Identification in 2 Independent Samples of a Novel Schizophrenia RiskHaplotype of the Dystrobrevin Binding Protein Gene (DTNBP1) FREE

Schizophrenia is a common disorder with a lifetime morbidity risk of1%, more if spectrum disorders are included.¹ Thereis a large genetic epidemiological literature showing that individual differencesin liability are largely genetic, the heritability is approximately 80%,² and it is a complex genetic disorder, although theprecise mode of inheritance is unknown.² Morethan 20 genome-wide linkage scans have revealed several promising linkagefindings,³ and of these, one of the best-supportedregions is 6p24-22.^4- 8 Toidentify the specific gene(s) responsible, Straub and colleagues⁹ haverecently undertaken detailed linkage disequilibrium (LD) across the linkedregions of 6p22 in their sample of Irish families in whom linkage was initiallyobserved. Significant association was found between schizophrenia and severalindividual markers and haplotypes (DNA sequence defined by multiple polymorphicsites) across the gene encoding dystrobrevin binding protein (DTNBP1).¹⁰ Although the findings suggested DTNBP1 as a susceptibility gene for schizophrenia, Strauband colleagues were unable to identify the specific susceptibility variantsin the original analysis or in a reanalysis of the same sample based on extramarker information.¹¹

A follow-up study of DTNBP1 was performed basedon 78 German and Israeli families who showed evidence for linkage to 6p and127 proband-parent trios, mainly from Germany but including a small numberof subjects from Hungary.¹² Once again, evidencefor association was obtained with individual markers, a finding that was strengthenedby analysis of haplotypes. Although the same haplotype was independently associatedin each of the 2 samples (linkage families and trios), it was defined by themost common allele at each of the 6 markers examined, a finding that contrastswith the earlier study¹⁰ in which the riskhaplotype was defined by the least common allele at all loci. Nevertheless,taken with the initial report, the data are consistent with the hypothesisthat DTNBP1 is a susceptibility gene for schizophrenia.

Both of the previous studies were based on a family-based associationdesign. Here, association is detected by finding that a marker or haplotypeis transmitted by parents to their affected offspring more often than wouldbe expected by chance (0.5). Rarely (for example, if particular combinationsof markers are not favorable to survival), markers may show segregation distortionin family-based studies that relates to some aspect of ascertainment unrelatedto the phenotype (in this example, simply being alive) rather than true association.The authors of the first study reported that this was not a likely explanationfor their data, because no excess transmission was noted for one of the markersto unaffected offspring, but it would be reassuring to find additional evidencefor association in case-control study designs in which segregation distortiondoes not apply. However, the first published case-control study based on 219Irish cases and 231 Irish controls failed to find evidence for associationbetween DTNBP1 and schizophrenia.¹³ Failuresto replicate are to be expected for loci of small effect, especially whenthe individual susceptibility variants are unknown, and single reports ofthis nature do not amount to rejection of the hypothesis. Nevertheless, furthersupport, preferably including case-control data, is required to make the casefor DTNBP1 as a susceptibility gene for schizophreniaunassailable.

In this study, we have used a large case-control sample (708 cases and711 controls) to seek confirmation of association between DTNBP1 and schizophrenia by typing the most informative set of markersreported by Straub and colleagues. In addition, we attempted to identify aspecific nucleotide variant or variants that confer susceptibility by screeningall known and predicted exons of DTNBP1 for variantsand have tested all the single nucleotide polymorphisms (SNPs) we identifiedfor association with schizophrenia using a DNA pooling approach.¹⁴ Moreover,following up on indirect evidence that there are polymorphisms that influence DTNBP1 expression,¹⁵ wehave screened the putative promoter regions of DTNBP1 forsequence variants and tested these for evidence for association. We foundno evidence for association between schizophrenia and the markers studiedpreviously^10,12 and also failedto replicate either of the reported findings with regard to the specific riskhaplotypes. However, when we included novel variants in our analysis, we obtainedsuggestive evidence for single marker association, and when these were addedto some of the previous markers to construct haplotypes, we obtained highlysignificant evidence for association. When the critical marker that was requiredto make our sample informative was typed in a sample that previously did notshow evidence for association with DTNBP1,¹³ the specific haplotypes that we observed to be significantlymore and less common in cases were again significantly associated with schizophrenia.Our data provide compelling evidence that DTNBP1 isindeed a susceptibility gene for schizophrenia; identify for the first time,to our knowledge, haplotypes that replicate across independent samples; andstrongly reject the possibility of segregation distortion confounding theprevious family-based studies. Finally, exploration of the relationship betweengenetic variation in DTNBP1 and the phenotype ofschizophrenia suggests that the common protective haplotype is associatedwith higher educational attainment. The latter is a crude index of generalintelligence (which was not directly assessed), and we therefore postulatethat variation DTNBP1 may modulate risk of schizophreniaby influencing broadly defined cognitive ability.

