Introduction

Chromosome 8p has been implicated as a locus for schizophrenia in several linkage studies, which has been confirmed by meta-analyses (Badner and Gershon 2002; Lewis et al. 2003). Stefansson et al. (2002) conducted an extensive fine mapping of the locus using Icelandic pedigrees, which identified an associated haplotype (HAPICE) in the 5′ region of the neuregulin 1 (NRG1) gene. NRG1 is a plausible candidate gene for schizophrenia considering its involvement in neurodevelopment, regulation of glutamate and other neurotransmitter receptor expression, and synaptic plasticity (Falls 2003). The association was soon replicated in a Scottish sample by the same group (Stefansson et al. 2003). However, results of subsequent replication studies were inconsistent, and no susceptible variant has yet been identified (Tosato et al. 2005; Li et al. 2006; Munafo et al. 2008). Some studies observed the association of variants located in the 3′ end region of NRG1, mainly from exon 3 to intron 7 (Petryshen et al. 2005; Yang et al. 2003).

The inconsistency may be explicable by the ethnic difference in allele and haplotype frequencies among the studies. However, it is equally possible that one or more genes interacting with NRG1 may also be implicated in schizophrenia and attribute to the inconsistency. The mechanism of signaling by NRG1 involves binding to the extracellular domain of the tyrosine kinase receptors ERBB3 or ERBB4, which leads to the formation of ERBB homo- or heterodimers, often including ERBB2 (Falls 2003). In the study by Stefansson et al. (2002), mutant mice heterozygous for either Nrg1 or Erbb4 showed a behavioral phenotype that overlaps with mouse models for schizophrenia. Thus, NRG1–ERBB signaling pathway may be a candidate for genetic network involved in the pathogenesis of schizophrenia (Corfas et al. 2004). Two studies have investigated the association between ERBB3 and schizophrenia using a Japanese sample, although no association was observed (Kanazawa et al. 2007; Watanabe et al. 2007). Conversely, four studies provided evidence for an association between ERBB4 and schizophrenia (Nicodemus et al. 2006; Norton et al. 2006; Silberberg et al. 2006; Benzel et al. 2007), two of which observed that interaction between NRG1 and ERBB4 increased susceptibility for schizophrenia (Norton et al. 2006; Benzel et al. 2007). However, no other study to date has examined the interaction between the two genes. In addition, no study has investigated the association between ERBB4 and schizophrenia in Asian populations. In this study, we conducted association analysis of the two genes, NRG1 and ERBB4, in Japanese schizophrenics. With respect to NRG1, polymorphisms in the 3′ end region were investigated in addition to the HAPICE region. The interaction analysis was also conducted between NRG1 and ERBB4 polymorphisms in the same sample.

Subjects and methods

Subjects comprised 416 unrelated Japanese patients with schizophrenia [225 men and 191 women; age, 45.0 ± 14.7 years, mean ± standard deviation (SD)] recruited from several psychiatric hospitals around Tokyo. All patients met the Diagnostic and Statistical Manual for Mental Disorders, Fourth Edition (DSM-IV) criteria for schizophrenia. Patient diagnosis was confirmed by at least two experienced psychiatrists independently through repeated unstructured interviews and clinical chart review. Controls comprised 520 gender-matched unrelated healthy volunteers (281 men and 239 women; age: 37.5 ± 11.3 years) residing in the same area of Japan. All patients and controls were ethnically Japanese. The objective of the study was clearly explained, and written informed consent was obtained from all subjects. The study was approved by the Ethical Committee of the Faculty of Medicine, the University of Tokyo.

Genomic DNA was extracted from leukocytes using the standard phenol–chloroform method. We genotyped 13 polymorphisms in NRG1 and five in ERBB4, as listed in Table 1. Markers that form HAPICE were all genotyped, except for two single nucleotide polymorphisms (SNPs), SNP8NRG221132 and SNP8NRG433E1006, whose minor allele frequencies were extremely low in our Japanese population [0 and 1.1%, respectively, in Iwata et al. (2004)]. We also genotyped rs1081062, located in the HAPICE region, which was observed to be significantly associated with schizophrenia in the previous Japanese study (Fukui et al. 2006). To investigate the association of the 3′ end region, we genotyped seven SNPs, five of which showed a significant association with schizophrenia in the previous two studies (Yang et al. 2003; Petryshen et al. 2005). With respect to ERBB4, we covered SNPs that showed a significant association with schizophrenia in the previous two studies (Norton et al. 2006; Silberberg et al. 2006), except for two SNPs, IVS12 − 15C>T and IVS2 + 67C>G, which showed no polymorphism in a Japanese population in the National Center for Biotechnology Information (NCBI) dbSNP database. To cover IVS2 + 67C>G, the nearest tag SNP, rs7560730, was genotyped.

