The contribution of three strong candidate schizophrenia susceptibility genes in demographically distinct populations


*M. Karayiorgou, Human Neurogenetics Laboratory, The Rockefeller University, 1230 York Avenue, Box #313, New York, NY 10021, USA. E-mail:


Here we characterize and compare the contribution of three recently identified strong candidate schizophrenia susceptibility genes; G72, neuregulin 1 (NRG1) and dystrobrevin-binding protein 1 (DTNBP1) in two independent datasets of patients with distinct genetic backgrounds. On the basis of corrected P-values from single- and multilocus transmission distortion tests our analysis provides no support for a contribution of G72, NRG1 or DTNBP1 in the tested samples. When transmission of individual haplotypes was considered, a picture more consistent with the original studies emerged, where transmission distortions in the same direction as the original samples and involving the same core haplotypes were observed for G72 and NRG1. Interestingly, whereas the NRG1 gene analysis was dominated by the presence of over-transmitted haplotypes, the G72 gene analysis was consistently dominated in both datasets by under-transmissions. Negative transmissions involved a core haplotype complementary to the originally detected over-transmitted haplotype, suggesting the presence of a protective variant within the G72 locus.

Schizophrenia [MIM 181500] is a severe psychiatric disorder with a lifetime prevalence of ∼1% in most studied populations. Similar to many common, complex disorders, schizophrenia is a multifactorial disorder characterized by the contribution of several susceptibility genes, which may act in conjunction with epigenetic processes and environmental factors. Over the past few years the field of schizophrenia genetics has moved from the analysis of candidate genes to systematic efforts of positionally cloning susceptibility genes from chromosomal regions first identified by linkage approaches.

These systematic positional cloning efforts involving relatively large numbers of markers resulted in the identification of strong positional candidate genes from 6p (Straub et  al. 2002), 8p (Stefansson et al. 2002a), 13q (Chumakov et al. 2002) and 22q11 (Liu et al. 2002a; Liu et al. 2002b). Support for all of these findings is rather strong, in terms of the degree of statistical significance and the reproducibility of the associations in the original studies, the identification of independent rare risk alleles (Jacquet et al. 2002; Liu et al. 2002a) and the consistent findings from animal models studies (Gogos et al. 1999; Stefansson et al. 2002a).

We present here a first attempt to characterize and compare the contribution of three of these recently proposed strong candidate genes; G72 on 13q [MIM607408] (Chumakov et al. 2002), NRG1 on 8p [MIM142445] (Stefansson et al. 2002a) and DTNBP1 on 6p [MIM607145] (Straub et al. 2002), in two independent datasets of schizophrenic patients with distinct demographic characteristics.

The G72 gene is located within a 65-kb region where strong evidence of association with schizophrenia was reported (Chumakov et al. 2002). This region is within a larger area extending from 13q32 to q34, where prior evidence of linkage to schizophrenia and bipolar disorder exists (Badner & Gershon 2002; Blouin et al. 1998; Brzustowicz et al. 1999; Detera-Wadleigh et al. 1999; Liu et al. 2001). Significant association with schizophrenia was found with several SNPs and haplotypes at this locus in a case-control study analyzing French-Canadian samples (213 cases/241 controls), with the association of two SNPs being replicated in a Russian cohort (183 cases/183 controls). Expression and functional studies pointed to the G72 gene as the best candidate from this region, and suggested a potential interaction with D-amino-acid oxidase and modulation of its enzymatic activity. Interestingly, a subsequent study provided evidence for an association between variants at the G72 locus and bipolar disorder (Hattori et al. 2003).

The neuregulin 1 (NRG1) gene encoding for a protein involved in a wide variety of neuronal functions, ranging from neuronal survival to myelination and synaptic plasticity, was identified as a susceptibility gene for schizophrenia through a genome-wide linkage scan of 33 large extended families with schizophrenia that pointed to a locus on chromosome 8p (Stefansson et al. 2002a). Fine-mapping of the 8p locus along with haplotype association analysis of 478 patients with schizophrenia and 394 control individuals narrowed the region of interest to the 5′ end of the NRG1 gene. A core haplotype at the 5′ end of the gene made up of several markers within a 290-kb block of linkage disequilibrium showed highly significant association with an estimated relative risk of 2.2. In a subsequent study the same group replicated the association in 609 unrelated Scottish patients (Stefansson et al. 2002b).

