Association of the NRG1 gene and schizophrenia ... - Semantic Scholar

Molecular Psychiatry (2006) 11, 539–546 & 2006 Nature Publishing Group All rights reserved 1359-4184/06 $30.00 www.nature.com/mp

FEATURE REVIEW

Association of the NRG1 gene and schizophrenia: a meta-analysis MR Munafo`1, DL Thiselton2, TG Clark3 and J Flint4 1 Department of Experimental Psychology, University of Bristol, Bristol, UK; 2Department of Psychiatry, Virginia Institute of Psychiatric and Behavioral Genetics, Virginia Commonwealth University, Richmond, VA, USA; 3Department of Epidemiology and Public Health, Imperial College London, London, UK and 4Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, UK

We investigated the association of the NRG1 gene and schizophrenia using meta-analytic techniques, combining all published data while restricting our analysis to studies investigating the most commonly reported single marker (SNP8NRG221533). We also investigated whether ancestry (European vs East Asian) and study design (family-based vs case–control) moderated any association. We found no evidence for an association of SNP8NRG221533 with schizophrenia, and significant between-study heterogeneity, which persisted when familybased studies were combined separately. However, when haplotype-based P-values were combined, there was evidence in support of an association of NRG1 with schizophrenia, and no evidence of between-study heterogeneity. Our meta-analysis provides support for the association of NRG1 with schizophrenia, but indicates that firmly establishing the role of NRG1 gene in schizophrenia by genetic association requires much larger sample sizes than have hitherto been reported. Association analyses and replications should take place at the level of the gene, rather than at the level of SNP, haplotype, or functional variant. Meta-analysis would then be carried out on the basis of the combination of P-values. Molecular Psychiatry (2006) 11, 539–546. doi:10.1038/sj.mp.4001817; published online 7 March 2006 Keywords: neuregulin; NRG1; schizophrenia; meta-analysis; genetic association

Schizophrenia is a severely disabling psychiatric disease that affects up to 1% of the population worldwide.1 It is now well established that susceptibility to schizophrenia is influenced by genetic factors, and over recent years significant progress has been made towards identification of specific alleles that confer risk.2 Association studies of candidate genes, both positional (i.e., located in genomic regions linked to schizophrenia) and functional (e.g., involved in brain development, synaptic connectivity and neurotransmission), have revealed several promising candidates which are thought might contribute to disease liability, such as dysbindin (DTNBP1),3 neuregulin 1 (NRG1),4 G72,5 a regulator of G-protein signalling 4 (RGS4),6 catecholO-methyl transferase (COMT),7 proline dehydrogenase (PRODH),8 metabotropic glutamate receptor 3 (GRM3),9 protein kinase AKT1 (AKT1)10 and disrupted-in-schizophrenia 1 (DISC1).11 Despite growing support for these associations, a substantial number of studies have failed to replicate Correspondence: Dr MR Munafo`, Department of Experimental Psychology, University of Bristol, 8 Woodland Road, Bristol BS8 1TN, UK. E-mail: [email protected] Received 6 February 2006; accepted 6 February 2006; published online 7 March 2006

the positive findings. This may be the result of insufficient power to subsequently detect the original effect, lack of linkage disequilibrium (LD) between the susceptibility allele and SNPs used in the population under study, genetic heterogeneity or a true lack of association.12 Some reports also suggest that the genetic basis of familial and sporadic schizophrenia is different, as has been proposed for DTNBP113 and NRG1.14 Of the candidate schizophrenia susceptibility genes so far identified, only two, COMT and NRG1, map to the two genomic regions (22q and 8p) identified in both recent meta-analyses of genome-wide linkage scans.12 The functional COMT Val158Met polymorphism has been investigated in multiple studies for association to disease, but with equivocal results. A meta-analysis of the association between the functional COMT Val158Met polymorphism and schizophrenia case status suggested that the association may be moderated by study sample ancestry,7 although a more recent meta-analysis failed to find any evidence for this, and did not find strong evidence for association.15 NRG1 was first suggested as a candidate susceptibility gene for schizophrenia in a linkage study carried out in Iceland.4 The gene is large, consisting of at least 15 alternative transcripts spanning B1.4 Mb.16 In the original association study of NRG1 and

