Human Genetics

, Volume 119, Issue 5, pp 558–570

Genome scans and gene expression microarrays converge to identify gene regulatory loci relevant in schizophrenia

Authors

    • Department of Psychiatry and Human Behavior, Functional Genomics Laboratory, College of MedicineUniversity of California
  • Mary E. Atz
    • Department of Psychiatry and Human Behavior, Functional Genomics Laboratory, College of MedicineUniversity of California
  • Brandi L. Rollins
    • Department of Psychiatry and Human Behavior, Functional Genomics Laboratory, College of MedicineUniversity of California
  • Kathleen M. Cooper-Casey
    • Department of Biological ChemistryUniversity of California Irvine
  • Ling Shao
    • Department of Psychiatry and Human Behavior, Functional Genomics Laboratory, College of MedicineUniversity of California
  • William F. Byerley
    • Veterans Administration Medical CenterUniversity of California
Original Investigation

DOI: 10.1007/s00439-006-0172-7

Cite this article as:
Vawter, M.P., Atz, M.E., Rollins, B.L. et al. Hum Genet (2006) 119: 558. doi:10.1007/s00439-006-0172-7

Abstract

Multiple linkage regions have been reported in schizophrenia, and some appear to harbor susceptibility genes that are differentially expressed in postmortem brain tissue derived from unrelated individuals. We combined traditional genome-wide linkage analysis in a multiplex family with lymphocytic genome-wide expression analysis. A genome scan suggested linkage to a chromosome 4q marker (D4S1530, LOD 2.17, θ=0) using a dominant model. Haplotype analysis using flanking microsatellite markers delineated a 14 Mb region that cosegregated with all those affected. Subsequent genome-wide scan with SNP genotypes supported the evidence of linkage to 4q33–35.1 (LOD=2.39) using a dominant model. Genome-wide microarray analysis of five affected and five unaffected family members identified two differentially expressed genes within the haplotype AGA and GALNT7 (aspartylglucosaminidase and UDP-N-acetyl-alpha-D-galactosamine: polypeptide N-acetylgalactosaminyltransferase 7) with nominal significance; however, these genes did not remain significant following analysis of covariance. We carried out genome-wide linkage analyses between the quantitative expression phenotype and genetic markers. AGA expression levels showed suggestive linkage to multiple markers in the haplotype (maximum LOD=2.37) but to no other genomic region. GALNT7 expression levels showed linkage to regulatory loci at 4q28.1 (maximum LOD=3.15) and in the haplotype region at 4q33–35.1 (maximum LOD=2.37). ADH1B (alcohol dehydrogenase IB) was linked to loci at 4q21–q23 (maximum LOD=3.08) and haplotype region at 4q33–35.1 (maximum LOD=2.27). Seven differentially expressed genes were validated with RT-PCR. Three genes in the 4q33–35.1 haplotype region were also differentially expressed in schizophrenia in postmortem dorsolateral prefrontal cortex: AGA, HMGB2, and SCRG1. These results indicate that combining differential gene expression with linkage analysis may help in identifying candidate genes and potential regulatory sites. Moreover, they also replicate recent findings of complex trans- and cis- regulation of genes.

Introduction

Genome-wide linkage analysis is an important tool for mapping complex genetic disorders such as schizophrenia. While linkage has been predictably more successful for Mendelian disorders, linkage analysis has identified a number of complex disease loci. To date, multiple linkage regions have been reported in schizophrenia, and some appear to harbor schizophrenia susceptibility genes (O’Donovan et al. 2003; Harrison and Owen 2003). With the exception of a few genes, such as neuregulin 1 (Stefansson et al. 2002) and dysbindin (Straub et al. 2002), there have been no other replicated genes found to be associated with schizophrenia in linkage regions. With a sufficient number of informative meioses (∼100 to 200) linkage can narrow a disease gene search for a Mendelian-based disorder to an interval of ∼1 cM. For complex diseases, due to incomplete penetrance and the inability to distinguish phenocopies from true recombination events, typically linkage can only narrow a disease gene search to an area of 10–30 cM. Such regions contain hundreds of potential candidate genes. Additional methods are needed to help prioritize the selection of candidate genes for mutation searches in broad linkage regions. It is known that noncoding variants can alter gene transcription (Knight 2005). Thus, genes that are differentially expressed in linkage regions may be attractive candidates for mutation searches and postmortem expression studies.

While access to brain tissue is not possible for living subjects, a readily available and commonly used source of mRNA for expression studies are lymphocytes, due to their availability from living participants. Lymphocyte tissue is of a different embryological origin than brain and does not express all brain relevant genes, while lymphocytes do show abundant expression of genes found in neural tissues (M. P. Vawter, unpublished results). Another advantage of lymphocyte gene expression is the relative absence of confounding factors, such as postmortem interval, agonal factor, or pH (Vawter et al. 2004; Tsuang et al. 2005). Medications can affect in vivo gene expression, but these effects might be mitigated in transformed lymphocytes utilized in the present study. Moreover, regulatory mutations that alter gene expression in the brain may lead to altered expression in other tissues. It is now possible to study gene expression on a genome-wide basis using high-density microarrays, and such investigations may help identify genes underlying illness within linkage regions. In addition, microarray investigation of gene expression also makes it possible to study whether a gene (or genes) within a linkage region is coregulated with transcription of other genes in the genome. Such studies may identify other loci important for schizophrenia as well as regulatory or metabolic pathways underlying the pathophysiology. We have combined genome linkage analysis and genome-wide expression in the present study.