Straub et al. (2002) reported association to schizophrenia of genetic variants in the gene for the dystrobrevin-binding protein 1 (DTNBP1), dysbindin, which is located within a broad region on chromosome 6p showing evidence for linkage in Irish families (Straub et al. 1995). A subsequent study replicated the association in 203 families from Germany and Israel (Schwab et al. 2003), although with a different and common haplotype. Surprisingly, one case/control study testing an independent sample drawn from the same original Irish population using the published variants failed to confirm the association of the DTNBP1 gene to schizophrenia (Morris et al. 2003). DTNBP1 is localized to presynaptic terminals, and may participate in the formation and maintenance of synapses (Benson et al. 2001).

Materials and methods

Family samples

USA sample

This family sample includes 210 triads from the United States recruited through advertisements in advocate newspapers (85%) or from area day treatment clinics (15%). The sample is Caucasian of European origin. The study was approved by the Rockefeller University Institutional Review Board (IRB) and all participants signed written informed consent. Participants were interviewed by specially trained clinicians with a minimum of 3 years of doctoral-level clinical experience who used the Diagnostic Instrument for Genetic Studies (DIGS) (Nurnberger et al. 1994). On the basis of information gathered in the DIGS, the clinical interviewers assigned appropriate diagnoses according to the Diagnostic and Statistical Manual, 4th Edition (DSM-IV) (American Psychiatric Association 1994). In a reliability study conducted by our five trained study interviewers, diagnostic agreement for DSM-IV schizophrenia was 97% and for DSM-IV schizoaffective disorder 92%.

SA Sample

This family sample includes 233 families from the genetically homogeneous Afrikaner population of South Africa. The Afrikaners are Caucasian and descend from approximately 2000 initial settlers from the Netherlands and other parts of Northern Europe who arrived at South Africa in consecutive waves, starting at 1652. These families were collected for genetic studies in a collaborative effort between the Laboratory of Human Neurogenetics at the Rockefeller University and two major psychiatric hospitals in South Africa (Weskoppies Hospital in Pretoria and Valkenberg Hospital in Cape Town). All subjects signed written informed consent and the study was approved by the IRBs at all participating sites (Rockefeller University, University of Pretoria and University of Cape Town). The DIGS, after it was translated and back translated into Afrikaans, was the diagnostic instrument of choice. All diagnostic interviews were conducted in person by specially trained clinicians. Specifically, two psychiatrists and one clinical psychologist with a minimum of 10 years clinical experience each were specially trained in the use of the DIGS and the research application of DSM-IV. On the basis of information gathered in the DIGS, the clinical interviewers assigned appropriate diagnoses according to the DSM-IV.

Ongoing reliability studies between the Afrikaner clinicians and the clinicians who conduct the diagnostic evaluations for our USA sample (see above) show >90% agreement for DSM-IV diagnoses. Reliability exercises consist of yearly reliability interviews between all the clinicians, as well as by review of videotapes of an additional two interviews per interviewer. A clinical comparison of cases of schizophrenia diagnosed in South Africa and the USA showed that with regard to basic sample descriptors and cardinal symptoms of disease, the two populations are equivalent (Karayiorgou et al. 2004).

For the statistical analysis of the SA sample we considered both a ‘SA trios’ subsample composed by family triads only (n = 169 families) and a ‘SA total’ sample, which in addition to the triads includes 64 extended families with at least two affected individuals (n = 233 families). All probands in the ‘SA trios’ subsample fall into the narrowly defined phenotypic Liability Class (LC) I, which comprises the ‘core schizophrenic phenotypes’ of schizophrenia and schizoaffective disorder, depressed type only [family data suggest that the two diagnoses are alternative expressions of the same genotypes (Cloninger 1989)]. In the ‘SA total’ sample, 64 additional extended families with 98 individuals who meet the strict diagnostic criteria of LC I are included. An additional 128 affected relatives in these families are considered in a separate LC, LC III. This group includes 59 relatives with schizoaffective disorder of affective course, as well as 69 relatives with other psychiatric disorders. Consideration of LC III accommodates the hypothesis that genes influencing susceptibility to schizophrenia may have pleiotropic effects (Collier & Sham 1997), may lead to other psychiatric disorders in the relatives of schizophrenic patients and the clinical manifestation of these genes may depend on interactions with the environment, other genes and/or stochastic factors.


