Gray matter endophenotypes associatedwith genetic risks for schizophrenia and bipolar disorder. A, Map of graymatter volume deficits associated with genetic risks for schizophrenia (redvoxels) and bipolar disorder (blue voxels) superimposed onto a single brainin standard stereotactic space. Green indicates overlapping voxels. Clusterwiseprobability of type 1 error, P = .004 forboth schizophrenia and bipolar disorder; that is, <1 false-positive testresult. The z coordinate for each axial slice inthe plane of the Talairach atlas is given in millimeters, and the right sideof each panel represents the right side of the brain. B, Linear associationsbetween systemic gray matter volume deficits in regions associated with geneticrisk for schizophrenia (y-axis) and genetic liability score (x-axis) estimatedseparately for patients with schizophrenia, unaffected relatives of patientswith schizophrenia, and unaffected relatives of patients with bipolar disorder.First principal component scores (y-axis) summarize correlated gray matterdeficits in all frontal, temporal, and subcortical regions for each individual.C, Linear associations between gray matter volume deficits in regions associatedwith genetic risk for bipolar disorder (y-axis) and genetic liability score(x-axis) estimated separately for patients with bipolar disorder, unaffectedrelatives of patients with bipolar disorder, and unaffected relatives of patientswith schizophrenia. Genetic liability scores are adjusted to the sample meanfor age, sex, and subject group.
White matter endophenotypes associatedwith genetic risks for schizophrenia and bipolar disorder. A, Map of whitematter volume deficits associated with genetic risks for schizophrenia (redvoxels) and bipolar disorder (blue voxels) superimposed onto a single brainin standard stereotactic space. Green indicates overlapping voxels. Clusterwiseprobability of type 1 error, P=.01 for both schizophreniaand bipolar disorder; that is, <1 false-positive test result. The z coordinate for each axial slice in the plane of the Talairachatlas is given in millimeters, and the right side of each panel representsthe right side of the brain. B, Linear associations between systemic whitematter volume deficits in regions associated with genetic risk for schizophrenia(y-axis) and genetic liability score (x-axis) estimated separately for patientswith schizophrenia, unaffected relatives of patients with schizophrenia, andunaffected relatives of patients with bipolar disorder. First principal componentscores (y-axis) summarize correlated white matter deficits in all regionsfor each individual. C, Linear associations between white matter volume deficitsin the left temporoparietal region identified as endophenotypic for bipolardisorder (y-axis) and genetic liability score (x-axis), estimated separatelyfor patients with bipolar disorder, unaffected relatives of patients withbipolar disorder, and unaffected relatives of patients with schizophrenia.Genetic liability scores are adjusted to the sample mean for age, sex, andsubject group.
McDonald C, Bullmore ET, Sham PC, Chitnis X, Wickham H, Bramon E, Murray RM. Association of Genetic Risks for Schizophrenia and Bipolar DisorderWith Specific and Generic Brain Structural Endophenotypes. Arch Gen Psychiatry. 2004;61(10):974-984. doi:10.1001/archpsyc.61.10.974
For more than a century, it has been uncertain whether or not the major
diagnostic categories of psychosis—schizophrenia and bipolar disorder—are
distinct disease entities with specific genetic causes and neuroanatomical
To investigate the relationship between genetic risk and structural
variation throughout the entire brain in patients and their unaffected relatives
sampled from multiply affected families with schizophrenia or bipolar disorder.
Analysis of the association between genetic risk and variation in tissue
volume on magnetic resonance images.
Psychiatric research center.
Subjects comprised 25 patients with schizophrenia, 36 of their unaffected
first-degree relatives, 37 patients with bipolar 1 disorder who experienced
psychotic symptoms during illness exacerbation, and 50 of their unaffected
Main Outcome Measures
We used computational morphometric techniques to map significant associations
between a continuous measure of genetic liability for each subject and variation
in gray or white matter volume.
Genetic risk for schizophrenia was specifically associated with distributed
gray matter volume deficits in the bilateral fronto-striato-thalamic and left
lateral temporal regions, whereas genetic risk for bipolar disorder was specifically
associated with gray matter deficits only in the right anterior cingulate
gyrus and ventral striatum. A generic association between genetic risk for
both disorders and white matter volume reduction in the left frontal and temporoparietal
regions was consistent with left frontotemporal disconnectivity as a genetically
controlled brain structural abnormality common to both psychotic disorders.
Genetic risks for schizophrenia and bipolar disorder are associated
with specific gray matter but generic white matter endophenotypes. Thus, Emil
Kraepelin’s pivotal distinction was neither wholly right nor wholly
wrong: the 2 major psychoses show both distinctive and similar patterns of
brain structural abnormality related to variable genetic risk.