Methods

Subjects

All subjects provided informed consent for genetic mapping studies, and the institutional review board approved all protocols. The multiplex pedigree consisted of 20 members (Fig. 1). Seventeen individuals in the pedigree were initially scanned with microsatellite markers for genome-wide linkage analysis. Ten informative subjects were selected for final screening: five individuals with schizophrenia and the haplotype, and five unaffected controls without the haplotype (Table 1). These 10 subjects (Fig. 1) were analyzed by microarray, SNP microarray genotyping, Q-PCR, and mutation scanning.
https://static-content.springer.com/image/art%3A10.1007%2Fs00439-006-0172-7/MediaObjects/439_2006_172_Fig1_HTML.gif
Fig. 1

Multiplex-pedigree showing schizophrenia (black diamond) and unaffected family members (open diamond). Individuals 1–10 were studied by microarray, SNP genotype, and Q-PCR. Individuals 1–17 were included in the genome-wide microsatellite marker linkage scan

RNA and DNA isolation from lymphocytes

The procedure for generating cell lines and extraction of total RNA was followed as previously described (Vawter et al. 2004). RNA was extracted from ∼5×107 lymphoblastic cells using the standard TRIzol isolation protocol (Invitrogen, Carlsbad, CA, USA). The total RNA was cleaned by passing over silica-based mini-spin columns (Qiagen RNeasy Mini Kit, Valencia, CA, USA) and analyzed on a 2100 Bioanalyzer (Agilent, Palo Alto, CA, USA) for quantification of 28 and 18S ribosomal RNA peaks. One flask of 5×107 cells was used for genomic DNA extraction using the standard chloroform/isoamyl alcohol isolation procedure. DNA was precipitated from the aqueous layer using a final concentration of 0.3 M sodium acetate and 1.5 volumes of ice-cold 95% isopropyl alcohol, and the precipitate was collected by centrifugation. DNA was resuspended in 1 ml of TE, and the purity and concentration were determined by spectrophotometric absorbance at 260 and 280 nm (BioRad SmartSpec 3000).

Postmortem DLPFC microarray study

A microarray study of dorsolateral prefrontal cortex (DLPFC) from the Stanley Foundation Microarray collection used 34 subjects with bipolar disorder, 36 subjects with schizophrenia, and 36 controls was run on the Codelink 20 K platform (GE Health, AZ, USA). The demographics of the subjects for each group are shown (Table 3), the study is being reported separately (M. P. Vawter, unpublished results). The Codelink 20 K platform uses a 30-mer probe, and fluorescent glass slides, to interrogate approximately 19,300 human Genbank accessions from cDNA, RefSeq mRNA, and Unigene clusters. The labeling, hybridization, and scanning of the microarrays were followed using the manufacturer’s protocol (GE Health). The expression of genes that were located in the 4q33–35.1 region are shown for controls, bipolar disorder, and schizophrenia (Table 4).

Genome scan with microsatellite markers

An initial genome scan of nine families, each containing three to five cases of schizophrenia, was previously conducted with 329 polymorphic DNA markers spaced at approximately 15–20 cM intervals as described (Coon et al. 1994). We further analyzed in the present study one pedigree with 5 affected and 12 unaffected individuals that showed suggestive linkage on chromosome 4q. Fine mapping was completed with microsatellite markers spaced at approximately 2 cM intervals. Radiolabeled PCR-products were separated on a 6% denaturing polyacrylamide gel followed by autoradiography. Two technicians blind to diagnosis scored genotypes independently.

GeneChip mapping assay

A linkage scan of this multiplex pedigree was performed using the GeneChip® Mapping10 K Xba Array containing 11,529 SNP markers (Affymetrix Inc., Santa Clara, CA, USA). Briefly, 250 ng of genomic DNA was digested with the restriction endonuclease XbaI (0.5 U/μl) for 2 h in the Applied Biosystems 9700 GeneAmp PCR System. The digested DNA was ligated using 0.25 μM Adaptor Xba and 1× T4 DNA ligase for 2 h. Ligated DNA was amplified by PCR in four separate reactions using 0.75 μM Xba PCR Primer and 0.1 U/μl AmpliTaq Gold polymerase. Each PCR product was run on a 2% TBE agarose gel to ensure successful amplification. The PCR products were pooled and cleaned using Qiagen QIAquick columns and DNA concentration determined by spectrophotometric absorbance at 260 nm (BioRad SmartSpec 3000). The purified samples were fragmented with 0.048 U/μl Fragmentation Reagent (Affymetrix) for 30 min. To check for proper fragmentation, 4 μl of fragmented PCR was run on a 4% TBE agarose gel. Fragments were end-labeled using 0.143 mm GeneChip DNA labeling reagent and 1.5 U/μl terminal deoxynucleotidyl transferase for 2 h. The labeled DNA was hybridized to a GeneChip® Mapping10 K array at 48°C for 16–18 h at 60 rpm in the Affymetrix 640 hybridization oven. After hybridization, the arrays were washed, stained, and scanned using an Affymetrix Fluidics Station F450 with images obtained by the Affymetrix GeneArray® scanner 3000. Affymetrix MicroArray Suite software was used to generate raw microarray feature intensities (RAS scores). RAS scores were processed using Affymetrix Genotyping Tools software GCOS/GDAS (Affymetrix Inc.) to derive SNP genotypes, marker order, and linear chromosomal location.

The SNP genotype data were formatted for dCHIP SNPLinkage software (Leykin et al. 2005). There were a total of 11,229 SNPs genotyped with a 90% average SNP detection rate across 10 arrays (range 85–95%). Of 11,229 SNPs on the Affymetrix SNP chip that were genotyped, 3,235 SNPs were not informative mainly due to homozygosity in all family members. A small fraction (160) were omitted due to Mendelian errors in the pedigree leaving 7,994 SNP markers for linkage analysis.

Linkage analysis

Parametric analysis of microsatellite markers to the disease assumed a dominant model, (θ recursively varied from 0.05–0.45) and was calculated using LINKAGE software. dChip SNP Linkage and MERLIN (Abecasis et al. 2002) were both used to analyze linkage of phenotype to SNP genotype data. SNP data for autosomal chromosomes were used for linkage analysis in the pedigree by a variant of the Lander–Green algorithm in the dChip Linkage module of dChip to perform multipoint parametric linkage analysis and compute a LOD score at each SNP position (Lander and Green 1987; Kruglyak et al. 1996, 1995). We verified linkage found with dChip Linkage with MERLIN version 1.0.1 software. The MERLIN software package version 1.0.1 (Abecasis et al. 2002) was used to analyze linkage to the disease with a dominant parametric model for both SNP and MSM markers. In using either MERLIN or dChip Linkage, there was only one suggestive linkage region found in the genome scan at chromosome 4q.