SNPs were typed either by PCR-restriction fragment length polymorphism genotyping or by fluorescence polarization template-directed dye-terminator incorporation genotyping (FP-TDI) (Chen et al. 1999) (for SNPs M12, M15, M25, P1655, SNP8NRG221533). All primers are available on request.

Samples were digested for 3 h with the enzymes HinfI (M7), EcoNI (M13), MfeI (M14), AflIII (M23), BsaAI (hcv7460562), RsaI (rs1040410), HpyCH4IV (P1765), FokI (P1763), RsaI (SNP8NRG241930), HpyCH4IV (SNP8NRG243177). For both microsatellite markers 478B14848 and 420M91395 the reverse primer was ‘PIG-tailed’ (i.e. the sequence GTTTCTT was added on the 5′ end) to facilitate accurate genotyping (Brownstein et al. 1996). Gel electrophoresis was carried out on an ABI 3700 automated DNA sequencer (Perkin Elmer, Foster City, CA) and genotypes were assigned using the Genotyper v2.5 software (Perkin Elmer). Genotyping results were examined for unlikely recombinations and within-family incompatibilities by the association test program TRANSMIT v2.5.2 (Clayton 1999). Data points of Mendelian inconsistencies or recombinations were removed. Across all genotypes from all three genes we detected 0.04% Mendelian inconsistencies. Genotyping error according to duplicate genotyping for 24 samples included for internal quality checks with each 384-well plate was 0.5%.

Statistical analysis

We performed single-locus and multi–locus haplotype association analysis using two programs, pdtphase v2.33 (Dudbridge 2002; Dudbridge 2003) a test of linkage disequilibrium in general pedigrees, and transmit v2.5.2 (Clayton 1999) a test of linkage disequilibrium in nuclear families. pdtphase v2.3 includes extensions to deal with haplotypes and missing data of the program pdt (Martin et al. 2000). Maximum-likelihood gametic frequencies under the null are calculated using an EM algorithm.

transmit uses a score test based upon the CPC (Conditional on Parental Genotype) likelihood for the estimate of unknown haplotype phase and missing data. With this program the analysis was restricted to one affected individual from each extended family (the index case) in order to perform a valid test for association.

In all analyses, we retained only those haplotypes that are common (more than 3% frequency in our sample) to avoid application of statistics to a small number of chromosomes.


We genotyped 17 markers across the 3 candidate genes, in two large family collections with distinct genetic backgrounds, expected to offer different advantages in association testing. One collection of families is from the outbred population of the USA and the second is from a South-African founder population of European descent, the Afrikaners (SA).

In each gene, we selected those loci that capture most of the associations reported previously. Of the 17 markers, 7 are close to or at the G72 locus (Chumakov et al. 2002), 5 are located in the DTNBP1 gene (Straub et al. 2002) and 5 in the NRG1 gene (Stefansson et al. 2002a) (Table 1). The linkage disequilibrium relationship among the tested loci is indicated in Table 2.

Table 1.  Single-locus and multi–locus association analysis
     USA trios
(n = 210)
SA trios
(n = 169)
SA total LC I/LC III
(n = 233)
  • *

    Distance in Kb between consecutive markers was estimated from UCSC.

  • Minor allele frequency in the SA sample. Allele frequency differences between the SA sample and the USA sample don't exceed 0.05%.

  • Sizes and frequencies of the common alleles (≥ 3%) of the microsatellite marker 478B14848 are: 217 = 0.12; 219 = 0.36; 221 = 0.16; 223 = 0.18; 225 = 0.17.

  • §

    Sizes and frequencies of the common alleles (≥ 3%) of the microsatellite marker 420M91395 are: 272 = 0.22; 274 = 0.44; 276 = 0.24.