Association of the NRG1 gene and schizophrenia MR Munafo` et al

540

schizophrenia,4 a haplotype of five SNPs and two microsatellites was significantly over-represented in patients compared to controls. Of the markers typed and included in the haplotype, one (SNP8NRG221533: a T > C single nucleotide polymorphism) was the single most significant SNP. Since the initial report, there have been numerous studies assessing association between schizophrenia and the original 7-marker NRG1 ‘high-risk’ haplotype4 (referred to as HAPICE), as well as with other sequence variants in the gene. The original 7-marker haplotype,4 and a 3-marker haplotype, has been found to be associated with schizophrenia in a Scottish case– control study17 and in a small sample of familial cases in a mixed North European case–control sample,14 while this association has not been replicated in a large Irish family sample18 and a Japanese case– control study.19 However, two studies performed in East Asian samples, one case–control and trio sample,20 and one case–control sample only,21 confirmed association to the same 50 portion of NRG1. The former suggested association to a different version of HAPICE, comprising the same markers but different alleles,20 while the latter could not be compared directly owing to non-overlapping marker sets.21 A European case–control study initially showed association with a haplotype overlapping HAPICE at the 30 end, and covering a novel EST cluster in intron 1 (HAPBIRE),22 although in a more recent report with an extended sample the association was no longer observed.23 Finally, a recent study found no association in two geographically-distinct trio samples of European ancestry with any single SNP from HAPICE,24 although a 4-marker haplotype comprising markers thereof (SNP8NRG241930, SNP8NRG243177, 478B14-848, 420M9-1395) showed nominal significance in one sample. More recently, studies incorporating markers located elsewhere in the NRG1 gene have provided evidence that other regions may also be associated with schizophrenia, and raise the question of long-range LD as a factor to be considered. In one report, where SNP8NRG221533 showed significant association with schizophrenia in a large East Asian trio sample, even stronger association was seen with markers located in exon 2 and intron 5, in a separate haplotype block covering the main body of the gene.25 However, the overtransmitted 38Arg allele of the exon 2 cSNP was found to be undertransmitted in an independent Chinese sample of trios and bios,26 and no significant difference in frequency between 38Gln or 38Arg was observed between cases and controls. Two recent reports, one in a European case–control and trio sample,16 and one in an East Asian family and case– control sample,27 demonstrated association with markers in three distinct blocks of LD spread throughout the gene, with some differentiation between familybased vs case–control samples as to the strength of the association. In an attempt to investigate the hypothesis that clinical heterogeneity may underlie inconsistency in replication, Bakker and co-workers28

Molecular Psychiatry

examined NRG1 markers of HAPICE for association with schizophrenia in European patients with and without deficit syndrome, compared to controls, and found the non-deficit group to be significantly associated for 2 HAPICE markers SNP8NRG221533 and 478B14-848, but with different risk alleles than in the original reports.4,17 An increasingly common phenomenon among genetic association studies is therefore emerging for NRG1 and schizophrenia, where, while no single causative allele has yet been identified that confers susceptibility to schizophrenia, some studies demonstrate association of disease with a variety of haplotypes located throughout the gene,17,20,21 and others do not.18,19 Overall, there is now substantial support both for and against genetic variation in NRG1 contributing to susceptibility to schizophrenia. The differences in allele frequencies, regional LD patterns and schizophrenia-associated risk haplotypes between various studies and ancestry groups (e.g., European vs East Asian) suggests that there may be more than one functional variant in a region spanning several hundred kilobases, and substantial allelic heterogeneity at the NRG1 locus. If NRG1 variation contributes only modestly to the overall risk of developing the disease, it is also possible that false negative findings are included in the body of data now available. We therefore attempted to combine all published data on the association of NRG1 and schizophrenia using meta-analytic techniques, restricting our analysis to studies investigating the most commonly reported single marker (SNP8NRG221533), which was also that most strongly associated in the original study.4 We also investigated whether this association differs between individuals of differing ancestry (European vs East Asian), and depends on the study design employed (family-based vs case–control). Finally, given the limitations associated with investigating a single marker, we combined haplotype-based P-values from included studies, to investigate the evidence for association at the level of the NRG1 gene.29

Methods Selection of studies for inclusion Case–control and family-based genetic association studies of the NRG1 gene in healthy control groups and clinically diagnosed schizophrenic patients were included. Studies reporting data on either single-sex or both male and female participants of any ethnic origin were included. Studies with data for only schizophrenic patients, with no corresponding healthy controls, were excluded. The principal outcome measure was the allele frequency odds ratio (OR) for the SNP8NRG221533 polymorphism and schizophrenia case status. The secondary outcome measure was the haplotype P-value for the strongest reported haplotype association with schizophrenia case status.


