The table in the center shows the 32 (haplotype-)tagging single-nucleotide polymorphisms (SNPs) and the global simulated P values for haplotype association for the linkage disequilibrium (LD) blocks of the HapMap Utah residents with ancestry from northern and western Europe (CEU) and the northern Swedish (N-Sw) population. Boldface type is used for the SNPs that showed single-marker association. *Statistically significant P values are indicated in boldface type. At the left is the LD figure for the core haplotype (HapICE) region with 142 HapMap SNPs in the CEU population. At the right is the LD figure for the 32 selected SNPs in the northern Swedish population.
This table shows sliding-window association results. *In the analyses, only single-nucleotide polymorphisms (SNPs) are incorporated. The bold line separates the 57-kilobase associated region at the 3′ side of the core haplotype (HapICE). SNPs that showed single-marker association are in boldface type. †Statistically significant values are indicated in boldface type. ‡HapICEmarkers. §The 3 tagging SNPs selected using Tagger. ∥For this polymorphism, 8 alleles were found with the following lengths: 216, 218, 220, 222, 224, 226, 228, and 230 base pairs (bp). ¶For this polymorphism, 9 alleles were found with the following lengths: 273, 275, 277, 279, 281, 283, 285, 287, and 291 bp.
This figure is a detailed haplotype analysis of the 57-kilobase associated region (N-Sw [northern Swedish population] LD [linkage disequilibrium] Block 7). The alleles of rs4458836 that split the first and second haplotype are italicized (see text), and the 2 alleles that divide the haplotypes almost perfectly into those occurring more frequently in schizophrenic (SZ) patients and those occurring more frequently in control individuals are indicated in boldface type (see text). *Single-nucleotide polymorphisms (SNPs) that showed single-marker association are in boldface type. †Statistically significant values are indicated in boldface type.
Alaerts M, Ceulemans S, Forero D, Moens LN, De Zutter S, Heyrman L, Lenaerts A, Norrback K, De Rijk P, Nilsson L, Goossens D, Adolfsson R, Del-Favero J. Support for NRG1 as a Susceptibility Factor for Schizophrenia in a Northern Swedish Isolated Population. Arch Gen Psychiatry. 2009;66(8):828–837. doi:10.1001/archgenpsychiatry.2009.82
Neuregulin 1 (NRG1), a growth factor involved in neurodevelopment, myelination, neurotransmitter receptor expression, and synaptic plasticity, first joined the list of candidate genes for schizophrenia when a 7-marker haplotype at the 5′ end of the gene (HapICE) was shown to be associated with the disorder in the Icelandic population. Since then, more genetic and functional evidence has emerged, which supports a role for NRG1 in the development of schizophrenia.
To determine the contribution of NRG1 to susceptibility for schizophrenia in a northern Swedish isolated population.
Detailed linkage disequilibrium (LD)–based patient-control association study. This is the first study to type and analyze the 7 HapICE markers and a set of 32 HapMap tagging single-nucleotide polymorphisms (SNPs) that represents variants with a minor allele frequency of at least 1% and fully characterizes the LD structure of the 5′ part of NRG1.
Outpatient and inpatient hospitals.
A total of 486 unrelated patients with schizophrenia and 514 unrelated control individuals recruited from a northern Swedish isolated population.
Main Outcome Measures
Association between markers and disease.
Analysis of the HapICE markers showed the association of a 7-marker and 2-microsatellite haplotype, different from the haplotypes associated in the Icelandic population and overrepresented in northern Swedish control individuals. Subsequently, a more detailed analysis that included all 37 genotyped SNPs was performed by investigating haplotypic association, dependent and independent of LD block structure. We found significant association with 5 SNPs located in the second intron of NRG1 (.007 ≤ P ≤ .04). Also, 2-, 3-, and 4-SNP windows that comprise these SNPs were associated (P < 3 × 10−4). One protective haplotype (0% vs 1.8%; P <5 × 10−5) and 1 disease risk–causing haplotype (40.4% vs 34.9%, P = .02) were defined.