Table 1 Allele frequencies of polymorphisms in the NRG1 and ERBB4 genes
Full size table

All SNPs were analyzed using the ABI PRISM 7900HT Sequence Detection System (Applied Biosystems, CA, USA). Two microsatellites were analyzed using the ABI 3130xl Genetic Analyzer (Applied Biosystems) following polymerase chain reaction (PCR) amplification using the same primer sets as Stefansson et al. (2002). Deviation from the Hardy–Weinberg equilibrium was assessed using the χ2 test comparing observed genotype frequencies with those expected from allele frequencies. χ2 was used to compare SNP frequencies between patients and controls. For comparison of microsatellite allele frequencies, Fisher’s exact test was used. Lewontin’s D′ was used to analyze pairwise linkage disequilibrium (LD) (Lewontin 1964). Haplotype block analysis was conducted with the Gabriel method and the four-gamete method (Gabriel et al. 2002; Wang et al. 2002). Haplotypes of the polymorphisms and their frequencies were estimated by the maximum-likelihood method, with an expectation–maximization algorithm (Excoffier and Slatkin 1995). Permutation P values were calculated in comparison with haplotype frequencies between patients and controls (Fallin et al. 2001). The SNPAlyze 5.1 Standard (DYNACOM, Japan) and Haploview 3.32 (Barrett et al. 2005) were used to conduct LD, haplotype block, and haplotype analyses. Statistical power was calculated using a Web-based statistical program, Genetic Power Calculator (Purcell et al. 2003). To test interactions between two polymorphisms, standard logistic regression analyses were performed using the model by Cordell (2002):

$$ \log \left(p/( {1 - p}) \right) = \mu + a_{1} x_{1} + d_{1} z_{1} + a_{2} x_{2} + d_{2} z_{2} + i_{\text{aa}} x_{1} x_{2} + i_{\text{ad}} x_{1} z_{2} + i_{\text{da}} z_{1} x_{2} + i_{\text{dd}} z_{1} z_{2} $$

where x i and z i are dummy variables related to the underlying genotype at locus i. We set x i as a proportional component to the amount of major (or minor) alleles, and z i as a component corresponding to dominancy, i.e., x i = 1 and z i = −0.5 for major homozygote, x i = 0 and z i = 0.5 for heterozygote, and x i = −1 and z i = −0.5 for minor homozygote, respectively. The coefficients μ, a 1,a 2,d 1,d 2,i aa,i ad,i da, and i dd are parameters to be estimated. Lack of interaction in this model implies that all interaction coefficients (i aa,i ad,i da, and i dd) are zero. If there is no interaction, the resulting model is

$$ \log \left( {p/\left( {1 - p} \right)} \right) = \mu + a_{1} x_{1} + d_{1} z_{1} + a_{2} x_{2} + d_{2} z_{2} . $$

Statistical significance was assessed by a likelihood ratio test, comparing the full model with the reduced model. When significant interaction was observed, odds ratio (OR) was calculated for each genotype combination. The χ2 test or Fisher’s exact test was used to test the statistical significance of the OR. The statistical package SPSS for Windows (SPSS Inc. 1999) was used for interaction analyses.

Results

Polymorphism allele frequencies and genotypic distributions compared between patients and controls are shown in Tables 1 and 2, respectively. All polymorphism distributions followed Hardy–Weinberg equilibrium in both patients and controls. A nominally significant difference was observed in microsatellite 478B14-848 allele frequencies between patients and controls (Fisher’s exact test after unifying the alleles in which frequencies <2%, P = 0.042) (Table 1), although the statistical level became insignificant after correction for multiple testing. No significant difference was observed in allele frequencies of other polymorphisms or genotypic distributions of all SNPs between patients and controls.

Table 2 Genotypic distributions of polymorphisms in the NRG1 and ERBB4 genes
Full size table

The strength of LD denoted as D′ between pairs of polymorphisms is shown in Fig. 1 (values of r 2 are shown in supplementary material Fig. 1). The LD pattern was broadly similar in the patient and control groups. In haplotype block analysis, a block consisting of SNP8NRG241930 and SNP8NRG243177 was suggested in NRG1 HAPICE region; two blocks, rs2466058–rs2954041 and rs7823899–rs6988339, in NRG1 3′ end region; and a bock rs7598440–rs707284 in ERBB4 (Fig. 1). We then compared frequencies of the suggested four haplotypes (SNP8NRG241930–SNP8NRG243177, rs2466058–rs2954041, rs7823899–rs6988339, and rs7598440–rs707284) between patients and controls in addition to three haplotypes consisting of all polymorphisms in each region (i.e., SNP8NRG221533–420M9-1395, rs3924999–rs2919381, and rs7424835–rs7560730). No significant difference was observed in the distributions of haplotypes between patients and controls (data not shown). The three-marker haplotype consisting of SNP8NRG221533 and the two microsatellites did not show significant difference.

