Genetic utility of broadly defined bipolar schizoaffective disorder as a diagnostic concept

Background

Psychiatric phenotypes are currently defined according to setsof descriptive criteria. Although many of these phenotypesare heritable, it would be useful to know whether any of thevarious diagnostic categories in current use identify casesthat are particularly helpful for biological–geneticresearch.

Aims

To use genome-wide genetic association data to explore the relativegenetic utility of seven different descriptive operationaldiagnostic categories relevant to bipolar illness within alarge UK case–control bipolar disorder sample.

Method

We analysed our previously published Wellcome Trust Case ControlConsortium (WTCCC) bipolar disorder genome-wide associationdata-set, comprising 1868 individuals with bipolar disorderand 2938 controls genotyped for 276 122 single nucleotide polymorphisms(SNPs) that met stringent criteria for genotype quality. Foreach SNP we performed a test of association (bipolar disordergroup v. control group) and used the number of associated independentSNPs statistically significant at P<0.00001 as a metricfor the overall genetic signal in the sample. We next comparedthis metric with that obtained using each of seven diagnosticsubsets of the group with bipolar disorder: Research DiagnosticCriteria (RDC): bipolar I disorder; manic disorder; bipolarII disorder; schizoaffective disorder, bipolar type; DSM–IV:bipolar I disorder; bipolar II disorder; schizoaffective disorder,bipolar type.

Results

The RDC schizoaffective disorder, bipolar type (v. controls)stood out from the other diagnostic subsets as having a significantexcess of independent association signals (P<0.003) comparedwith that expected in samples of the same size selected randomlyfrom the total bipolar disorder group data-set. The strongestassociation in this subset of participants with bipolar disorderwas at rs4818065 (P = 2.42x10^–7). Biological systemsimplicated included gamma amniobutyric acid (GABA)_A receptors.Genes having at least one associated polymorphism at P<10^–4included B3GALTS, A2BP1, GABRB1, AUTS2, BSN, PTPRG, GIRK2 and CDH12.

Conclusions

Our findings show that individuals with broadly defined bipolar schizoaffective features have either a particularly stronggenetic contribution or that, as a group, are genetically morehomogeneous than the other phenotypes tested. The results pointto the importance of using diagnostic approaches that recognisethis group of individuals. Our approach can be applied to similardata-sets for other psychiatric and non-psychiatric phenotypes.

INTRODUCTION

TOP

INTRODUCTION

Recent molecular genetic genome-wide association studies havemade a major contribution to understanding the biological contributionsto many common human diseases including diabetes, heart disease,Crohn’s disease and rheumatoid arthritis.^1–4 Theseapproaches provide psychiatry with powerful research tools thatoffer the opportunity to identify the biological systems involvedin the pathogenesis of major psychiatric illness.⁵ However, psychiatric phenotypes are currently defined according to setsof descriptive criteria that were developed to have acceptablereliability but with no expectation that the resulting categorieswould represent valid entities. This uncertainty over the validityof the diagnostic categories (phenotypes) that we use in psychiatrymakes psychiatric research more challenging than for most non-psychiatric disorders^6,7 and, thereby, has an adverse impact on the evidenceavailable on which to base clinical practice. In this paperwe use genome-wide genetic association data on people withbipolar disorder (and controls) to explore the relative genetic support (a form of biological validity) for different descriptiveoperational diagnostic categories within a published, largeUK case–control bipolar disorder study of 4806 individualsgenotyped for 469 557 genetic polymorphisms that met basicquality control metrics (Wellcome Trust Case Control Consortium (WTCCC)).³