G72M7104.6G/A0.39 (G)0.187      0.534      0.553/0.449      
      0.444      0.121      0.168/0.162     
 M127.4A/G0.39 (A)0.702 0.584    0.060 0.147    0.062/0.055 0.202/0.157    
      0.658 0.410    0.070 0.147    0.069/0.074 0.236/0.169   
 M136.6A/C0.40 (C)0.463 0.421 0.243  0.055 0.070 0.067  0.079/0.106 0.081/0.047 0.144/0.090  
      0.352 0.497 0.258  0.052 0.043 0.226  0.083/0.083 0.058/0.025 0.341/0.246 
 M141.9A/G0.39 (G)0.553 0.369 0.397 0.4850.053 0.022 0.168 0.4120.081/0.076 0.033/0.015 0.154/0.065 0.540/0.334
      0.503 0.427 0.708  0.059 0.091 0.399  0.092/0.054 0.087/0.035 0.364/0.280 
 M1566.3A/G0.39 (A)0.938 0.531 0.870  0.045 0.201 0.184  0.060/0.055 0.242/0.121 0.150/0.110  
      0.576 0.919    0.201 0.276    0.184/0.123 0.229/0.174   
 M2360.6T/C0.47 (C)0.261 0.835    0.525 0.276    0.603/0.555 0.180/0.163    
      0.406      0.776      0.686/0.687     
 M25–-G/T0.37 (T)0.542      0.902      0.633/0.549      
DTNBP1hcv746056224.6T/C0.14 (C)0.918      0.766      0.963/0.748      
      0.798      0.732      0.578/626     
 rs104041073.1T/C0.08 (T)0.541 0.985    0.327 0.949    0.128/0.223 0.807/0.824    
      0.584 0.942    0.644 0.669    0.249/0.420 0.810/0.871   
 P165529.2C/G0.50 (C)0.937 0.750 0.856  0.837 0.469 0.555  1.00/0.798 0.572/0.669 0.496/0.538  
      0.992 0.668    0.691 0.607    0.685/0.743 0.269/0.286   
 P17653.8A/G0.20 (A)1.00 0.843    0.398 0.587    0.316/0.401 0.535/0.513    
      0.452      0.294      0.193/0.205     
 P1763–-G/T0.21 (G)0.855      0.270      0.107/0.178      
NRG122153320.2T/C0.35 (C)0.553      0.569      0.715/0.715      
      0.527      0.729      0.848/0.845     
 2419301.2T/G0.32 (T)0.586 0.207    0.441 0.794    0.893/1.00 0.885/0.983    
      0.233 0.811    0.566 0.214    0.677/0.843 0.642/0.365   
 24317793.4T/C0.40 (T)0.443 0.936 0.414  0.691 0.120 0.095  0.703/0.842 0.573/0.342 0.170/0.120  
      0.497 0.619    0.258 0.028    0.057/0.374 0.082/0.033   
 478B1484828.0(CA)n–-0.808 0.714    0.754 0.103    0.091/0.782 0.320/0.130    
      0.707      0.101      0.253/0.329     
 420M91395–-(CA)n§–-0.310      0.931      0.078/0.093      
Table 2.  Background LD for the markers tested
GeneLocusD′ values*
  • *

    D′ values for untransmitted chromosomes of nuclear families were calculated with the program UNPHASED v2.33 (Dudbridge 2003).

  • D′ values from the USA sample are above the diagonal, D′ values from the SA sample are below the diagonal.


We analyzed the data using two statistical programs, the pdt (Martin et al. 2000) in the version pdtphase v2.33 (Dudbridge 2002; Dudbridge 2003), which includes extensions to deal with haplotypes and missing data, and the program transmit v2.5.2 (Clayton 1999). Both of these programs are family-based tests of association, for single- and multilocus analysis. Here, we present the data only from the pdt test (Table 1), which unlike transmit uses the entire dataset and is therefore more powerful. However, results obtained using the program transmit were qualitatively very similar. To increase the power of the analysis and facilitate comparisons to previous reports, in addition to single markers we also tested successive sets of adjacent markers. pdtphase v2.33 calculates P-values at 1df, comparing a given haplotype with all other haplotypes combined. Therefore, in addition to the global test, we also considered transmission distortions of specific haplotypes at 1 df. In Fig. 1, we report the -log(P-value) of each haplotype as provided by pdt. While none of the P-values withstands correction for multiple testing, this representation allows us to rank the strength of the association and at the same time visualize over- vs. under-transmission distortions and compare them with previous studies. The value of such comparisons even in the absence of formal statistical significance has been demonstrated previously (Altshuler et al. 2000). The allelic composition of representative haplotypes along with the associated relative risk is presented in Table 3.

Figure 1.

Blue label indicates under-transmitted haplotypes; orange label indicates over-transmitted haplotypes. In the ‘SA total’ sample, where two different liability classes (LCs) are used, solid lines with filled circles indicate results obtained using LC I, while dashed lines with empty circles indicate results obtained using LC III. (a)  G72 haplotypes: all blue labeled haplotypes in the middle and right panels are variable sizes of the haplotype AACGATT. Haplotype A = GACGAT, haplotype B = GACGATG. (b)  NRG1 haplotypes: the over-transmitted orange labeled haplotypes are variable sizes of the haplotype YGT-219–274, the under-transmitted blue labeled haplotypes are variable sizes of the haplotype CGT-225. Haplotype A = GC-219–274, haplotype B = TC-223–276, haplotype C = ATC-223–276.