The NRG1 gene contributes to the susceptibility for schizophrenia in the northern Swedish population.
Neuregulin 1 (NRG1; OMIM 1142445, UniGene Hs.453951, NCBI GeneID 3084) was first identified as a susceptibility gene for schizophrenia in the Icelandic population by the use of a combined linkage and association approach.1 Fine mapping and haplotype analysis of a linkage peak on chromosome arm 8p detected in 33 extended families, followed by patient-control and family-based association analyses, identified a core haplotype (HapICE) that was significantly overrepresented in patients with schizophrenia compared with control individuals (15.4% vs 7.5%, 1-sided P < .001). This core at-risk haplotype is defined by 5 single-nucleotide polymorphisms (SNPs) (SNP8NRG221132, SNP8NRG221533, SNP8NRG241930, SNP8NRG243177, and SNP8NRG433E1006) and 2 microsatellite markers (478B14-848 and 420M9-1395). It covers 290 kilobases (kb) and contains the first 5′ exon of the type IV isoform and the first 5′ exon of the type II isoform GGF2 of NRG1 (exons E187 and E1006, respectively2). The NRG1 gene spans approximately 1.2 Mb, encodes approximately 30 different isoforms, and gives rise to 6 types of protein variants (types I-VI), classified based on their structural features.
The NRG1 protein is a pleiotropic growth factor that, through binding with its ERBB4 tyrosine kinase receptors, is involved in neuronal migration, synaptogenesis, synaptic plasticity, gliogenesis, neuron-glia communication, myelination, and neurotransmission in the central nervous system, mainly in glutamatergic and γ-aminobutyric acidergic signaling.3- 6 As such, it fits perfectly into the neurodevelopmental hypothesis7- 9 and the myelin dysfunction10 and glutamate dysfunction hypotheses11,12 for schizophrenia. In addition, results from NRG1 and Erbb4 receptor mutant mice1,13 and messenger RNA expression changes in the prefrontal cortex and hippocampus of patients with schizophrenia14,15 support NRG1 as a promising functional candidate gene for schizophrenia that deserves special attention. The genomic location of the gene also corresponds to a linkage region for schizophrenia (8p22-11, OMIM SCZD6) identified by multiple genomewide linkage scans and meta-analyses.16,17
Since the original report, there have been numerous studies that confirm an association between schizophrenia and NRG1 in several populations. Some research groups detected an association with HapICE,18- 20 some with other haplotypes or other variants at the 5′ end,21- 25 and some with other parts of the gene.26- 28 Two meta-analyses29,30 that combined P values of the haplotypes with the strongest reported association and 1 meta-analysis31 that focused on the frequently typed Icelandic markers provided evidence for a positive but weak association. As with most candidate genes for complex disorders, the strength of evidence for association of NRG1 with schizophrenia has weakened with time, and negative association results have also been reported.32- 34 Recently, genomewide association studies have become feasible; thus far, they are the most extensive and systematic approach for testing common variations across the entire human genome. At the time of this writing, 5 genomewide association studies had been reported for schizophrenia,35- 39 but not a single SNP exceeded the genomewide significance threshold of P < 5 × 10−8. With the focus on NRG1, in none of the studies were SNPs in or in the vicinity of the gene found among the top hits or even achieved the threshold of P < 1 × 10−5 of “moderately strong evidence for association” (set by the Wellcome Trust Case Control Consortium40). Sullivan and colleagues38 had a more detailed look at their results for 15 candidate genes, and 19 of the 293 genotyped SNPs in NRG1 showed a value of P < .05, of which the most significant was P = .0009 for rs16879809. This is not convincing evidence in light of a genomewide association study, but it gives some suggestive support for NRG1 as a candidate gene for schizophrenia.
Inconsistent gene-disease associations constitute a real problem in the current literature regarding complex disorders. They might point to a true lack of association, but they can also exist owing to insufficient power of the studies to detect risk variants with small effects, genetic heterogeneity, or low genetic variation coverage with a lack of linkage disequilibrium (LD) between the SNPs used in the study and the actual susceptibility allele. This last problem can and should be avoided by the typing of sufficient SNPs, which are carefully selected to comprehensively cover as much variation as possible in a given genomic region.
