Association between SNPs and gene expression in multiple regions of the human brain

Abstract

Identifying the genetic cis associations between DNA variants (single-nucleotide polymorphisms (SNPs)) and gene expression in brain tissue may be a promising approach to find functionally relevant pathways that contribute to the etiology of psychiatric disorders. In this study, we examined the association between genetic variations and gene expression in prefrontal cortex, hippocampus, temporal cortex, thalamus and cerebellum in subjects with psychiatric disorders and in normal controls. We identified cis associations between 648 transcripts and 6725 SNPs in the various brain regions. Several SNPs showed brain regional-specific associations. The expression level of only one gene, PDE4DIP, was associated with a SNP, rs12124527, in all the brain regions tested here. From our data, we generated a list of brain cis expression quantitative trait loci (eQTL) genes that we compared with a list of schizophrenia candidate genes downloaded from the Schizophrenia Forum (SZgene) database (http://www.szgene.org/). Of the SZgene candidate genes, we found that the expression levels of four genes, HTR2A, PLXNA2, SRR and TCF4, were significantly associated with cis SNPs in at least one brain region tested. One gene, SRR, was also involved in a coexpression module that we found to be associated with disease status. In addition, a substantial number of cis eQTL genes were also involved in the module, suggesting eQTL analysis of brain tissue may identify more reliable susceptibility genes for schizophrenia than case-control genetic association analyses. In an attempt to facilitate the identification of genetic variations that may underlie the etiology of major psychiatric disorders, we have integrated the brain eQTL results into a public and online database, Stanley Neuropathology Consortium Integrative Database (SNCID; http://sncid.stanleyresearch.org).

Citation: Transl Psychiatry (2012) 2, e113, doi:10.1038/tp.2012.42

& 2012 Macmillan Publishers Limited All rights reserved 2158-3188/12 http://www.nature.com/tp

Web End =www.nature.com/tp

Association between SNPs and gene expression in multiple regions of the human brain

S Kim1, H Cho2, D Lee2 and MJ Webster1

Identifying the genetic cis associations between DNA variants (single-nucleotide polymorphisms (SNPs)) and gene expression in brain tissue may be a promising approach to nd functionally relevant pathways that contribute to the etiology of psychiatric disorders. In this study, we examined the association between genetic variations and gene expression in prefrontal cortex, hippocampus, temporal cortex, thalamus and cerebellum in subjects with psychiatric disorders and in normal controls. We identied cis associations between 648 transcripts and 6725 SNPs in the various brain regions. Several SNPs showed brain regional-specic associations. The expression level of only one gene, PDE4DIP, was associated with a SNP, rs12124527, in all the brain regions tested here. From our data, we generated a list of brain cis expression quantitative trait loci (eQTL) genes that we compared with a list of schizophrenia candidate genes downloaded from the Schizophrenia Forum (SZgene) database (http://www.szgene.org/

Web End =http:// http://www.szgene.org/

Web End =www.szgene.org/ ). Of the SZgene candidate genes, we found that the expression levels of four genes, HTR2A, PLXNA2, SRR and TCF4, were signicantly associated with cis SNPs in at least one brain region tested. One gene, SRR, was also involved in a coexpression module that we found to be associated with disease status. In addition, a substantial number of cis eQTL genes were also involved in the module, suggesting eQTL analysis of brain tissue may identify more reliable susceptibility genes for schizophrenia than casecontrol genetic association analyses. In an attempt to facilitate the identication of genetic variations that may underlie the etiology of major psychiatric disorders, we have integrated the brain eQTL results into a public and online database, Stanley Neuropathology Consortium Integrative Database (SNCID; http://sncid.stanleyresearch.org

Web End =http://sncid.stanleyresearch.org ).

Translational Psychiatry (2012) 2, e113; doi:http://dx.doi.org/10.1038/tp.2012.42

Web End =10.1038/tp.2012.42 ; published online 8 May 2012

Introduction

Schizophrenia, bipolar disorder and severe depression are common and highly disabling brain diseases caused by an interaction of genetic and environmental factors.1,2 How

ever, despite enormous efforts, the genetic variations that contribute to these diseases and their environmental risk factors remain elusive. Genome-wide association studies have frequently been employed to identify susceptibility genes and single-nucleotide polymorphisms (SNPs) that may be associated with these mental disorders.35 A number

of candidate genes for the disorders have been reported. For instance, a web resource for schizophrenia, the Schizophrenia Forum (SZgene) database (http://www.szgene.org/

Web End =http://www.szgene.org/), includes results from 1727 genetic association studies and reports 1008 candidate genes and 8788 polymorphisms in the update on 15 April 2011.6 Despite the numerous candidate genes reported for schizophrenia, the effect size of each variant is small or moderate and most associated SNPs have failed to be replicated. The need for independent and systematic validation to prioritize further examination of possible candidate genes for mental disease is widely acknowledged.

Identication of DNA sequence variants that regulate gene expression levels in a relevant tissue is one of the most

promising approaches used to initially scan for candidate genes as well as to prioritize previously identied candidate genes that are associated with complex disease such as psychiatric disorders.79 The identication of a cis association of a SNP with gene expression levels has been previously used to validate candidate genes for complex traits mapped to the same chromosomal locations.10 Our recent study using an integrative approach that combined results from genome-wide SNP scans for the cytoarchitectural traits and cis expression quantitative trait loci (eQTL) analysis in the brain tissue revealed two novel candidate genes associated with cellular abnormalities in the prefrontal cortex of major psychiatric disorders.11 Limited availability of human postmortem brain tissues is a major obstacle to obtaining detailed brain expression complex trait loci (eQTL) mapping. Utilization of publicly available resources is an effective alternative strategy that may overcome such limitation. The Stanley Neuropathology Consortium Integrative Database (SNCID; http://sncid.stanleyresearch.org

Web End =http://sncid.stanleyresearch.org ) is a publicly available and web-based tool that integrates expression microarray data sets from ve brain regions including frontal cortex, temporal cortex, thalamus, cerebellum and hippocampus and genome-wide SNP genotype data sets of subjects in the Stanley Neuropathology Consortium (SNC) and the Array Collection (AC).12 A total of 1749 neuropathology data sets using the

1Stanley Brain Research Laboratory, Stanley Medical Research Institute, Rockville, MD, USA and 2Department of Bio and Brain Engineering, KAIST, Yuseong-gu, Daejeon, Republic of KoreaCorrespondence: Dr MJ Webster, Stanley Medical Research Institute, 9800 Medical Center Drive, Rockville, MD 20850, USA or Dr D Lee, KAIST, 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Republic of Korea.