Search strategy The search was performed on three databases: PubMed, PsycInfo and Medline. These databases were searched from the first date available in each database up to 31st December 2005, using the search terms ‘schizophrenia’, ‘NRG1’, ‘neuregulin’. Once articles had been collected bibliographies were then hand-searched for additional references. The abstracts of studies identified by these search strategies were then examined with reference to the inclusion and exclusion criteria. Duplications were deleted, and the whole text of each reference was then checked to further establish whether the study met the study inclusion criteria. Studies that reported previously published data were excluded. Data extraction For each study, the following data were extracted using standard forms: (1) Author(s) and year of publication; (2) Methods (country of origin, dominant ancestry of sample, case and control sample size, diagnostic criteria for schizophrenic case status, statement of Hardy–Weinberg equilibrium, method of genotyping); (3) Data (allele frequency in control and case groups, mean age, sex ratio, haplotype Pvalue). Ancestry was coded as European, East Asian, African, or Mixed. Discrepancies were resolved by mutual consent. Statistical analysis Odds ratios and their s.e. for individual studies were calculated from 2 2 tables in a case/control format. For the family-based trio data, the cases came from the affected children and data for the controls was based on the difference between the parents and child using the approach of Kazeem and Farrall.30 When a single study included both case–control and familybased data, or samples drawn from distinct populations, these were considered as separate study samples. Pooled ORs were calculated using fixed effects and random effects approaches,31 and the significance of the pooled OR determined using a Z-test. Data were initially analysed within a fixed effects framework, using inverse variance methods, which assumes that the effect of allele frequency is constant across studies, and between-study variation is due to random variation. This assumption was checked using a w2 test of goodness of fit for homogeneity. Where there was evidence of a significant association in the presence of significant between-study heterogeneity, a random effects framework was employed, using DerSimonian and Laird methods.31 This assumes that between-study variation is due to both random variation and an individual study effect. Random effects models are more conservative than fixed effects models and generate a wider confidence interval. Stratified analyses were conducted to assess any moderating effects of ancestry (European, East Asian) and study design (case–control, family-based), and

the difference in pooled OR compared using a Z-test. The OR of the first published study was compared to the pooled OR of the remaining studies using a Z-test, as there is evidence for a substantially greater estimate of effect size in the first published study.32 Each individual study was systematically removed separately from the meta-analysis in a sensitivity analysis, to determine whether any single study disproportionately influenced the results of the meta-analysis. Publication bias was assessed by means of a funnel plot of individual study log OR against 1/s.e. log OR, and formally by means of Egger’s test,33 which is based on a weighted linear regression of log OR on s.e. log OR, where the weights are inversely proportional to the variance of log OR. We combined the lowest (i.e., most highly significant) individual haplotype P-values reported in individual studies by first transforming them to Zstatistics and then pooling using a Stouffer method,34 either unweighted or weighted by sample size, to ascertain evidence for association of NRG1 with schizophrenia at the level of the gene rather than the individual SNP. The selection of haplotypes for inclusion was based solely on the strongest result reported in each study. Data were analysed using the S-Plus (Version 6.1) statistical software package.

541

Results Description of studies A total of 14 studies published between 2002 and 2006, comprising nineteen independent samples, were identified by the search strategy, met the inclusion criteria and contributed to the meta-analysis. Each sample was included independently in the analysis. The characteristics of these samples are described in Table 1. Two studies that did not utilize NRG1 SNP8NRG221533,21,26 one study that investigated schizotypal personality in adolescents,37 and one study that did not report allele frequencies38 were excluded. Eleven samples reported data on participants of predominantly European ancestry, seven on participants of predominantly East Asian ancestry, and one on participants of predominantly African ancestry. Six samples included a statement of Hardy–Weinberg equilibrium, while a further six samples reported data that enabled formal testing of whether SNP8NRG221533 genotype frequencies deviated significantly from Hardy–Weinberg Equilibrium. In two samples, there was evidence of deviation.25,28 Nine samples used DSM-IV criteria for assessing schizophrenia case status, while four used either DSM-IV or DSM-III-R criteria, three used DSM-III-R criteria only, one used ICD-10 criteria and one used RDC criteria. In one case, the diagnostic criteria used were not stated. Meta-analysis When all samples (k = 19) were included there was no evidence for association of NRG1 SNP8NRG221533 Molecular Psychiatry


