Subjects and assessment

All patients were admitted to the Adolescent Clinic of the Academic Medical Center of the University of Amsterdam and attended a special program for adolescents with FEP. After complete description of the study to the subjects, written informed consent was obtained. The study was approved by the human subjects review board of our institution. Exclusion criteria were organic brain disease, endocrine disorder, or learning disability. Diagnosis was established according to DSM-IV [37] criteria and was based on the Mini International Neuropsychiatric Interview (MINI) [38]. At time of testing patients were clinically stable for at least four weeks (i.e. no changes in antipsychotic medication in the four weeks before testing). Additionally, healthy siblings were recruited, and unrelated healthy controls from the general population matched for age, educational level, and ethnicity. Both groups were screened for psychopathology using the MINI and were excluded when they had an Axis I disorder, organic brain disease, or learning disability. All subjects were administered the Wechsler Adult Intelligence Test – third version (WAIS III). Blood was collected from all subjects for DNA isolation. Where available, parents (of patients) were asked to provide blood for DNA isolation.

DNA extraction and genetic analysis

Genomic DNA was extracted using a filter-based method (QIAamp DNA Mini Kit, Qiagen Ltd, United Kingdom). Genotyping was performed using single base primer extension and analyzed by matrix-assisted laser-desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) on a Bruker III Daltonics Mass Spectrometer as described previously [39] (details available upon request). All DNA samples were genotyped in duplicate to ensure reliability.

Choice of markers

Statistical analysis

1. Demographic data

We included 76 patients with FEP, 31 healthy siblings, and 31 unrelated control subjects (see Table 3). Ethnicity counts are shown in Table 4. The patients met DSM-IV criteria for schizophrenia (N = 57; 75%), schizoaffective disorder (N = 13; 17%), brief psychotic disorder (N = 3; 4%), and psychotic disorder not otherwise specified (N = 3; 4%). Mean duration of psychosis was 20 months (SD = 15.06). Patients were receiving antipsychotic medication (80.2%), mostly atypical antipsychotics. Mean dose of antipsychotic medication in chlorpromazine equivalents was 291.06 (SD = 266.70). Six out of 76 patients (8%) were also diagnosed with mild to moderate depressive disorder and used an SSRI. None of the healthy siblings and none of the unrelated controls used any psychotropic medication. Means and numbers for age, educational level, gender, VIQ, PIQ, and FSIQ for cases, siblings and controls are presented in Table 3. All data were normally distributed. There was no difference between patients receiving typical and atypical neuroleptic treatment on IQ scores. We found no significant correlation between chlorpromazine equivalents and FSIQ (Pearson's Rho=-0.13; p = 0.3). Also, educational level did not differ significantly between patients, siblings and controls (p = 0.11). The number of males and females differed significantly between groups (p <0.01). Mean VIQ, PIQ, and FSIQ scores differed significantly (p <0.001) between patients, siblings, and controls (Figure 1). IQ scores in patients (mean VIQ = 89.2 ± 14.3, mean PIQ = 85.7 ± 13.0, mean FSIQ = 86.7 ± 13.8) were significantly lower than IQ scores in siblings (mean VIQ = 95.7 ± 11.1, mean PIQ = 98.6 ± 11.4, mean FSIQ = 96.7 ± 11.6) and controls (mean VIQ = 105.6 ± 16.9, mean PIQ = 106.4 ± 14.4, mean FSIQ = 107.0 ± 15.5). Post-hoc analyses showed that siblings had significantly lower IQ scores than controls (p = 0.005) and significantly higher IQ scores than patients (p = 0.001) (Figure 1).

	Patients N = 76	Siblings N = 31	Controls N = 31	Statistic	p-value
Age (SD)	21.5 (2.9)	22.8 (3.8)	21.5 (3.1)	F = 2.27	p = 0.11
Gender (M/F)	66/10	13/18	23/8	χ2 = 23.03	p <0.001
Educational level (SD)	4.45 (0.91)	4.76 (1.15)	4.83 (1.09)	F = 1.94	p = 0.15
VIQ (SD)	89.2 (14.3)	95.7 (11.1)	105.6 (16.9)	F = 14.68	p <0.001*
PIQ (SD)	85.7 (13.0)	98.6 (11.4)	106.4 (14.4)	F = 31.35	p <0.001*
FSIQ (SD)	86.7 (13.8)	96.7 (11.6)	107.0 (15.5)	F = 25.12	p <0.001*

Subject characteristics (age, gender, and educational level) and verbal, performance, and full scale IQ scores (VIQ, PIQ, and FSIQ). Between-group differences in age, educational level, and between VIQ, PIQ, and FSIQ are calculated using ANOVA. Between-group differences in gender are calculated using a χ² test. The p-values marked with * result from an overall ANOVA; all pairwise comparisons are significant.