Method

TOP

Method

Samples
Our sample comprised the 1868 participants with bipolar disorderand 2938 controls investigated in the WTCCC study. A detaileddescription of the sample has been provided elsewhere.³ All individuals were from the UK and over the age of 16 years.Clinical assessment included semi-structured interview andreview of case notes. Ratings of symptom occurrence and courseof illness were made including the operational criteria (OPCRIT)item checklist.^8,9 Diagnoses were based on all available data.The primary diagnostic system used for classifying participantswas the Research Diagnostic Criteria (RDC)¹⁰ because it providesmore delineation between individuals on the basis of the patternof mood and psychotic symptomatology than do DSM–IV¹¹or ICD–10.¹² Participants with bipolar disorder hadexperienced at least one episode of clinically significantelevated mood according to RDC: bipolar I disorder (n = 1316),schizoaffective disorder, bipolar type (n = 279), bipolar IIdisorder (n = 171) and manic disorder (n = 102). The diagnosesaccording to DSM–IV for the same set of participants were:bipolar I disorder (n = 1594), schizoaffective disorder, bipolartype (n = 98), bipolar II disorder (n = 134), bipolar disorder,not otherwise specified (n = 42). (Each individual has a diagnosisaccording to both RDC and DSM–IV. The joint distribution of diagnoses is shown in online Table DS1. Diagnostic reliabilityas measured by mean kappa ( ${kappa}$ ) coefficients for a set of 20 individualswas 0.86 for RDC and 0.84 for DSM–IV.) With the exceptionof DSM–IV bipolar disorder, not otherwise specified (n= 42), all definitions of diagnosis were of large enough samplesize to warrant further investigation. The controls, who werenot screened for psychiatric illness, came from two sources:the UK 1958 birth cohort longitudinal epidemiological sample (n = 1458) and the UK Blood Donor Service (n = 1480). It has previously been shown that it is valid to combine these twocontrol samples for use as controls in genetic associationstudies using UK disease samples, including the current bipolardisorder sample.³

Genotypic data
Polymorphisms used in analyses
The WTCCC data-set comprised 469 557 single nucleotide polymorphisms(SNPs) distributed across the genome. For the current analysiswe selected autosomal SNPs for analysis that had a minor allelefrequency of at least 5% in our total sample and met stringentlevels of genotyping quality. The large number of genotypesscored in a study such as this requires the use of generic approachesto quality control, allowing SNPs to be excluded where the quality of genotyping is in question. We used the following qualityfilter for inclusion of SNPs:

call rate (i.e. the proportion of genotypes that could be scored confidently, out of all the genotypes that were attempted forthat SNP) >99.5% in WTCCC participants with bipolar disorderand in controls;
Hardy-Weinberg P>0.001 in participantswith bipolar disorder (i.e. to exclude significant deviationfrom Hardy–Weinbergequilibrium at significance P<0.001);
Hardy–Weinberg P>0.01 in controls.

We have demonstrated that SNPs meeting these criteria showeda very high level of genotype agreement with genotypes scoredindependently in our laboratory using the Sequenom platform(of over 67 000 genotypes typed for 140 SNPs we found 99.95%agreement; data not shown). Using these stringent quality filtersthere were autosomal 276 122 SNPs selected for analysis.

Statistical analyses
Principle of the analysis
In this analysis we were interested in whether any particulardiagnostic phenotype definition(s) provided strong associationevidence when a set of participants with bipolar disorder meetingthis definition was analysed against controls for the fullset of SNPs within the genome-wide analysis. We used a genome-wideanalysis (i.e. a case–control comparison for each of the276 122 SNPs) as the basic unit of study and summarised theoverall association evidence in that analysis by counting thenumber of independent SNPs that showed association exceedinga specified significance threshold (chosen as P<10^–5,the significance benchmark in the WTCCC study used to designateat least ‘moderately strong’ evidence for association).We refer to these associated SNPs as ‘hits’ (whichis a shorthand term used in molecular genetics to indicatean independent association signal that meets a specified levelof statistical significance). Our basic aim was to determineif one or more definitions of diagnosis possessed greater utilityby virtue of identifying more hits. When comparing the numberof hits from different diagnostic phenotype definitions thereis an important complicating factor – the case samplesize varies for the different diagnostic sets. The sample size affects the power to detect associations and must, therefore,be taken into account. Having provided this orientation towardsour analysis, we will now explain the details of the methodsused.