With this in mind, we conducted a detailed LD analysis of the HapICE region based on information provided by the International HapMap Project (http://www.hapmap.org). A set of 32 (haplotype-)tagging SNPs that fully characterizes the LD structure and variation content of this genomic region was selected. For replication purposes and consistency with other studies, we also included the 7 Icelandic markers in this study. All of the polymorphisms were genotyped and subsequently analyzed in a sample comprising 486 patients with schizophrenia and 514 matched control individuals recruited from a geographically isolated northern Swedish population.
The patient sample comprised 486 unrelated individuals (180 females and 306 males) who fulfilled the DSM-IV41 criteria for schizophrenia and who originated from a geographically isolated population living in the Västerbotten region in northern Sweden. The mean (SD) age at disease onset was 24.8 (7.3) years and at inclusion was 53.1 (15.1) years. The patients were initially identified through inpatient hospital registers, and all had on at least 1 occasion received a discharge diagnosis of schizophrenia. Ascertainment of the patients was performed between Janaury 1, 1992, and December 31, 2005. Clinical characterization was assessed by trained research nurses and research psychiatrists by the use of register data, data from psychiatric records, and semistructured interviews, namely, the Mini International Neuropsychiatric Interview,42 the Diagnostic Interview for Genetic Studies43 (http://zork.wustl.edu/nimh/home/m_DIGS2.0_Interview.html), the Family Interview for Genetic Studies (http://zork.wustl.edu/nimh/home/FIGSInterview.html), and the Schedules for Clinical Assessment in Neuropsychiatry,44 a method that proved to yield high reliability and validity of a DSM-IV diagnosis of schizophrenia.45 The final diagnosis was determined by the consensus of 2 research psychiatrists (1 of whom, R.A., is a coauthor of this article), and only patients for whom full consensus was reached were included in the study.
The control population consisted of 514 unrelated individuals (275 females and 239 males) with a mean (SD) age of 58.0 (13.0) years at inclusion. They originated from the same geographic area as the patients and were randomly selected from the Betula study.46,47 None of the control individuals were reported to have a diagnosis of schizophrenia based on studies of psychiatric records or interviews of the nurses.
All the participants were white, and none were of Finnish, Norwegian, or Lappish descent. All the participants gave written informed consent, and the study was approved by the regional medical ethical committees of the universities of Umeå and Antwerp.
The association sample was controlled for population stratification by the genotyping of 37 microsatellite markers via the use of standard genotyping and scoring methods. Statistical tests for population stratification were performed using the program Structure (http://pritch.bsd.uchicago.edu/structure.html). No population substructure was observed in the association sample (data not shown).
Genomic DNA was extracted from peripheral blood using standard methods. The polymerase chain reaction (PCR) was used to amplify all genomic regions of interest. Primer sequences are given in eTable 1 ,and detailed information on protocols is available on request.
The genotyping of all SNPs except SNP8NRG243177 and SNP8NRG433E1006 was performed using the MassARRAY iPLEX Gold technology (Sequenom Inc, San Diego, California), following the protocol provided by Sequenom (http://www.sequenom.com). The PCR and extension primers were designed using Assay Design 3.0 (Sequenom Inc). Analysis and scoring were performed using Typer 3.3 (Sequenom Inc).
Primer3 (http://frodo.wi.mit.edu/primer3/) was used to design the PCR primers for the other 2 SNPs. SNP8NRG243177 was sequenced using the BigDye Terminator v3.1 Cycle Sequencing Kit, according to the instructions of the manufacturer (Applied Biosystems Inc, Foster City, California). Sequencing reactions were run on an ABI 3730XL automated sequencer (Applied Biosystems Inc), and the resulting trace files were analyzed with novoSNP.48 Genotyping of SNP8NRG433E1006 was performed using a pyrosequencer (PSQ HS96; Biotage AB, Uppsala, Sweden), preceded by a nested PCR.
