Molecular Evidence for Increased Expression of Genes
Related to Immune and Chaperone Function in
the Prefrontal Cortex in Schizophrenia
Dominique Arion, Travis Unger, David A. Lewis, Pat Levitt, and Károly Mirnics
Background: Schizophrenia is characterized by complex gene expression changes. The transcriptome alterations in the prefrontal cortex
have been the subject of several recent postmortem studies that yielded both convergent and divergent findings.
Methods: To increase measurement precision, we used a custom-designed DNA microarray platform with long oligonucleotides and
multiple probes with replicates. The platform was designed to assess the expression of ? 1800 genes specifically chosen because of their
14 matched pairs of schizophrenia and control subjects were analyzed with two technical replicates and four data mining approaches.
Results: In addition to replicating many expression changes in synaptic, oligodendrocyte, and signal transduction genes, we uncovered
and validated a robust immune/chaperone transcript upregulation in the schizophrenia samples.
in schizophrenia subjects represents a long-lasting and correlated signature of an early environmental insult during development that
actively contributes to the pathophysiology of prefrontal dysfunction.
postmortem brain, prefrontal cortex, qPCR, schizophrenia
roles in the etiology of schizophrenia, the molecular substrate for
the pathophysiology of the disease remains elusive and is
recognized as complex (Lewis and Levitt 2002). To address this
issue, a number of DNA microarray studies have been conducted
(for review see Iwamoto and Kato 2006; Mirnics et al. 2006).
These high throughput transcriptome profiling experiments have
identified several altered molecular processes in schizophrenia,
including downregulation of genes in the ?-aminobutyric acid
(GABA) and glutamate systems (Mirnics et al. 2000, 2001; Vawter
et al. 2002), gene expression changes related to synaptic and
mitochondrial functions (Altar et al. 2005; Iwamoto et al. 2005;
Middleton et al. 2005; Mirnics et al. 2000; Prabakaran et al. 2004),
and a repression of oligodendrocyte messenger RNAs (mRNAs)
in the prefrontal cortex (PFC) (Aston et al. 2004; Hakak et al.
2001; Pongrac et al. 2002; Tkachev et al. 2003).
There have been both convergent and divergent results from
the transcriptome profiling studies (Mirnics et al. 2006). Differ-
ences could be attributed to: 1) systematic cohort biases arising
from brain collection procedures (e.g., hospitalized subjects vs.
subjects who died suddenly, divergent inclusion/exclusion crite-
ria); 2) sample processing differences in RNA isolation proce-
chizophrenia is a severe mental disorder that affects ap-
proximately 1% of the population (Jablensky 2000). Al-
though both genetic and environmental factors play critical
dures, complementary DNA (cDNA) synthesis, in vitro transcrip-
tion, or fluorescent labeling; 3) microarray platform differences;
or 4) differences in segmentation, data extraction, normalization,
and statistical analyses (Hollingshead et al. 2005). Thus, the
combined experimental variation and noise in every microarray
experiment is considerable, and the microarray data can be less
precise and accurate than desired.
Given these limitations, the present study was aimed at
analyzing putative schizophrenia-associated molecular pathways
at increased measurement precision. We postulated that the ideal
microarray would: 1) be built with longer oligonucleotides to
achieve a higher level of probe specificity; 2) contain multiples
probes against the same gene; 3) have a high probe replicate
redundancy; and 4) incorporate microarray probes on the basis
of published schizophrenia datasets and hypotheses. In addition,
we established that the comprehensive microarray experiments
would analyze only samples with the highest RNA integrity using
two technical replicates. Finally, we used several analytical
approaches to maximize the identification of convergent data.
The present study reports on a cohort composed of 14 pairs of
control and schizophrenic brain samples matched for age, gen-
der, and postmortem interval (PMI). We identified significant
changes in expression for 67 genes and expressed sequence tags
(ESTs) between schizophrenia and control samples. In addition
to confirming some of the previously reported gene expression
changes, we found a highly correlated alteration in the expres-
sion of genes involved in immune/chaperone function as part of
the transcriptome disturbances in schizophrenia. These results
are discussed in the context of the environmental influences that
might predispose to the disease.
Methods and Materials
Human Brain Samples and Tissue Preparation
Fresh-frozen human tissue was obtained from the University of
Pittsburgh’s Center for the Neuroscience of Mental Disorders Brain
Bank. In this study, 14 pairs of schizophrenia and control postmor-
tem brains were matched for gender and as closely as possible for
From the Departments of Psychiatry (DA, DAL, TU, KM) and Neuroscience
(DAL), University of Pittsburgh, Pittsburgh, Pennsylvania; and the De-
Kennedy Center for Human Development (PL, KM), Vanderbilt Univer-
sity, Nashville, Tennessee.
Address reprint requests to Károly Mirnics, M.D., Department of Psychiatry,
Vanderbilt University, 8130 MRB 3, 465 21stAve South, Nashville, TN
37232; E-mail: firstname.lastname@example.org.
Received November 1, 2006; revised December 20, 2006; accepted Decem-
ber 24, 2006.
BIOL PSYCHIATRY 2007;62:711–721
© 2007 Society of Biological Psychiatry
Table 1. Subjects Studied
Race AgePMICause of DeathpH
RaceAgePMI DSM IV DiagnosisCause of Death pH
1 592cM/B 4122.1 ASCVD 6.729 533sM/W40 29.1Chronic undifferentiated
Accidental asphyxiation6.82 8.4
2 567cF/W 46 15Mitral valve
Gun shot wound to
6.728.9 537s F/W 3714.5
Suicide by hanging6.688.6
3 516cM/B 2014 6.868.4547sM/B 2716.5 Heat stroke6.957.4
4 630cM/W 65 21.2 6.959 566sM/W63 18.3 Chronic undifferentiated
5 604cM/W39 19.3 Hypoplastic
7.08 8.6 581sM/W4628.1 Accidental combined
6 546c F/W37 23.56.74 8.6587sF/B38 17.8 7.029
7 1047cM/W43 12.4 ASCVD6.639 722sM/B 459.1 Gastrointest hemorrhage6.719.2
8 551cM/W61 16.4Cardiac tamponade 6.638.3625sM/B4923.5ASCVD 7.327.6
9685c M/W56 14.5 Hypoplastic
6.578.1622s M/W58 18.9 Right MCA infarction6.78 7.4
10 681c M/W5111.6 7.15 8.9 640sM/W 495.2Pulmonary embolism 6.868.4
11806c M/W 5724 6.947.8 665sM/B5928.1 Intestinal hemorrhage6.92 9.2
Suicide by gun shot
Suicide by drug overdose
14871cM/W 2816.5Trauma 7.14 8.5878s M/W 3310.8Myocardial fibrosis6.72 8.9
Demographic data of these 14 matched pairs of schizophrenic and control subjects. PMI, postmortem interval (in hours); ASCVD, atherosclerotic coronary vascular disease; RNA RIN, RNA integrity
number measured by BioAnalyzer 2100 (Agilent Technologies); StDev, standard deviation of means; ?, schizophrenic subjects not receiving medications at time of death.