The Cardiff, Wales, case-control sample consisted of 708 subjects withschizophrenia from the United Kingdom and Ireland (478 males and 221 females)matched for age, sex, and ethnicity to 711 blood donor controls (478 malesand 233 females). Of the cases, 141 had at least 1 affected first-degree relativewhose diagnosis had been confirmed by identical methods to that of the proband.The samples of familial cases (n = 141; 47 females and 94 males; mean ±SD age, 49.1 ± 12.9 years) were all ascertained for a sib-pair linkagestudy,¹⁶ which revealed no evidence for linkageto 6p,¹⁷ with analyses in the region yieldingmaximum logarithm of odds (LOD) scores of less than 0.4.¹⁸ Allpatients had a consensus diagnosis of schizophrenia according to DSM-IV criteria made by 2 independent raters following a semistructuredinterview by trained psychiatrists or psychologists using the Schedules forClinical Assessment in Neuropsychiatry interview¹⁹ andreview of case records. The operational criteria checklist (OPCRIT) and globalassessment scale (GAS) were also completed.^20- 21 Highlevels of reliability (κ>0.8) were achieved between raters for diagnosesand rating scale items. The mean ± SD age at first psychiatric contactfor the sample was 23.6 ± 7.7 years, and the mean age at ascertainmentwas 41.8 ± 13.5 years.

The Dublin, Ireland, case-control sample consisted of 219 cases and231 controls from the Republic of Ireland. All cases were interviewed by apsychiatrist or psychiatric nurse trained to use the Structured Clinical Interviewfor DSM. Diagnosis was based on DSM-III-R criteria using all available information (interview, familyor staff report, and medical record review). All cases were older than 18years, were of Irish origin, and met criteria for schizophrenia or schizoaffectivedisorder. The control sample, obtained from Irish blood donors, was not specificallyscreened for psychiatric illness, but individuals were not taking regularlyprescribed medications. In neither country are blood donors remunerated evenfor expenses. The mean ± SD age of the Irish sample was 46.0 ±8.5 years: 45.1 ± 13.1 years and 46.9 ± 10.4 years for malesand females, respectively. Mean age at first psychiatric contact for caseswas 25.2 ± 12.4 years.

For both samples, all cases were screened to exclude substance-inducedpsychotic disorder or psychosis due to a general medical condition. Ethicscommittee approval was obtained in all regions where patients were recruited,and informed written consent was obtained from all participants.

Phenotype analysis was performed using the Cardiff sample for the maincommon risk and protective haplotypes. We examined (1) age of onset, definedas the age at which psychiatric help for psychotic symptoms was first sought;(2) severity, defined by worst GAS score; (3) course of illness (ranging fromsingle episode with full recovery to continuous illness with gradual deterioration);and (4) symptom dimensions obtained by factor analysis of OPCRIT psychosisitems. As previously observed in a smaller sample,²² principalcomponents analysis of the OPCRIT psychosis items using direct oblimin rotationresulted in 4 factors being extracted after examination of both eigenvaluesand the scree plot. The 4 factors correspond to OPCRIT items regarding paranoiddelusions and hallucinations, disorganized symptoms, negative symptoms, andfirst-rank delusions. Given that low cognitive ability is a risk factor forschizophrenia,²³ we also investigated for associationwith level of education achieved. This was assessed by the self-declarationof the highest level of qualifications attained. Subjects with no qualificationswere scored 0, those with qualifications obtained at school up to but excludingA level (a university entry examination in the United Kingdom) were scored1, and those with A levels and any further educational qualification werescored 2. Dividing subjects into more categories (for example, affording higherscores for graduate and postgraduate degrees) yielded similar results to theanalyses presented in this article.

Of the 8 SNPs that showed significant associations in the study by Straubet al¹⁰, we restricted our analysis to SNPsP1655, P1635, P1325, and P1757. We excluded the other SNPs because they werein very strong LD (D′>0.9 and r²>0.9),with at least 1 of the genotyped SNPs.¹⁰ Testsfor single-marker allelic association in our case-control sample were performedby χ² analysis. Tests for haplotype association with schizophreniawere performed using the software EHPLUS,²⁴ andstatistical significance was estimated using the permutation test PMPLUS.²⁵ Associations between haplotypes and age at onset,GAS score, and symptom dimensions were analyzed using linear regression models.Course of illness and education were analyzed using ordinal regression.