	Subject
Country of origin	Patients	Siblings	Controls
Netherlands (white Caucasian)	48	25	22
Turkey	3	1	0
Morocco	4	0	4
Surinam or Dutch Antilles	17	3	5
Missing	4	2	0

Country of origin did not differ significantly between subject groups (Fisher's Exact = 9.44, p = 0.24).

2. Association analysis with single SNPs and marker combinations in cases, siblings, and unrelated controls

When we analyzed association between IQ scores and 2-, 3-, or 4-marker combinations, none of the marker combinations showed significant association with VIQ, PIQ, or FSIQ scores (data not shown). The multiple SNP models showed that a model including only rs760761 fitted the data equally well compared to a model including all 7 SNPs, suggesting that the observed differences may be caused by a single mutation.

3. QPDT results

We found significant association between FSIQ scores and 4 single SNPs in the DTNBP1 gene using the QPDT (rs 2619528: p = 0.044; rs 760761: p = 0.022; rs 2619522: p = 0.022; rs 2619538: p = 0.026). None of the 2-, 3-, or 4-marker combinations showed significant association with FSIQ scores. Three single SNPs showed significant association with VIQ (rs760761: p = 0.027; rs2619522: p = 0.027; rs2619538: p = 0.030). The 4 SNPs associated with FSIQ were also associated with PIQ (rs2619528: p = 0.033; rs760761: p = 0.023; rs2619522: p = 0.024; rs2619538: p = 0.023). Using UNPHASED, we found LD between the 4 SNPs associated with FSIQ to be high (D' ranging from 0.7–1.0), therefore we think the SNPs are part of an LD block and do not contribute to variance in FSIQ individually.

Previous work on DTNBP1 and cognition

Our data are partly consistent with a recent report of association between genetic variation in DTNBP1 and measures of prefrontal brain function in healthy adults [43]. In this study the polymorphisms rs2619528 and rs760761 showed strong association with prefrontal electrophysiology as measured by event-related potentials (ERPs) both individually and as part of a haplotype. In particular the GG variant of rs2619528 and the CC variant of rs760761 were associated with impaired prefrontal brain electrophysiology. In our study population, the rs760761 CC genotype was associated with better intellectual functioning. A possible explanation for this discrepancy is that genetic variation in DTNBP1 may differentially affect electrophysiology and IQ measures. Alternatively, effects of rs760761 on prefrontal brain electrophysiology in healthy controls may differ from those in patients with schizophrenia; therefore future studies should investigate whether associations between DTNBP1 and electrophysiological measures are also present in patients with schizophrenia.

Support for a role of DTNBP1 in cognition and in schizophrenia came first from Burdick [32] who recently reported association between a DTNBP1 haplotype and general cognitive ability, and from Donohoe et al [33] who reported association between a DTNBP1 haplotype and spatial working memory in patients with schizophrenia. Our data are consistent with these studies although we used a different approach. The risk haplotype used by Donohoe et al [33] was previously identified in same cohort by Williams et al [24]; similarly, the risk haplotype used by Burdick et al [32, 44] and by DeRosse et al [35] was previously identified in same sample by Funke et al [21]. We however tested SNPs in a hypothesis-free fashion. Also, the SNPs we typed did not include the tagging SNP (rs1018381) from the risk haplotype used in the sample of Funke et al [21]. And although we did type the three SNPs that formed the risk haplotype in the Donohoe et al [33] study, our study focused on general cognitive ability similar to Burdick et al [32], whereas Donohoe [33] investigated a specific cognitive process, namely spatial working memory. Therefore, results may be partly discrepant from previous studies due to differences in approach, cognitive measures, and genotyped SNPs.