aAlcohol dependence, current at time of death.
bAlcohol abuse, current at time of death.
cAlcohol abuse, in remission at time of death.
dOther substance dependence, current at time of death.
eOther substance abuse, current at time of death.
fOther substance abuse, in remission at time of death.
712 BIOL PSYCHIATRY 2007;62:711–721
D. Arion et al.
control (CTR) and schizophrenia (SCZ) subjects (Table 1) did not
differ in mean (? SD) age at the time of death (42.21 ? 14.8 years
and 42.57 ? 12.31 years, respectively), PMI (17.34 ? 5.39 hours
and 17.43 ? 8 hours, respectively), tissue storage time (67.4 ?
24.8 months and 73.7 ? 18.2 months, respectively), brain pH
(6.88 ? .21 and 6.88 ? .20, respectively), and RNA integrity
number (RIN) (8.6 ? .40 and 8.4 ? .67, respectively). Consensus
DSM-IV diagnoses for all subjects were made with data from clinical
records, toxicology studies, and structured interviews with surviving
relatives, as described previously (Volk et al. 2000). The details of
the demographic and clinical features of this subject cohort have
been previously described in (Hashimoto et al. 2003, 2007).
Nimblegen Array Design and Processing
Our custom-designed Nimblegen DNA microarrays were
composed of 60-mer single-stranded oligonucleotides synthe-
sized by maskless in situ photolithographic synthesis. Each
microarray contained probes against approximately 1800 genes.
The expression of each gene was assessed by five independent
probe sequences and each of the probes was printed at four
technical replicates on a sub-array (Figure 1). Each microarray
consisted of 5 sub-arrays, resulting in a total of 100 measure-
ments/gene/array (5 probes ? 4 replicate spots ? 5 identical
sub-arrays). In addition, the experiment was performed in two
technical replicates with independent cDNA and cRNA synthesis
steps. As a result, our dataset consisted of 56 microarrays.
The approximately 1800 gene expression probes were chosen
on the basis of: 1) previous microarray studies conducted in our
laboratory, 2) previously published gene expression and genetic
data in schizophrenia, and 3) genes participating in cellular
processes implicated in the pathophysiology of PFC dysfunction
in schizophrenia. As a result, the list of microarray probes was
enriched in GABA, glutamate, synaptic, glial, dopamine, seroto-
nin system, and other transcripts. Furthermore, on the basis of
previous microarray studies of subjects with schizophrenia and
matched control subjects, an additional approximately 400 genes
with unaltered expression were chosen to ensure proper nor-
malization across the microarrays.
Coronal blocks containing area 9 of the PFC were cut on a
cryostat at 20 ?m thickness. The sections were collected into
tubes containing Trizol reagent (Invitrogen, Carlsbad, California),
and RNA was isolated according to the manufacturer’s instruc-
tions. The RNA quality was assessed with an Agilent BioAnalyzer
2100 system (Agilent, Palo Alto, California). After cDNA synthesis
(cDNA SuperScript Custom Kit by Invitrogen Corporation, Carls-
bad, California) and in vitro transcription (MEGAscript T7 Kit,
Ambion, Austin, Texas), the resulting cRNA was labeled accord-
ing to Nimblegen recommendations. Hybridization was per-
formed at the Nimblegen Systems facility with established pro-
tocols (for further information, see http://www.nimblegen.com/
Figure 1. DNA microarray design. For each of the ? 1800 genes, five inde-
pendent, non-overlaping 60-mer DNA probes were designed. Each array
blocks (sub-arrays). Each block contained four probe replicates for each of
the five probes. Fluorescent intensity values were first averaged within the
five intensity values, one for each probe (P1–P5). Because the experiment
of the investigated genes (5 for R1 ? 5 for R2). These Robust Multi-array
Analysis (RMA) normalized intensities entered the data mining process de-
scribed in Figure 2.
D. Arion et al.
BIOL PSYCHIATRY 2007;62:711–721 713
The microarray images were segmented and analyzed with a
customized system named NimbleScan (http://www.nimblegen.
com/products/software/nimblescan.html). Data were then nor-
malized and analyzed with Robust Multi-array Analysis (RMA)
(Irizarry et al. 2003). For analysis purposes and for each gene, the
five probes were treated as independent measurements, and for
each probe the result of the 20 measurements/array were
averaged into a single value. To define differentially expressed
genes between SCZ and CTR brain samples and across the two
technical replicates (R1 and R2), data were analyzed with four
different strategies (Figure 2).
A: Groupwise Strategy With R1 and R2 Analyzed Indepen-
dently. In this strategy, the normalized expression levels for all
14 SCZ samples and all 14 CTR samples were first averaged and
then compared. This analysis was performed for both R1 and R2
independently. A gene was then considered differentially ex-
pressed if: 1) it showed an absolute average log2 ratio (ALR)
between SCZ and CTR samples of more than 20% (|ALR| ? .263;
1.2-fold); 2) at least one gene probe showed a groupwise Student
t test significance of p ? .05 between SCZ and CTR samples; and
3) the differential expression was observed in both R1 and R2.