Fig. 1
figure 1

The strength of linkage disequilibrium (LD) between pairs of polymorphisms in patients and controls. The values of D′ for patients are shown in the upper portion and those for controls in the lower portion (the empty box means D′ value of 1.0). The heavy-line frames show suggested haplotype blocks. aNRG1 HAPICE region. Two microsatellites were treated as binary polymorphisms consisting of major allele and others. bNRG1 3′ end region. cERBB4

Full size image

Interaction analyses were performed between two polymorphisms, one in the NRG1 and another in the ERBB4 gene (Table 3). As a result, significant interactions were observed between rs2954041 in NRG1 and rs7424835 in ERBB4 and between rs2919381 in NRG1 and rs7560730 in ERBB4 [2-log (likelihood ratio) = 12.6 and 19.2, P = 0.013 and 0.00073, respectively]. The statistical level of interaction between rs2919381 in NRG1 and rs7560730 in ERBB4 remained significant after Bonferroni correction (P = 0.047, corrected for the 65 interaction analyses). Table 4 shows distribution and OR for each genotype combination between NRG1 rs2919381 and ERBB4 rs7560730. Increased risk for schizophrenia was observed in subjects with genotype combinations of GA–CT, GG–TT, and AA–TT for NRG1–ERBB4 (OR = 2.70, 3.56, and 7.47 compared with AA–CC combination, P = 0.049, 0.016, and 0.018, respectively) (Table 4). No significant interaction was observed between the other two pairs of polymorphisms in the NRG1 and ERBB4 genes.

Table 3 Interaction analyses between two polymorphisms of NRG1 and ERBB4 genes
Full size table
Table 4 Distribution and odds ratio for each genotype combination between NRG1 rs2919381 and ERBB4 rs7560730
Full size table

Discussion

Three Japanese studies have investigated the association between the HAPICE region of NRG1 and schizophrenia (Iwata et al. 2004; Fukui et al. 2006; Ikeda et al. 2008). The first study observed no significant association (Iwata et al. 2004), whereas the second observed the association of rs1081062 located in intron 1 of NRG1 (Fukui et al. 2006). Thus, the results contradicted each other. However, the third study, which is a recently published two-stage association study with 60 SNPs along the whole gene and large sample sets (1,126 cases and 1,022 controls for screening, and 1,262 cases, 1,022 controls, and 166 trios for confirmation), did not observe a significant association between SNPs and schizophrenia (Ikeda et al. 2008). Our results did not observe the association between NRG1 itself and schizophrenia, which correlates with the results of Ikeda et al. (2008).

The interaction analysis in our study, however, suggested the possibility that NRG1 may be implicated in schizophrenia through epistatic interaction with ERBB4. In a Japanese population, ERBB4 may be required as an additional factor for NRG1 to be a susceptibility locus for schizophrenia. It is unclear at this point how the two SNPs interact with each other and contribute to the pathogenesis of schizophrenia from the biological point of view. Both SNPs, rs2919381 in NRG1 and rs7560730 in ERBB4, are located in the intron of each gene. There is a possibility that these SNPs may not themselves be the etiological variants but, rather, in LD with the true susceptible variants. However, to our knowledge, no coding mutation has been identified in the ERBB4 gene to date. A portion of intronic variants could be related to altered splice-variant expression, as shown in other intronic SNPs in ERBB4, such as rs4673628, rs7598440, rs707284, and rs839523 (Law et al. 2007).

There are differences in details in the results between our study and others (Norton et al. 2006; Benzel et al. 2007), although all studies observed interaction between NRG1 and ERBB4. Norton et al. (2006) observed a significant interaction between HAPICE in NRG1 and IVS12 − 15C>T in ERBB4 in a Caucasian sample. On the other hand, no evidence for interaction was found for IVS2 + 67C>G in their study, the nearest tag SNP of which is rs7560730, and we observed a significant interaction between this SNP and NRG1 rs2919381. This might be partly attributable to the difference in ethnicity. The IVS12 − 15C>T showed no polymorphism in a Japanese population, and differences in allele and haplotype frequencies of NRG1 are remarkable between the populations (Gardner et al. 2006). The difference in ethnicity or selection of SNPs may account for the difference between our result and that in Benzel et al. (2007), which suggested 45 pair-wise SNP interactions between NRG1 and ERBB4 in Caucasian.

Several limitations may be acknowledged in our study. One is the possibility of type II error regarding the relationship between schizophrenia and each gene. Power analysis showed that the statistical power was more than 80% when genotype relative risk was set at 1.75–2.7 for SNPs used in the study (minor allele frequency = 0.111–0.474) under a dominant inheritance model (α = 0.05). This may not be enough to detect susceptibility loci of modest effects for schizophrenia. Second, we did not investigate the entire region of each gene. We selected markers that showed a significant association with schizophrenia in previous studies (Stefansson et al. 2002; Stefansson et al. 2003; Yang et al. 2003; Iwata et al. 2004; Petryshen et al. 2005; Fukui et al. 2006; Norton et al. 2006; Silberberg et al. 2006). Larger sample size and a gene-based approach using tag SNPs may be needed to obtain the conclusion, especially for the relationship between ERBB4 itself and schizophrenia in the Japanese population. Third is subject attributes. Patient diagnosis was not assessed using the standard structured interview; therefore, it may be possible that not all subjects completely satisfied the DSM-IV criteria for schizophrenia. Also, controls were not age-matched, which could also affect the results.

In conclusion, our results suggest that interaction between variants in NRG1 and ERBB4 might contribute to susceptibility for schizophrenia in the Japanese population. However, larger sample size and a systematic LD-based marker selection may be necessary to confirm our results and clarify the role of ERBB4 in the Japanese population.