Genome-wide association analysis
Each set of participants with bipolar disorder was comparedagainst the set of 2938 controls for all of the 276 122 SNPsthat met our stringent quality control filter (see above).According to the various diagnostic definitions, the numberof cases varied from 102 to 1868. For each SNP we employed the Cochran–Armitage trend test of genotype distributionsto test association with disease (i.e. we compared the groupwith bipolar disorder with the 2938 controls). Assuming noassociation with disease, the trend test statistic followsthe ${chi}$ ² distribution on 1 degree of freedom (as implemented withinthe PLINK analytic suite of programs).¹³ This method is astandard approach to analysis of genetic association data andis robust to departures of the data from Hardy–Weinbergequilibrium. To allow for any systematic inflation of the teststatistics we adjusted the trend test statistic of each SNPby ${lambda}$ , where ${lambda}$ is the genomic control inflation factor, estimatedto be the median of all 276 122 test statistics divided by 0.456.¹⁴ Because SNPs that are near to each other can showcorrelated association signals (due to linkage disequilibrium)we filtered that set of association signals from each analysisto remove non-independent SNPs using the clumping facility withinPLINK.¹³ (The non-independent SNPs lie within 250 kb and arein linkage disequilibrium at r²>0.2 with the index SNP.)For each such genome-wide analysis we used the number of ( ${lambda}$ -corrected)independent SNPs showing an association signal at P<10^–5 (number of ‘hits’) as a summary measure of thegenome-wide analysis.

Testing significance of observed number of hits
We formulated the null hypothesis that, for a subset of participantswith bipolar disorder v. controls, the number of hits we observedfollows the distribution expected by chance. This hypothesisassumes that the bipolar disorder sample is genetically homogeneous(i.e. minimal variation between individuals), and that thesubset of participants with bipolar disorder under investigationis a truly random selection from the full bipolar disorder sampleof 1868 individuals. It is important to note that the null hypothesis here is that the genetic effects within the sample with bipolardisorder are homogeneous in the sense that they do not varyaccording to diagnostic subset. It is not an assumption thatthere are no genetic effects (i.e. we do not assume that thereare no differences between those with bipolar disorder and controls).The one-sided alternative is that we observed more hits thanwould be expected by chance, i.e. the total sample with bipolardisorder is genetically heterogeneous. Under the alternativehypothesis, the subset under investigation is postulated tohave properties that facilitate the detection of genetic effects.Such properties include a higher genetic loading or, perhapsmore plausibly since bipolar disorder broadly defined has alreadya high heritability, a greater degree of genetic homogeneityfor the subset. Although it may seem counter-intuitive thatmore homogeneity (i.e. fewer risk genes) would lead to thedetection of greater numbers of risk genes, such a scenariois entirely expected. This is because the reduction in the numberof risk loci in the subset compared with that in bipolar disorderas a whole would effectively increase the effect size at anyone locus and thus the ease of detection.

In order to allow for the effect of differing sample size weused a simulation procedure to generate the distribution ofthe number of hits for subsets of the sample with bipolar disorderv. controls under the null hypothesis. We randomly selectedsamples of individuals with bipolar disorder, where the numberof individuals is the same size as the diagnostic set of interest,and used each in a case–control genome-wide analysis againstthe total set of controls. For each of the subsets we undertooka genome-wide analysis as described above (including both genomiccontrol adjustment and filtering for independent signals) andcounted the number of SNPs that exceeded the threshold of significance (P<10^–5). This whole procedure was repeated 1000 timesto produce a distribution of the number of SNPs expected bychance when testing individuals with bipolar disorder v. controls.

Determining empirical significance levels
The simulation procedure allowed us to compare the observednumber of hits with the simulated distribution to determinewhether there was evidence to accept or reject the null hypothesisat the 5% significance level. The empirical P was given bythe proportion of times that the simulated sets achieved atleast as many hits as were observed in the test set.

Results

TOP

Results

Genome-wide analyses of diagnostic subsets
We undertook genome-wide association analyses for seven (non-independent) diagnostic subsets of the 1868 individuals with bipolar disorderv. controls. The number of hits at P<10^–5 is shown in Table 1. It can be seen that the largest number of hitsoccurred with the RDC schizoaffective disorder, bipolar typeset even though it was one of the smaller sets tested (n =279 individuals).