B: Groupwise Strategy With R1 and R2 Averaged. In B, the
normalized expression levels for R1 and R2 were averaged for
each of the SCZ and CTR samples. Then, the resulting expres-
sion levels for all 14 SCZ and 14 CTR samples were averaged
and compared. A gene was then considered differentially
expressed if: 1) at least one gene probe showed an absolute
average log2 expression difference between SCZ and CTR of
more than 20% (|ALR| ? .263; 1.2-fold); and 2) it showed a
groupwise Student t test significance of p ? .05 between SCZ
and CTR samples.
C: Pairwise Strategy With R1 and R2 Analyzed Indepen-
dently. In C, the normalized expression levels were compared
in a pairwise fashion for each of the 14 SCZ and CTR samples.
Then, the results for all 14 pairs were averaged. The same
comparison was performed for each R1 and R2 independently. A
gene was then considered differentially expressed if: 1) at least
one gene probe showed an absolute average log2 expression
difference of more than 20% (|ALR| ? .263; 1.2-fold) between
SCZ and CTR samples; 2) it showed a pairwise Student t test
significance of p ? .05 between SCZ and CTR samples; and 3) the
differential expression was observed in both R1 and R2.
D: Pairwise Strategy With R1 and R2 Averaged. In D, the
normalized expression levels for all CTR and SCZ pairs in R1
and R2 were averaged. Then, each of the matched SCZ and
CTR brain samples were compared in a pair wise fashion to
determine gene expression differences. A gene was then
considered differentially expressed if: 1) at least one probe
showed an absolute average log2 expression difference of
more than 20% (|ALR| ? .263; 1.2-fold); and 2) it showed a
pairwise Student t test significance of p ? .05 between SCZ
and CTR samples.
A gene was considered differentially expressed if it reported
significant and unidirectional differential expression in all four
Two-way cluster analyses (genes and samples) were per-
formed on normalized log2 transformed signal levels across the
28 subjects with Euclidian distance analysis in Genes@Work
developed by IBM (Armonk, New York) (Lepre et al. 2004).
Figure 2. Data analysis design. A total of four data analysis
or averaged. (A) In the first approach (groupwise analysis
with independent replicates), the data from the first repli-
control subjects (C1–C14). A similar comparison was per-
differentially expressed gene, a corresponding microarray
probe had to show |ALR| ? .263 and p ? .05 in both repli-
cates. (B) In the second approach (groupwise analysis with
averaged across the two technically replicated originating
from a same subjects, and the averaged values were com-
pared in the groupwise fashion between the control sub-
(C) In the third approach (pairwise analysis with indepen-
dent replicates), the 14 matched pairs were compared in a
pairwise fashion, generating two ALRs for each of the pairs
ilar expression changes across the R1 and R2 analysis were
considered differentially expressed. (D) In the fourth ap-
R2 for each of the five probes. Then, these average values
714 BIOL PSYCHIATRY 2007;62:711–721
D. Arion et al.
Testing for Confounds
The two potential confounding factors in our study are
antipsychotic medication and ethanol abuse. For post hoc testing
for antipsychotic medication effects, we divided the subjects into
two groups: 1) pairs of subjects where the schizophrenic subjects
were not taking medication at the time of death (Pairs 2, 9, and
13), and 2) pairs of subjects where the schizophrenic subjects
were taking antipsychotic medication at the time of death. For
post hoc testing of effects of ethanol dependence on the
observed gene expression differences, we also separated the
subject pairs into two groups (EtOH group: pairs 5, 9, 11, 13, and
14), this time by history of ethanol dependence, and compared
the expression ALRs of the two cohorts. For testing of both
confounds, we assessed the gene expression of the 20 critically
changed genes across the two newly created pair groups. Data
were not available about smoking history of the subjects.
Pearson’s correlation testing was performed on the normal-
ized log2 values for eight selected genes (SERPINA3, IFITM1,
IFITM3, CHI3L1, CD14, HSPA1B, HSPB1, and HSPA1A). The
correlation was calculated on the averaged R1R2 data and across
all 28 human subjects. We obtained five correlation values for
each gene (one for each probe). The highest across-gene and
within-gene correlation is reported.
False Discovery Assessment
The false discovery rate (FDR) assessment, a built-in feature of
this experimental design, was assessed by establishing how often
the independent probes from the same gene reported opposite
expression changes. In all of our analyses, two or more different
probes against the same gene never reported a statistically
significant expression change across the SCZ-CTR comparison in
opposite directions (e.g., increase vs. decrease). This suggests a
very high measurement precision and a negligible FDR for the
differentially expressed genes. However, we acknowledge that
our converging data mining strategy might carry an increased
amount of type II error (false negative observations), which are
not possible to address in current microarray experiments.
Real-Time Quantitative Polymerase Chain Reaction
For selected genes showing differential expression between
CTR and SCZ subjects, cDNA synthesis was performed with two
independent reverse transcriptions for each sample with the
High Capacity cDNA Archive Kit from Applied Biosystems
(Foster City, California). For each reaction, we used 50 ng of total
RNA from each subject. Priming was performed with random
hexamers according to the manufacturer’s recommendations. For
each sample, amplified product differences were measured with
four replicates with SYBR Green chemistry-based detection
(Mimmack et al. 2004). Mitochondrial adenosine triphosphate
(ATP) synthase 6 (MTATP6) and glyceraldehyde-3-phosphate
dehydrogenase (GAPDH) were used as the endogenous refer-
ence genes, because they did not display significant variation in
gene expression between CTR and SCZ samples. The efficiency
for each primer set was assessed before quantitative polymerase
chain reaction (qPCR) measurements, and a primer set was
considered valid if its efficiency was between 92% and 100%. The
qPCR reactions were carried out in an ABI Prism 7000 thermal
cycler (Applied Biosystems) with the ABI Prism 7000 SDS
software with the automatic baseline and threshold detection
options selected. These quantified data were exported to Mi-
crosoft Excel for establishing SCZ and CTR ?Ct, determining
SCZ?Ct? CTR?Ct(??Ct) and significance testing. A one-tailed
Student t test was used to determine the significance of the
qPCR-reported SCZ-CTR expression differences.
All microarray data will be made publicly available at the
authors’ Web site at the time of publication.
In this study, a Nimblegen custom microarray analysis al-
lowed us to compare the expression levels of ? 1800 different
genes, including ? 400 control genes between SCZ and CTR
samples from area 9 of human postmortem PFC. The combina-
tion of four independent analytical strategies reduced the dataset
to 67 transcripts reporting significant differential expression
between PFC-harvested samples from SCZ and CTR (|ALR| ?