E-mail: mailto:[email protected]

Web End [email protected] or mailto:[email protected]

Web End [email protected] Keywords: cis SNP; eQTL; post-mortem brain; psychiatric disorders; schizophrenia; SNCID

Received 12 March 2012; accepted 10 April 2012

Cis eQTLs in various brain regions S Kim et al

2

SNC are integrated into the database, which thereby enables one to further explore the correlations between gene expression levels and quantitative measures of neuropathological markers in the various brain regions. The specic aims of this study are twofold. First, we explore the candidate genes that may be functionally relevant for major psychiatric disorders by identifying cis associations between SNPs and gene expression in various brain tissues. Second, we examine the possible functional role of schizophrenia candidate genes that were previously identied in genetic association studies. Thus, we explored cis eQTLs in the four brain regions, frontal cortex, temporal cortex, thalamus and cerebellum, of SNC subjects and in hippocampus of AC subjects. We also repeated the analysis in frontal cortex data from the AC as a replication study to examine the overall consensus of cis eQTLs between the two frontal data sets. We then examined whether the expression levels of any candidate genes from the SZgene database meta-analysis (http://www.szgene.org/

Web End =http://www.szgene.org/) were regulated by cis expressed SNPs (eSNPs) in brain tissues, in order to determine if there were any functional effects on gene expression of the previously identied schizophrenia susceptibility genes. Finally, we performed a coexpression network analysis between the genes in the frontal cortex that were differentially expressed between schizophrenia and normal controls and the cis eQTL genes in an attempt to identify the potential role of these genes in a disease-specic coexpression module.

Materials and methods

Data used in this study. Gene expression microarray data from frontal cortex,13 cerebellum, thalamus and temporal cortex14were generated by multiple independent groups using samples from the SNC (N 60), which contains 15

well-matched cases in each of four groups: schizophrenia, bipolar disorder, major depression and unaffected controls.15

Other sets of microarray data from frontal cortex16,17and

hippocampus were generated using samples from the AC (N 105). The AC is an independent tissue collection

containing 35 cases in each of three groups: schizophrenia, bipolar disorder and unaffected controls. The groups from both tissue collections are matched for descriptive variables such as age, gender, race, post-mortem interval, mRNA quality, brain pH and hemisphere. Outlier chip data were excluded in this analysis based on previous quality-control analyses for chip-level parameters such as scaling factor, gene call and average correlation.18 Information for the microarray studies such as tissue collection, brain region and number of outlier chips is listed in the Supplementary Table S1 online. The confounding effects on the Frozen Robust Multiarray Analysis (fRMA)-normalized microarray gene expression data were identied using Surrogate Variable Analysis (SVA).19 To adjust disease effect on the gene expression data, we randomly assign 0 or 1 for the primary variable in the SVA. All covariates from SVA were used in the linear regression to adjust the confounding effects on the gene expression data. The standardized residuals from the linear regression were used to evaluate the effectiveness of this method on removing confounding variables on two

microarray data sets from both the SNC and AC. Transcripts correlated with potential confounding variables were identied using nonparametric analysis. The continuous variables such as age, brain pH, post-mortem interval and lifetime exposure to antipsychotics were examined by correlation analysis using R (open source program from Comprehensive R Archive Network (CRAN)). Two categorical variables such as microarray batch and sex were tested using variance analysis. Adjusted P-values, based on the Hochberg method that were o0.05, were considered signicant. Although all cases and controls were included in the analysis, only the disorder cases were used for the correlation analysis for the effect of lifetime exposure to antipsychotics. SNP genotyping data using DNA samples from the SNC and the AC were generated by Dr Chun-Yu Liu and colleagues (University of Chicago, IL, USA) using the Human SNP Array 5.0 chips (Affymetrix, Santa Clara, CA, USA).20

eQTL analysis. Raw image les from SNP chips, quality-control analysis and identication of ethnic outliers were performed as previously described.11 Briey, genotypes were called using the BRLMM algorithm (Affymetrix). SNPs with a call rate of o90%, minor allele frequency o5% or extreme deviation from HardyWeinberg equilibrium test (Po0.05) were ltered out for further eQTL analyses. A total of 309 531 SNPs passed this lter. For examination of population stratication, clustering was initially performed using the pairwise identity-by-state (IBS) calculator in the PLINK.21 IBS pairwise distances were then plotted and examined by multidimensional scaling analysis and Z statistical analysis. Samples of 43 s.d. compared with the group mean were considered outliers. Four ethnic outliers from the SNC and three outliers from the AC were excluded in the eQTL analysis. One additional sample from AC was excluded because of a nal diagnosis of CADASIL (cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy). We only used genotyped SNP data from chips for our association analysis rather than imputing genotypes because SNP imputation can often result in errors in genotyping and cause false-positive associations.22 The standardized residuals from the linear regression were used as traits in PLINK for eQTL analyses. We dened cis eSNPs as those that were localized within 1 Mb of either the 50 or the 30 end of the gene. The trans eSNPs were dened as all SNPs that reached genome-wide signicance level, except those in a cis position. We employed a conservative Bonferroni method to correct multiple testing for controlling false positives.7 Adjusted P-values of o0.05 (unadjusted P-value; 1.6E 07 0.05/309 531) were con

sidered genome-wide signicant for eQTL analyses.