542

Table 1

Characteristics of included samples

Study and year Stefansson et al. (2002)4 Stefansson et al. (2003)4 Williams et al. (2003)14 Yang et al. (2003)25 Bakker et al. (2004)28 Corvin et al. (2004)22 Hall et al. (2004)24 Hall et al. (2004)24 Iwata et al. (2004)19 Li et al. (2004)27 Li et al. (2004)27 Thiselton et al. (2004)18 Zhao et al. (2004)20 Zhao et al. (2004)20 Fukui et al. (2005)35 Petryshen et al. (2005)16 Petryshen et al. (2005)16 Lachman et al. (2006)36 Lachman et al. (2006)36

Case n

Case N

Control n

Control N

269 440 463 328 171 180 132 105 644 285 361 335 396 385 372 214 67 120 19

740 1146 1284 492 520 486 404 318 1212 516 664 996 733 675 652 642 222 352 278

229 315 472 NA 453 170 NA NA 530 250 NA 100 263 NA 456 145 NA 111 34

772 1030 1316 NA 1170 444 NA NA 1030 423 NA 312 506 NA 784 484 NA 318 270

Ancestry

Method

Diagnosis

European European European East Asian European European European European East Asian East Asian East Asian European East Asian East Asian East Asian European European European African

Case–control Case–control Case–control Family-based Case–control Case–control Family-based Family-based Case–control Case–control Family-based High density Case–control Family-based Case–control Case–control Family-based Case–control Case–control

RDC DSM III-R DSM IV ICD 10 DSM IV DSM IV DSM IV DSM IV DSM IV DSM III-R/IV DSM III-R/IV Not stated DSM III-R DSM III-R DSM IV DSM IV DSM IV DSM III-R/IV DSM III-R/IV

n = number of C alleles; N = total number of alleles. Abbreviation: NA, not applicable.

allele frequency and schizophrenia case status (Z = 1.57, P = 0.116, OR = 1.04, 95% CI 0.99–1.10). There was evidence of significant between-study heterogeneity (w2 [18] = 51.13, P < 0.001). Stratified analyses When samples that recruited participants of predominantly European ancestry only were included (k = 11) there was no evidence for association of NRG1 SNP8NRG221533 allele frequency and schizophrenia case status (Z = 1.60, P = 0.108, OR = 1.06, 95% CI 0.99–1.14). There was evidence of significant between-study heterogeneity (w2 [10] = 29.41, P = 0.001). When samples that recruited participants of predominantly East Asian ancestry only were included (k = 7) there was no evidence for significant association of NRG1 SNP8NRG221533 allele frequency and schizophrenia case status (Z = 0.88, P = 0.380, OR = 1.04, 95% CI 0.95–1.13). There was evidence of significant between-study heterogeneity (w2 [6] = 15.98, P = 0.014). The difference in pooled OR between samples that recruited samples of predominantly European ancestry and those that recruited samples of predominantly East Asian ancestry was not significant (Z = 0.39, P = 0.699). When samples that employed a family-based study design only were included (k = 6) there was no evidence for significant association of NRG1 SNP8NRG221533 allele frequency and schizophrenia case status (Z = 0.30, P = 0.761, OR = 1.02, 95% CI 0.91–1.14). There was evidence of significant between-study heterogeneity (w2 [5] = 16.55, P = 0.005). When samples that employed a case–control study design only were included (k = 13) there was no evidence for association of NRG1 SNP8NRG221533 Molecular Psychiatry

allele frequency and schizophrenia case status (Z = 1.62, P = 0.104, OR = 1.05, 95% CI 0.99–1.12). There was evidence of significant between-study heterogeneity (w2 [12] = 34.57, P < 0.001). The difference in pooled OR between samples that employed a family-based design and those that employed a case– control design was not significant (Z = 0.51, P = 0.610). When one case–control sample that recruited cases from high density families18 was excluded the results did not change substantially (Z = 1.55, P = 0.123), and the between-study heterogeneity remained significant (P < 0.001). Sensitivity analysis The results of the fixed-effects meta-analysis remained non-significant when each study was systematically removed from the analysis, with the exception of two studies27,28 where removal rendered the association of SNP8NRG221533 with schizophrenia significant (P < 0.05). In both cases, there remained evidence of significant between-study heterogeneity (P < 0.001), and when the analysis was re-run within a random effects framework the evidence for association was non-significant (P > 0.4). In addition, when two studies25,28 that reported SNP8NRG221533 genotype frequencies which deviated significantly from Hardy–Weinberg Equilibrium were removed from the analysis, the result of the main analysis did not alter substantially (Z = 1.67, P = 0.095, OR = 1.05, 95% CI 0.99–1.11). Date of publication When the first published study4 was removed from the analysis there was no evidence for association of

14

0.4

12

0.2

10

1/SE Log OR

0.6

0 -0.2

6 4

-0.6

2

-0.8 2002

543

8

-0.4

0 2003

2004 2005 Publication Year

2006

2007

Figure 1 Individual sample log OR and publication date. Individual sample log ORs are presented, plotted against publication date. Correlational analysis indicates a significant negative association between publication data and log OR (P = 0.007), suggesting a trend towards a smaller log OR for the SNP8NRG221533 polymorphism over time.