MERLIN was used to analyze each quantitative gene expression phenotype from microarray analysis to genetic microsatellite and SNP markers. The linkage of an individual gene expression value as a continuous trait and a microsatellite marker was calculated with MERLIN variance component linkage analysis program. The potential regulatory regions were mapped as cis-regulatory if within 5 Mb of the gene, and trans-regulatory, if >5 Mb from the gene (Morley et al. 2004). Two nominal differential expressed genes within the haplotype (set 1: AGA aspartylglucosaminidase, GALNT7 UDP-N-acetyl-alpha-D-galactosamine:polypeptide N-acetylgalactosaminyltransferase 7) were tested for linkage to markers on chromosome 4. Potential trans-regulators of 200 of the most significant differentially expressed genes with high coefficient of variation and located outside the schizophrenia haplotype (set 2: 200 genes) were examined for linkage to chromosome 4q genetic markers. To investigate trans-regulation further, all the remaining differentially expressed probesets throughout the genome (set 3: 1,327 genes) were analyzed for linkage to chromosome 4 markers. There were 153 SNPs from 150–191 Mb on chromosome 4 and 29 microsatellite markers spanning the entire chromosome that were used to detect linkage of gene expression to the 4q33–35.1 haplotype. Although these methods have application to genome-wide studies, this report focuses on linkage studies conducted mainly on the chromosome 4 results. A separate report of genome-wide linkage to genome-wide gene expression is in preparation to analyze the results of the more extensive calculations involving ∼5×108 comparisons of genetic markers to gene expression.

Direct DNA sequencing of candidate genes

Candidate genes within the linkage peak were selected for mutational analysis. Individual genes (AGA, FLJ22649, HAND2) were scanned for known and unknown SNPs. Affected individuals with the chromosome 4q haplotype and unaffected individuals without the haplotype were used for sequencing. DNA was amplified by PCR followed by fluorescent sequencing. Briefly, the primers were designed using Oligo 6.0 (Molecular Biology Insights) and purchased from Applied Biosystems. DNA fragments were resolved via electrophoresis on a 1% agarose gel, and bands were visualized (ChemiDoc digital imaging system, BioRad, CA, USA) and extracted from the gel with the Qiagen Gel Extraction Kit (Qiagen Inc., CA, USA). Each fragment was sequenced using fluorescently labeled dye terminator sequencing chemistry (Big Dye Terminator Cycle Sequencing v.3.1, Applied Biosystems) using capillary gel electrophoresis (ABI Prism 3100).

Oligonucleotide expression microarray

Total RNA from human lymphoblastoid cells was reverse-transcribed to cDNA using a T7-(dT)24 (HPLC purified) primer (Superscript cDNA Synthesis Kit, Invitrogen). The cDNA was cleaned (Qiagen Gene Chip Sample Cleanup Module) and synthesized into cRNA (MegaScript T7, Ambion, TX, USA), incorporating biotinylated-UTP and biotinylated-CTP (PE Lifesciences, MA, USA). Biotinylated cRNA was cleaned (Qiagen Gene Chip Sample Cleanup Module), quality-checked (Bioanalyzer 2100, Agilent), fragmented, and hybridized to the Affymetrix Human Genome U133A array. Arrays were washed, stained, and scanned on the Affymetrix Fluidics Station and Gene Scanner. The gene expression traits were derived from the U133A chips and analyzed with robust multiarray condensation algorithm (RMA; Irizarry et al. 2003). Differential gene expression (gene expression trait for the purpose of this analysis) was defined as a gene that displayed a significant two-tailed t test (P<0.05) in schizophrenia compared to unaffected family members.

Q-PCR

Primers were designed within a 3′ exon using the Primer Express program (Applied Biosystems). A nucleotide BLAST (NCBI) was performed to check the specificity of the primers. cDNA for quantitative real-time PCR was synthesized from the same human total RNA extracted from lymphocytes and used in microarray analysis. A 1:100 dilution of cDNA (5 μl) was added to each Q-PCR reaction containing SYBR Green PCR Master Mix (Applied Biosystems), and amplification was carried out on the 7000 Sequence Detection System (Applied Biosystems). A standard curve used to quantitate the products was generated from dilutions of known concentrations of genomic DNA. All samples and standards were run in triplicate. The average values were used for t tests to compare control and schizophrenia copy number before and after correction for a housekeeping gene (SLC9A1). The overall results did not change after normalization.

Results

Genome-wide scan with microsatellite and SNP markers

First, a low-resolution genome-wide linkage analysis was carried out with ∼329 RFLP and microsatellite markers. The multiplex pedigree (Table 1, Fig. 1) showed a suggestive linkage peak in the 4q region (D4S1530, LOD 2.17, θ=0) using a dominant model.
Table 1

Demographics for each subject used in microarray gene expression, Q-PCR, and SNP genotype studies

Group

Age (years)

Gender

Age of onset

RNA 28S/18S

Affymetrix microarray U133

Present calls

3′/5′ GAPDH

3′/5′ β-actin

Schizophrenia

 A

60

F

32

1.9

45.4

0.94

1.13

 A

48

F

18

2.0

43.5

0.94

1.05

 A

42

M

19

1.3

46.0

0.92

1.15

 A

24

M

20

1.2

46.7

1.02

1.35

 A

18

M

17

1.4

45.4

0.95

1.21

 Average

38.4

2F/3M

21.2

1.6

45.4

1.0

1.2

 Standard deviation

17.3

 

6.1

0.3

1.2

0.04

0.1

Control

 U

22

F

NA

1.6

43.8

1.68

2.73

 U

21

F

NA

1.4

45.3

1.03

1.34

 U

16

M

NA

1.4

46.2

1.02

1.39

 U

48

M

NA

2.1

40.7

1.87

4.11

 U

34

F

NA

1.4

43.5

0.95

1.18

 Average

28.2

3F/2M

 

1.6

43.9

1.3

2.2

 Standard deviation

12.9

  

0.3

2.1

0.4

1.3

Schizophrenia–control (t test P value)

0.3

  