Coexpression network analysis. Unsupervised and supervised coexpression network analyses were performed using the Weighted Correlation Network Analysis (WGCNA) in R.23 The coexpression network was generated using expression values of all genes in the frontal cortex of schizophrenia and normal controls from the AC (unsupervised WGCNA). A second coexpression network was generated using signicant cis eQTL genes and genes that were differentially expressed in the frontal cortex between

Translational Psychiatry

Cis eQTLs in various brain regions S Kim et al

3

schizophrenia and normal controls from the AC samples (supervised WGCNA).24 A total of four microarray data sets (at http://www.stanleygenomics.org

Web End =www.stanleygenomics.org ; study no. 1, 3, 5 and 7) were generated from prefrontal cortex. Three of these (study no. 1, 3 and 7)16,17 were generated using the same platform, Affymetrix 133a, and hence to avoid variations between platforms we pooled the data from these three data sets. The pooled data were then subjected to median normalization with the biometric research branch (BRB)-array tools (http://linus.nci.nih.gov/BRB-ArrayTools.html

Web End =http:// http://linus.nci.nih.gov/BRB-ArrayTools.html

Web End =linus.nci.nih.gov/BRB-ArrayTools.html ) to remove systematic variations. After median normalization, confounding effects were adjusted using SVA and a linear regression method as described in the previous section. However, disease effect was not removed. Standardized residuals that were signicantly associated with disease (nominal P-value o0.05) and standardized residuals of cis eQTL genes were then used as input for the WGCNA.23 The minimum module size and the minimum height for merging modules were set at 30 and 0.25, respectively. The coexpression module was visualized using VisANT.25

Functional annotation. The cis eQTL genes and genes that were involved in the coexpression module were functionally annotated using the Database for Annotation, Visualization and Integrated Discovery (DAVID) database (http://david.abcc.ncifcrf.gov/home.jsp

Web End =http://david.abcc.ncifcrf.gov/home.jsp) and by the over-representational analysis method.26 The biological processes of Gene Ontology Consortium (http://www.geneontology.org

Web End =http://www.geneontology.org) were used for functional annotations. The P-values of o0.05 were considered signicant.

Results

eQTL analysis in various human post-mortem brain tissues. Gene expression microarray data derived from post-mortem brain tissue are often confounded by uncontrolled biological, clinical and technical variables.27

Batch effect is particularly problematic and has been shown

to signicantly affect gene expression levels in microarray data.28,29 To remove the effect of batch and other con

founding variables in our gene expression microarray data, we normalized the data using the newly developed method, fRMA, followed by the SVA.19,30 We evaluated how effective this method was at removing confounding variables using two microarray data sets from both the SNC and AC (Supplementary Table S1 online). Using the data set from SNC temporal cortex (study 18) we found that microarray batch was the most signicant confounding variable in both the RMA and fRMA-normalized data sets, with 947 and 1031 transcripts signicantly correlated with batch, respectively (Supplementary Table S2 online). Using the data set from AC frontal cortex (study 1) we found that microarray batch and brain pH were both major confounding variables (Supplementary Table S3 online). The SVA successfully adjusted the effects of the confounding variables on both microarray data sets (Supplementary Table S2 and S3 online).

Using the SVA we obtained the standardized residuals from the linear regression with covariates and conducted a genome-wide eQTL analysis of various brain tissues. We used the standardized residuals as traits. We initially analyzed gene expression microarray data from frontal cortex, temporal cortex, thalamus and cerebellum from the SNC (Supplementary Table S1 online). Expression levels of a total of 53, 11, 84 and 27 genes were correlated with cis SNPs in the frontal cortex, temporal cortex, thalamus and cerebellum at genome-wide signicance level, respectively (nominal Po1.6E 07;

Figure 1a and Supplementary Table S4 online). Among the cis eQTL genes, expression levels of 16, 0, 20 and 5 genes were also signicantly associated with trans SNPs in the frontal cortex, temporal cortex, thalamus and cerebellum, respectively (Supplementary Table S5 online). In addition, correlations between the expression levels of 31, 1, 69 and 15 genes and cis SNPs were unique in the frontal cortex, temporal cortex, thalamus and cerebellum, respectively (Figure 1a). The expression level of only one gene, phosphodiesterase 4D interacting protein (PDE4DIP), was associated with a SNP, rs12124527, in all the brain regions tested here. We then

Array Collection Data

Consortium

Data

Temporal Thalamus

Cerebellum

69

(C12)

2

(C15)

1

(C4)

8

(C8)

1

(C11)

3

(C6)

Frontal

1

(C7)

426 (Q2)

34

(Q1)

19 (Q3)

31

(C10)

0

(C3) 15

(C13)

1

(C1)

2

(C5)

1

(C9)

Frontal Hippocampus

2

(C2)

5

(C14)

313

(A2)

147

(A1)

134

(A3)

Figure 1 Number of cis expression quantitative trait loci (eQTL) genes in various brain regions. Venn diagram shows common and unique cis eQTL genes across multiple brain regions of the Stanley Neuropathology Consortium (SNC) samples (a) and of the Array Collection (AC) samples (c). Overlapped cis eQTL genes in the frontal cortex between the SNC samples and the AC samples are shown (b).

Translational Psychiatry

Cis eQTLs in various brain regions S Kim et al

4

Table 1 Biological processes (Gene ontology) signicantly associated with cis eQTL genes in the frontal cortex of the Stanley Neuropathology Consortium samples

Biological process categories Count Fraction (%) P-value

GO:0007155cell adhesion 7 1.361868 0.013 GO:0022610biological adhesion 7 1.361868 0.013 GO:0007601visual perception 4 0.77821 0.022 GO:0050953sensory perception of light stimulus 4 0.77821 0.022 GO:0046907intracellular transport 6 1.167315 0.036 GO:0035249synaptic transmission, glutamatergic 2 0.389105 0.039

Abbreviations: eQTL, expression quantitative trait loci; GO, Gene Ontology.

replicated the cis eQTLs of the frontal cortex using the larger AC collection. The replication study revealed associations between cis SNPs and expression levels of 460 genes and replicated 34 cis eQTL genes out of 53 (64%) that were identied in the SNC study (Figure 1b and Supplementary Table S6 online). Moreover, 281 cis eQTL genes were identied in the AC hippocampus data and 147 cis eQTLs were common to both the frontal cortex and hippocampus (Figure 1c and Supplementary Table S7 online). Among the cis eQTL genes, expression levels of 43 and 46 genes were also signicantly associated with trans SNPs in the frontal cortex and hippocampus, respectively (Supplementary Table S5 online). The association between PDE4DIP expression and the rs12124527 SNP was replicated in the AC frontal cortex and hippocampal data.