View this table:
[in this window]
[in a new window]

Table 1 The number of independent single nucleotide polymorphisms (SNPs) that exceeded a significance threshold of P<10^–5 when the diagnostic subset is compared with the controls (corrected for genomic control, ${lambda}$ )

Empirical significance levels for diagnostic sets
The observed number of hits in the RDC schizoaffective disorder,bipolar type set was achieved only once in any of the 2500simulations of 279 participants randomly selected from thebipolar disorder data-set and compared with the controls. Thiscorresponds to an empirical significance of P = 0.0004 forthe null hypothesis that the RDC schizoaffective disorder, bipolartype set does not differ genetically from the bipolar disordersample as a whole. A conservative Bonferroni correction formultiple testing of seven subsets gives an empirical significanceof P = 0.0028. None of the other diagnostic sets differed fromchance expectation (P>0.05).

Genome-wide analysis in the RDC schizoaffective disorder, bipolar type group v. controls
As our analysis provided evidence that the RDC schizoaffectivedisorder, bipolar type diagnostic criteria identified a bipolarphenotype with a particularly strong utility for genetic studies,which can be considered a form of genetic validity, we presentthe results of the genome-wide analysis for the RDC schizoaffectivedisorder, bipolar type diagnostic set. For the strongest hitsin our observed data, we visually inspected genotype clusterplotsas a further check of genotyping quality. Independently associatedSNPs that exceed a significance threshold of P<10^–5(all of which have good quality clusterplots) are shown inTable 2. For each independent signal, the table also showsthe number of nearby SNPs that are in linkage disequilibrium(i.e. closely correlated) with the index SNP and thereby providequality control criteria to check for any additional stronglysupported association signals. This revealed one additionalindependent signal (rs4786811 on chromosome 16p13.3) at a significancethreshold of P<10^–5 that hadbothanacceptable clusterplotandalso support from a closely correlated SNP. The clusterplotsfor these SNPs can be found in the online supplement, as can a list of all SNPs (online Table DS2) showing nominally significant association (P<0.05).

View this table:
[in this window]
[in a new window]

Table 2 Independent association signals at P<10^–5 for comparison of participants in the Research Diagnostic Criteria (RDC) schizoaffective disorder, bipolar type group against controls^a

Discussion

TOP

Discussion

The aim of our analysis was to use the WTCCC bipolar disordergenome-wide association data-set to determine whether any ofthe bipolar-spectrum diagnostic subsets within the sample providedenhanced genetic utility, compared with the bipolar disordersample as a whole and compared with the other diagnostic subsets.Of the four RDC and three DSM–IV diagnostic categoriesexamined, only the RDC schizoaffective disorder, bipolar type category showed a significantly enhanced number of associationsignals above expectation based on the null hypothesis thatall the diagnostic subsets were simply a random selection fromthe bipolar disorder sample as a whole, i.e. that the geneticcontributions do not vary according to the diagnostic subset. It is important to stress that this null hypothesis allowsfor there being genetic differences between participants withbipolar disorder and controls – it is not a null hypothesisthat there are no genetic effects operating in bipolar disorder.Thus, our analysis provides evidence that, within our data-setand at the association threshold considered (i.e. ${lambda}$ -correctedP<10^–5), the RDC schizoaffective disorder, bipolartype sample is a particularly valuable phenotype for geneticstudies. Indeed, the RDC schizoaffective disorder, bipolartype sample had more hits than the sum of the hits in the otherthree RDC diagnostic sets (P = 0.022).