.263; 1.2-fold and p ? .05). Of these 67 genes, 22 genes (33%)
were found upregulated in SCZ samples compared with CTR
samples with a mean ALR value of .68 (1.6-fold; Table 2A) and
45 (67%) were found downregulated with a mean ALR value of
.58 (1.5-fold; Table 2B). These results are in agreement with
previous studies reporting a global downregulation of the ex-
pression of a variety of functional gene groups in schizophrenia
(for review see Iwamoto and Kato 2006). The 67 differentially
expressed genes between SCZ and CTR samples represented a
heterogenic group of transcripts, some of which have been
previously associated with the pathophysiological processes in
schizophrenia. Downregulation of transferrin (TF), synapsin 2
(SYN2), synaptojanin 1 (SYNJ1), regulator of G-protein signaling
4 (RGS4), mitogen-activated protein kinase 1 (MAPK1), glutamic-
oxaloacetic transaminase 1 (GOT1), potassium channel, subfam-
ily K, member 1 (KCNK1), and mu-crystallin (CRYM) have all
been previously reported to be downregulated in subjects with
schizophrenia (Aston et al. 2004; Hakak et al. 2001; Iwamoto et
al. 2005; Middleton et al. 2002; Mirnics et al. 2000; Pongrac et al.
2002; Vawter et al. 2002). Finally, this experimental series also
confirmed and extended previous findings of altered GABAergic
system transcriptome in schizophrenia (Lewis et al. 2005). Owing
to the volume and complexity of these findings, the GABA
system-related expression changes are the subject of a separate
manuscript (Hashimoto et al. 2007).
The analyses revealed an unexpected upregulation of genes
related to immune function and chaperone system. Of the 19
fully annotated transcripts that reported increased expression in
subjects with schizophrenia, 10 (? 50%) have been linked to
functions in immune/chaperone systems (Table 2A). For exam-
ple, in subjects with schizophrenia, ? 50% expression upregulation
(HSPB1), 70kDa heat shock protein 1B (HSPA1B), 70kDa heat
shock protein 1A (HSPA1A), metallothionein 2A (MT2A), interferon
induced transmembrane protein 1 (IFITM1), interferon induced
transmembrane protein 3 (IFITM3), CD14 antigen (CD14), chitinase
3-like1 (CHI3L1), serine-cysteine protease inhibitor A3 (SERPINA3),
and paired immunoglobulin-like receptor beta (PILRB). These tran-
scripts and their protein products are known to exhibit increased
expression in response to cellular stress and/or immune stimulation
(Chung et al. 2003b, 2004; Gosslau and Rensing 2000; Kohler et al.
2003; Lewin et al. 1991; Mosser et al. 2000; Mousseau et al. 2000;
Penkowa and Hidalgo 2001; Potter et al. 2001; Seidberg et al. 2003;
Shiratori et al. 2004; Smith et al. 2006; Takekawa and Saito 1998; van
Bilsen et al. 2004; Xia et al. 2006).
A two-way unsupervised clustering (genes and subjects) of
the annotated gene probe signal intensities resulted in the
D. Arion et al.
BIOL PSYCHIATRY 2007;62:711–721 715
separation of the samples in two different clusters (Figure 3). Of
the 14 SCZ subjects, 9 clustered together, and these subjects
clearly were major contributors to the statistical significance
across the dataset. In contrast, the remaining 5 SCZ subjects
seemed to be indistinguishable from the control subjects, sug-
gesting a molecular sub-stratification within the SCZ samples.
The clustering also yielded an interesting separation along the
dimension of expressed genes; here, the immune/chaperone-
related transcripts (MT2A, IFITM1, IFITM3, CHI3L1, SERPINA3,
CD14, TNC, GADD45G, HSPB1, HSPA1B, and HSPA1A) strongly
clustered together as a distinct sub-group, suggesting a potential
co-regulation of these molecules.
The pre-established criterion of the differentially expressed
transcripts needing to fulfill a combination of four distinct data
analysis strategies resulted in the reporting of only the most
reliable expression changes (Table 2B). Owing to the conser-
vative nature of this data mining approach, however, a
number of true differential expressions might have been
eliminated from our dataset. In Supplement 1 we report all
other non-GABA system-related transcripts that were differen-
tially expressed by at least one of the four analysis strategies.
Of these additional 170 putatively changed transcripts, ap-
proximately 15% were found changed in previous schizophre-
nia studies, suggesting that these data might also contain
findings for further exploration.
Testing for Effects of Medication and Ethanol Abuse
The reported expression findings did not seem to be a result
of administration of chronic antipsychotic medication: subjects
with schizophrenia that were not taking medication at time of
death showed similar expression changes to subjects that were
taking medication at the time of death. For the subset of the 21
most changed genes, subjects with schizophrenia receiving
medication showed an average ALR ? ?.53 for decreased and
.60 for increased expression (Supplement 2A). Similarly, subjects
not receiving medication at the time of death showed for the
same genes an average ALR of ?.80 and 1.11, respectively, with
an overall correlation of r ? .94 (p ? .01). Furthermore, in an
unsupervised clustering of the gene expression changes in
Figure 3, two of the three subjects with schizophrenia that were
not taking medication (829s,622s,537s) clustered together with
the subjects that were taking medication at the time of death.
Similarly, sample pairs where the schizophrenic subject had
current alcohol abuse/dependence at the time of death showed
the similar gene expression changes as the rest of the pairs where
the schizophrenic subject had no active EtOH abuse/depen-
dence. The expression differences for the critical genes were
highly correlated (r ? .90, p ? .01) between these two groups,
with a trend of more robust expression changes in pairs without
EtOH abuse/dependence (average ALRs .59 vs. .77 and ?.26 vs.
?.77; Supplement 2B). Thus, on the basis of these data, it seems
that antipsychotic medication and EtOH consumption were not
confounding factors in our dataset.