Next, we performed a functional annotation analysis to identify biological processes that were overrepresented in the brain cis eQTL genes. Whereas several processes such as cell adhesion, visual perception and glutamatergic transmission were overrepresented in the genes with cis eSNPs in the SNC frontal cortex (Table 1), metabolic processes such as glutamine metabolic process and protein transport and targeting and antigen processing were overrepresented in the AC frontal cortex (Table 2). Amino acid metabolic process, nucleotide biosynthesis and enzyme-linked receptor protein signaling pathways were signicantly overrepresented in cis eQTL genes in the AC hippocampus (Supplementary Table S8 online).

Comparison between schizophrenia susceptibility candidate genes and brain cis eQTL genes. Genetic association studies have yielded numerous candidate genes that may increase the risk for schizophrenia. However, most candidate genes have not been replicated nor functionally validated. To examine the possible functional role of schizophrenia candidate genes, we compared the list of the candidate genes in the SZgene database meta-analysis (updated 12/1/2010) to our list of cis eQTL genes. The SZgene meta-analysis identied 45 genetic variants and 42 linked genes. After excluding the non-SNP variants from their data set, we were left with 39 SNPs and 39 linked candidate genes. Because only 6 SNP markers out of the 39 SNPs were included in our Affymetrix SNP 5.0 data set, we conducted a gene-level comparison instead of SNP-level comparison. We determined whether there were cis associations between the expression levels of the 39 candidate genes and SNPs within 1 Mb of the genes. Among the 39 candidate genes, we found that the expression levels of four genes, HTR2A, PLXNA2, SRR and TCF4, were signicantly

associated with cis SNPs in at least one brain region tested (Table 3). The expression levels of HR2A and PLXNA2 were associated with cis SNPs in the frontal cortex, whereas the expression levels of SRR (serine racemase) and TCF4 were associated with cis SNPs in two brain regions. The cis eSNPs of these genes are located at least 25 kb from the SNPs that were signicantly associated with schizophrenia in the SZgene meta-analysis. Thus the SZgene casecontrol genetic association analyses for schizophrenia may not have identied the most functionally relevant genetic variations that contribute to the etiology of psychiatric disorders.

Coexpression network analysis in the frontal cortex. To further examine whether or not the four schizophrenia candidate genes (HTR2A, PLXNA2, SRR and TCF4) and genes of which expression levels were regulated by cis SNPs may be involved in the etiology of schizophrenia, we performed both unsupervised and supervised gene coexpression network analyses using the AC frontal cortex data. We were unable to construct a coexpression module that was signicantly associated with schizophrenia disease status using the unsupervised analysis. One module was associated with disease (P 0.05); however, it was also associated with

post-mortem interval (P 0.01). We then constructed a

supervised coexpression network using genes differentially expressed between schizophrenia and normal controls (Supplementary Table S9 online) and the cis eQTL genes obtained from the pooled data of three Affymetrix 133a microarray data sets that measured gene expression in the frontal cortex. We constructed one coexpression module that was signicantly associated with schizophrenia disease status (P 2E 08; Figure 2a). Age, sex, post-mortem

interval, brain pH and lifetime antipsychotic treatment were not signicantly associated with this module (all P40.05).

Genes associated with apoptosis, chromatin organization, RNA splicing, cell cycle, regulation of nucleic acid metabolism and endocytosis were overrepresented in this module (Figure 2b and Supplementary Table 10 online). A previous coexpression network analysis that used gene expression microarray data from prefrontal cortex from schizophrenia subjects and controls31 also identied a module (module 16) with similar overrepresentation of biological processes such as chromatin organization, cell cycle, endocytosis and regulation of nucleic acid metabolism. Apoptosis and endocytosis have previously been associated with the pathophysiology of the frontal cortex in schizophrenia,3244 and recent studies

have also indicated that aberrant RNA splicing and epi-genetic alterations may be involved in the pathophysiology

Translational Psychiatry

Cis eQTLs in various brain regions S Kim et al

5

Table 2 Biological processes (Gene ontology) signicantly associated with cis eQTL genes in the frontal cortex of the Array Collection samples

Biological process categories Count Fraction (%) P-value

GO:0006541glutamine metabolic process 6 0.13 5.00E 05

GO:0009064glutamine family amino acid metabolic process 8 0.17 1.50E 04

GO:0006412translation 19 0.41 5.10E 04

GO:0046907intracellular transport 28 0.60 1.80E 03

GO:0006414translational elongation 9 0.19 2.00E 03

GO:0009069serine family amino acid metabolic process 5 0.11 2.50E 03

GO:0034613cellular protein localization 19 0.41 5.50E 03

GO:0070727cellular macromolecule localization 19 0.41 5.80E 03

GO:0002474antigen processing and presentation of peptide antigen via MHC class I 4 0.09 6.00E 03

GO:0002483antigen processing and presentation of endogenous peptide antigen 3 0.06 7.10E 03

GO:0019885antigen processing and presentation of endogenous peptide antigen via MHC class I 3 0.06 7.10E 03

GO:0006508proteolysis 37 0.79 7.60E 03

GO:0006605protein targeting 12 0.26 9.50E 03

GO:0006886intracellular protein transport 17 0.36 1.10E 02

GO:0019882antigen processing and presentation 7 0.15 1.10E 02

GO:0046777protein amino acid autophosphorylation 7 0.15 1.20E 02

GO:0015031protein transport 28 0.60 1.30E 02

GO:0019883antigen processing and presentation of endogenous antigen 3 0.06 1.30E 02

GO:0045184establishment of protein localization 28 0.60 1.40E 02

GO:0051603proteolysis involved in cellular protein catabolic process 23 0.49 1.70E 02

GO:0044257cellular protein catabolic process 23 0.49 1.70E 02

GO:0043632modication-dependent macromolecule catabolic process 22 0.47 2.00E 02

GO:0019941modication-dependent protein catabolic process 22 0.47 2.00E 02

GO:0001510RNA methylation 3 0.06 2.00E 02

GO:0044265cellular macromolecule catabolic process 26 0.56 2.30E 02

GO:0030163protein catabolic process 23 0.49 2.40E 02

GO:0009070serine family amino acid biosynthetic process 3 0.06 2.40E 02

GO:0048002antigen processing and presentation of peptide antigen 4 0.09 2.40E 02

GO:0008104protein localization 30 0.64 2.50E 02

GO:0007143female meiosis 3 0.06 2.90E 02

GO:0044271nitrogen compound biosynthetic process 14 0.30 3.20E 02

GO:0006607NLS-bearing substrate import into nucleus 3 0.06 3.30E 02

GO:0034660ncRNA metabolic process 11 0.24 3.50E 02

GO:0006625protein targeting to peroxisome 3 0.06 3.80E 02

GO:0043574peroxisomal transport 3 0.06 4.40E 02

GO:0009057macromolecule catabolic process 26 0.56 4.70E 02

GO:0006399tRNA metabolic process 7 0.15 5.00E 02

Abbreviations: eQTL, expression quantitative trait loci; GO, Gene Ontology; MHC, major histocompatibility complex; NLS, nuclear localization signal; tRNA, transfer RNA.

