequeira & Tim Klempan &
Neil Ernst & Jarlath ffrench-Mullen & Gustavo Turecki
Received: 7 November 2006 /Accepted: 16 February 2007 / Published online: 21 March 2007
# Springer-Verlag 2007
Abstract The human brain is divided and categorized in
different ways, yet a molecular genetic approach to region
specificity does not exist. Using data from 12 healthy
control subjects across 18 brain regions, we performed a
microarray analysis using both the HG-U133AB and HG-
U133 plus 2 chips for each subject to determine molecular
targets showing region specificity. Using a previously
published data as our guide, we confirm SIX3, GPR6,
SH3RF2, and hSyn as molecular markers of the nucleus
accumbens and gamma-aminobutyric-acid A receptor al-
pha-6, Nik-related kinase, and eomesodermin as molecular
markers of the cerebellum.
Keywords Brain .Microarray . Genetic marker
Introduction
To date, our understanding of the human brain is limited;
regions are defined by gross anatomy, external morphology,
or the location of specific nuclei and neuronal pathways [1,
2]. Brain region identification has progressed further with
the introduction of antibodies that have led to the labeling
of specific proteins in specific brain areas [3]. Still, given
the complexity of the brain, much work needs to be done to
understand how and if neural regions differ.
Gene expression arrays can provide some clues as to
how different regions of the brain differ at the molecular
level [4]. The difficulty arises in deciding what constitutes a
brain region and how well the technology can detect any
difference in global gene expression on a region-by-region
basis. Most researchers use either Brodmann’s definitions
of cortical areas or gross anatomical features such as
fissures or gyri to define regions.
Recently, Roth et al. [5] published a study comparing
gene expression levels in a series of central nervous system
(CNS) structures using microarray technology. The
strengths of that paper were the quality of the sample used
and the quality of the analysis performed. Their suggestions
suffered somewhat in light of the fact that, for a marker to
be region-specific or define a region, that marker had to be
absent in any other brain region. Furthermore, the human
cortex was considered like one region, while arguably, this
region could be divided up into many different regions,
suggesting that some of their potential markers may not be
specific to the region they highlighted.
The purpose of this study was to further evaluate and
confirm the claims of Roth et al. [5] in regions that overlap
with our own work using our fully clinically characterized
sample of normal controls. Using the same microarray
platform in a different sample, we provide independent
Neurogenetics (2007) 8:219–224
DOI 10.1007/s10048-007-0084-2
Electronic supplementary material The online version of this article
(doi:10.1007/s10048-007-0084-2) contains supplementary material,
which is available to authorized users.
C. Ernst : A. Sequeira : T. Klempan : G. Turecki
McGill Group for Suicide Studies, McGill University,
Montreal, QC, Canada
N. Ernst
Department of Computer Science, University of Toronto,
Toronto, ON, Canada
J. ffrench-Mullen
Gene Logic, Inc,
Gaithersburg, MD, USA
G. Turecki (*)
Douglas Hospital Research Centre, Pavilion Frank B Common,
Rm. F-3125, 6875 LaSalle Blvd.,
Verdun, Montreal, QC H4H 1R3, Canada
e-mail: gustavo.turecki@mcgill.ca

support for a number of region-specific markers suggested by
Roth et al. and provide additional evidence of the molecular
specificity of brain regions in humans.
Materials and methods
All subjects were of French–Canadian origin, a well-
characterized genetically homogenous population [6], and
died without a prolonged agonal state. All subjects were
victims of motor vehicle accidents or cardiac arrest. For all
12 subjects (mean age=36 years; SEM=3.4 years), psy-
chological autopsies were performed as described elsewhere
[7] and, together with medical chart information, were free
of any psychopathology or psychotropic medication.
The postmortem interval ranged from 20–36 h. Brains
were extracted, sectioned, and snap frozen and stored at −80°C.
Cortical brain regions extracted were based on the
Brodmann areas (BA): 4, 6, 8/9, 10, 11, 20, 21, 24, 29,
38, 44, 45, 46, and 47. Subcortical regions were also used,
including the nucleus accumbens, hippocampus, and the
amygdala. The final area examined was the cerebellum.