The genomic structure of all known spliced forms of DTNBP1 (Figure 1) was determined in silico by combining the reference exonic sequence inthe online genetic databases LocusLink and Ensembl (available at: http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=84062 and http://www.ensembl.org/Homo_sapiens/geneview?gene=ENSG00000047579) with the 12 DTNBP1 transcripts in AceView(an integrated view of human genes as reconstructed by alignment of all publiclyavailable messenger RNA and expressed sequence tags on the genome sequence,available at: http://www.ncbi.nlm.nih.gov/IEB/Research/Acembly/av.cgi?c=locusid&l=84062). All available exons were aligned according to the University of California,Santa Cruz, human genome reference sequence (July 2003 freeze [the versionof the genomic sequence that this analysis was based on]) using BLAST 2 sequences²⁶ (an Internet-based program to perform gapped alignmentsbetween DNA sequences) (available at: http://www.ncbi.nlm.nih.gov/blast/bl2seq/bl2.html) and the genomic structure and sequence used to design amplicons formutation discovery across all exons.

There are 4 predicted alternative first exons, indicating that it islikely there are 4 independent promoters (Figure 1). Only promoters 1, 2, and 4 were found to contain a characteristicpromoter sequence as determined in silico using thesoftware Cister: Cis-element Cluster Finder²⁷ (availableat http://zlab.bu.edu/~mfrith/cister.shtml). However, we screened2 kb of sequence upstream of the predicted start of transcription of eachof the 4 potential promoters.

The derived genomic sequences were used to design primers using Primer3software (available at: http://www-genome.wi.mit.edu/cgi-bin/primer/primer3_www.cgi).²⁸ Large exons were amplified usingsets of amplimers of no more than 600 bases that overlapped by no less than50 bases. All polymerase chain reactions (PCRs) were performed using standardtouchdown protocols previously described.²⁹ (Detailsof the PCR conditions are available from the authors.)

The sample for mutation screening consisted of 14 unrelated white subjectsfrom the United Kingdom who met DSM-IV criteria forschizophrenia, each of whom had at least 1 affected sibling. The PCR productsfrom each were screened for sequence variation by denaturing high-performanceliquid chromatography using a sensitive protocol we have described elsewhere.^29- 30 The PCR products from individualsshowing chromatograms suggestive of heteroduplex formation were sequencedin both directions using Big-Dye terminator chemistry and an ABI3100 sequenceraccording to the manufacturer's instructions (Applied Biosystems, Foster City,Calif). All variants were confirmed by allele-specific primer extension usingSNaPshot reagents and an ABI3100 sequencer according to the manufacturer'sinstructions (Applied Biosystems).

In Cardiff, individual genotyping was performed by means of single nucleotideprimer extension using either the Acycloprime (Perkin Elmer Life Science Products,Boston, Mass) or SNaPshot methods according to manufacturer's instructions,with alleles being determined by fluorescence polarization measurement usingan Analyst (LJL Biosystems Ltd, Surrey, England) or an ABI3100 sequencer,respectively. In Dublin, the Irish sample was genotyped using SNaPshot andan ABI377 DNA Sequencer (Applied Biosystems). All polymorphisms that we identifiedby denaturing high-performance liquid chromatography were genotyped by primerextension in DNA pools constructed from a subset of our cases (n = 552) andcontrols (n = 552) taken from the first case-control sample. Analysis wasperformed on 6 different DNA pools, each containing a different set of 184cases or controls. Pools were created from DNA that had been quantified usingthe PicoGreen dsDNA Quantitation Reagent (Molecular Probes, Eugene, Ore) anda Labsystems Fluoroskan Ascent (LifeSciences International, Basingstoke, Hampshire,England) fluorometer. Each DNA pool was amplified in 2 separate PCR reactionsand the products subjected to allele-specific primer extension using SNaPshotas described.¹⁴ Estimated allele frequencieswere converted to numbers and were tested for approximate statistical significanceby χ² analysis. Any differences where P<.10were then confirmed by individual genotyping.