Table 3 Schizophrenia candidate genes (from the SZgene database) with expression levels signicantly associated with cis eSNPs in brain tissue

Symbol Chr. Positiona Study_ID Brain_Region SNP S_Position P-valueb Adjusted P-value

HTR2A 13 (R): 47407513-47470175 Study01 Frontal cortex rs1923882 47411661 3.08E 10 9.45E 05

HTR2A 13 (R): 47407513-47470175 Study07 Frontal cortex rs1923882 47411661 2.54E 08 7.80E 03

PLXNA2 1 (R): 208195587-208417665 Study01 Frontal cortex rs6659522 208199100 3.26E 10 1.00E 04

PLXNA2 1 (R): 208195587-208417665 Study03 Frontal cortex rs6659522 208199100 1.03E 09 3.15E 04

PLXNA2 1 (R): 208195587-208417665 Study07 Frontal cortex rs6702082 208206946 4.04E 08 1.24E 02

SRR 17 (F): 2207248-2228553 Study01 Frontal cortex rs16952025 2116798 3.24E 08 9.94E 03

SRR 17 (F): 2207248-2228553 Study03 Frontal cortex rs16952025 2116798 8.86E 08 2.72E 02

SRR 17 (F): 2207248-2228553 Study07 Frontal cortex rs16952025 2116798 1.65E 13 5.06E 08

SRR 17 (F): 2207248-2228553 Study17 Hippocampus rs16952025 2116798 9.60E 11 2.95E 05

TCF4 18 (R): 52889562-53255860 Study01 Frontal cortex rs1261085 52889967 1.08E 08 3.33E 03

TCF4 18 (R): 52889562-53255860 Study03 Frontal cortex rs1261134 52931763 1.98E 15 6.09E 10

TCF4 18 (R): 52889562-53255860 Study05 Frontal cortex rs1261073 52907820 2.18E 12 6.68E 07

TCF4 18 (R): 52889562-53255860 Study07 Frontal cortex rs1261073 52907820 1.57E 15 4.82E 10

TCF4 18 (R): 52889562-53255860 Study16 Thalamus rs1261134 52931763 3.86E 08 1.18E 02

Abbreviation: eSNP, expressed single-nucleotide polymorphism.

Genome build;hg19. aF and R represent the forward orientation or reverse orientation on a chromosome. bAdjusted P-value using Bonferroni method.

of schizophrenia.35,36 Several genes associated with GABAergic neurons, including g-aminobutyric acid (GABA)

A receptor, d (GABRD) and parvalbumin (PVALB), were found in the coexpression module. However, out of the four schizophrenia candidate genes common to both the SZgene

meta-analysis and our cis eQTL gene list, only one, SRR, was found in the module. The biological process, response to drug, was overrepresented in the module and was enriched with 11 genes including SRR. A substantial number of cis eQTL genes were involved in the coexpression module

Translational Psychiatry

Cis eQTLs in various brain regions S Kim et al

6

Apoptosis Cytoskeleton organization

Chromatin organization RNA splicing Response to drug

Endocytosis

0 5 10 15 20 25 30

No. of genes

Figure 2 Coexpression network analysis in the frontal cortex. The coexpression module that is signicantly associated with schizophrenia in frontal cortex of the Array Collection (AC) (a) and biological processes (Gene ontology) overrepresented in the genes in the coexpression module (b). Network connections with topological overlap above the threshold of 0.02 were visualized using VisANT.25 The cis expression quantitative trait loci (eQTL) genes are pink. The candidate gene, SRR (serine racemase), derived from the meta-analyses of genetic studies in the SZgene database (http://www.szgene.org/) is in blue.

that was signicantly associated with schizophrenia disease status and were also associated with the biological processes. This result indicates that cis eQTL analysis in brain tissue may more reliably identify susceptibility genes for schizophrenia as compared with the current casecontrol genetic association studies.

Discussion

Identifying genetic variations that affect gene expression in the brain may be a promising approach for nding molecular pathways that are functionally relevant to the etiology and/or treatment of mental disease. In this study, we conducted an eQTL analysis of 315 440 transcripts in 5 different brain regions from two different tissue collections and identied cis associations between 648 transcripts and 6725 SNPs. The expression of one gene, PDE4DIP, was associated with one SNP, rs12124527, in all brain regions examined. This association was also previously described in the frontal cortex.20 The protein encoded by PDE4DIP serves to anchor phosphodiesterase 4D to the Golgi/centrosome region of the cell. A number of abnormalities in the phosphodiesterase signaling system have been described in the brains of subjects with schizophrenia, bipolar disorder and depression,3742 indicating that molecules within this system could be potential targets for therapeutic intervention.38,43

Approximately 14% of cis eQTL genes were also correlated with trans SNPs in various brain regions, suggesting that the expression levels of a subset of cis eQTL genes may be regulated by multiple variants. However, when we examined whether or not the expression levels of candidate genes from the SZgene database meta-analysis were signicantly associated with cis SNPs, we found only 4 genes that overlapped between the SZgene database and our eQTL gene list. Furthermore, only one candidate gene, SRR, was involved in

a coexpression module that was associated with schizophrenia. SRR maps to chromosome 17p13 and encodes an enzyme that synthesizes D-serine from L-serine.44 The

D-serine is an endogenous co-agonist of the N-methyl-D-aspartate (NMDA) receptor.45 Hypofunction of the NMDA receptor is potentially a major underlying pathophysiology of schizophrenia.46,47 Our results support this hypothesis and

suggest that abnormal NMDA receptor-mediated signaling may be inuenced by genetic variations. A SNP, rs16952025, localized in an intron of the neighboring gene, SMG6, was signicantly associated with the expression level of SRR. However, there was no signicant association between this SNP and the expression level of SMG6. Several post-mortem studies have examined levels of SRR mRNA and serine racemase protein in schizophrenia48 and found abnormalities in schizophrenia, although the results have been inconsistent. Although SRR mRNA levels appear to be unchanged in frontal cortex of schizophrenia,41 the protein levels have been reported to be either decreased,49 increased50 or unchanged.51 The inconsistent results are most likely because of different methodologies, different cohorts (often with small numbers) and the different brain areas used. Consequently further study will be required in the future when larger cohorts become available to conrm changes in SRR levels in the brain of subjects with mental illness.