NRG1 SNP8NRG221533 allele frequency and schizophrenia case status (Z = 0.91, P = 0.365, OR = 1.03, 95% CI 0.97–1.08). There was evidence of significant between-study heterogeneity (w2 [17] = 45.53, P < 0.001). The difference in OR of the first published sample and the pooled OR of remaining samples was significant for both fixed effects (Z = 2.44, P = 0.014) and random effects (Z = 2.52, P = 0.012) models. In addition, publication date was significantly negatively correlated with the individual sample log OR (rs = 0.59, P = 0.007), indicating larger reported effect sizes of SNP8NRG221533 in earlier published studies. A plot of publication date against individual sample log OR is presented in Figure 1. Publication bias A funnel plot of individual sample log OR against 1/ s.e. log OR is presented in Figure 2. This did not indicate any evidence of publication bias. Egger’s test also did not indicate any evidence of publication bias (P = 0.060). Haplotype analysis Combining the most highly significant haplotype P-values reported in individual studies indicated a significant association of NRG1 with schizophrenia in both the unweighted (P = 0.020) and weighted (P = 0.016) analyses. There was no evidence of significant between-study heterogeneity (w2 [17] = 20.94, P = 0.229). Analyses separately stratified by study design and ancestry, and excluding the first published study, all remained significant (P < 0.05) in both the unweighted and weighted analyses, and in all cases there was no evidence of between-study heterogeneity (P > 0.08). Removing the two studies25,28 that reported SNP8NRG221533 genotype frequencies which deviated significantly from Hardy–Weinberg Equilibrium did not substantially alter the results of either

0

5

10

15

20

25

Log OR

Figure 2 Individual sample log OR and 1/s.e. log OR. Individual sample log ORs (effect size) are presented, plotted against 1/s.e. log OR (accuracy). Ascertainment bias (e.g., caused by publication bias) would be evidenced by asymmetry in this plot. Egger’s test indicated no evidence of ascertainment bias (P = 0.060).

1 0.9 0.8 0.7 p-value

Log OR


0.6 0.5 0.4 0.3 0.2 0.1 0 2002

2003

2004 2005 Publication Year

2006

2007

Figure 3 Individual haplotype P-value and publication date. Individual haplotype P-values, corresponding to the strongest reported haplotype association of NRG1 with schizophrenia in an individual study, are presented, plotted against publication date. Correlational analysis indicates no association between publication data and haplotype P-value (P = 0.135).

the unweighted (P = 0.022) or the weighted (P = 0.024) analysis. In addition, publication date was not significantly correlated with the haplotype P-value (rs = þ 0.36, P = 0.135), indicating no association between publication date and statistical significance of reported haplotype associations. A plot of publication date against individual haplotype P-value is presented in Figure 3. Haplotype blocks that have been reported to be significantly associated with schizophrenia in the included studies are presented in Table 2.

Discussion Since the neuregulin gene NRG1 was first proposed as a susceptibility gene for schizophrenia in 2002,4 Molecular Psychiatry

544


Table 2

NRG1 haplotype blocks reported to be associated with schizophrenia


Studies included in the meta-analysis are presented, with significant haplotype blocks reported to be associated with schizophrenia in each individual study highlighted. Markers typed (o), markers significantly associated with schizophrenia as part of a haplotype block (x), and markers significantly associated with schizophrenia independently (x) are also presented. Intronic SNPs rs385396 and rs32016134 have high FST values and lie in the 50 region of Hs. 97362, indicative of positive selection and regulatory potential for this putative transcript (Gardner et al., 2006).39