Table 3.  Odds ratios and frequencies of haplotypes showing evidence for transmission distortion
   M7M12M13M14M15M23M25OR [95%CI]*Frequency
  • *

    Odds ratios calculated using one affected per family.

  • Frequency in under-transmitted chromosomes.

  • ‡ 

    SA total sample LC I.

G72SAUnder-trans.AACGATT0.42 [0.19–0.92]0.08
    ACGATT0.47 [0.25–0.88]0.13
      GATT0.41 [0.22–0.77]0.13
       ATT0.45 [0.25–0.83]0.13
 USAUnder-trans.GACGATG0.32 [0.13–0.82]0.05
   GACGAT 0.40 [0.22–0.73]0.11
NRG1SAOver-trans. GT219274  1.77 [0.98–3.22]0.09
     T219274  1.71 [1.00–2.92]0.10
      219274  1.35 [0.95–1.92]0.27
 USAUnder-trans.CGT    0.71 [0.52–0.99]0.35

G72 gene results

With the exception of locus M15 (global P = 0.045) in the SA sample, no single-locus analysis provided nominal P-values <0.05 (Table 1). Multi-locus analysis in the SA trios sample resulted in two tests with nominal P-values <0.05 (M13-M14-M15 P = 0.022; M12-M13-M14-M15 P = 0.043). The addition of extended SA families to the SA trios sample (SA total), resulted in an increase of the number of tests with P-values <0.05 when LC III was considered (M12-M13-M14 P = 0.047; M13-M14-M15 P = 0.015; M12-M13-M14-M15 P = 0.025; M13-M14-M15-M23-M25 P = 0.035). By contrast, when LC I was considered, only one test with P-values <0.05 was observed (M13-M14-M15 P = 0.033).

When we analyzed transmission of individual haplotypes, a picture more consistent with the original study emerges where transmission distortions in the same direction as the original samples and involving the same core haplotypes are observed in both USA and SA samples. Specifically, the allelic composition of multilocus haplotypes showing transmission disequilibrium at the G72 gene in both SA samples defines one consistently under-transmitted haplotype AACGATT (M7-M12-M13-M14-M15-M23-M25) (Fig. 1, Table 3), which appears at variable lengths according to the combinations of loci tested [SA trios: M13-M14-M15-M23-M25; SA total sample LCI/III: M15-M23-M25]. In the USA sample, two single haplotypes are under-transmitted (M7-M12-M13-M14-M15-M23; M7-M12-M13-M14-M15-M23-M25) with an allelic composition almost identical to that observed in SA except for the first (M7) and last (M25) locus (Fig. 1a, haplotypes labeled A, B). Interestingly, in the report of Chumakov et al. (2002) the multilocus analysis showed positive transmissions of a haplotype complementary to the one observed in our SA samples, with the exceptions of the alleles at locus M7 and M23. This may suggest the conservation across ethnic groups of a core risk/protective-haplotype defined by the loci M12-M13-M14-M15 that is in strong linkage disequilibrium in both populations (Table 2). Further analysis will be necessary to confirm this.

NRG1 gene results

No single-locus analysis provided nominal P-values <0.05 across the five tested NRG1 loci (Table 1). Multi-locus analysis, however, resulted in tests with nominal P-values <0.05 in the SA samples, but not the USA sample. In particular, in the SA trios, one multilocus test showed P-value <0.05 (SNP8NRG241930-SNP8NRG243177–478B14848–420M91395 P = 0.028). Unlike the G72 gene, addition of extended SA families to the SA trios sample did not result in an increase of tests with P-values <0.05.

Analysis of transmission of individual haplotypes revealed distortions in the same direction as the original samples involving the same core haplotypes, but only in the SA sample. Specifically, analysis of transmission of individual haplotypes detected variable lengths of one unique over-transmitted haplotype YGT-219–274 (Fig. 1b, Table 3), that often includes the two microsatellite markers. Interestingly, this allelic composition reflects that of haplotypes over-transmitted in the original studies (Stefansson et al. 2002a; Stefansson et al. 2002b). The best result was observed in the SA total sample with LC I (SNP8NRG243177–478B14848–420M91395) (Fig. 1, Table 3). Only one under-transmitted haplotype GC-219–274 was observed in the SA trios (SNP8NRG221533-SNP8NRG241930-SNP8NRG243177–478 B14848). A smaller portion of it represents the only weak transmission distortion in the USA sample (SNP8NRG 221533-SNP8NRG241930-SNP8NRG243177).