0.9

0.2

0.1

0.2

Haplotype analyses using flanking microsatellite markers delineated a 14 Mb region that cosegregated with all schizophrenic cases. In order to determine whether other linkage signals may have been missed with the initial low-resolution scan, we carried out a higher-resolution SNP scan by microarray similar to another linkage study that also used both microsatellite markers and SNP microarray (Middleton et al. 2004). A linkage peak from 171.2–185.1 Mb (NCBI physical map) was mapped with a multipoint maximum LOD of 2.39 with the Affymetrix 10 K SNP markers (Fig. 2). The SNP linkage results showed a haplotype that segregated with the disease in this pedigree (Table 2). The linkage analysis was conducted with both MERLIN and dCHIP Linkage, and the results agreed for the region on 4q that showed suggestive linkage. MERLIN linkage using a dominant parametric model showed a maximum LOD=2.56. To correct the SNP linkage data for linkage disequilibrium (LD), we reanalyzed the SNP linkage markers with several cluster options using MERLIN; however, there was a recombination in the region that prevented correction for LD in this pedigree.
https://static-content.springer.com/image/art%3A10.1007%2Fs00439-006-0172-7/MediaObjects/439_2006_172_Fig2_HTML.gif
Fig. 2

Linkage analysis of schizophrenia for Affymetrix 10 K SNP markers is shown for all chromosomes (a). The maximum LOD peak was 2.39 (θ=0) at 4q33.1–34.3 assuming a dominant model. Genotypes within the highest LOD peak are enlarged (U unaffected, A affected) (b). The shaded area represents an LOD interval of (−15 min to 3 max), and the red line threshold is LOD=0. The color codes for each subject shown depict genotype: red box = A, blue box = B, yellow box = AB, white box = no call

Table 2

Linkage of SNP genotype markers to schizophrenia in pedigree shown in Fig. 1 using dChip Linkage. Similar linkage results were obtained with MERLIN parametric dominant linkage testing

CHR

NCBI

cM

LOD

dbSNP RS ID

4

167.973

161.3

−0.3

rs1246507

4

169.018

162

−0.31

rs2319145

4

170.007

162.9

−0.3

rs497179

4

170.099

163.1

−0.3

rs1387330

4

170.534

163.7

−0.31

rs1318822

4

171.983

165.3

−0.31

rs1403225

4

172.293

165.6

−0.3

rs402841

4

172.945

166.2

0.36

rs2332211

4

172.963

166.2

0.36

rs958355

4

172.963

166.2

0.36

rs958357

4

172.963

166.2

0.36

rs1038813

4

173.079

166.3

0.41

rs952008

4

173.171

166.4

0.46

rs3907555

4

173.625

166.9

0.64

SNP_A-1508870

4

173.625

166.9

0.64

SNP_A-1509404

4

173.629

166.9

0.64

rs718913

4

173.968

167.3

0.74

rs1812424

4

174.091

167.4

1.54

rs2200257

4

174.511

167.6

1.96

rs1347703

4

174.587

167.7

2.08

rs1370576

4

174.587

167.7

2.08

rs1370577

4

174.893

167.9

2.25

rs3905287

4

175.367

168.1

2.37

rs2119788

4

176.767

170

2.38

rs985261

4

177.437

170.3

2.39

rs2333244

4

177.454

170.3

2.39

rs727917

4

178.123

171.4

2.39

rs723818

4

178.123

171.4

2.39

rs723819

4

178.123

171.4

2.39

rs723820

4

178.262

171.6

2.39

rs309753

4

178.326

171.7

2.39

rs404409

4

178.326

171.7

2.39

rs448261

4

178.343

171.7

2.39

rs309773

4

178.349

171.7

2.39

rs1375749

4

178.516

171.9

2.39

rs1485765

4

179.029

172.5

2.39

rs1821970

4

179.029

172.5

2.39

rs1371231

4

179.029

172.5

2.39

rs722832

4

179.206

172.8

2.39

rs1395477

4

179.575

173.2

2.39

rs1902018

4

179.589

173.2

2.39

rs1454214

4

179.858

173.7

2.39

rs966108

4

180.001

174

2.39

rs722387

4

180.101

174.1

2.39

rs1397413

4

180.504

174.9

2.39

rs727353

4

180.569

175

2.39

rs720096

4

180.574

175

2.39

rs1368870

4

180.574

175

2.39

rs1368871

4

180.9

175.6

2.39

rs4103290

4

181.438

176.6

2.39

rs1455691

4

181.68

177.2

2.39

rs1379996

4

181.978

177.8

2.39

rs1112857

4

181.978

177.8

2.39

rs6815286

4

181.996

177.8

2.39

rs1343863

4

182.35

178.4

2.38

rs2309248

4

182.859

179.8

2.36

rs1969398

4

182.86

179.8

2.36

rs1879908

4

183.346

180.6

2.35

rs724658

4

183.346

180.6

2.35

rs724659

4

183.638

181

2.14

rs726856

4

183.799

181.3

1.85

rs1914369

4

183.882

181.6

0.69

SNP_A-1519703

4

184.275

183.1

−0.43

rs955638

4

185.301

186

−2.36

rs1113122

4

185.301

186

−2.36

rs868082

4

186.462

190.3

−3.26

rs1077767

4

187.227

192.9

−4.28

rs2310160

4

187.583

193.6

−4.39

rs726466

4

187.888

194.9

−4.63

rs1378149

Overall, the region on chromosome 4q implicated by both microsatellite 171.2–185.1 Mb and SNP linkage 171.8–184.6 analyses was consistent. Recombination inferred by both microsatellite marker and SNP methods defined the haplotype interval. The SNP haplotype that segregated with the affected in the 4q33–35.1 region is shown (Fig. 2).

Differentially expressed genes lymphocytes

In the latest build of Unigene (March 2006) there are approximately 125 Unigene clusters in the 4q33–35.1 region. The Affymetrix U133A GeneChip chip definition file incorporated 12 known genes within the region. The expression results were condensed using RMA and evaluated for statistical significance with a simple unmoderated t test. There were 1,340 genes significantly different (P<0.05) between the affected and unaffected groups. Two genes were differentially expressed, AGA and GALNT7, and were both within the chromosome 4q33–35.1 region.