a number of studies have reported an association between this gene and schizophrenia. A recent review of schizophrenia genetics concluded that the evidence ‘argue[s] convincingly that NRG1 is a likely susceptibility gene for schizophrenia’.12 Our metaanalysis of studies reporting data on SNP8NRG221533 does not support the existence of an association between this individual marker and the illness. However, when haplotype-based P-values were combined there was evidence of significant association of the NRG1 gene and schizophrenia, which was robust to the removal of the first published study and stratification by both study type and sample ancestry. The removal of studies which reported SNP8NRG221533 genotype frequencies that deviated significantly from Hardy–Weinberg Equilibrium did not alter these results substantially. In the case of the single-marker analysis, we could not discern any variable that accounted for the observed between-study heterogeneity. Heterogeneity was still observed when studies employing familybased designs were combined separately, and when studies recruiting participants of European and East Asian ancestry were considered separately, which suggests that population stratification is not the sole explanation for this heterogeneity. One possible source of between-study heterogeneity is the difference between studies in the genotyping completion rate; allele and genotype frequencies from called samples may differ from those in the entire sample. Unfortunately it was not possible to test this possibility directly, as not all studies systematically report the genotyping completion rate. We did find evidence of a significant difference between the effect size estimate in the first published study compared to the pooled OR for subsequent studies, and there was a significant negative correlation between date of publication and log OR reported. This is consistent with evidence that initial reports of genetic association typically provide a substantially greater estimate of effect size than later attempts to replicate the association,32 although this may be misleading as in relatively recently-published studies conducted in China reported associations tended to be with more 30 markers. When studies that recruited samples of predominantly European ancestry and studies that recruited samples of predominantly East Asian ancestry were analysed separately, there was no evidence for significant association in either case, and no evidence for a significant difference in pooled ORs between these two groups of studies. Although there was no formal evidence of publication bias, Egger’s test did approach statistical significance, and it should be noted that this is a relatively weak test, suggesting that there may be publication bias present. In the case of the haplotype-based analysis, the association of NRG1 remained significant when case– control and family-based studies were analysed separately, and when samples comprising participants of predominantly European ancestry and

predominantly East Asian ancestry were analysed separately. In all cases there was no evidence of between-study heterogeneity, and there was no significant correlation between haplotype P-value and publication date. Although we only included studies that reported data on the SNP8NRG221533 marker, it should be noted that this criterion resulted in the exclusion of only two studies. These results argue against the possibility that SNP8NRG221533 ‘tags’ a risk haplotype, but suggest that variation in the NRG1 gene is associated with schizophrenia risk. Meta-analytic techniques require the combination of comparable data, which was only possible for SNP8NRG221533, due to the use of disparate marker sets across studies. While it would have been theoretically possible to have combined individual SNP data for multiple SNPs, it is unclear how informative this would be without compelling evidence for a single SNP ‘tagging’ a haplotype block. Most studies of NRG1 find association with haplotypes, rather than with individual SNPs, which, although potentially capturing more of the genetic variation within the locus, and increasing the likelihood of ‘tagging’ a functional variant, raises issues of what should be considered a replication. Nevertheless, our haplotype-based analysis supports the candidature of NRG1 as a schizophrenia risk gene, and does not suggest that this association is moderated by ancestry. Several of the haplotype blocks reported to be associated with schizophrenia overlap substantially with the original HAPICE block, which implicates the 50 end of the gene, suggesting that altered mRNA expression or splicing may mediate this association.40 However, it should be borne in mind that NRG1 encodes approximately 15 separate proteins, with a diverse range of functions in the brain, and any one of these might conceivably influence susceptibility to schizophrenia.40 Our meta-analysis can only partially help to resolve the inconsistent genetic association findings so far published; new ways of combining haplotype-based association results are required, and, ideally, analyses of much larger sample sets. It is likely that current studies are under-powered to detect genetic effects of modest magnitude. While the original finding that the SNP8NRG221533 variant ‘tags’ the NRG1 risk haplotype was not supported by our meta-analysis, we did find support for the association of NRG1 with schizophrenia, suggesting that analyses at the level of the SNP, specific haplotype, or functional variant may not be as reliable as analyses at the level of the gene. Therefore, as suggested by Neale and Sham,29 association analyses and replications should take place at the level of the gene, rather than at the level of SNP, specific haplotype, or functional variant. Meta-analysis would then be carried out on the basis of the combination of P-values, as originally proposed by Fisher.41 It remains to be seen whether this approach will support the candidature of susceptibility genes in schizophrenia.

545



546

Acknowledgments Dawn Thiselton is an Essel Foundation NARSAD Young Investigator. Jonathan Flint is supported by the Wellcome Trust.