We individually genotyped 96 individuals from the first case-controlsample for all SNPs to estimate marker-marker LD. The program ldmax (http://www.sph.umich.edu/csg/abecasis/GOLD/)³¹ wasused to reconstruct haplotypes and calculate D′ and r². An α version of the program Haploview (developedin and maintained by Mark Daly's lab at the Whitehead Institute, Cambridge,Mass, by Jeffrey Barrett; software available for download at http://www-genome.wi.mit.edu/personal/jcbarret/haploview/) was also used to reconstruct haplotypes and to examine haplotype blockstructure.

The results we obtained in the Cardiff case-control sample for SNPsP1655, P1635, P1325, and P1757¹⁰ are presentedin Table 1. Only minor differenceswere found between cases and controls in allele (Table 1) or genotype (not shown) frequencies, none of which approachedstatistical significance. All makers were in Hardy-Weinberg equilibrium. Nofurther evidence for association was obtained when the markers were used toconstruct haplotypes (Table 2).The specific haplotype implicated by Straub and colleagues,¹⁰ correspondingto GGCA at markers 1655-1635-1325-1757, respectively, was not significantlyin excess in cases (0.097 in cases, 0.102 in controls, χ²₁ = 0.186, P = .67). More recently, it hasbeen shown that the risk haplotype in that sample can be fully defined byalleles GCA at the latter 3 loci.¹¹ In oursample, that specific haplotype was also not significantly more frequent incases and controls (10.0%:10.1%, χ²₁ = 0.004, P = .95). Similarly, we did not find the haplotype implicatedby Schwab and colleagues¹² (corresponding toalleles ACG at markers 1635-1325-1757) to be significantly in excess in patients(0.71 in cases, 0.71 in controls).

DTNBP1 is predicted to have 13 exons and hasat least 12 different known messenger RNA transcripts (Figure 1). We initially screened 6669 bases of genomic sequenceof which 3161 were exonic. Of those 3161 bases, 1077 encode amino acids, and495 and 1589 bases represent the 5′ and 3′ untranslated regions,respectively. The remaining 3508 base pairs were intronic. We found 7 SNPs,2 of which were exonic but untranslated, whereas the other 5 were in introns.All are aligned with the genomic structure in Figure 1 and listed in Table 3. None of the changes are predicted to alter the amino acid sequenceof the protein.

Given that we previously obtained evidence that an unknown polymorphismaffected DTNBP1 expression in the cerebral cortex,¹⁵ we extended our mutation screen to the 4 putativepromoter regions of DTNBP1, because promoters areamong (though are not the only) the most important regulatory elements ina gene. A total of 7920 bases of putative promoter sequence were screened,resulting in a further 9 SNPs.

The details of each SNP together with allele frequencies determinedfrom the case-control pools are presented in Table 3. Three SNPs had a difference in allele frequency of P<.10, and these were taken to individual genotypingin the Cardiff case-control sample. Given their exonic position, we also genotypedthe 2 untranslated region SNPs, despite the fact that neither had shown evidencefor association on pooled genotyping. The individual genotype data are similarto those obtained from the pooled genotyping and provide only nonsignificanttrends for association (Table 4).

We then constructed the 9-marker haplotype that was composed of allthe markers we had individually genotyped. This revealed modest evidence forassociation (global P = .045). Five of the 9 possible8-marker haplotypes also yielded globally significant evidence for association,as did 20 of 36 of the 7-marker haplotypes and 39 of 84 of the 6-marker haplotypes.In total, approximately one third of all possible haplotypes revealed resultsthat gave global significance at P<.05. That ahigh proportion of haplotypes was associated is not surprising given the nonindependenceof both individual markers and haplotypes (Figure 2), and Bonferroni correctionis clearly not appropriate. Marker combinations yielding global evidence forassociation at P<.003 are presented in Figure 3. The haplotypes that yielded globalevidence for significant association at this level include SNP A in promoter1 and at least 1 of P1635 and P1655. The haplotype that yielded the strongestglobal evidence for association consisted of markers P1655, P1635, and SNPA (χ² = 31.19, empirical P<.001,100 000 simulations). The frequencies of each of the individual haplotypes(with frequency >0.01) for this combination of markers in cases and controlsare given in Table 5. Post hocinspection of the individual haplotypes shows that those consisting of allelesCAT are significantly in excess in our cases, whereas those consisting ofCAA and GGT are significantly more common in controls.