After ANCOVA adjustment of expression means for RNA quality and age, the number of significantly differentially expressed genes was reduced to 451. We are using a custom chip definition file, so that only 11,282 Unigene clusters are represented on the U133A chips after a stringent BLAST of each Affymetrix designed probe (Dai et al. 2005). The significant 451 genes did not survive correction for multiple testing. Three genes in the chromosome 4q haplotype showed a trend for differential expression (AGA; GALNT7; vascular endothelial growth factor C VEGFC; P≤0.1) after ANCOVA adjustment for age and microarray chip quality (3′/5′ GAPDH ratio) as covariates.

Differentially expressed genes postmortem dorsolateral prefrontal cortex

A microarray study of DLPFC from the Stanley Foundation (Table 3) used 34 subjects with bipolar disorder, 36 subjects with schizophrenia, and 36 controls was run on the Codelink 20 K platform (GE Health). The 88 highest quality microarray samples were used for an analysis of covariance using group and gender as main factors, and age and pH as continuous covariates. The expression of all genes on the microarray platforms within the 4q33–35.1 region for both brain and lymphocytes is shown (Table 4). There were three genes differentially expressed in DLPFC in schizophrenia in the 4q33–35.1 region: AGA, HMGB2 (high-mobility group box 2), and SCRG1 (Scrapie responsive protein 1). AGA was differentially expressed in DLPFC and lymphocyte in schizophrenia, although in lymphocyte the expression was not significant after ANCOVA.
Table 3

Demographics of postmortem Stanley microarray dorsolateral prefrontal cortex samples

 Demographics

Group

Bipolar

Control

Schizophrenia

Mean

SD

Mean

SD

Mean

SD

 Age

45.4

10.7

43.3

7.4

42.6

8.4

 PMI

37.8

18.4

29.6

12.5

29.9

14.8

 pH

6.45

0.25

6.59

0.27

6.46

0.24

Race

 White

32

36

35

 Black

1

0

0

 Native american

1

0

0

 Hispanic

 0

0

1

Gender

 Female

17

11

9

 Male

17

25

27

Total subjects

34

36

36

After processing all microarrays, the highest quality microarrays (n=88) were used to conduct analysis of covariance using group and gender as main factors, with age and pH used as covariates. The results for all genes in the chr 4q33–35.1 region are shown in Table 4

Table 4

Comparison of lymphocyte and DLPFC (Stanley Foundation) gene expression for 4q33–35.1

Lymphocyte U133A chip

DLPFC CodeLink 20 K

UGCluster

P (schizoprenia vs. control) ANCOVA

Fold change (schizoprenia vs. control)

Symbol

Cytoband

P (schizoprenia vs. control)

Fold change (schizoprenia vs. Control)

P value (bipolar vs. control)

Fold change (bipolar vs. control)

Hs.126838

0.840

0.989

ADAM29

4q34

NA

NA

NA

NA

Hs.207776

0.100

0.851

AGA

4q32–q33

0.046

1.263

0.274

1.135

Hs.127407

0.170

1.394

GALNT7

4q31.1

0.890

0.986

0.744

1.033

Hs.413099

0.879

1.002

GLRA3

4q33–q34

0.991

0.999

0.766

1.031

Hs.75819

0.798

0.958

GPM6A

4q34

0.493

0.844

0.739

0.921

Hs.75819

0.798

0.958

GPM6A

4q34

0.546

0.947

0.637

0.959

Hs.75819

0.798

0.958

GPM6A

4q34

0.889

1.008

0.068

0.895

Hs.388245

0.763

1.036

HAND2

4q33

0.205

0.898

0.176

0.891

Hs.388245

0.763

1.036

HAND2

4q33

0.437

0.857

0.594

0.902

Hs.434953

0.771

0.927

HMGB2

4q31

0.005

1.346

0.005

1.344

Hs.77348

0.347

0.979

HPGD

4q34–q35

0.066

1.619

0.166

1.444

Hs.77348

0.347

0.979

HPGD

4q34–q35

0.976

0.997

0.408

0.917

NA

NA

NA

MORF4

4q34.1

0.823

1.024

0.007

0.743

Hs.413835

0.700

1.076

SAP30

4q34.1

0.652

1.034

0.702

0.972

Hs.7122

0.165

0.950

SCRG1

4q31–q32

0.021

1.260

0.571

1.057

Hs.42194

0.823

0.974

SPCS3

4q34.2

NA

NA

NA

NA

Hs.435215

0.074

1.120

VEGFC

4q34.1–q34.3

0.790

1.064

0.306

1.268

Both datasets were analyzed with an analysis of covariance to adjust for age. There were multiple probes for the same gene on Codelink, so GPM6A, HAND2, and HPGD results are shown. There were three genes differentially expressed in DLPFC in schizophrenia are shown in bold: AGA aspartylglucosaminidase, HMGB2 high-mobility group box 2, and SCRG1 Scrapie responsive protein 1. AGA was differentially expressed in DLPFC and lymphocyte in schizophrenia, although in lymphocyte the expression was not significant after ANCOVA

NA gene not on array platform

Linkage analysis of microarray gene expression

The quantitative gene expression values were used for calculation of genome-wide linkage with a variance components analysis in MERLIN (Table 5).
Table 5

Two genes within the haplotype that showed differential expression were analyzed for linkage to genetic markers on a genome-wide scan. GALNT7 showed linkage to markers outside the chromosome 4q33–35.1 region (LOD=3.15), and AGA showed suggestive linkage to markers within the haplotype region (LOD=2.37). Bold LOD scores >2.3 are shown for GALNT7 and AGA

Gene symbol

Gene physical position (Mb)

Genetic marker (microsatellite or SNP)

