In this manuscript we present confirmation of earlier published associations between schizophrenia and the genes NRG1 and ERBB4. We build a new, compelling picture of pathway involvement by showing that associations exist with additional gene family-members, namely EGFR, NRG2, and NRG3. Lastly, we present statistical evidence of gene epistasis, consistent with a highly complex biological pathway of schizophrenia susceptibility. This is the first time that the roles and inter-relationships of the ERBB and NRG gene families as a whole have been investigated in relation to schizophrenia and our approach sets a precedent in the investigation of common complex diseases.
Clear evidence for gene-gene interaction was detected for gene combinations NRG1-NRG2, NRG1-NRG3 and EGFR-NRG2, and suggestive evidence was also seen for ERBB4-NRG1, ERBB4-NRG2, ERBB4-NRG3 and ERBB4-ERBB2. Evidence of intragenic interaction was seen for SNPs in ERBB4. Of the 2640 tests for pair-wise interaction, 212 were significant at the 5% level, which is substantially more than the 131 expected by chance. This large number of low P- values may be partly explained by non-independence among tests. However, it may also indicate that some of the 79 SNPs studied interact to modulate susceptibility to schizophrenia.
Numerous studies have investigated the association between schizophrenia and NRG1 (reviewed in ). A comparison of the associations generated in the current study with those previously published shows good agreement. Associations that we observed at the 5' region of NRG1 overlap with those found by Stefansson et al. [3, 4], Corvin et al. , Li et al. , Zhao et al.  and Petryshen et al. . Similarly, the associations we observed at the 3' end of the gene overlap with those found by Yang et al. , Li et al. , Petryshen et al.  and Lachmann et al. . Two polymorphisms in our study directly replicate these associations: the putatively functional SNP rs3924999 was found to be associated with schizophrenia in a Han Chinese family study conducted by Yang et al. . This SNP was also tested in a Caucasian study of 136 families by Duan et al. , but was not significant. Rs2466049 was significant as part of a haplotype found in a Caucasian family by Petryshen et al. .
For the ERBB4 gene, four polymorphisms (rs3748962, rs839523, rs2054617 and rs6435711) screened in this study were also analysed by Silberberg et al. . However, none of the individual SNP associations observed in the previous study was replicated here. As no underlying functional variant has been identified for ERBB4, this lack of association at the SNP level is unsurprising and is consistent with a heterogeneous haplotypic background across studies. It is worth noting also that non-coding polymorphisms might lead to dysregulation of ErbB4 expression levels, a hypothesis corroborated by the Silberberg et al.  study, which found enhanced expression of the ERBB4 Jm-a CYT-1 isoform (one of four splice variants, see ) in schizophrenic patients. The altered balance of ErbB4 isoforms could lead to variations in homo- and heterodimer formation and thus in downstream signalling pathways.
In addition to providing further evidence of NRG1 and ERBB4 associations to schizophrenia, our study is the first to suggest possible involvement of three additional genes from the NRG and ERBB gene families, i.e. NRG2, NRG3 and EGFR (ERBB1). It is interesting to note that we only detected evidence for association of NRG genes which are expressed in the central nervous system, namely NRG1, NRG2 and the nervous-system-specific NRG3 , whereas NRG4, which is mainly expressed in pancreas and absent from the brain  was not found to be associated to schizophrenia.
To our knowledge, an association of EGFR to schizophrenia has not previously been investigated. Association of a polymorphism in EGF (one of the six EGFR ligands, Fig. 1) with schizophrenia was reported in Finnish men , but not replicated in a Japanese population . Further support for a possible role of EGF and EGFR in the disease comes from findings of abnormal expression of these genes in the forebrain and serum of schizophrenic patients . EGFR is usually omitted in reviews on NRG-ERBB signalling, although it is well-established that ErbB4 can interact with and cross-phosphorylate EGFR upon neuregulin-binding [36–39]. Furthermore, EGFR and ERBB4 show partially overlapping temporal and spatial mRNA expression in several brain areas, such as cortex, striatum, cerebellum and hippocampus, and are coexpressed in GABAergic interneurons [40, 41]. Therefore, it is very likely that EGFR/ErbB4 heterodimers mediate important functions in the developing and adult central nervous system in vivo. Whilst we limited this study to the NRG and ERBB gene families, it would be interesting to investigate association of the additional EGFR and ErbB4 ligands (see Fig. 1).
The significant gene-gene interactions reported in this study support our hypothesis of pathway involvement in schizophrenia. These genes have related functions and, for example, the NRG1-NRG2 and NRG1-NRG3 interactions may suggest competition between these ligands for a receptor. The NRG2-EGFR interaction is superficially more difficult to explain since NRG2 is a ligand for ErbB3 and ErbB4 and does not usually recognize EGFR directly or with high affinity [42–44]. However, a recent study reports that a single amino acid change in the EGFR ligand-binding domain dramatically increases the receptor affinity for NRG2β which can then induce potent stimulation of EGFR signalling . The interactions within the ERBB4 gene suggest that combinations of particular genotypes may result in modified receptor activity, leading to altered risk of schizophrenia.
Such suggestions concerning possible mechanisms of interaction remain speculative until the reality of these interactions has been established biologically and the effect of each two-locus genotype on the phenotype is understood. A follow-up study currently underway, using independent cases and controls, will clarify which of these interactions can be replicated and may generate further hypotheses regarding the nature of the interactions. It will be interesting also to investigate whether observed associations or their interactions correspond to particular phenotypically defined patient sub-groups. The strong evidence of genetic associations and interactions reported here provides ample justification for such an enquiry.