Two recent reports have shown evidence for association between single-nucleotide polymorphisms (SNPs) in ErbB4 and schizophrenia.^{1, 2} In addition, one found support for statistical interaction between NRG1 and ErbB4 resulting in increased risk of schizophrenia,¹ although evidence for statistical interaction was only observed in ErbB4 heterozygotes, which is difficult to interpret. Knockout NRG1/ErbB4 mice exhibit behavioral abnormalities thought to coincide with selected aspects of schizophrenia in humans.³ Based on these intriguing statistical and biobehavioral findings, we sought to assess association between ErbB4 and schizophrenia status in three independent family-based samples. In addition, we examined whether ErbB4 was associated with cognitive test scores in healthy controls that have been shown to be related to increased genetic risk for schizophrenia.^{4, 5} A thorough examination of epistasis between ErbB4, NRG1, and other protein complex-binding partners is currently underway,⁶ and will be reported in the future.

A family-based sample (N=296 families) was ascertained as part of the Clinical Brain Disorders Branch/National Institute of Mental Health Sibling Study (SS), the details of which are described elsewhere.⁷ All subjects were diagnosed using the Structured Clinical Interview (SCID). Probands met Diagnostic and Statistical Manual-IV (DSM-IV) criteria for broad schizophrenia diagnosis. Control individuals (N=370) were ascertained from the National Institutes of Health Normal Volunteer Office and were screened by SCID diagnosis for psychiatric disorders, again as described previously.⁷ All participants gave informed consent and self-identified as Caucasian. Two additional independent family-based samples (Caucasian (N=71 families), African American (N=51 families)) were ascertained through the NIMH Genetics Initiative (NIMHGI). Genotypes for all SNPs were obtained using the Taqman 5′-exonuclease allelic discrimination assay. Hardy–Weinberg Equilibrium (HWE) was tested using exact tests. Analyses of single SNPs were conducted using Family Based Association Test (FBAT) in nuclear families⁸ and linear regression in the healthy controls using STATA, version 8.2 (College Park, TX, USA). Haplotype analysis was performed using FBAT for families and the R package haplo.stats⁹ in healthy controls. P-values have not been adjusted for multiple testing in any sample. Cognitive intermediate phenotypes were tested in control subjects and were based on seven normalized derived factor scores. In a factor analysis of 24 variables from nine neuropsychological tests shown to be associated with increased genetic risk for schizophrenia,^{4, 5, 10} seven factors were identified that explained 68% of the variance on these measures (Genderson M, Aper-Diaz CM, Weinberger DR, Goldberg TE, unpublished data). These factors were labeled: visual memory (comprised of specific subscales of the (Weschler memory scale and the California Verbal Learning Test (CVLT)), working memory (N back test), verbal memory (story recall), processing speed Wechsler Adult Intelligence Scale (WAIS), Trail Making Test (TRAILS), Fluency), executive function (Wisconsin Card Sorting Test (WCST)), attention (Continuous Performance Test (CPT) and digit span). SNPs were selected to replicate prior reports, so previously positively associated SNPs and several related tag SNPs selected from HapMap (www.hapmap.org)^{1, 2} were chosen for genotyping. Several previously reported SNPs were not available with our assay.

rs4673628 showed genotype frequencies that departed from HWE in both controls (P=0.0098) and in probands (P=0.0002). Family-based affection status analyses in the SS revealed two 3-SNP haplotypes significantly positively associated with schizophrenia status: rs3748962–rs2289086–rs3791709 (G-A-A; global P=0.020, haplotype P=0.020) and rs7598440–rs839541–rs839523 (G-A-A; global P=0.042, haplotype P=0.032) (Table 1).

At rs7598440-rs839541-rs839523, the same haplotype showing positive association with schizophrenia in the SS Caucasian families also showed marginal evidence for positive association in the NIMHGI African-American families (global P=0.061, haplotype P=0.012). The NIMHGI Caucasian set showed marginal association with a haplotype overlapping one significantly positively associated in the SS families at rs2289086–rs3791709, but the association with schizophrenia was in the opposite direction as in SS.

Haplotype rs3791709–rs4673628–rs2272024 (T-T-G) was found to be negatively associated with both verbal memory factor scores (global P=0.012, haplotype P=0.055) and digit span factor scores (global P=0.047, haplotype P=0.013) scores in healthy controls (Table 1). Digit span factor scores showed the strongest association with ErbB4. Minor allele homozygous genotype carriers at rs4673628 and rs2272024 were found to have poorer digit span scores than those with homozygous major allele genotypes (P=0.005 and 0.008, respectively), and the C-A haplotype at two 3-SNP windows containing these SNPs were positively associated with digit span scores (global P=0.047, haplotype P=0.047). This same pattern was observed for the next sliding haplotype window.

We report evidence for association between a single haplotype in ErbB4 and schizophrenia in two independent family samples. Our primary haplotype overlaps with a haplotype reported to show strong association with schizophrenia in persons of Ashkenazi Jewish decent² and contains two of the three SNPs reported by Silberberg et al. (2006) (rs839523–rs7598440). However, our positively associated alleles at rs839523 and rs7598440 are the opposite of those reported by Silberberg et al.² SNP allele frequencies did not vary between our cases and controls, HapMap Centre d'Etude du Polymorphisme Humain (CEPH) samples, and the cases from Silberberg et al.² Interestingly, the controls reported by Silberberg and co-workers have very different allele frequencies (Table 2), which may explain the discrepancy. We were not able to detect association between the Silberberg et al.² 3-SNP or 2-SNP haplotypes in our three family sets (data not shown), indicating that the additional SNP genotyped in our sample (rs839541) is necessary but not sufficient to detect association in our SS sample and the NIMHGI African-American sample. We were also able to show that ErbB4 haplotypes and genotypes influence selected cognitive functions in a sample of healthy controls, suggesting that these markers monitor a functional element that impacts on ErbB4 signaling and cell function.