Marker physical position

LOD

P value

Cis-regulatory position

GALNT7

174,785,845

rs538307

4:83,182,269

1.51

0.004

91,603,576

GALNT7

174,785,845

rs1103021

4:117,447,717

2.95

0.00011

57,338,128

GALNT7

174,785,845

rs724749

4:126,184,175

2.75

0.0002

48,601,670

GALNT7

174,785,845

rs951530

4:126,242,981

3.15

0.00007

48,542,864

GALNT7

174,785,845

rs724156

4:130,903,653

2.9

0.00013

43,882,192

GALNT7

174,785,845

rs1876054

4:133,497,030

2.87

0.00014

41,288,815

GALNT7

174,785,845

rs1378149

4:187,549,919

1.56

0.004

12,764,074

GALNT7

174,785,845

rs1280100

4:188,234,023

1.52

0.004

13,448,178

GALNT7

174,785,845

rs952178

4:189,221,787

1.55

0.004

14,435,942

GALNT7

174,785,845

D16S83

16:1,054,534

2.3

0.0004

GALNT7

174,785,845

UT581

16:3,310,331

2.01

0.001

AGA

179,047,814

rs1318822

4:170,460,379

2.03

0.0011

8,587,435

AGA

179,047,814

rs1403225

4:171,867,785

1.85

0.002

7,180,029

AGA

179,047,814

dbSNP_12a

4:173,507,703

1.85

0.002

5,540,111

AGA

179,047,814

dbSNP_13b

4:173,507,717

1.9

0.002

5,540,097

AGA

179,047,814

rs1812424

4:173,851,484

1.85

0.002

5,196,330

AGA

179,047,814

D4S2431

4:175,095,302

1.52

0.004

3,952,512

AGA

179,047,814

UT1950

4:181,860,360

2.37

0.0005

2,812,546

GALNT7 UDP-N-acetyl-alpha-D-galactosamine:polypeptide N-acetylgalactosaminyltransferase 7 (GalNAc-T7); Affymetrix probeset-U133A 218313_s_at

AGA aspartylglucosaminidase; Affymetrix probeset-U133A 204332_s_at

adbSNP_12 corresponds to Affymetrix SNP probe 1508870

bdbSNP_13 corresponds to Affymetrix SNP probe 1509404

Genome-wide linkage of the two differentially expressed genes in the haplotype (set 1) showed a maximum LOD score of 2.37 for AGA to a single marker (UT1950) in the haplotype (Table 5). Linkage for AGA was computed with SNP markers and AGA showed evidence of cis-regulatory influence. Figure 3 demonstrates the association between AGA expression and SNP rs723819 genotypes (177.6 Mb) in the unaffected (genotype 1) compared to the affected (genotypes 2 and 3). However, AGA also showed evidence of trans-regulation (Table 5) on chromosome 4 with an LOD of 2.03 to SNP rs1318822 located 8.9 Mb upstream of the gene.
https://static-content.springer.com/image/art%3A10.1007%2Fs00439-006-0172-7/MediaObjects/439_2006_172_Fig3_HTML.gif
Fig. 3

The AGA microarray gene expression is plotted by genotype for SNP rs723819 (1 homozygous for the major allele; 2 heterozygous; 3 homozygous for the minor allele). There was no overlap in genotype or microarray expression

The second differentially expressed gene, GALNT7, displayed evidence of suggestive linkage to an area near the 4q33–35.1 region, and evidence of linkage (LOD=3.15) to a regulatory control locus at SNP marker rs951530 (chromosome 4:126,141,720). GALNT7 also showed suggestive linkage to chromosome 16 (Table 5). Thus, AGA was the only gene within the schizophrenia haplotype region that showed linkage (LOD=2.37) to a potential cis-regulator.

To discover additional gene expression traits that might be coregulated within the linkage region, a set of differentially expressed genes was used for a similar linkage analysis. The top 200 differentially expressed genes (set 2) with the highest coefficients of variation across all samples were selected (Cheung et al. 2003; Cheung and Spielman 2002). This second subset was tested for linkage to chromosome 4q33–35.1 markers. Linkage results for the top 200 dysregulated genes to markers within the chromosome 4q haplotype showed one gene, alcohol dehydrogenase IB (ADH1B), with an LOD of 2.27 (Table 6). However, another trans-regulatory locus for ADH1B showed an LOD score of 3.08 (rs719880 located at chr4: 162,504,718). We cannot presently resolve whether these regulatory loci on chromosome 4 at 162 and 177 Mb represent the same potential control locus or these are two different regulatory loci. There was no overlap in expression between affected and unaffected individuals for gene expression of ADH1B (Fig. 4).
Table 6

Two hundred differentially expressed genes were scanned for linkage to markers within the haplotype. ADH1B showed evidence of a strong trans-regulatory loci at 162 Mb with a maximum LOD=3.08. ADH1B also showed evidence of suggestive linkage within the 4q33–35.1 region shown in bold in physical position

Gene symbol

Chromosome 4 NCBI (Mb)

Genetic marker (microsatellite or SNP)

Physical position

LOD

P value

Cis-regulatory position

ADH1B

100,537,000

FGA

155,800,000

2.7

0.0002

55,263,000

ADH1B

100,537,000

rs1343937

156,388,728

2.84

0.0002

55,851,728

ADH1B

100,537,000

rs719880

162,724,739

3.08

0.00008

62,187,739

ADH1B

100,537,000

rs355191

167,251,943

2.07

0.001

66,714,943

ADH1B

100,537,000

rs191644

167,252,139

2.07

0.001

66,715,139

ADH1B

100,537,000

rs2319145

168,956,566

2.11

0.0009

68,419,566

ADH1B

100,537,000

rs2333244

177,246,433

2.27

0.0006

76,709,433

ADH1B

100,537,000

rs1368871

180,251,308

2.26

0.0006

79,714,308

ADH1B

100,537,000

rs1368870

180,251,647

2.26

0.0006

79,714,647

ADH1B alcohol dehydrogenase 1B (class I), beta polypeptide; Affymetrix probeset-U133A 209614_at

https://static-content.springer.com/image/art%3A10.1007%2Fs00439-006-0172-7/MediaObjects/439_2006_172_Fig4_HTML.gif
Fig. 4

The expression levels of AGA and ADH1B genes did not overlap between the affected and the unaffected, and showed linkage to chromosome 4 markers in the schizophrenia locus. AGA is located within the putative schizophrenia locus on 4q and showed evidence of cis-regulation