To validate the microarray findings, we selected 12 genes for
real-time qPCR analysis. Six of these transcripts were overex-
pressed (SERPINA3, CHI3L1, HSPB1, MT2A, IFITM1, IFITM3),
whereas six other genes were underexpressed (DIRAS2, CRYM,
TF, MOG, MAPK1, RGS4) in the SCZ samples. The agreement
between the ??Ct of qPCR and DNA microarray ALR datasets
was striking: the differential expression for the 12 investigated
genes was correlated at r ? .93 (p ? .001) between the two
methods (Figure 4).
In addition, because there were no probes against IFITM2 on
the microarray, in the context of IFITM1 and IFITM3 transcript
increases, we quantified the expression of the third family
member, IFITM2. By qPCR measurement the IFITM2 overexpres-
sion in the SCZ samples was comparable to that seen in the other
two IFITM family members (???Ct ? .54 or 1.45 fold increase,
p ? .05).
The successful verification of the immune/chaperone system
transcript findings raised the question of the origin of these
robust transcriptome changes. To determine whether the ob-
Table 2A. Genes With Increased Expression in the PFC of Subjects With Schizophrenia
UniGene Gene NameChr SymbolmALRp
Serine (or cysteine) proteinase inhibitor A3 (?-1 antitrypsin)
Chitinase 3-like 1 (cartilage glycoprotein-39)
Heat shock 70kDa protein 1B
28 kDa heat shock protein
Interferon induced transmembrane protein 3 (1-8U protein)
Interferon induced transmembrane protein 1 (9-27 protein)
Growth arrest and DNA-damage-inducible, ?
Heat shock 70kD protein 1A
Leucine-zipper-like transcriptional regulator 1
Paired immunoglobulin-like receptor ?
ATP synthase, H? transporting, mitochondrial F1 complex, ? 1
Tenascin C (hexabrachion)
Orphan purinergic receptor P2X-like 1
PHD finger protein 1 variant 2
Transmembrane protein 63C
ATP, adenosine triphosphate; PHD, plant homeodomain. The table has been culled of expressed sequence tags (ESTs), unannotated DNA fragments, and
?-aminobutyric acid (GABA) system transcripts.
716 BIOL PSYCHIATRY 2007;62:711–721
D. Arion et al.
served transcriptome changes were the result of an active
immune process in the brains of subjects with schizophrenia, we
investigated the levels of 2-prime,5-prime oligoadenylate syn-
thetase 1 (OAS1), a critical marker of acute viral infection and
interferon ? (INF?), a critical first activator of multiple down-
stream immune cascades, including the OAS system. Subjects
with schizophrenia, when compared with matched control sub-
jects, did not show altered transcript levels of OAS1 or INF? (data
Immune/Chaperone Transcript Co-Regulation
The clustered data strongly suggested that the upregulated
immune- and HSP-related transcripts might represent a robust,
selectively regulated gene expression network. To test this, we
performed an expression level cross-correlation of five immune
genes with three HSP family members. This correlational analysis
was performed across the whole dataset (28 human PFC samples)
for the five probes/gene (Figure 5). The data revealed that the
expression levels for the five immune markers tested (SERPINA3,
IFITM1, IFITM3, CHI3L1, and CD14) were highly correlated
within the human PFC (max r ? .73–.93, all p ? .01). Similarly,
the three HSP family members (HSPA1B, HSPB1, and HSPA1A)
showed a high within-group correlation (max r ? .85–.99, all p ?
.01). In contrast, only a weak-to-moderate correlation was ob-
served between the members of the chaperone and immune
genes, suggesting that the modulation of the changed immune-
related genes and altered chaperone-related transcripts might
occur independently and not as a direct adaptive relationship
between these two systems.
Detailed analysis of the custom DNA microarray data from a
cohort of 14 matched pairs of CTR and SCZ brain samples from
area 9 of human PFC revealed a number of important observa-
tions related to changes in gene expression in schizophrenia.
First, with rigorous and convergent statistical approaches, we
identified 67 genes that were differentially regulated across CTR
and SCZ samples. Second, the number of downregulated tran-
scripts outnumbered the upregulated ones by an approximately
2:1 margin, confirming previous reports that gene expression
changes in the PFC of subjects with schizophrenia are predom-
inantly characterized by transcript reductions rather than in-
creases (Iwamoto and Kato 2006; Mirnics et al. 2000). Third, we
confirmed many of gene expression changes reported in previ-
ous postmortem studies, including downregulation of TF (Hakak
et al. 2001);, RGS4 (Mirnics et al. 2001); SYN2, SYNJ1, MAPK1,
GOT1, KCNK1, and CRYM (Mirnics et al. 2000); and upregulation
of CHI3L1 (Chung et al. 2003a), HSPB1 (Kuromitsu et al. 2001),
HSPA1B (Hakak et al. 2001), and MT2A (Aston et al. 2004).
Table 2B. Genes With Decreased Expression in the PFC of Subjects With Schizophrenia
UniGeneGene NameChrSymbol mALRp
Regulator of G-protein signaling 4
Ecotropic viral integration site 2A (EVI2A)
Sodium channel, voltage-gated, type III, ?
Mitogen-activated protein kinase 1, transcript variant 2
GTP-binding RAS-like 2
Secretory carrier membrane protein 1, variant 2
Ephrin-A3 precursor (EHK-1 tyrosine kinase ligand EFL-2)
Sparc/osteonectin (testican) 3
Cyclin-dependent kinase 5, regulatory subunit 1 (p35)
Potassium channel, subfamily K, member 1
DnaJ (Hsp40) homolog, subfamily C, member 14
Ephrin type-A receptor 4 precursor
CAP, adenylate cyclase-associated protein, 2
Zinc finger, CCCH-type with G patch domain
Heat shock 70kDa protein 8, transcript variant 1
ADP-ribosylation factor-like 6 interacting protein
Glutamic-oxaloacetic transaminase 1, soluble
Platelet-activating factor acetylhydrolase Ib, ? 45kDa
Signal recognition particle 72kDa
Kinesin-associated protein 3
LIM domain only 2 (rhombotin-like 1)
NADH dehydrogenase Fe-S protein 4 (NADH Q reductase)
Small glutamine rich protein with tetratricopeptide repeats 2
ADP-ribosylation factor domain protein 1, 64kDa
Synapsin II (SYN2), transcript variant Iia
Non-metastatic cells 1, protein (NM23A)
Discs, large homolog 2, chapsyn-110
GTP, guanosine triphosphate; ADP, adenosine diphosphate; NADH, adenosine 5’-(trihydrogen diphosphate); other abbreviations as in Figure 2A.