For microsatellite markers 478B14-848 and 420M9-1395, PCR amplifications were performed using fluorescent-labeled primers, and PCR products were sized using an ABI 3730XL sequencer (Applied Biosystems Inc). Genotypes were assigned and scored using TracI, an in-house–developed software program (http://www.vibgeneticservicefacility.be/index.htm?soft/traci.php). Two researchers (M.A. and S.D.Z.) manually scored the genotypes independently, with a correlation of 0.98. A randomly chosen one-tenth of the sample was typed in duplo for all markers, with an error rate of 0.0028.
The SNPs were selected from the HapMap Data Release No. 20 (Phase II) of January 2006. The central ethnicity, Utah origin (CEU) population was used as a reference population, and the LD measures D′ and r2 were calculated using Haploview (http://www.broad.mit.edu/mpg/haploview). Markers with a minor allele frequency of at least 1% were included, and blocks were defined based on confidence intervals as proposed by Gabriel et al.49 Haploview LD analysis of the HapICE region resulted in 7 distinct haplotype blocks that range from 7-kb 5′upstream of SNP8NRG221132 to microsatellite marker 420M9-1395 (chromosome 8: 31586151-31748954). All haplotypes with an estimated overall frequency of 1.5% or greater were considered in the analyses, and haplotype-tagging SNPs (htSNPs) were chosen to represent these haplotypes. Additional tagging SNPs (tSNPs) were selected by the use of the Tagger program (http://www.broad.mit.edu/mpg/tagger50) to perform pairwise tagging, with an r2 threshold of 0.8 and a logarithm of the odds threshold for multimarker tests of 3.0. After the genotyping of all of the polymorphisms, Haploview was used to calculate LD blocks specific to the northern Swedish population.
To calculate Hardy-Weinberg equilibrium and to investigate allelic and genotypic association for the 2 microsatellites and all of the SNPs, GENEPOP version 3.3 (http://genepop.curtin.edu.au) was used. Haplotype associations and haplotype frequencies for marker combinations were calculated using haplo.stats v1.2.1 (http://mayoresearch.mayo.edu/mayo/research/biostat/schaid.cfm). To avoid false-positive findings owing to multiple testing, empirical simulated P values were calculated by the use of a Markov chain method with 5000 dememorization steps and 50 batches of 1000 iterations for the single-marker allelic and genotypic tests and by the use of 5000 random permutations of patient and control labels for the P values of sliding-window analyses, global haplotype, and haplotype-specific P values. The level of significance for all statistical tests was defined as P ≤ .05.
The HapMap Generic Genome Browser (http://www.hapmap.org/cgi-perl/gbrowse) was used to view the genomic region around HapICE. The LD structure of the region was analyzed in the CEU population by the use of Haploview and resulted in 7 haplotype blocks (Figure 1). We selected a panel of 30 htSNPs that represented all haplotypes with a frequency greater than 1.5% in the blocks. Repeat sequences hampered an assay design for 1 htSNP in block 3 that represented a haplotype with a frequency of 4.9%. This haplotype then became indistinguishable from the second most common haplotype with a frequency of 22.6%, which augmented it to 27.5%. To perform additional analyses independent of LD block structure while catching as much variation as possible, we analyzed the HapICE region and previously chosen htSNPs by the use of the Tagger program. A set of 3 extra tSNPs was selected, which ensured that the final panel of 32 (h)tSNPs represents a maximal number of polymorphisms in the region (Figure 1).