Our comprehensive brain eQTL analysis functionally validated only 4 genes out of 39 candidate genes positively identied in the SZgene meta-analysis. We were unable to identify any signicant associations between the expression levels of the remaining genes and cis SNPs in any of the brain regions we tested. In fact, the 39 candidate genes were derived from 1008 candidate genes that were obtained from 1727 original genetic association studies. Such a low functional validation rate raises the possibility that the current casecontrol genetic association studies may not effectively

Translational Psychiatry

Cis eQTLs in various brain regions S Kim et al

7

identify genetic variations that underlie the etiology of schizophrenia. However, there are other reasons that may contribute to a low functional validation rate. For example, the probes on the microarray platforms used to analyze gene expression in this study mainly bind to sequences in the 30-untranslated regions and do not distinguish between various alternative splicing isoforms. Indeed, tissue-specic alternative RNA-splicing is very predominant in the brain.52

Furthermore, intronic SNPs can be associated with the altered expression of specic alternative splicing isoforms of certain schizophrenia candidate genes, for example, ErbB4 and GRM3.53,54 Therefore, comprehensive expression proling that includes various alternative splicing isoforms using deep mRNA-sequencing technology may aid in the identication of novel cis eQTL genes in human post-mortem brain tissues in the future.

The frontal cortex is one of the most thoroughly examined brain regions in post-mortem studies and many neuropathology abnormalities have been identied in this region in schizophrenia.55,56 Previous gene expression microarray studies in the frontal cortex identied several biological processes that were overrepresented in the genes differentially expressed between schizophrenia and normal controls; for example, decreased presynaptic function, abnormal mitochondrial function and altered expression of apoptosis-related genes are all major ndings from microarray studies of frontal cortex in schizophrenia.33,34,57 However, glutamater

gic transmission, amino acid metabolism, proteolysis and protein targeting were all overrepresented in the eQTL genes in the frontal cortex in our current study. Thus, the abnormalities described in the biological pathways from the eQTL study may be more directly related to genetic variation, whereas the pathways identied by gene expression studies are likely to be inuenced by factors in addition to genetic variation, including epigenetics and environmental factors.

Although our study reveals a number of associations between cis SNPs and gene expression in multiple brain regions, the results should be interpreted with caution. First, the SNC, which we used for the initial eQTL analyses, contains a relatively small sample size (N 56). Small sample

size is known to generate higher false-positive associations as well as to be a cause of low detection power in genome-wide association analysis. Thus, the cis eQTL results from frontal cortex, cerebellum, thalamus, and temporal cortex using SNC samples should be viewed as exploratory. However, we subsequently performed a second analysis using an independent collection (AC) with a larger sample size (N 101).

Our previous power analysis using AC as well as a previous eSNP association study indicated that a relatively small sample size (N 100) has 480% power to detect an

association of gene expression traits with moderate effect size (R2 0.35).11,58 We therefore attempted to replicate the

results of cis eQTLs in frontal cortex from SNC using the AC data. A total of 34 (64%) of the cis eQTL genes identied in the SNC frontal cortex data were also found in the AC frontal data. However, we identied 450 additional cis eQTL genes in the AC frontal cortex samples, which were not identied in the SNC frontal samples, suggesting that some signicant cis eQTL genes may have been missed in the SNC analysis.

Second, using whole tissues for gene expression traits may dilute the effect of some genetic variants that may only act on cell type-specic gene expression. Although this phenomena has not been explored in the brain, there are numerous cell type-specic abnormalities in the brain of subjects with psychiatric disorders.32,5961 Thus, the use of cell type-

specic expression traits in future studies may increase the power to identify cis eQTLs in the brain.

In this study, we investigated the associations between SNPs and gene expression in various human brain regions. Although previous brain eQTL studies focused on cortex,8,62

we have extended the analysis to include the hippocampus, thalamus and cerebellum. These data can be used to identify genetic variations associated with psychiatric disorder and can be used to identify genetic variations that affect neuropatho-logical abnormalities and gene expression changes. As we show in this study, the data can be used to functionally validate candidate genes to determine if they are affecting changes in gene expression in subjects with neuropsychiatric disorders. In order to facilitate further studies, we have integrated the genome-wide eQTL results from this study into the SNCID, which is a web-based database that also includes 1747 neuropathological markers measured in the same SNC samples. The update will allow users to investigate associations between SNPs and genes of interests in various brain regions and to further explore associations between SNPs and neuropathological markers and gene expression traits that are correlated with neuropathological markers in the various brain regions of subject with major psychiatric disorders.

Conict of interest

The authors declare no conict of interest.

Acknowledgements. We thank all the investigators who generated the original data in the SNCID, and their many collaborators, who made this database possible. We also thank all the technicians in the SMRI brain laboratory who prepared the brain tissues and extracted the RNA and DNA from the tissues. We specially thank the Keymind Company for their technical assistance with the database, in particular Marvin Suo. We thank Dr Horvath for helpful comments on network analysis using the WGCNA. HC and DL were supported by the World Class University program (R32-2008-000-10218-0) of the Ministry of Education, Science and Technology through the National Research Foundation of Korea.

1. Mueser KT, McGurk SR. Schizophrenia. Lancet 2004; 363: 20632072.2. Sullivan PF, Neale MC, Kendler KS. Genetic epidemiology of major depression: review and meta-analysis. Am J Psychiatry 2000; 157: 15521562.

3. Purcell SM, Wray NR, Stone JL, Visscher PM, ODonovan MC, Sullivan PF et al. Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature 2009; 460: 748752.