Biological mechanism

Our data add to the growing evidence that genetic variation in the DTNBP1 gene modifies schizophrenia susceptibility by influencing general cognitive abilities. Not only is the 6p22–24 locus implicated in cognition and neurocognitive deficit in schizophrenia by recent genome scans [26, 27] and has DTNBP1 been associated with more severe negative symptoms which are highly correlated to cognitive dysfunction [34, 45], but also recent neuropathological studies have revealed reduced expression of DTNBP1 in the prefrontal cortex [28] and in the hippocampus [29] in postmortem brain tissue of schizophrenia patients. Both structures are involved in cognitive functions, the prefrontal cortex merely in working memory and executive function, the hippocampus mainly in memory formation. Also, risk haplotypes in the DTNBP1 gene have been associated with reduced expression of DTNBP1 mRNA in brain tissue of healthy subjects and in patients with schizophrenia, whereas 'protective' haplotypes were associated with high DTNBP1 expression [46]. Changes in DTNBP1 expression are probably not caused by antipsychotic drug treatment [47] but instead may reflect an increased genetic vulnerability for schizophrenia. Thus, genetic variation in the DTNBP1 gene may modulate schizophrenia susceptibility by altering DTNBP1 expression in areas of the brain crucial for cognitive functions. This may occur already in an early stage, since genetic variation in DTNBP1 has been associated with poor premorbid function in children developing childhood-onset schizophrenia [48], suggesting that DTNBP1 contributes to early neurodevelopmental impairment. A possible mechanism by which dysbindin contributes to schizophrenia susceptibility, may be by modulating neurotransmitter systems in regions of the brain that are crucial for cognitive functioning and that are probably disturbed in schizophrenia: the glutamatergic system in the hippocampal formation and the dopaminergic system in the prefrontal cortex.

Recently, Talbot [29] investigated expression and localization of DTNBP1 in postmortem brain tissue of schizophrenia patients and reported evidence for a relationship between DTNBP1 and hippocampal glutamate neurotransmission: DTNBP1 expression was reduced in terminal fields of intrinsic glutamatergic connections and these reductions were related to glutamatergic alterations in intrinsic hippocampal formation connections [29]. Glutamatergic neurotransmission in the hippocampus is important for memory formation [49], therefore reductions in DTNBP1 in the hippocampus of schizophrenia patients as observed by Talbot [29] may disturb normal information processing and diminish synaptic plasticity in the hippocampus. The hippocampus has been implicated in schizophrenia pathogenesis previously, both by structural neuroimaging studies reporting reductions of hippocampal size, and by neuropathological studies reporting abnormal clustering of neurons [50, 51].

Another possible mechanism by which DTNBP1 could modulate schizophrenia susceptibility, is by modulating prefrontal cortex (PFC) function. Reduced DTNBP1 expression has been reported in the dorsolateral PFC (DLPFC) in postmortem brain tissue of patients with schizophrenia [28]. Disturbed development of the DLPFC probably underlies cognitive dysfunction in schizophrenia and may contribute to schizophrenia pathogenesis [52]. Recently Kumamoto et al [30] reported that DTNBP1 also may be involved in the dopaminergic system, and may be capable of activating the midbrain dopaminergic system via indirect pathways [30]. Dopaminergic systems in the PFC and midbrain form a feedback mechanism and are modulated by genetic variation in dopamine clearance pathways; tuning of these systems is of critical importance for optimal cognitive functioning [53–56]. Thus, genetic variation in DTNBP1 may impact on dopaminergic and glutamatergic systems in the PFC and hippocampus, thereby disturbing memory and cognition and increasing schizophrenia susceptibility.

Strengths and limitations

Strength of our study is the relatively young age of the patients (mean age was 21 years in our sample versus approximately 40 years in the sample of Burdick et al [32]), therefore potential confounders, e.g. disease duration or chronic antipsychotic treatment, are unlikely to influence IQ measurements. Limitations of our study are small sample size and the use of an ethnically mixed sample. To reduce the risk for population stratification, we (1) adjusted for ethnicity in the case-control analyses; (2) used unaffected siblings as controls; (3) matched the unrelated controls for ethnicity; and (4) used additional family-based analyses which confirmed results of the case-control analyses. Therefore, we think it is unlikely our results are due to admixture. Multiple testing effects could also allow the occurrence of type I errors. Because we expect the contribution of individual genes to the risk of complex genetic disorders to be small, and because the tests we performed are not completely independent (LD between SNPs is high and VIQ, PIQ, and FSIQ are highly correlated) we consider the Bonferroni method too strict to correct for type I error. However, we acknowledge that our results would not survive other – less stringent – methods to correct for multiple testing. Therefore our findings are preliminary and need to be replicated in a large and genetically homogeneous sample. Also, most of our included patients are male; therefore results may not generalize to female patients. Finally, although we found no correlation between chlorpromazine equivalents and FSIQ, the use of antipsychotic medication is a potential confounding factor in our study since all patients used antipsychotics whereas neither the siblings nor the unrelated controls used any psychotropic medication.

Future directions

Future research will need to further investigate the complex genetic background of cognitive phenotypes in schizophrenia, including involvement of multiple susceptibility genes and gene-environmental interactions. Additionally, the use of combined techniques for measuring cognitive phenotypes, e.g. acquiring functional brain images during neuropsychological testing, may improve reliability and resolution of cognitive phenotypes in schizophrenia, thereby increasing power for molecular genetic studies.