References 1 Rossler W, Salize HJ, van Os J, Riecher-Rossler A. Size of burden of schizophrenia and psychotic disorders. Eur Neuropsychopharmacol 2005; 15: 399–409. 2 Owen MJ, Williams NM, O’Donovan MC. The molecular genetics of schizophrenia: new findings promise new insights. Mol Psychiatry 2004; 9: 14–27. 3 Straub RE, Jiang Y, MacLean CJ, Ma Y, Webb BT, Myakishev MV et al. Genetic variation in the 6p22.3 gene DTNBP1, the human ortholog of the mouse dysbindin gene, is associated with schizophrenia. Am J Hum Genet 2002; 71: 337–348. 4 Stefansson H, Sigurdsson E, Steinthorsdottir V, Bjornsdottir S, Sigmundsson T, Ghosh S et al. Neuregulin 1 and susceptibility to schizophrenia. Am J Hum Genet 2002; 71: 877–892. 5 Chumakov I, Blumenfeld M, Guerassimenko O, Cavarec L, Palicio M, Abderrahim H et al. Genetic and physiological data implicating the new human gene G72 and the gene for D-amino acid oxidase in schizophrenia. Proc Natl Acad Sci USA 2002; 99: 13675–13680. 6 Chowdari KV, Mirnics K, Semwal P, Wood J, Lawrence E, Bhatia T et al. Association and linkage analyses of RGS4 polymorphisms in schizophrenia. Hum Mol Genet 2002; 11: 1373–1380. 7 Glatt SJ, Faraone SV, Tsuang MT. Association between a functional catechol O-methyltransferase gene polymorphism and schizophrenia: meta-analysis of case–control and family-based studies. Am J Psychiatry 2003; 160: 469–476. 8 Liu H, Heath SC, Sobin C, Roos JL, Galke BL, Blundell ML et al. Genetic variation at the 22q11 PRODH2/DGCR6 locus presents an unusual pattern and increases susceptibility to schizophrenia. Proc Natl Acad Sci USA 2002; 99: 3717–3722. 9 Egan MF, Straub RE, Goldberg TE, Yakub I, Callicott JH, Hariri AR et al. Variation in GRM3 affects cognition, prefrontal glutamate, and risk for schizophrenia. Proc Natl Acad Sci USA 2004; 101: 12604–12609. 10 Emamian ES, Hall D, Birnbaum MJ, Karayiorgou M, Gogos JA. Convergent evidence for impaired AKT1-GSK3beta signaling in schizophrenia. Nat Genet 2004; 36: 131–137. 11 Millar JK, Wilson-Annan JC, Anderson S, Christie S, Taylor MS, Semple CA et al. Disruption of two novel genes by a translocation co-segregating with schizophrenia. Hum Mol Genet 2000; 9: 1415–1423. 12 Harrison PJ, Weinberger DR. Schizophrenia genes, gene expression, and neuropathology: on the matter of their convergence. Mol Psychiatry 2005; 10: 40–68; image 5. 13 Van Den Bogaert A, Schumacher J, Schulze TG, Otte AC, Ohlraun S, Kovalenko S et al. The DTNBP1 (dysbindin) gene contributes to schizophrenia, depending on family history of the disease. Am J Hum Genet 2003; 73: 1438–1443. 14 Williams NM, Preece A, Spurlock G, Norton N, Williams HJ, Zammit S et al. Support for genetic variation in neuregulin 1 and susceptibility to schizophrenia. Mol Psychiatry 2003; 8: 485–487. 15 Munafo MR, Bowes L, Clark TG, Flint J. Lack of association of the COMT (Val158/108 Met) gene and schizophrenia: a meta-analysis of case–control studies. Mol Psychiatry 2005; 10: 765–770. 16 Petryshen TL, Middleton FA, Kirby A, Aldinger KA, Purcell S, Tahl AR et al. Support for involvement of neuregulin 1 in schizophrenia pathophysiology. Mol Psychiatry 2005; 10: 366–374, 328. 17 Stefansson H, Sarginson J, Kong A, Yates P, Steinthorsdottir V, Gudfinnsson E et al. Association of neuregulin 1 with schizophrenia confirmed in a Scottish population. Am J Hum Genet 2003; 72: 83–87. 18 Thiselton DL, Webb BT, Neale BM, Ribble RC, O’Neill FA, Walsh D et al. No evidence for linkage or association of neuregulin-1 (NRG1) with disease in the Irish study of high-density schizophrenia families (ISHDSF). Mol Psychiatry 2004; 9: 777–783, ; image 729. Molecular Psychiatry