4. Stefansson H, Ophoff RA, Steinberg S, Andreassen OA, Cichon S, Rujescu D et al. Common variants conferring risk of schizophrenia. Nature 2009; 460: 744747.

5. Sklar P, Smoller JW, Fan J, Ferreira MA, Perlis RH, Chambert K et al. Whole-genome association study of bipolar disorder. Mol Psychiatry 2008; 13: 558569.

6. Allen NC, Bagade S, McQueen MB, Ioannidis JP, Kavvoura FK, Khoury MJ et al. Systematic meta-analyses and eld synopsis of genetic association studies in schizophrenia: the SzGene database. Nat Genet 2008; 40: 827834.

7. Dixon AL, Liang L, Moffatt MF, Chen W, Heath S, Wong KC et al. A genome-wide association study of global gene expression. Nat Genet 2007; 39: 12021207.

8. Myers AJ, Gibbs JR, Webster JA, Rohrer K, Zhao A, Marlowe L et al. A survey of genetic human cortical gene expression. Nat Genet 2007; 39: 14941499.

9. Stranger BE, Nica AC, Forrest MS, Dimas A, Bird CP, Beazley C et al. Population genomics of human gene expression. Nat Genet 2007; 39: 12171224.

10. Emilsson V, Thorleifsson G, Zhang B, Leonardson AS, Zink F, Zhu J et al. Genetics of gene expression and its effect on disease. Nature 2008; 452: 423428.

Translational Psychiatry

Cis eQTLs in various brain regions S Kim et al

8

11. Kim S, Webster MJ. Integrative genome-wide association analysis of cytoarchitectural abnormalities in the prefrontal cortex of psychiatric disorders. Mol Psychiatry 2010; 16: 452461.

12. Kim S, Webster MJ. The stanley neuropathology consortium integrative database: a novel, web-based tool for exploring neuropathological markers in psychiatric disorders and the biological processes associated with abnormalities of those markers. Neuropsycho-pharmacology 2010; 35: 473482.

13. Tkachev D, Mimmack ML, Ryan MM, Wayland M, Freeman T, Jones PB et al. Oligodendrocyte dysfunction in schizophrenia and bipolar disorder. Lancet 2003; 362: 798805.

14. Aston C, Jiang L, Sokolov BP. Microarray analysis of postmortem temporal cortex from patients with schizophrenia. J Neurosci Res 2004; 77: 858866.

15. Torrey EF, Webster M, Knable M, Johnston N, Yolken RH. The stanley foundation brain collection and neuropathology consortium. Schizophr Res 2000; 44: 151155.

16. Iwamoto K, Bundo M, Kato T. 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 Mol Genet 2005; 14: 241253.

17. Ryan MM, Lockstone HE, Huffaker SJ, Wayland MT, Webster MJ, Bahn S. Gene expression analysis of bipolar disorder reveals downregulation of the ubiquitin cycle and alterations in synaptic genes. Mol Psychiatry 2006; 11: 965978.

18. Higgs BW, Elashoff M, Richman S, Barci B. An online database for brain disease research. BMC Genomics 2006; 7: 70.

19. Leek JT, Storey JD. Capturing heterogeneity in gene expression studies by surrogate variable analysis. PLoS Genet 2007; 3: 17241735.

20. Liu C, Cheng L, Badner JA, Zhang D, Craig DW, Redman M et al. Whole-genome association mapping of gene expression in the human prefrontal cortex. Mol Psychiatry 2010; 15: 779784.

21. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 2007; 81: 559575.

22. Pei YF, Li J, Zhang L, Papasian CJ, Deng HW. Analyses and comparison of accuracy of different genotype imputation methods. PLoS One 2008; 3: e3551.

23. Langfelder P, Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 2008; 9: 559.

24. Gargalovic PS, Imura M, Zhang B, Gharavi NM, Clark MJ, Pagnon J et al. Identication of inammatory gene modules based on variations of human endothelial cell responses to oxidized lipids. Proc Natl Acad Sci USA 2006; 103: 1274112746.

25. Hu Z, Mellor J, Wu J, DeLisi C. VisANT: an online visualization and analysis tool for biological interaction data. BMC Bioinformatics 2004; 5: 17.

26. Dennis Jr G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC et al. DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol 2003; 4: P3.

27. Mirnics K, Levitt P, Lewis DA. DNA microarray analysis of postmortem brain tissue. Int Rev Neurobiol 2004; 60: 153181.

28. Fare TL, Coffey EM, Dai H, He YD, Kessler DA, Kilian KA et al. Effects of atmospheric ozone on microarray data quality. Anal Chem 2003; 75: 46724675.

29. Leek JT, Scharpf RB, Bravo HC, Simcha D, Langmead B, Johnson WE et al. Tackling the widespread and critical impact of batch effects in high-throughput data. Nat Rev Genet 2010; 11: 733739.

30. McCall MN, Bolstad BM, Irizarry RA. Frozen robust multiarray analysis (fRMA). Biostatistics 2010; 11: 242253.

31. Torkamani A, Dean B, Schork NJ, Thomas EA. Coexpression network analysis of neural tissue reveals perturbations in developmental processes in schizophrenia. Genome Res 2010; 20: 403412.

32. Uranova NA, Vostrikov VM, Orlovskaya DD, Rachmanova VI. Oligodendroglial density in the prefrontal cortex in schizophrenia and mood disorders: a study from the Stanley Neuropathology Consortium. Schizophr Res 2004; 67: 269275.

33. Mirnics K, Middleton FA, Marquez A, Lewis DA, Levitt P. Molecular characterization of schizophrenia viewed by microarray analysis of gene expression in prefrontal cortex. Neuron 2000; 28: 5367.

34. Kim S, Webster MJ. Correlation analysis between genome-wide expression proles and cytoarchitectural abnormalities in the prefrontal cortex of psychiatric disorders. Mol Psychiatry 2010; 15: 326336.

35. McInnes LA, Lauriat TL. RNA metabolism and dysmyelination in schizophrenia. Neurosci Biobehav Rev 2006; 30: 551561.

36. Huang HS, Matevossian A, Whittle C, Kim SY, Schumacher A, Baker SP et al. Prefrontal dysfunction in schizophrenia involves mixed-lineage leukemia 1-regulated histone methylation at GABAergic gene promoters. J Neurosci 2007; 27: 1125411262.