We typed the 7 HapICE markers in the northern Swedish association sample for schizophrenia. Eight alleles were found for microsatellite 478B14-848 and 9 for microsatellite 420M9-1395. Alleles with lengths of 218 and 279 correspond with the Icelandic “0” alleles (associated alleles). None of the HapICE markers showed single-marker allelic or genotypic association (Table 1). With the analysis of 5-SNP, 2-microsatellite, and 7-marker haplotype association, no significant global P value was obtained (Table 2 and eTable 2). The haplotypes associated in the Icelandic population did not show significant haplotype-specific association in the northern Swedish population (P > .05); they were even more frequent in control individuals than in patients with schizophrenia. One 2-microsatellite haplotype (216-279) and one 7-marker haplotype (GTGTA-216-279) were significantly overrepresented in control individuals (14.5% vs 11.2%, P = .03 and 2.3% vs 1.1%, P = .03, respectively) (Table 2).
We performed single-marker allelic and genotypic association analyses in the northern Swedish association sample for schizophrenia. All polymorphisms were in Hardy-Weinberg equilibrium in the control population (P ≥ .01). Significant association with schizophrenia was observed for 5 SNPs located in the second intron of NRG1 (.007 ≤ P ≤ .04) (Table 1).
Subsequently, we performed haplotype association analyses, independent and dependent of LD block structure. For the independent analysis, we used sliding windows of 2-, 3- and 4-SNP haplotypes for all 37 genotyped SNPs. Significant associations were found with six 2-SNP, six 3-SNP, and seven 4-SNP windows (.0003 ≤ P ≤ .046) (Figure 2). All these associated windows contained at least 1 of the associated SNPs except for one 2-SNP window (Figure 2).
For the analyses dependent on LD blocks, we first calculated the LD block structure specific for the northern Swedish population by introducing the genotype data for the 32 (h)tSNPs from the 514 control individuals into the Haploview program. The LD structure from the northern Swedish population was different from that for the CEU population (Figure 1), so we performed a haplotype association analysis based on the haplotype blocks defined for each population (herein referred to as CEU LD blocks and N-Sw LD blocks). A significant difference in haplotype distribution between patients with schizophrenia and control individuals was observed for CEU LD blocks 5 and 6 (P = .005 and P = .001, respectively) and for N-Sw LD block 7 (P = .003) (Figure 1 and eTables 3 and 4). Each of the associated LD blocks also contained 1 or more of the associated SNPs.
A more detailed analysis of the haplotypes in sliding windows and LD blocks in the associated region resulted in a comprehensive explanation of the observed associations (Figure 3). Investigation of the complete 13-SNP N-Sw LD block 7 showed that the AATCGGGCGAGA haplotype occurs in 1.8% of control individuals but in no patients with schizophrenia (P < 5 × 10−5). Another haplotype, AATCGGGCGAGA, which differed only in the allele for rs4458836, was also overrepresented in control individuals, so analysis shows that the haplotype AAT-G/T-CGGGCGAGA occurs in 14% of control individuals compared with 10.8% of patients (P = .02). On this haplotype, the associated microsatellite haplotype 216-279 (AAT-G/T-CGG-216-GCGAGA-279) is found.
The 2 other SNPs that showed single-marker association, rs7014221 and rs7014410, divide the haplotypes almost perfectly into those occurring more frequently in patients and those that occur more frequently in control individuals (Figure 3), which causes their 2-SNP haplotype (CEU LD block 6) to be significantly associated (P = .001), with the TA haplotype overrepresented in patients (76.9% vs 71.1%, P = .003) and the CG haplotype overrepresented in control individuals (28.4% vs 23.0%, P = .007).
Haplotype analysis of the second to the eighth SNP of N-Sw LD block 7 reveals significant global association (P = .003), and a GAGTAGGC haplotype shows up that is significantly overrepresented in patients with schizophrenia (40.4% vs 34.9%, P = .02).