It has often been argued that schizoaffective disorder may beclosely related to schizophrenia – indeed, DSM–IVclassifies schizoaffective disorder within diagnostic codecategory 295, as a subtype of schizophrenia. Thus, one importantquestion is the following: is the enhanced number of hits seenin the RDC schizoaffective disorder, bipolar type group in ourbipolar disorder data-set relatively specific to the RDC schizoaffective disorder, bipolar type group, or is it a general property ofindividuals with schizophrenia-like features (in which caseit could be more usefully thought of as providing support forthe genetic utility of schizophrenia). We have recently undertakena genome-wide association study of schizophrenia using the samemethodology as that used in the WTCCC study of bipolar disorder(same genotyping platform, same laboratory, same set of SNPscalled at the same time using the same algorithm and with case–controlcomparisons made using the same set of 2938 controls).¹⁵ When we apply our analytic approach to our set of people with ‘schizophrenia spectrum’ (i.e. schizophrenia and schizoaffective disorders– the latter set includes the 279 individuals from thebipolar disorder sample) we similarly observe that RDC schizoaffectivedisorder, bipolar type stands out in having more hits (RDCdiagnoses: schizoaffective disorder, bipolar type (n = 299),hits 7, P = 0.023; schizoaffective disorder, depressed type(n = 114), hits 2, P = 0.81; schizophrenia (n = 257), hits3, P = 0.53). Thus, within our data, and at the significancethreshold considered, the participants with RDC schizoaffectivedisorder, bipolar type stand out from both the other participantswith bipolar disorder and the schizophrenia groups.

We note that classical family, twin and adoption genetic studieshave long been used as one of the key methods for validatingpsychiatric diagnoses.¹⁶ Classical approaches require twinand family samples, the availability of which is limited bythe substantial difficulties and high costs inherent in theirrecruitment. Classical genetic studies estimate the contributionof all inherited forms of variation. Here we have used availablegenome-wide association study data to estimate the relativecontribution of the common genetic variation to susceptibilityto different phenotypes. This may be a useful approach fordelineating phenotypic sets that could be particularly fruitfulfor study using genome-wide association studies. The major limitation of using genome-wide association study data is that currentgenome-wide association study genotype data-sets do not captureall genetic variation that may be relevant to illness. Rather,they provide information about a substantial proportion ofcommon genetic variation within the genome. They do not currentlyprovide information about the contribution of rare variantsor mechanisms such as structural variation.

The choice of threshold used for counting the top hits is inevitably,to some extent, arbitrary. In general, the signal to noiseratio is expected to be higher for more significant P-valuethresholds but there are progressively fewer hits that achievemore stringent thresholds. Thus, there is a trade off in choosinga threshold that is reasonably stringent but allowing a reasonablenumber of hits. A threshold of P<10^–5 was a benchmarkthreshold used for reporting signals within the WTCCC studyand we adopted this as an appropriate compromise. However,we note that our findings remain unchanged with a higher (P<5x10^–6)or lower (P<5x10^–5) threshold (data not shown). At these thresholds, RDC schizoaffective disorder, bipolar typeagain stood out as the only diagnostic subset having a significantlyincreased number of hits, although as would be expected, theactual number of hits was less or more respectively.

The number of hits at a certain threshold gives a guide to theease of detection of genes contributing to the phenotype. Thesignificance level of the individual association signals isalso important. Broadly, the first issue relates to the numberof loci of given effect size, whereas the second relates towhether some loci have particularly large effect sizes. We notethat the strongest signal in the RDC schizoaffective disorder,bipolar type subset of participants was more significant thanany hit in either the bipolar disorder set as a whole or theother subsets considered, which further supports the geneticutility of such cases.

Our findings, thus, suggest that the RDC schizoaffective disorder,bipolar type set of cases is likely to be particularly fruitfulfor genetic investigations aimed at identifying common polymorphismsthat influence risk (i.e. the type of genetic variants thatgenome-wide association studies are designed to detect). Ifour findings generalise to other samples, it would suggestthat careful phenotypic selection of participants could enhancepower to identify genes conferring susceptibility to illness.At present there is a great deal of effort being invested inrapidly assembling the large samples of individuals with psychiatricillness that are expected to be necessary to provide powerto detect susceptibility genes. The current findings suggest that it will also be important to pay sufficient attentionto the phenotype so that genetically relevant distinctionscan be made.