In the third set, the remaining 1,327 differentially expressed probesets from the RMA t test genome wide were analyzed and produced three gene probesets showing suggestive linkage (LOD>2.3) to SNPs on chromosome 4 within the haplotype. The three genes that showed linkage to SNPs within the chromosome 4 haplotype are: ZNF291, CSNK1A1, and MSCP (Table 7). However these genes did not show differential expression following ANCOVA adjustment. A complete analysis of all gene expression linkages to SNP genotype markers would require ∼5.2×108 calculations, the main focus for this paper was in the schizophrenia haplotype region.
Table 7

The remaining differential expressed probesets in schizophrenia compared to the unaffected were also analyzed for linkage to the chromosome 4 haplotype. There were no genes that showed above an LOD score >2.5

Gene

Cytogenetic band

Affymetrix probeset

dbSNP ID

Kong and Cox LODa

P value

SNP location

ZNF291

15q24

215848_at

DBSNP_13b

2.32

0.0005

172.8

CSNK1A1

5q32

208867_s_at

RS2333244

2.42

0.0004

176.4

MSCP

8p21

218978_s_at

RS1902018

2.48

0.0004

178.5

MSCP

8p21

218978_s_at

RS1454214

2.46

0.0004

178.5

aCalculated in MERLIN linkage (Abecasis et al. 2002)

bD4S1308 /// D4S2328. dbSNP_13 corresponds to Affymetrix SNP probe 1509404

Q-PCR validation of differentially expressed genes

Genes with nominal differential expression between unaffected and affected family members were tested by real-time Q-PCR with SybrGreen. The Q-PCR fold changes for schizophrenia compared to controls were larger than Q-PCR fold-changes that were observed with microarray (Table 8). For example, ADH1B was validated by Q-PCR (Table 8) and showed a 9.5-fold decrease in the affected versus unaffected family members. However, AGA and GALNT7 displaying robust fold changes consistent with microarray did not reach statistical significance. The overall validation of candidate genes tested in this study was 7 out of 14 genes tested indicating about 50% validation rate of microarray data.
Table 8

Q-PCR of genes that showed differential microarray expression

Gene symbol

Microarray fold change

Q-PCR fold change

Microarray (P value t test)

Q-PCR (P value t test)

Cytogenetic band

Gene name

ADAM9

1.45

1.70

0.024

0.04

8p11.22

A disintegrin and metalloproteinase domain 9 (meltrin gamma)

ADH1B

0.89

0.10

0.023

0.02

4q21–q23

Alcohol dehydrogenase IB (class I), beta polypeptide

AGA

0.88

0.77

0.022

0.42

  

CTLA4

0.82

0.14

0.036

0.04

2q33

Cytotoxic T-lymphocyte-associated protein 4

FLJ20637

0.81

0.68

0.014

0.08

4q22.1

Hect domain and RLD 6

G3BP2

1.28

2.04

0.045

0.07

4q21.21

Ras-GTPase activating protein SH3 domain-binding protein 2

GALNT7

1.34

2.81

0.041

0.46

4q34.1

UDP-N-acetyl-alpha-D-galactosamine:polypeptide N-acetylgalactosaminyltransferase 7 (GalNAc-T7)

HCK

2.58

4.71

0.030

0.02

20q11–q12

Hemopoietic cell kinase

HERC3

0.89

0.90

0.023

0.34

4q21

Hect domain and RLD 3

IGJ

0.45

0.63

0.038

0.21

4q21

Immunoglobulin J polypeptide, linker protein for immunoglobulin α and μ polypeptides

NR2F2

1.74

4.15

0.047

0.02

15q26.2

Nuclear receptor subfamily 2, group F, member 2

NS3TP2

1.27

1.75

0.030

0.08

5q23.3

HCV NS3-transactivated protein 2

SMARCA2

1.75

2.27

0.028

0.03

9p22.3

SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 2

SNX2

1.43

0.95

0.042

0.45

5q23

Sorting nexin 2

TMOD3

1.21

1.39

0.031

0.12

15q21.1–q21.2

Tropomodulin 3 (ubiquitous)

WSB2

1.49

2.18

0.003

0.02

12q24.23

WD repeat and SOCS box containing protein 2

ADH1B showed a significant difference between the affected and unaffected by Q-PCR. ADH1B was confirmed by Q-PCR and showed an LOD score of 3.0 for regulatory control locus just outside of the haplotype and a secondary locus inside of the haplotype. Of 14 candidate genes tested 7 were confirmed (shown in bold) indicating a 50% validation rate of the microarray data, an additional three genes showed a trend at P value <0.1 for validation

Mutation screening

Direct sequencing of all coding regions, exon–intron junctions and possible regulatory regions for FLJ22649, HAND2, and AGA was performed on the five affected and five unaffected family members. There was no potential disease predisposing mutation found in these genes that segregated with schizophrenia in this pedigree. Nine previously published exonic SNPs were genotyped in the same sample set. There was no clear allelic segregation detected based upon illness.

Discussion

This study used a combined approach of linkage to identify a schizophrenia susceptibility region and gene expression to identify candidate genes within the linkage region. The genome-wide scan with microsatellite markers and SNPs suggested linkage of schizophrenia to chromosome 4q33–35.1 in a high-density multiplex pedigree. The pedigree was followed up with microarray screening of gene expression. We were not able to demonstrate a significant linkage between the putative haplotype for disease and for gene expression. This may be due to a suggestive but nonsignificant LOD score generated for schizophrenia in this single pedigree. Thus, we emphasize that these results indicate the methods of using gene expression and linkage with a disease to attempt to pinpoint regulatory and possibly causative loci.

We found that ADH1B differential gene expression is validated by Q-PCR, and showed a significant LOD score to a regulatory locus. After considering all 1,327 nominally significant gene expression differences for linkage to chromosome 4, both ADH1B and GALNT7 showed significant LOD scores for linkage to chromosome 4 (LOD>3.0). Interestingly, both genes showed suggestive linkage of a smaller magnitude to the haplotype region (LOD>2.0). We focused our search on chromosome 4q haplotype, as an exhaustive analysis of all gene expression and SNP markers would require ∼5×108 statistical tests. However, with additional subjects it would be worthwhile to pursue a genome-wide scan of gene expression regulatory loci.