The NRG1 gene is reported to be 1 of the best-replicated candidate genes for schizophrenia located in the vicinity of replicated linkage loci for schizophrenia. In addition to association and linkage findings, NRG1 and Erbb4 receptor mutant mice showing a behavioral phenotype that overlaps with mouse models for schizophrenia1,13 and messenger RNA expression changes in the prefrontal cortex and hippocampus of patients with schizophrenia14,15 constitute functional evidence of the involvement of NRG1 in the cause of the disorder. Along with that evidence, its functions in neurodevelopment, myelination, regulation of neurotransmitter receptor expression, and synaptic plasticity strongly support NRG1 as a susceptibility gene for schizophrenia.4
However, even this best-replicated candidate gene has not been immune to the typical problem of inconsistencies in genetic association findings. The original association has been replicated, but the strength of association has weakened in subsequent studies, and negative association results have been reported.32- 34 Because schizophrenia is a complex disorder, genetic heterogeneity might partly explain these findings. It is possible that NRG1 is a risk factor in some populations whereas it is not in others. Yet, insufficient genetic variation coverage by studying only a few polymorphisms in the gene can also produce such inconsistencies. The fact that association with schizophrenia has been demonstrated with a variety of haplotypes located throughout NRG1 clearly shows that in different populations, susceptibility alleles might be located on different haplotypic backgrounds. Also, differences in LD structure between populations can cause different SNPs to be in LD with the true susceptibility allele. Therefore, it is important in an association study to type many SNPs, which are carefully selected to comprehensively cover as much variation as possible in a given genomic region, even for replication of a previous finding. In this study, we decided to type the 7 HapICE markers and a set of 32 (h)tSNPs, selected to fully characterize the LD structure and variation content of the HapICE genomic region, in a northern Swedish association sample consisting of 486 patients with schizophrenia and 514 control individuals from a geographically isolated population. People from an isolated population are genetically more homogeneous than are individuals from an outbred population, so the disease under study might have a more homogeneous genetic background, and risk variants could be found with higher frequency, which increases the power to detect risk genes in an association study.
In a HapICE “replication” analysis, we investigated whether the 7 Icelandic markers also index risk of schizophrenia in the northern Swedish population. We did not replicate the finding of Stefansson and colleagues1 in the strict sense of the word, by which we mean that we did not detect association with the same HapICE or any of the single markers (Table 2). HapICE and the single strongest-associated SNP8NRG221533 C allele were even more frequent in northern Swedish control individuals. The 2 HapICE SNPs with possible functional consequences—SNP8NRG433E1006, a nonsynonymous SNP in type II exon E1006 (bp position 31616782-31617787), and SNP8NRG243177 5′ of type IV exon E187 (bp position 31616303-31616489) in the promoter region of a type IV isoform—did not show association, so they are not likely to play a role in the development of schizophrenia in the northern Swedish population. Also, no significant global haplotype association was found in an analysis of 5-SNP, 2-microsatellite, and 7-marker haplotypes (Table 2 and eTable 4), so there is no significant difference in global haplotype frequency distribution between patients with schizophrenia and control individuals. With the investigation of haplotype-specific P values, one 2-microsatellite haplotype (216-279) and one 7-marker haplotype (GTGTA-216-279) were significantly overrepresented in control individuals (Table 2), which indicates that variation in NRG1 could also have a protective effect against schizophrenia. We also found that SNP8NRG221533 should not be typed as the only SNP to represent the other HapICE SNPs (performed in different studies) because then several low-frequency haplotypes (1%-5%) are missed (eTable 4). This would have been the case in the northern Swedish population for the associated haplotype GTGTA-216-279. Although statistically significant, these results are not very compelling, especially because the global haplotype P values were greater than .3.
Because we carefully selected more polymorphisms based on LD, we were able to analyze the 5′ region of NRG1 in detail, and this allowed us to get more refined and comprehensive results, that produced more significant evidence for association. Only 3 other research groups have previously performed an LD-based association study with HapMap (h)tSNPs in the HapICE region. Thomson and coworkers28 analyzed the whole NRG1 gene region by the use of Haploview, with the inclusion of only SNPs with a minor allele frequency of at least 10% and 36 htSNPs that represented only haplotypes with a frequency greater than 10%. They performed a sliding-window analysis and found association in the 5′ and 3′ parts of the gene. Ikeda and colleagues34 selected 15 tSNPs (7 of these overlap with the present (h)tSNP panel) in the HapICE region, with the inclusion of only SNPs with a minor allele frequency of at least 5%, but did not find an association with schizophrenia in the Japanese population in a sliding-window analysis. Petryshen et al27 selected 18 htSNPs in the first 2 exons and introns of NRG1 such that haplotypes with a frequency greater than 2% were represented, but they did not detect association with those SNPs, although a haplotype that consisted of2 HapICE SNPs was overrepresented in their control individuals (15% vs 12%, P = .05). We selected 32 (h)tSNPs in the HapICE region that tag all SNPs with a minor allele frequency of at least 1%, and the htSNPs represent all haplotypes with a frequency greater than 1.5% in the LD blocks. We performed LD block–based haplotype analysis and sliding-window analysis that were not dependent on LD blocks so as to be robust to misdefinition of blocks.