37. Millar JK, Pickard BS, Mackie S, James R, Christie S, Buchanan SR et al. DISC1 and PDE4B are interacting genetic factors in schizophrenia that regulate cAMP signaling. Science 2005; 310: 11871191.

38. Wong ML, Whelan F, Deloukas P, Whittaker P, Delgado M, Cantor RM et al. Phosphodiesterase genes are associated with susceptibility to major depression and anti-depressant treatment response. Proc Natl Acad Sci USA 2006; 103: 1512415129.

39. Fatemi SH, King DP, Reutiman TJ, Folsom TD, Laurence JA, Lee S et al. PDE4B polymorphisms and decreased PDE4B expression are associated with schizophrenia. Schizophr Res 2008; 101: 3649.

40. Fatemi SH, Reutiman TJ, Folsom TD, Lee S. Phosphodiesterase-4A expression is reduced in cerebella of patients with bipolar disorder. Psychiatr Genet 2008; 18: 282288.

41. Fatemi SH, Folsom TD, Reutiman TJ, Vazquez G. Phosphodiesterase signaling system is disrupted in the cerebella of subjects with schizophrenia, bipolar disorder, and major depression. Schizophr Res 2009; 119: 266267.

42. Numata S, Iga J, Nakataki M, Tayoshi S, Taniguchi K, Sumitani S et al. Gene expression and association analyses of the phosphodiesterase 4B (PDE4B) gene in major depressive disorder in the Japanese population. Am J Med Genet B Neuropsychiatr Genet 2009; 150B: 527534.

43. Hennah W, Porteous D. The DISC1 pathway modulates expression of neuro-developmental, synaptogenic and sensory perception genes. PLoS One 2009; 4: e4906.

44. Wolosker H, Blackshaw S, Snyder SH. Serine racemase: a glial enzyme synthesizing D-serine to regulate glutamate-N-methyl-D-aspartate neurotransmission. Proc Natl Acad Sci USA 1999; 96: 1340913414.

45. Leeson PD, Iversen LL. The glycine site on the NMDA receptor: structure-activity relationships and therapeutic potential. J Med Chem 1994; 37: 40534067.

46. Olney JW, Newcomer JW, Farber NB. NMDA receptor hypofunction model of schizophrenia. J Psychiatr Res 1999; 33: 523533.

47. Belforte JE, Zsiros V, Sklar ER, Jiang Z, Yu G, Li Y et al. Postnatal NMDA receptor ablation in corticolimbic interneurons confers schizophrenia-like phenotypes. Nat Neurosci 2010; 13: 7683.

48. Verrall L, Burnet PW, Betts JF, Harrison PJ. The neurobiology of D-amino acid oxidase and its involvement in schizophrenia. Mol Psychiatry 2010; 15: 122137.

49. Bendikov I, Nadri C, Amar S, Panizzutti R, De Miranda J, Wolosker H et al. A CSF and postmortem brain study of D-serine metabolic parameters in schizophrenia. Schizophr Res 2007; 90: 4151.

50. Verrall L, Walker M, Rawlings N, Benzel I, Kew JN, Harrison PJ et al. d-Amino acid oxidase and serine racemase in human brain: normal distribution and altered expression in schizophrenia. Eur J Neurosci 2007; 26: 16571669.

51. Steffek AE, Haroutunian V, Meador-Woodruff JH. Serine racemase protein expression in cortex and hippocampus in schizophrenia. NeuroReport 2006; 17: 11811185.

52. Grabowski PJ, Black DL. Alternative RNA splicing in the nervous system. Prog Neurobiol 2001; 65: 289308.

53. Sartorius LJ, Weinberger DR, Hyde TM, Harrison PJ, Kleinman JE, Lipska BK. Expression of a GRM3 splice variant is increased in the dorsolateral prefrontal cortex of individuals carrying a schizophrenia risk SNP. Neuropsychopharmacology 2008; 33: 26262634.

54. Law AJ, Kleinman JE, Weinberger DR, Weickert CS. Disease-associated intronic variants in the ErbB4 gene are related to altered ErbB4 splice-variant expression in the brain in schizophrenia. Hum Mol Genet 2007; 16: 129141.

55. Knable MB, Barci BM, Bartko JJ, Webster MJ, Torrey EF. Molecular abnormalities in the major psychiatric illnesses: Classication and Regression Tree (CRT) analysis of postmortem prefrontal markers. Mol Psychiatry 2002; 7: 392404.

56. Knable MB, Torrey EF, Webster MJ, Bartko JJ. Multivariate analysis of prefrontal cortical data from the Stanley Foundation Neuropathology Consortium. Brain Res Bull 2001; 55: 651659.

57. Prabakaran S, Swatton JE, Ryan MM, Huffaker SJ, Huang JT, Grifn JL et al. Mitochondrial dysfunction in schizophrenia: evidence for compromised brain metabolism and oxidative stress. Mol Psychiatry 2004; 9: 684697, 643.

58. Cheung VG, Spielman RS, Ewens KG, Weber TM, Morley M, Burdick JT. Mapping determinants of human gene expression by regional and genome-wide association. Nature 2005; 437: 13651369.

59. Beasley CL, Zhang ZJ, Patten I, Reynolds GP. Selective decits in prefrontal cortical GABAergic neurons in schizophrenia dened by the presence of calcium-binding proteins. Biol Psychiatry 2002; 52: 708715.

60. Lewis DA, Hashimoto T, Morris HM. Cell and receptor type-specic alterations in markers of GABA neurotransmission in the prefrontal cortex of subjects with schizophrenia. Neurotox Res 2008; 14: 237248.

61. Vostrikov VM, Uranova NA, Orlovskaya DD. Decit of perineuronal oligodendrocytes in the prefrontal cortex in schizophrenia and mood disorders. Schizophr Res 2007; 94: 273280.

62. Webster JA, Gibbs JR, Clarke J, Ray M, Zhang W, Holmans P et al. Genetic control of human brain transcript expression in Alzheimer disease. Am J Hum Genet 2009; 84: 445458.

Translational Psychiatry is an open-access journal published by Nature Publishing Group. This work is licensed under the Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/3.0/

Supplementary Information accompanies the paper on the Translational Psychiatry website (http://www.nature.com/tp

Web End =http://www.nature.com/tp)

Translational Psychiatry