Gene expression in peripheral white blood cells has been tested in schizophrenia, bipolar, and controls (Tsuang et al. 2005; Middleton et al. 2005). Linkage regions in schizophrenia and bipolar disorder have been further scanned for changes in gene expression of peripheral white blood cells (Middleton et al. 2005). We have also found that three genes within the 4q33–35.1 region are differentially expressed in DLPFC (AGA, SCRG1, and HMGB2). Thus, another potential use for this type of investigation is for determining whether the lymphocyte expression differences serve as potential biomarkers, especially if the gene is also expressed in the brain.

Several genetic studies of schizophrenia have reported positive linkage findings on chromosome 4q (Hovatta et al. 1999; Paunio et al. 2001; Levinson et al. 1998; Mowry et al. 2000; Kennedy and Macciardi 1998; Ekelund et al. 2000; Kaufmann et al. 1998; Table 9). Evidence of suggestive linkage was reported to the same 4q region as the present study (Straub et al. 2002; Fallin et al. 2003). However, in a meta-analysis of schizophrenia linkage studies, chromosome 4q was not significant (Lewis et al. 2003); thus, the 4q region may only account for a small fraction of disease liability.
Table 9

Summary of genetic studies in schizophrenia reporting findings on chromosome 4q

Marker

NCBI (Mb)

LOD (Zmax)

Reference

D4S2361

85

1.3

Paunio et al. (2001)

D4S35

91–96

>1.0

Coon et al. (1994)

D4S2623

111

1.36

Levinson et al. (1998)

D4S2623

111

1.43

Mowry et al. (2000)

4q22–31

111

3.0

Gurling et al. (2001)

D4S2917

109

2.64

Reported in Kennedy and Macciardi (1998)

D4S2394

130

1.04

Paunio et al. (2001)

D4S2394

130

0.76, 1.40, 1.19

Ekelund et al. (2000)

D4S2639

185

0.62, 0.83, 1.13

Stages I, II, and III

D4S1564

118.96

(1.54)

Kaufmann et al. (1998)

D4S2395

134.97

(1.68)

 

D4S1644

151.48

(1.54)

 

D4S424

142

(0.34)

Hovatta et al. (1999)

D4S1586

158

(2.74)

 

D4S3046

163

(0.29)

 

D4S3250–D4S3335

124–185

1.84

Straub et al. (2002)

D4S1625

144

1.67

 

D4S2417

147

1.50

 

D4S2631

180.8

1.2

 

D4S3335

156.3

1.2

 

185

0.96

 

D4S1535

185

2.3

Fallin et al. (2003)

D4S415

179

  

D4S1539

176

  

We have used Q-PCR to validate the microarray data. However, we were unable to validate two candidate genes in the 4q33–35.1 region. In reanalysis of the data using unequal variance, the t test for both genes became nonsignificant. Thus, the Q-PCR failure was due to high variation in samples. However, we were able to validate 7 out of 14 genes tested by Q-PCR.

ADH1B might be a candidate gene based upon gene expression differences in schizophrenia and linkage to regulatory loci within the schizophrenia haplotype and outside of the haplotype. ADH1B metabolizes substrates in pathways involving serotonin, norepinephrine, dopamine, and alcohol (Consalvi et al. 1986; Helander et al. 1994; Svensson et al. 1999; Matsuo and Yokoyama 1989; Matsuo et al. 1989) and thus is relevant to psychiatric disorders. Although we did not find a predisposing mutation in any exons of AGA for schizophrenia, there are known mutations of AGA that cause the lysosomal storage disease, Aspartylglucosaminuria (OMIM# 208400). Aspartylglucosaminuria is characterized by predominant cognitive deterioration, mental retardation, and emotional lability.

Potential regions for control of gene expression were mapped to cis- and trans-regulatory sites. The results of the study are consistent with prior reports demonstrating that gene expression traits in lymphocytes are heritable in multiplex pedigrees (Morley et al. 2004; Monks et al. 2004; Schadt et al. 2003a, 2003b). These results suggest that studying gene expression traits in combination with linkage studies helps identify regulatory regions for candidate genes. A high LOD score between a quantitative trait (gene expression) and a genetic marker suggests that possible regulatory elements within the linkage region may regulate the quantitative trait (gene expression) (Schadt et al. 2003a). Schadt et al. (2003a, 2003b) reported that 423 genes were linked to a chromosome 2 locus of interest, which was a quantitative trait of subcutaneous fat deposition. Notably, only four genes were within 2 cM of the peak where gene expression traits showed linkage. “Most of the genes linked to the chromosome 2 locus do not physically reside on chromosome 2, and so, are at least partially regulated by one or more loci in the chromosome 2 hotspot region” (Schadt et al. 2003a). This reasoning can apply to the present results, as most genes with differential expression showed regulatory loci in a trans- location, while some genes also showed evidence of being partially regulated by two distinct loci.

The differentially expressed gene, such as ADH1B, located outside of the haplotype but linked to the haplotype may not confer risk to illness, but may modify gene expression. This modification in gene expression could confer a variation in symptoms, onset, or subtype in the schizophrenia syndrome. Consistent with the Schadt et al. (2003a, 2003b) study, we find differentially expressed candidate genes on other chromosomes with suggestive linkage to the haplotype (Table 5). Both cis- and trans- acting factors regulate genes, perhaps in combination in a complex disorder, as suggested by our results and others (Morley et al. 2004). Improvements in the analytical methods are needed to integrate differential gene expression as a phenotype and linkage to genomic markers (Schadt et al. 2003a, 2003b; Kraft et al. 2003; Sham et al. 2002; Horvath and Baur 2000; Middleton et al. 2005). However, taken together, linkage analysis of the inheritability of gene expression traits will be useful for mapping regulators of gene expression.

Acknowledgements

This work was supported by the William Lion Penzner Foundation (UCI Department of Psychiatry and Human Behavior) and the NIMH RMH074307A award (MPV). Postmortem brain tissue was donated by The Stanley Medical Research Institute’s brain collection courtesy of Dr. Michael B. Knable, Dr. E. Fuller Torrey, Dr. Maree J. Webster, Dr. Serge Weis, and Dr. Robert H. Yolken.

Copyright information

© Springer-Verlag 2006