For the haplotype analysis based on LD blocks, we first analyzed LD block structure specific for the northern Swedish population and found that it was different from that obtained for the CEU population (Figure 1), so we performed a haplotype association analysis based on the haplotype blocks defined for each population. The LD block structure can differ between populations. For the CEU population, all 142 SNPs in the region were used to define the blocks, whereas for the northern Swedish population, only the 32 genotyped (h)tSNPs were used. With the comparison of the larger LD pattern in the whole region, 2 identical major blocks are seen in the northern Swedish and CEU populations (thin black lines in Figure 1). The stronger and more long-range LD seen in the northern Swedish population is in line with what is theoretically expected for a young isolated population. The difference in LD block structure between the 2 populations does not mean that the (haplotype-)tagging SNPs selected in the CEU population do not cover the variation in the northern Swedish population. Recent studies that evaluate tSNP transferability proved the usefulness of the HapMap CEU population as a reference sample for tSNP selection for further investigations into genomic variation of northern European populations.51- 53 Smith and colleagues53 demonstrated that tSNPs selected based on HapMap CEU data capture more than 98% of the variants at an r2 > 0.8 in their studied family-based cohort of northern European descent. So, the present (h)tSNP panel is most likely suited to capture a maximum of known and unknown variations in the northern Swedish population.
Significant single-marker association with schizophrenia was observed for 5 SNPs located in the second intron of NRG1 (Table 1). Sliding-windows analyses of 2-, 3- and 4-SNP haplotypes that contain these SNPs also showed association (Figure 2). In the haplotype block–based analysis, significant global P values were observed for CEU LD blocks 5 and 6 and for northern Swedish LD block 7 (Figure 1), which each contained associated SNPs. Therefore, all the present analyses, independent and dependent on LD block structure, pointed to a 57-kb region at the 3′ side of HapICE that showed strong association with schizophrenia.
Subsequently, we performed a more detailed analysis of the haplotypes in that 57-kb region (Figure 3). All 13 SNPs show strong LD and form 1 LD block in the northern Swedish population (N-Sw LD block 7). The AATCGGGCGAGA haplotype occurs in 1.8% of control individuals but is absent in patients with schizophrenia. The haplotype AAT-G/T-CGGGCGAGA occurs in 14% of control individuals compared with 10.8% of patients. This result is in line with the previous findings: the AAT-G/T-CGGGCGAGA haplotype can be distinguished from all the other haplotypes by the minor allele of rs7017348, rs6468061, or rs17601950 (Figure 3), which each show single-marker association, and the associated microsatellite haplotype 216-279 is found on this SNP haplotype. A variant with a protective effect might have originated on the AATCGGGCGAGA haplotype (or in LD with it), which would cause it to occur now in approximately 2% of the control population. Another possibility is that the protective variant occurred on (or in LD with) the common AATCGGGCGAGA haplotype not long before the G allele of rs4458836 originated in such a way that some of the AATCGGGCGAGA haplotypes and all the AATCGGGCGAGA haplotypes now carry (or are in LD with) the protective variant.