RNA from brain extractions was processed on the HG-
U133AB and HG-U133 plus 2 chips. Analysis was done
using MAS 5.0 and Genesis 2.0 (Gene Logic, Gaithersburg,
MD). Subjects were excluded based on glyceraldehyde 3-
phosphate dehydrogenase (GAPDH) and B-actin 3′:5 ratios.
Only those subjects with B-actin 3′:5′ ratios above 0.3 and
GAPDH ratios above 0.6 on every one of these markers
were used. Cluster analysis was performed using average-
linkage hierarchical cluster analysis with a correlation
metric (Avadis; Strand Genomics, Redwood City, California).
To ensure the reliability of our data, we processed every
sample on two separate microarray platforms. For every
probe set, we generated a Pearson correlation coefficient
between platforms across all subjects; only those probe sets
with r2 values above 0.3 were used to confirm previously
suggested markers.
To determine how specific these markers are across
species, we compared positive markers from this confirma-
tory study to that of the mouse brain. We used the Allen
brain atlas (http://www.brain-map.org) for marker compar-
ison [8]. The Allen brain atlas is an online system in which
global gene expression can be verified on a gene-by-gene
basis throughout the entire mouse brain. Using only the
confirmed genes, we entered the gene names and down-
loaded all images in as many planes as possible for every
marker. We had specific criteria for anatomical regions
when comparing our data to mouse data. The cerebellum is
easily discerned due to its well-defined anatomy and specific
structure. Both the nucleus accumbens core and shell were
analyzed for this study. Using Paxinos and Watson criteria
[9], the regions we inspected from the Allen brain atlas
correspond to: anteroposterior +1.3 to +1.6 mm; medio-
lateral ±1.0 to ±1.8 mm from bregma; dorsoventral −6.8 to
−6.3 mm from dura.
To assess expression level of neural markers in non-
neural areas, we again relied on the Roth et al. [5] data,
available at gene expression omnibus GSE3526. To
compare these neural markers to other organs, we selected
only those organs where there were four or more subjects,
from five diverse areas and normalized using robust
multiarray analysis [10].
Results
Using the gene list from the study of Roth et al. [5] as our
guide, we filtered microarray data for each of the genes in
all 18 brain regions. Their study focused on the midbrain
and lower CNS structures, whereas our study focused on
cortical structures; however, both studies examined the
cerebellum, the nucleus accumbens, and the hippocampus/
amygdala complex. Therefore, we were able to verify their
suggested markers in only these structures.
Using region specific markers proposed by Roth et al.
[5], we examined the expression value for 54 potential
markers. Across all subjects, we calculated a mean
expression value for a given region for both the HG-U133
chip set and the HG-U133 plus 2 chip. Figure 1 is a heat
map that demonstrates the results for this comparison.
Following this first-pass analysis, we performed follow-up
analysis on a gene-by-gene basis.
We had two criteria for whether a gene was considered a
region-specific marker: First, the gene had to be expressed
and called present (based on the p value generated through
the comparison of a perfect match probe to a mismatch
probe on the microarray, MAS 5.0) in only the area of
interest and in no other region, or second, if the gene was
present in more than the area of interest, that it was
expressed at least threefold higher in the region of interest.
To begin our analysis on a gene-by-gene basis, we used
a measure of quality control to ensure our own data was
reproducible. To do this, we calculated the Pearson
correlation coefficient between HG-U133AB and HG-
U133 plus 2 chips. For this reason, only probe sets present
on both chips were investigated. As shown in Table 1,
results from both chips were highly reproducible as
suggested by the high Pearson correlation coefficient
between the chips.
Table 1 represents the findings from this study. Column
1 indicates the probe set ID and Column 2 is the gene
symbol. Column 3 indicates whether or not this study
replicates the findings from Roth et al. [5]. Those rows
indicated by ‘Yes’ with a BA moniker indicate that the gene
met criteria in all regions except for the area noted. The
220 Neurogenetics (2007) 8:219–224