It is of substantial interest that the existence of one or morerelatively discrete nosological entities with mixed mood–schizophreniafeatures is supported by latent class analyses (e.g. Kendleret al,^17,18 McGrath et al,¹⁹ Sham et al²⁰) and that geneticepidemiology supports a strong genetic component to schizoaffectiveillness (e.g. Andreasen et al,²¹ Bertelsen et al,²² Farmer et al,²³ Gershon et al,²⁴ Maier et al,²⁵ Slater & Cowie,²⁶Cardno et al²⁷) with similar heritabilities to those in schizophreniaand bipolar disorder. Our findings are consistent with moleculargenetic evidence for the existence of relatively specific geneticsusceptibility for a form of major psychiatric illness thathas features of both bipolar disorder and prominent psychosis.^28–30 This could be interpreted as specific support for a categoryof ‘schizoaffective’ illness or for the existenceof a region of overlap of schizophrenia and bipolar disorderclinical spectra in which the genetic variants that influencesusceptibility are easier to identify than are those that conferspecific risk to bipolar disorder or schizophrenia alone. In either case, this clinical entity has genetic utility and meritsexplicit recognition.

The RDC and other modern diagnostic criteria in psychiatry weredeveloped on largely descriptive grounds and we consider itmost unlikely that the schizoaffective disorder, bipolar typecategory will map directly onto the underlying biology. Wedo not believe that ‘schizoaffective disorder’in general, or RDC schizoaffective disorder in particular, is a neatly defined, discrete, biological diagnostic entity. Ourfindings do, however, show that it can be useful for the purposesof research (and perhaps also clinical practice) to identifyand classify together sets of cases with such clinical features.Whether, in the long run, this is best achieved by using categories,dimensions or some mixture of the two will require future study.Such further work aimed at refining the relationship betweenclinical phenotype and genetic risk factors has the potentialto help psychiatry move towards a system of classificationthat relates more closely to underlying pathogenesis.

Clinical features of RDC schizoaffective bipolar disorder
The RDC schizoaffective disorder, bipolar type describes individualsthat, in addition to clear-cut episodes of mania, display psychoticsymptoms (delusions and/or hallucinations) that are not easilyunderstood as being the result of extreme mood change and thatare often seen also in individuals diagnosed with schizophrenia.It does not include all the people with bipolar disorder withpsychosis. An analysis using the set of participants with bipolardisorder who had experienced psychotic symptoms during theirlifetime did not reveal significantly more independent hits(data not shown). Thus, we can be confident that our findingdoes not simply relate to a subset of people with bipolar disorderhaving psychotic features. It is of interest that we did notobserve evidence for increased hits in the DSM–IV schizoaffective disorder, bipolar type subset of individuals. It is possiblethat this may simply reflect the smaller sample size (n = 98)or it may be an indication that the RDC definition of schizoaffectivedisorder, bipolar type is more biologically useful than thatof DSM–IV (at least with respect to identifying the contributionof common genetic variation to disease susceptibility). TheRDC definition focuses on temporal co-occurrence of a majoraffective syndrome with specific types of psychotic features,whereas the focus of the DSM–IV definition is temporalseparation of mood and psychotic symptomatology without referenceto the quality of the psychotic features.

Implications for revisions of diagnostic classifications
We note that the imminent revision of official diagnostic classifications (i.e. DSM–V and ICD–11) may be influenced by theopinions articulated in several recent articles that the conceptof schizoaffective disorder is unreliable, unhelpful and shouldbe abandoned.^31–34 In contrast, our data suggest thatwhat is needed is better recognition of such cases. Abandoningthe schizoaffective concept is unlikely to be the optimal wayof achieving that goal.