To attempt to replicate our data, we analyzed the Dublin case-controlsample that did not support association when analyzed by the previously availablemarkers.¹³ The haplotype frequencies in thissample are remarkably similar to the Cardiff sample (Table 5), and the same CAT and the CAA and GGT haplotypes were significantlymore and less common, respectively, in the schizophrenic cases. On this occasion,analyses of the individual haplotypes were under the specific hypotheses generatedfrom the first sample rather than post hoc, and therefore although the P values obtained were more modest (in keeping with thesample size), their prior probability is greater. When we combine the samples,the specific haplotypes CAT, CAA, and GGT give highly significant evidencein favor of association (χ² = 10.42, P =.001; χ² = 10.80, P = .001; χ² = 35.96, P<.001, respectively). The estimatedodds ratio for the CAT risk haplotype is 1.40 (95% confidence interval, 1.13-1.74).

Estimates of LD between all of the SNPs identified in this study andthose associated with schizophrenia are presented in Figure 2.

We performed 8 tests for phenotype association with 2 haplotypes, atotal of 16 tests. Analysis of the CAT (risk) haplotype revealed no significantassociation with any aspect of the phenotype. Age was associated with theCAA (protective) haplotype in cases but not in the controls. Although thisis almost certainly a spurious finding, it was necessary to adjust for agein the phenotype analyses. Analysis of the CAA protective haplotype revealeda significant association only with higher educational attainment (likelihoodratio test χ²₁ = 10.6, P =.001, corrected P = .02). Although there was no evidencefor age at onset as a confounder, this analysis was restricted to subjectswith age at onset of younger than 21 years to remove any influence of earlyonset on education. Without this restriction, evidence for association remained(likelihood ratio test χ²₁ = 6.6, P = .01).

Using a large case-control sample, we attempted to replicate and extendthe original¹⁰ and the subsequent¹² reportsimplicating DTNBP1 as a susceptibility gene for schizophrenia.Analyses using the most significant and informative SNPs from those studiesrevealed no evidence for allelic, genotypic, or haplotype association. Moreover,despite using a screening sample with power of 0.8 to identify variants witha frequency of 0.05, sequence analysis of the exons revealed no nonsynonymousvariants in the coding sequence that might account for previous findings.It may be relevant that the previous studies included samples that contributedto positive linkage signals on 6p, whereas our samples did not.¹⁸

However, following up our report of polymorphic, cis-acting influences on DTNBP1 expression,we screened 4 putative DTNBP1 alternative promotersfor sequence variants. Pooled genotyping of the SNPs provided modest evidencefor allelic association for 3 SNPs, 2 of which are located in putative promoters.Although individual genotyping in an extended sample did not yield evidencefor allelic association, inclusion of these SNPs, particularly SNP A, withthe previous markers yielded strong evidence for association. Moreover, wewere able to exactly replicate the pattern of our findings in a completelyindependent sample. Because our primary data set is case control, our findingscannot be attributed to segregation distortion. The data from the 2 sampleswe report in this study when viewed in the context of the 2 previous studies^10,12 form an impressive and convincingbody of evidence that variation at the DTNBP1 locusconfers susceptibility to schizophrenia. Moreover, we also demonstrate thatthis effect generalizes to samples that do not display evidence for linkageto 6p. Although more genetic studies will be required before we understandprecisely how genetic variation at this locus confers susceptibility, thereis now sufficient evidence to justify intensive investigations into diseasemechanisms. Our data suggest the existence in our samples of a single riskhaplotype defined by CAT, several neutral haplotypes, and 2 protective haplotypes(CAA and GGT). The GGT haplotype was not observed in any schizophrenic subjectsand may therefore represent a protective factor that is rare but of substantialeffect. Identifying the mechanisms underlying such a strong protective effectmay offer an important opportunity for therapeutic intervention.

Although our data from both samples are internally consistent, our riskhaplotype has no overlap with that in the original Irish sample.^10- 11 Ourrisk haplotype carries allele A at P1635 like the (largely) German sample,¹² so it is possible that our markers offer a more powerfuldefinition of a haplotype that is common to both populations. However, thiscan only be established if the German group types our additional criticalmarkers.

We attempted to identify particular aspects of the phenotype that areassociated with the 2 common risk and protective haplotypes. Given that DTNBP1 is thought to be a member of the dystrophin proteincomplex (DPC),³² that mutations in the dystrophingene can result in lower IQ,³³ and that lowcognitive ability is a risk factor for schizophrenia,²³ wealso sought evidence for association with educational achievement. We usedthis as a proxy for general cognitive ability. Although educational achievementis only a crude measure of IQ, there is evidence for correlation (r = 0.4-0.6) between the 2 measures.³⁴ Ourfinding that higher educational achievement is associated with the protectivehaplotype suggests the hypothesis that the protectivehaplotype of DTNBP1 modifies the risk of schizophreniaby influencing cognitive ability. Further work will be required to resolvethis issue.