D. Arion et al.
BIOL PSYCHIATRY 2007;62:711–721 717
Figure 3. Hierarchical clustering of gene expression
changes. Normalized log2 intensities were clustered
vertical: samples) on the basis of Euclidian distance.
value in a single subject. The color intensity is propor-
tional to its relative expression level (green: underex-
pressed; red: overexpressed). Note that a subset of
tern that is distinct from the rest of the schizophrenic
and control subjects. Expressed sequence tags (ESTs)
and unannotated probesets were removed from the
dataset, whereas ?-aminobutyric acid (GABA)-ergic
transcripts are denoted with ***.
718 BIOL PSYCHIATRY 2007;62:711–721
D. Arion et al.
Fourth, we identified an unexpected and strongly correlated
upregulation of a subset of genes involved in immune/chaper-
one function. Fifth, we found that the immune/chaperone sig-
nature was primarily present in a subset of subjects with schizo-
phrenia. Although we recognize that chaperones can function in
a wide range of roles that are independent of immune processes
(e.g., response to cellular stress and signal transduction) (Garcia-
Osta et al. 2003; Kim et al. 2001; Pae et al. 2005), in our dataset
we favor the interpretation that the observed chaperone and
immune changes are of common origin and causally interrelated,
and they will be discussed in this context.
The neuroimmune hypothesis of schizophrenia has been
debated for decades (Giovannoni and Baker 2003; Hanson and
Gottesman 2005; Jones et al. 2005; Muller et al. 2000; Muller and
Schwarz 2006; Patterson 2002; Rothermundt et al. 2001; Strous
and Shoenfeld 2006). Relative risk for developing schizophrenia
is increased more than twofold compared with the general
population with 2nd trimester infection (reviewed by Cannon
and Clarke 2005). Additionally, epidemiological, serological,
gene expression, and pathological findings all suggest an infec-
tive-immune component of the disease, albeit replication across
different cohorts of patients has been elusive (Rothermundt et al.
2001). Nevertheless, the combined evidence suggests an infec-
tive-immune predisposition to schizophrenia and that this pre-
disposition is likely to interact with genetic susceptibility for
developing the disease. In this context, the changes related to
immune/chaperone functions can represent either a response to
an ongoing infective-immune challenge or a long-lasting signa-
ture of an immune system challenge that might have acted during
brain development, which in the human extends in a lengthy
fashion from 1st trimester through puberty. Most of the studies of
schizophrenia to date suggest that the observed neuroimmune
changes are a long-lasting consequence of a previous infective-
immune challenge (for review see Nawa and Takei 2006; Sper-
ner-Unterweger 2005). Here, we addressed this more directly by
determining that there is no change in the transcript levels of
OAS1 and INF?, two critical markers of acute immune response
(Rothwell and Hopkins 1995; Rothwell et al. 1996). Thus, we also
favor the interpretation of a developmentally based, long-term
alteration in the transcriptome of genes related to immune/
chaperone function, although one must be aware that certain
pre-mortem life stressors and adverse socioeconomic conditions,
which are highly prevalent in patients with schizophrenia, might
also contribute to some of the observed expression changes.
Long-term consequences of early immune challenge are not
unprecedented. Studies in rodents that undergo prenatal or
perinatal exposure to immune challenges (such as maternal
exposure to polyriboinosinic-polyribocytidylic acid [poly(I:C)], a
synthetic cytokine inducer), high levels of pro-inflammatory
cytokines, or viral infections develop post-adolescent behavioral
deficits that are similar in nature to clinical manifestations in
schizophrenia (Ashdown et al. 2006; Meyer et al. 2005; Tohmi et
al. 2004; Zuckerman and Weiner 2005). In the mouse, prenatal
exposure in mid-pregnancy to poly(I:C) reduces the number of
reelin positive cells in hippocampus (Meyer et al. 2006). Further-
more, poly(I:C) administration also causes increased dopamine
turnover, prepulse inhibition deficits, and cognitive impairments
in the adult offspring, and the latter is improved by administra-
tion of clozapine (Ozawa et al. 2006). Finally, in a rat model,
poly(I:C) administration during pregnancy also produced long-
lasting pathophysiological changes that are also observed in
schizophrenia, including dopaminergic hyperfunction and loss
of latent inhibition (Zuckerman et al. 2003).
In view of these data, we propose that the transcriptome
signature of altered genes related to immune/chaperone function
might be a consequence of early life tumor necrosis factor
(TNF)-?, interleukin (IL)-1, IL-6, and/or INF? brain activation. In
-1.00 -0.75 -0.50-0.25 0.00 0.250.50 0.751.00
Microarray ALR difference (S-C)
qPCR ddCt (C-S)
Figure 4. Quantitative polymerase chain reaction (qPCR) and microarray
data are highly correlated. We performed qPCR validation for 12 genes
Validation was performed on 13 of the 14 pairs of matched subjects. Differ-
axis, whereas Y axis denotes ??Ct in the qPCR experiment. Each symbol
corresponds to a single gene. Note that the microarray data are highly
correlated with the qPCR data (r ? .93, p ? .001).
Figure 5. Immune and chaperone transcripts are highly correlated. Microarray data from five immune and three chaperone genes were examined for
correlation across all schizophrenic and control samples. Numbers in black boxes report maximal self-correlation between the five different probes for the
the correlation between the immune and chaperone molecules was only moderate-to-weak.
D. Arion et al.
BIOL PSYCHIATRY 2007;62:711–721 719
this proposed mechanism, the elevated pro-inflammatory cyto-
kine levels during late embryonic development or perinatal
period could not only impair normal differentiation and/or
refinement of neural connectivity but also leave behind a specific
immune/chaperone signature primarily consisting of altered
IFITM, SERPINA3, and HSP transcript increases.