Genetic signals in the genome-wide analysis of RDC schizoaffective disorder, bipolar type data-set
As is expected in genome-wide association studies in a modestlysized sample,⁵ none of the association signals in our analysisof the RDC schizoaffective disorder, bipolar type data-setachieved accepted levels of genome-wide significance for Europeansamples (P<7.2x10^–8).³⁵ Independent replication andmeta-analysis will be required to confirm the role of any ofthe strongly associated loci in susceptibility to schizoaffective disorder. To date there has been no previous report of a systematicgenetic association analysis of a set of individuals with schizoaffectivedisorder in comparison with controls. However, it is interestingto note that an analysis of the WTCCC bipolar disorder data-setthat used a completely different analytic approach also identifiedthe RDC schizoaffective disorder, bipolar type subset of participantsas being of particular interest.³⁶ That analysis used phenotyperefinement of a specific genetic association signal of interestin the complete bipolar disorder and control data-set (at GABRB1)and found the signal to be maximal within the RDC schizoaffectivedisorder, bipolar type subset of participants. Testing of independentSNPs within genes encoding gamma amniobutyric acid (GABA)_Areceptors showed this set of individuals to have significantsystem-wide association with variation across the set of SNPsat these receptors (P = 6.6x10^–5)³⁶ with gene-wideevidence for association at GABRB1, GABRA4, GABRB3, GABRA5and GABRR1. This is consistent with the current analysis ofthe same data-set in which we observe a signal at P<0.00001at GABRB1 (Table 2), and multiple associated SNPs at theseother genes within our set of nominally significant associations(online Table DS2). The strongest association signal withinthe RDC schizoaffective disorder, bipolar type data-set (P = 2.32x10^–7) occurred on chromosome 21 with SNPs within the gene B3GALT5, a member of the beta-1,3-galactosyltransferase (beta3GalT) gene family which encode type II membrane-boundglycoproteins with diverse enzymatic functions. To our knowledge,these specific proteins have not been previously implicatedin pathophysiology of mood or psychotic illness. The strongassociation we observe at chromosome 16p13.3 is of interestbecause it lies within the gene A2BP1, encoding ataxin 2-bindingprotein 1 isoform 4, a protein that binds to ataxin-2 and may contribute to the restricted pathology of familial neurodegenerativedisease, spinocerebellar ataxia type 2. Disruption of A2BP1has been described in association with neuropsychiatric phenotypesincluding autism,³⁷ mental retardation and epilepsy.³⁸ Other genes of potential interest that show association signals atthe less stringent significance threshold of P<0.0001 (onlineTable DS3) include autism susceptibility candidate 2 (AUTS2)on chromosome 7q11.2; BSN, the gene encoding the protein bassoonwhich is thought to be involved in the organisation of thecytomatrix at the nerve terminals active zone which regulatesneurotransmitter release and which is essential in regulatedneurotransmitter release from a subset of brain glutamatergic synapses; PTPRG, encoding a member of the protein tyrosine phosphatase family which are signalling molecules that regulatea variety of cellular processes including cell growth, differentiation,mitotic cycle and oncogenic transformation; GRIK2 encodingglutamate receptor, ionotropic kainate 2 precursor (glutamatereceptor 6) (GluR-6) (GluR6); and CDH12 encoding cadherin 12,type 2 preproprotein, a type 2 classical cadherin from thecadherin superfamily of integral membrane proteins that mediatecalcium-dependent cell–cell adhesion. It will require substantial additional work by us and others in order to confirmwhich of the polymorphisms showing strong association withinthe current study influence risk of illness.

Clinical implications
We have used molecular genetic genome-wide case–controlassociation data to compare the genetic association signalsaccording to several different operational categories in thebipolar disorder spectrum. The participants meeting RDC criteriafor schizoaffective disorder, bipolar type (a broad definitionof schizoaffective disorder) received strongest support (i.e.this was the most genetically useful, and by this criterion,most biologically valid diagnostic subset). It is importantfor research, and may be important for clinical practice, thatsuch individuals are better recognised and distinguished fromother people with mood–psychosis disorders.

Strong consideration is currently being given to abolishingthe schizoaffective concept and category from the revisionsof the official psychiatric diagnostic classifications (DSM–Vand ICD–11). This is likely to be unhelpful to the progressof psychiatric knowledge, given that it is emerging as a diagnosticentity that receives strong research support. We hope thatpsychiatry is moving towards the time when our patients canbenefit from diagnostic concepts that are built on solid foundationsof empirical biological evidence rather than being perchedprecariously on the shifting sands of expert opinion.

ACKNOWLEDGMENTS

TOP

ACKNOWLEDGMENTS

We are indebted to all individuals who have participated inour research. We thank MDF-The BiPolar Organization for thehelp of its staff and members. The members of the WTCCC arelisted in the online supplement. We are grateful to WESC foruse of computing resources.

REFERENCES

TOP