19 Iwata N, Suzuki T, Ikeda M, Kitajima T, Yamanouchi Y, Inada T et al. No association with the neuregulin 1 haplotype to Japanese schizophrenia. Mol Psychiatry 2004; 9: 126–127. 20 Zhao X, Shi Y, Tang J, Tang R, Yu L, Gu N et al. A case control and family based association study of the neuregulin1 gene and schizophrenia. J Med Genet 2004; 41: 31–34. 21 Tang JX, Chen WY, He G, Zhou J, Gu NF, Feng GY et al. Polymorphisms within 50 end of the Neuregulin 1 gene are genetically associated with schizophrenia in the Chinese population. Mol Psychiatry 2004; 9: 11–12. 22 Corvin AP, Morris DW, McGhee K, Schwaiger S, Scully P, Quinn J et al. Confirmation and refinement of an ‘at-risk’ haplotype for schizophrenia suggests the EST cluster, Hs.97362, as a potential susceptibility gene at the Neuregulin-1 locus. Mol Psychiatry 2004; 9: 208–213. 23 Morris DW, McGhee K, Schwaiger S, Nangle J, Murphy K, Donohoe G et al. An update of association studies of DTNBP1, NRG1, RGS4 and MRDS1 in an extended Irish schizophrenia case–control sample. Am J Med Genet 2004; 130B: 143. 24 Hall D, Gogos JA, Karayiorgou M. The contribution of three strong candidate schizophrenia susceptibility genes in demographically distinct populations. Genes Brain Behav 2004; 3: 240–248. 25 Yang JZ, Si TM, Ruan Y, Ling YS, Han YH, Wang XL et al. Association study of neuregulin 1 gene with schizophrenia. Mol Psychiatry 2003; 8: 706–709. 26 Hong CJ, Huo SJ, Liao DL, Lee K, Wu JY, Tsai SJ. Case–control and family-based association studies between the neuregulin 1 (Arg38Gln) polymorphism and schizophrenia. Neurosci Lett 2004; 366: 158–161. 27 Li T, Stefansson H, Gudfinnsson E, Cai G, Liu X, Murray RM et al. Identification of a novel neuregulin 1 at-risk haplotype in Han schizophrenia Chinese patients, but no association with the Icelandic/Scottish risk haplotype. Mol Psychiatry 2004; 9: 698–704. 28 Bakker SC, Hoogendoorn ML, Selten JP, Verduijn W, Pearson PL, Sinke RJ et al. Neuregulin 1: genetic support for schizophrenia subtypes. Mol Psychiatry 2004; 9: 1061–1063. 29 Neale BM, Sham PC. The future of association studies: gene-based analysis and replication. Am J Hum Genet 2004; 75: 353–362. 30 Kazeem GR, Farrall M. Integrating case-control and TDT studies. Ann Hum Genet 2005; 69(Part 3): 329–335. 31 DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: 177–188. 32 Trikalinos TA, Ntzani EE, Contopoulos-Ioannidis DG, Ioannidis JP. Establishment of genetic associations for complex diseases is independent of early study findings. Eur J Hum Genet 2004; 12: 762–769. 33 Hayashino Y, Noguchi Y, Fukui T. Systematic evaluation and comparison of statistical tests for publication bias. J Epidemiol 2005; 15: 235–243. 34 Darlington RB, Hayes AF. Combining independent P-values: extensions of the Stouffer and binomial methods. Psychol Methods 2000; 5: 496–515. 35 Fukui N, Muratake T, Kaneko N, Amagane H, Someya T. Supportive evidence for neuregulin 1 as a susceptibility gene for schizophrenia in a Japanese population. Neurosci Lett 2005 (Epub ahead of print). 36 Lachman HM, Pedrosa E, Nolan KA, Glass M, Ye K, Saito T. analysis of polymorphisms in AT-rich domains of neuregulin 1 gene in schizophrenia. Am J Med Genet B Neuropsychiatr Genet 2006; 141: 102–109. 37 Lin HF, Liu YL, Liu CM, Hung SI, Hwu HG, Chen WJ. Neuregulin 1 gene and variations in perceptual aberration of schizotypal personality in adolescents. Psychol Med 2005; 35: 1589–1598. 38 Duan J, Martinez M, Sanders AR, Hou C, Krasner AJ, Schwartz DB et al. Neuregulin 1 (NRG1) and schizophrenia: analysis of a US family sample and the evidence in the balance. Psychol Med 2005; 35: 1599–1610. 39 Gardner M, Gonzalez-Neira A, Lao O, Calafell F, Bertranpetit J, Comas D. Extreme population differences across Neuregulin 1 gene, with implications for association studies. Mol Psychiatry 2006; 11: 66–75. 40 Owen MJ, Craddock N, O’Donovan MC. Schizophrenia: genes at last? Trends Genet 2005; 21: 518–525. 41 Fisher RA. Statistical Methods for Research Workers. Oliver and Boyd: Edinburgh, 1925.