How does elevation of these immune/chaperone system
molecules contribute to the symptoms of schizophrenia? The
present study does not address this directly, but we speculate
that this immune/chaperone signature extends beyond a corre-
lation with an early environmental insult and might actively
contribute to the clinical features of the illness. Many immune/
chaperone genes are known to be essential for the normal
functioning of the central nervous system, and immune function
genes are capable of altering cognitive performance (Blalock et
al. 2003; Heyser et al. 1997; Hoffman et al. 1998; Wilson et al.
2002; Ziv et al. 2006). The causality between the immune/
chaperone gene expression changes and altered cognitive per-
formance in schizophrenic patients needs to be addressed in
comprehensive clinical studies and in animal models.
In conclusion, our data suggest that a subgroup of subjects
with schizophrenia carries an immune/chaperone transcriptome
signature of an early environmental insult in the PFC. The
identification of this putative immune/chaperone molecular sub-
phenotype of schizophrenia, if it is validated across other cohorts
of subjects, will allow a separate, mechanistic follow-up in the
context of behavioral changes associated with the disease.
This work was supported by R01 MH067234 (KM), 2 P50
MH45156 CCNMD Project 2 (KM), and K02 MH070786 (KM).
We are thankful to Drs. Christine Konradi and Krassimira
Garbett for valuable comments on the manuscript. We also
thank Katherine C. Douglas, Annie Bedison, and Melissa Ma-
cioce for superb technical assistance with the experiments.
Supplementary material cited in this article is available
Altar CA, Jurata LW, Charles V, Lemire A, Liu P, Bukhman Y, et al. (2005):
Deficient hippocampal neuron expression of proteasome, ubiquitin,
and mitochondrial genes in multiple schizophrenia cohorts. Biol Psychi-
Ashdown H, Dumont Y, Ng M, Poole S, Boksa P, Luheshi GN (2006): The role
of cytokines in mediating effects of prenatal infection on the fetus:
Implications for schizophrenia. Mol Psychiatry 11:47–55.
Aston C, Jiang L, Sokolov BP (2004): Microarray analysis of postmortem
temporal cortex from patients with schizophrenia. J Neurosci Res 77:
PW (2003): Gene microarrays in hippocampal aging: Statistical profiling
identifies novel processes correlated with cognitive impairment. J Neu-
Cannon M, Clarke MC (2005): Risk for schizophrenia—broadening the con-
cepts, pushing back the boundaries. Schizophr Res 79:5–13.
Chung C, Tallerico T, Seeman P (2003a): Schizophrenia hippocampus has
elevated expression of chondrex glycoprotein gene. Synapse 50:29–34.
Chung RS, Adlard PA, Dittmann J, Vickers JC, Chuah MI, West AK (2004):
Neuron-glia communication: Metallothionein expression is specifically
up-regulated by astrocytes in response to neuronal injury. J Neurochem
Chung RS, Vickers JC, Chuah MI, West AK (2003b): Metallothionein-IIA pro-
motes initial neurite elongation and postinjury reactive neurite growth
and facilitates healing after focal cortical brain injury. J Neurosci 23:
Garcia-Osta A, Frechilla D, Del Rio J (2003): Reduced basal and phencyclid-
Hakak Y, Walker JR, Li C, Wong WH, Davis KL, Buxbaum JD, et al. (2001):
Hanson DR, Gottesman, II (2005): Theories of schizophrenia: A genetic-
Hashimoto T, Arion D, Unger T, et al. (2007): Analysis of GABA-related tran-
ular Psychiatry, Epub ahead of print 1 May 2007; doi:10.1038/
Hashimoto T, Volk DW, Eggan SM, Mirnics K, Pierri JN, Sun Z, et al. (2003):
cortex of subjects with schizophrenia. J Neurosci 23:6315–26.
Heyser CJ, Masliah E, Samimi A, Campbell IL, Gold LH (1997): Progressive
decline in avoidance learning paralleled by inflammatory neurodegen-
Hollingshead D, Lewis DA, Mirnics K (2005): Platform influence on DNA
microarray data in postmortem brain research. Neurobiol Dis 18:649–
Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U,
Speed TP (2003): Exploration, normalization, and summaries of high
density oligonucleotide array probe level data. Biostatistics 4:249–264.
Iwamoto K, Bundo M, Kato T (2005): Altered expression of mitochondria-
related genes in postmortem brains of patients with bipolar disorder or
schizophrenia, as revealed by large-scale DNA microarray analysis. Hum
Iwamoto K, Kato T (2006): Gene expression profiling in schizophrenia and
related mental disorders. Neuroscientist 12:349–361.
Jablensky A (2000): Epidemiology of schizophrenia: The global burden of
nia. Schizophr Res 52:127–35.
II in neuronal differentiation and survival. Brain Res 992:128–136.
Kuromitsu J, Yokoi A, Kawai T, Nagasu T, Aizawa T, Haga S, Ikeda K (2001):
Reduced neuropeptide Y mRNA levels in the frontal cortex of people
with schizophrenia and bipolar disorder. Brain Res Gene Expr Patterns
Lepre J, Rice JJ, Tu Y, Stolovitzky G (2004): Genes @Work: An efficient algo-
rithm for pattern discovery and multivariate feature selection in gene
expression data. Bioinformatics 20:1033–1044.
Lewin AR, Reid LE, McMahon M, Stark GR, Kerr IM (1991): Molecular analysis
of a human interferon-inducible gene family. Eur J Biochem 199:417–
Lewis DA, Hashimoto T, Volk DW (2005): Cortical inhibitory neurons and
Meyer U, Feldon J, Schedlowski M, Yee BK (2005): Towards an immuno-
precipitated neurodevelopmental animal model of schizophrenia. Neu-
Meyer U, Nyffeler M, Engler A, Urwyler A, Schedlowski M, Knuesel I, et al.
profiling reveals alterations of specific metabolic pathways in schizo-
phrenia. J Neurosci 22:2718–2729.
Middleton FA, Peng L, Lewis DA, Levitt P, Mirnics K (2005): Altered expres-
nia. Neuropsychopharmacology 30:974–983.
720 BIOL PSYCHIATRY 2007;62:711–721
D. Arion et al.
Mimmack ML, Brooking J, Bahn S (2004): Quantitative polymerase chain Download full-text
reaction: Validation of microarray results from postmortem brain stud-
ies. Biol Psychiatry 55:337–345.