Our data also provide a reminder that if allelic heterogeneity occursin complex diseases, failure to replicate association to specific markersshould be treated cautiously. In our study, positive findings only emergedwhen we generated novel SNPs. We had hoped that by examining 20 SNPs at anaverage density of 7.5 kb across DTNBP1, we wouldbe able to establish the haplotype block structure³⁵ ofthe region, because this is likely to facilitate further replication studies,help understanding of the relationship among the different patterns of associationamong studies, and allow more exact localization of the source(s) of the LDsignal. Even at this density, we were unable to formally define any haplotypeblocks that have not been subjected to ancestral recombination according tothe method of Gabriel and colleagues.³⁵ Likevan den Oord and colleagues,¹¹ we found a limitednumber of common haplotypes across dysbindin and many adjacent marker pairswere in strong LD (Figure 2), butour findings differ in that the markers that define the risk haplotype extendbeyond the region suggested by that group (corresponding to SNP E-P1635, Figure 1) to include the putative promoter.

Our data do not implicate individual SNPs or haplotypes as acting directlyto increase disease susceptibility. Moreover, although we have previouslyshown that cis-acting elements regulate DTNBP1 expression in native brain tissue,¹⁵ wecannot conclude that this is the mechanism underlying the association. Unfortunately,because the minor allele frequencies of the expressed SNPs are low, much largertissue sample sizes than we currently have available will be required if allele-specificexpression assays are to be used to relate specific haplotypes to gene expressionin the way we have done for COMT.³⁶

The link between abnormal dysbindin function, whatever the genetic mechanism,and schizophrenia is also unclear. Dysbindin is expressed widely in the brainand other tissues.¹⁰ In the brain, dysbindinbinds β-dystrobrevin. β-Dystrobrevin is a member of the DPC foundin postsynaptic densities.³² Straub and colleagues¹⁰ postulated that compatible with the DPC's roles insynaptic structure, maintenance and synaptic signaling, altered dysbindinor DPC function may lead to several of the structural and functional abnormalitiesthat have been reported in schizophrenia, including altered function at glutamatergicand γ-aminobutyric acid synapses and reduced synaptic density in frontalcortex and hippocampus. Recently, it has also been shown that mice that expressno dystrophin (and therefore have altered DPC function) have abnormal developmentof the posterior cerebellar vermis.³⁷ The authorsof that study³⁷ postulated that similar abnormalitiesin cerebellar development as a result of altered dysbindin function may resultin several other abnormalities that have been reported in schizophrenic patients,including altered working memory, eye movement, and cerebellar structure.

Schizophrenia in particular and psychiatric disorders in general havepreviously been thought of as relatively impervious to molecular genetic analysis.It is therefore ironic that with the recent positional cloning and subsequentconfirmation of DTNBP1,^10,12NRG1,^38- 40 andpossibly G30/G70,^41- 42 schizophrenia is now one of the fewdisorders in which genes of small-to-modest effect have been identified bypositional genetics. These successes are immensely encouraging. First, theexistence of several promising linkages suggests that other susceptibilitygenes for schizophrenia are likely to be found in the coming years. Second,the identification of novel genes and pathways in the pathogenesis of schizophreniawill open new vistas for neurobiological research⁴³ andthe prospect of improvements in diagnostics and treatment.

Submitted for publication July 24, 2003; final revision received October31, 2003; accepted November 18, 2003.

The Cardiff study was funded by grants from the Medical Research Council(MRC) (London, United Kingdom). Dr Zammit is an MRC (United Kingdom) TrainingFellow. Dr Morris is a Health Research Board (Ireland) Research Fellow. DrCorvin is a Wellcome Trust Research Fellow in Mental Health. Sample collectionin Ireland was performed by D. Meagher, MD, J. Quinn, MD, Paul Scully, MD,and Siobhan Schwaiger, RGN. The work in Ireland was supported by Science FoundationIreland (Dublin), the Higher Education Authority (Dublin), and the StanleyMedical Research Institute (Bethesda, Md).

Corresponding authors and reprints: M. C. O'Donovan, PhD, FRCPsych,and M. J. Owen, PhD, FRCPsych, FMedSci, Department of Psychological Medicine,University of Wales College of Medicine, Heath Park, Cardiff CF14 4XN, Wales(e-mail: odonovanmc@cf.ac.uk or owenmj@cf.ac.uk).