Haplotype analysis of the second to the eighth SNP of N-Sw LD block 7 reveals haplotype GAGTAGGC, which is significantly overrepresented in patients with schizophrenia (40.4% vs 34.9%). This haplotype might carry (or be in LD with) a variant that confers risk of schizophrenia. Any of the 2 other SNPs that showed single-marker association, rs7014221 and rs7014410, divides the 13-SNP haplotypes almost perfectly into those that occur more frequently in patients and those that occur more frequently in control individuals (Figure 3), which causes their 2-SNP haplotype (CEU LD block 6) to be significantly associated with the TA haplotype overrepresented in patients (76.9% vs 71.1%) and the CG haplotype overrepresented in control individuals (28.4% vs 23.0%).
Based on these results, we can conclude that NRG1 plays a role in the etiology of schizophrenia in the northern Swedish population. Association was found with SNPs and haplotypes in a 57-kb region in the second intron of NRG1, downstream of exon E1006 and approximately 100 kb upstream of expressed sequence tag cluster Hs.97362.22 Variants in the associated region or in LD with it might have an effect on the expression of different NRG1 isoforms. In accordance with this hypothesis, the detection of protective and risk-conferring haplotypes might potentially be explained. If overexpression of an isoform augments the risk of schizophrenia, underexpression might render an individual less susceptible to the disorder or the other way around. The complex gene structure of NRG1 gives rise to the presumption that sophisticated spatial and temporal regulation of NRG1 isoform expression is necessary to fine-tune the protein function at different stages during the development of the nervous system and in different signaling strategies in the adult brain. This is supported by biological evidence, mainly from mutant mice (reviewed by Falls54). Any factor that disturbs this regulation might disturb normal functioning of the brain and be implicated in the etiology of schizophrenia. Increased type I NRG1 transcripts have been found in the dorsolateral prefrontal cortex14 and the hippocampus of patients with schizophrenia,15 and transcription of type IV NRG1, a brain-specific isoform of the gene, is upregulated by the T allele of SNP8NRG24317715,55 that has been associated with schizophrenia. Individuals with the TT genotype also showed reduced white matter density and structural connectivity,56 impaired frontal and temporal lobe activation,57 and reduced spatial working memory capacity,58 all features that possibly underlie schizophrenia pathology. In the northern Swedish population, however, this SNP was not associated, and the T allele was even more frequent in control individuals; therefore, another disease-causing mechanism is present.
The present data support NRG1 as a susceptibility factor for schizophrenia and should promote replication studies in different populations and with variants covering the whole gene. This study should encourage increased functional research into the different isoforms of NRG1, the effect of variants on their expression, and the relation with schizophrenia and other psychiatric disorders. We also hope that it will encourage more detailed and well-structured LD-based studies that investigate the association of candidate genes with complex disorders.
Correspondence: Jurgen Del-Favero, PhD, Applied Molecular Genomics Group, Department of Molecular Genetics, Flanders Institute for Biotechnology (VIB), University of Antwerp–Campus Drie Eiken, Universiteitsplein 1, B-2610 Antwerpen, Belgium (email@example.com).
Submitted for Publication: July 23, 2008; final revision received December 4, 2008; accepted January 12, 2009.
Author Contributions: Dr Del-Favero takes responsibility for the integrity of the data and the accuracy of the data analysis and declares that all authors had full access to all the data in the study.
Financial Disclosure: None reported.
Funding/Support: This research was funded by grants 2003-5158 and 2006-4472 from the Swedish Research Council; by the Medical Faculty of Umeå University; by the County Councils of Västerbotten and Norrbotten, Sweden; and by grants from the Fund for Scientific Research—Flanders, the Industrial Research Fund, and the Special Research Fund of the University of Antwerp. Dr Alaerts holds a PhD fellowship of the Fund for Scientific Research—Flanders.
Additional Contributions: The personnel of the VIB Genetic Service Facility (http://www.vibgeneticservicefacility.be/) provided the genetic analyses. Researchers Gunnel Johansson, RN, Lotta Kronberg, RN, Tage Johansson, RN, Lisbeth Bertilsson, RN, Annelie Nordin, RN, and Eva Lundberg, RN, ascertained the sample of patients with schizophrenia.