Mirnics K, Levitt P, Lewis DA (2006): Critical appraisal of DNA microarrays in
psychiatric genomics. Biol Psychiatry 60:163–176.
Mirnics K, Middleton FA, Marquez A, Lewis DA, Levitt P (2000): Molecular
expression in prefrontal cortex. Neuron 28:53–67.
Mirnics K, Middleton FA, Stanwood GD, Lewis DA, Levitt P (2001): Disease-
in schizophrenia. Mol Psychiatry 6:293–301.
Mosser DD, Caron AW, Bourget L, Meriin AB, Sherman MY, Morimoto RI,
tion against stress-induced apoptosis. Mol Cell Biol 20:7146–7159.
a novel immunoreceptor tyrosine-based inhibitory motif-bearing pro-
tein, recruits SHP-1 upon tyrosine phosphorylation and is paired with
the truncated counterpart PILRbeta. J Biol Chem 275:4467–4474.
Muller N, Riedel M, Gruber R, Ackenheil M, Schwarz MJ (2000): The immune
system and schizophrenia. An integrative view. Ann N Y Acad Sci 917:
Muller N, Schwarz MJ (2006): Neuroimmune-endocrine crosstalk in schizo-
phrenia and mood disorders. Expert Rev Neurother 6:1017–1038.
Nawa H, Takei N (2006): Recent progress in animal modeling of immune
inflammatory processes in schizophrenia: Implication of specific cyto-
kines. Neurosci Res 56:2–13.
Ozawa K, Hashimoto K, Kishimoto T, Shimizu E, Ishikura H, Iyo M (2006):
Immune activation during pregnancy in mice leads to dopaminergic
opmental animal model of schizophrenia. Biol Psychiatry 59:546–554.
Pae CU, Kim TS, Kwon OJ, Artioli P, Serretti A, Lee CU, et al. (2005): Polymor-
phisms of heat shock protein 70 gene (HSPA1A, HSPA1B and HSPA1L)
and schizophrenia. Neurosci Res 53:8–13.
tions in fetal brain development and mental illness. Curr Opin Neurobiol
Penkowa M, Hidalgo J (2001): Metallothionein treatment reduces proin-
flammatory cytokines IL-6 and TNF-alpha and apoptotic cell death
during experimental autoimmune encephalomyelitis (EAE). Exp Neu-
Pongrac J, Middleton FA, Lewis DA, Levitt P, Mirnics K (2002): Gene expres-
sion profiling with DNA microarrays: Advancing our understanding of
psychiatric disorders. Neurochem Res 27:1049–1063.
ical chaperones ACT and apo-E are necessary catalysts of Alzheimer
amyloid formation. Neurobiol Aging 22:923–930.
Prabakaran S, Swatton JE, Ryan MM, Huffaker SJ, Huang JT, Griffin JL, et al.
(2004): Mitochondrial dysfunction in schizophrenia: evidence for com-
promised brain metabolism and oxidative stress. Mol Psychiatry 9:643,
Rothermundt M, Arolt V, Bayer TA (2001): Review of immunological and
immunopathological findings in schizophrenia. Brain Behav Immun 15:
and mechanisms of action. Trends Neurosci 18:130–136.
the central nervous system: Physiology, pharmacology, and pathology.
Seidberg NA, Clark RS, Zhang X, Lai Y, Chen M, Graham SH, et al. (2003):
Alterations in inducible 72-kDa heat shock protein and the chaperone
cofactor BAG-1 in human brain after head injury. J Neurochem 84:514–
Shiratori I, Ogasawara K, Saito T, Lanier LL, Arase H (2004): Activation of
natural killer cells and dendritic cells upon recognition of a novel CD99-
like ligand by paired immunoglobulin-like type 2 receptor. J Exp Med
Smith RA, Young J, Weis JJ, Weis JH (2006): Expression of the mouse fragilis
gene products in immune cells and association with receptor signaling
complexes. Genes Immun 7:113–121.
disorders: Evidence and therapeutic implications. Drugs 65:1493–1520.
Strous RD, Shoenfeld Y (2006): Schizophrenia, autoimmunity and immune
system dysregulation: A comprehensive model updated and revisited.
Takekawa M, Saito H (1998): A family of stress-inducible GADD45-like pro-
Tkachev D, Mimmack ML, Ryan MM, Wayland M, Freeman T, Jones PB, et al.
der. Lancet 362:798–805.
Tohmi M, Tsuda N, Watanabe Y, Kakita A, Nawa H (2004): Perinatal inflam-
matory cytokine challenge results in distinct neurobehavioral alter-
ations in rats: Implication in psychiatric disorders of developmental ori-
gin. Neurosci Res 50:67–75.
van Bilsen JH, van Dongen H, Lard LR, van der Voort EI, Elferink DG, Bakker
AM, et al. (2004): Functional regulatory immune responses against hu-
man cartilage glycoprotein-39 in health vs. proinflammatory responses
Vawter MP, Crook JM, Hyde TM, Kleinman JE, Weinberger DR, Becker KG,
cortex in schizophrenia: A preliminary study. Schizophr Res 58:11–20.
Volk DW, Austin MC, Pierri JN, Sampson AR, Lewis DA (2000): Decreased
prefrontal cortical gamma-aminobutyric acid neurons in subjects with
Xia Y, Yamagata K, Krukoff TL (2006): Differential expression of the CD14/
TLR4 complex and inflammatory signaling molecules following i.c.v.
administration of LPS. Brain Res 1095:85–95.
Ziv Y, Ron N, Butovsky O, Landa G, Sudai E, Greenberg N, et al. (2006):
Immune cells contribute to the maintenance of neurogenesis and spa-
tial learning abilities in adulthood. Nat Neurosci 9:268–725.
Zuckerman L, Rehavi M, Nachman R, Weiner I (2003): Immune activation
during pregnancy in rats leads to a postpubertal emergence of
disrupted latent inhibition, dopaminergic hyperfunction, and
altered limbic morphology in the offspring: A novel neurodevelop-
Zuckerman L, Weiner I (2005): Maternal immune activation leads to behav-
D. Arion et al.
BIOL PSYCHIATRY 2007;62:711–721 721