DTNBP1 gene results

No single-locus or multilocus analysis showed evidence for association to the disease across the five tested DTNBP1 loci in either the USA or the SA samples. In addition, unlike G72 and NRG1, we did not identify any single haplotype showing evidence of transmission disequilibrium, according to the criteria applied to the other two genes.


On the basis of corrected P-values from single- and multilocus transmission distortion tests, our analysis provides no support for a contribution of G72, NRG1 or DTNBP1 in the tested samples. Given that the total number of disease/control chromosomes examined in this and previous reports are comparable, our inability to achieve statistical significance (after correction for multiple testing) most likely results from the overall effect of alleles being modest and variable in different demographic groups, requiring larger datasets in our populations for a conclusive demonstration. When we considered transmission of individual haplotypes, a picture more consistent with the original studies emerged where transmission distortions in the same direction as the original samples and involving the same core haplotypes were observed in one (for the NRG1 gene), or both (for the G72 gene) tested samples. Interestingly, whereas the NRG1 gene analysis was dominated by the presence of over-transmitted haplotypes, the G72 gene analysis was dominated by primarily negative transmissions, involving a core haplotype complementary to the previously detected over-transmitted haplotype in the French-Canadian population (Chumakov et al. 2002). This unexpected result implies that the under-transmitted haplotype may contain a variant that is protective against the disease, and the previously described over-transmissions represent, at least partially, compensatory transmissions. It is conceivable that this ‘protective variant’ counteracts the effect of a susceptibility allele at another locus, which is particularly frequent or penetrant in the SA population. If, indeed, the G72 gene positively regulates the levels of D-serine in the brain (Chumakov et al. 2002) then the existence of a protective variant of the G72 gene (that would result in higher levels of this neuromodulatory aminoacid) is consistent with observations that administration of D-serine improves cognition and decreases negative symptoms in schizophrenic subjects when combined with typical antipsychotics (Tsai et al. 1998).

The results obtained for the DTNBP1 gene may be due to a complex pattern of allelic associations that are not sufficiently captured by the SNPs employed in the present study. This is consistent with the observation that different haplotypes from this gene seem to be contributing to different populations (Schwab et al. 2003). Alternatively, susceptibility alleles in this gene may not play a large role in the etiology of schizophrenia in the samples we tested. It has been suggested that genetic variation in the DTNBP1 gene is particularly involved in the development of schizophrenia in cases with a familial loading of the disease. Specifically, Van Den Bogaert et al. (2003) showed that stratification of their samples according to ‘positive family history’ (defined as having at least one first- or second-degree relative with schizophrenia) resulted in significant associations between DTNBP1 haplotypes and the disease in one out of three case/control samples they analyzed. Because the portion of cases with genetic loading may vary between samples, we performed an identical stratification of our samples that resulted in n = 47 and n = 74 cases with family history of schizophrenia in the USA and SA sample, respectively. However, analysis of these genetically loaded sample subsets did not provide any evidence of association either (data not shown).

Overall, our results showed somewhat stronger effects in the SA sample compared to the USA sample. This may reflect different causative genes segregating in various demographic groups. It may also be a consequence of the higher degree of genetic homogeneity/identity by descent in the founder population of the Afrikaners (Abecasis et al. 2004; Hall et al. 2002; Karayiorgou et al. 2004), which makes these families more informative for linkage disequilibrium-based tests when using similar numbers of patients. The degree to which clinical heterogeneity accounts for the differences observed among the two datasets we tested and the original samples also needs to be explored in future studies.

Electronic database information

Online Mendelian Inheritance in Man (OMIM),
Schizophrenia [MIM 181500];
G72 gene [MIM 607408];
Neuregulin 1 gene (NRG1) [MIM 142445];
Dystrobrevin-binding protein 1 or dysbindin gene (DTNBP1) [MIM 607145].


The authors thank all the families who participated to the study. Also, Sandra Demars for help with the genotypings. Support was provided in part by the New York Council Speaker's Fund (JAG) and NIH MH61399 (MK). Clinical work at the Rockefeller University Hospital GCRC was supported by grant M01 RR00102.