More than a century ago, Emil Kraepelin1 dividedpsychotic illness into 2 diagnostic categories: dementia praecox and manic-depressiveinsanity. The distinction between these disorders, now known as schizophreniaand bipolar disorder, is embedded in the major diagnostic systems in currentuse. However, the line of demarcation between these clinical phenotypes isblurred, with many patients demonstrating features of both putative diseases.Consequently, there is continued controversy regarding whether or not the2 disorders are indeed distinct disease entities caused by separable geneticand other risks.2
Twin and adoption studies have established that both disorders are highlyheritable.3- 7 Susceptibilitygenes likely act by causing abnormalities in adult brain structure and function,perhaps as a result of aberrant early neurodevelopmental control.8 It is clear that an inherited liability to developpsychosis reflects the combined effects of several susceptibility genes andtheir interactions with environmental risks such as perinatal complicationsand drug abuse.9 Psychotic disorders lack well-defined,quantitative phenotypes (even postmortem), and therefore genetic researchhas relied on clinical syndromes with imprecise boundaries and heterogeneousconstitutions. More valid phenotypes for genetic research into psychosis couldbe provided by endophenotypes; for example, quantitative deviations in brainstructure or function that underlie the clinical symptoms and are likely torepresent more direct effects of the action of susceptibility genes.10,11 The definition of such endophenotypesmay also provide neurobiological substrates for more accurate diagnosis andclassification of psychotic disorders than classical, clinical-syndromal phenotypes.11
Case-control studies of schizophrenia with magnetic resonance imaging(MRI) have demonstrated enlarged ventricles and subtle (<5.0%) volumetricdeficits in multiple cortical and subcortical regions, including medial temporallobe structures and the thalamus and frontal lobes, as well as volume deficitsin white matter tracts.12- 15 Brainabnormalities in bipolar disorder have been less thoroughly investigated,but there is some imaging evidence of ventricular enlargement and increasedrates of deep white-matter hyperin tensities.16- 18 There are conflicting findings from the few studies that have compared patientswho have schizophrenia or bipolar disorder with each other or the same controlgroup, with some studies reporting gray matter or medial temporal lobe volumedeficits only in schizophrenia19- 21 andothers finding such deficits in both disorders.22,23
If the seminal Kraepelinian dichotomy of psychosis is correct, the neuroanatomicalendophenotypes associated with genetic risks for schizophrenia and bipolardisorder should be distinct. To test this prediction, we conducted, to ourknowledge, the first large-scale comparative MRI study of adult patients withschizophrenia or bipolar 1 disorder and their unaffected first-degree relatives,all from multiply affected families (N = 148). We calculated a quantitativemeasure of genetic liability for each subject to model their likely exposureto genetic risk, and we used computational morphometric techniques to comprehensivelyand reliably map significant associations between genetic risk and variationin gray and white matter volume throughout the brain.
We recruited subjects through voluntary support groups or by directreferral from their mental health services. We successfully performed MRIon 25 patients with schizophrenia, 36 of their first-degree relatives withoutpsychosis, 37 patients with bipolar 1 disorder, and 50 of their first-degreerelatives without psychosis. The patients with bipolar disorder had all experiencedpsychotic symptoms during episodes of illness exacerbation. Patients and relativeswere assessed using the same clinical scales. Structured diagnostic interviewswere performed using the Schedule for Affective Disorders and Schizophrenia–LifetimeVersion,24 and additional information regardingthe timing and nature of symptoms was obtained to enable DSM-IV diagnoses. Information regarding history of psychiatric illnesswas obtained from the most reliable informants using the Family Interviewfor Genetic Studies25 and from medical noteswhen available. The Schedule for Schizotypal Personalities26 wasused to assess relatives without psychosis and controls for schizotypal traitsand to make DSM-IV diagnoses of schizotypal personalitydisorder.
The study sample was independent from that described previously by McDonaldet al.27 Subjects were not included if theyhad organic brain disease, had experienced head trauma resulting in loss ofconsciousness for more than 5 minutes, or fulfilled DSM-IV criteria for substance or alcohol dependence in the 12 months priorto assessment. No subjects were inpatients at the time of assessment. Thestudy was approved by the relevant local ethical committees, and all subjectsgave written informed consent to participate.
The patients with schizophrenia and their relatives were from 27 families(in some families the index patient did not successfully complete MRI), andin each family the index patient had at least 1 first- or second-degree relativeaffected with schizophrenia (20 families), another nonorganic psychotic disorder(3 families), or schizotypal disorder (4 families). Subjects with bipolardisorder and their relatives were from 32 families; in each family the indexpatient had at least 1 first- or second-degree relative affected with bipolardisorder accompanied by psychotic symptoms (24 families) or another nonorganicpsychotic disorder (8 families).
We modeled the likely variation in the level of genetic risk among subjectsusing a continuous quantitative measure of genetic liability based on eachindividual’s affection status and the number, affection status, andgenetic relatedness of all adult members of each family as far as second degreefrom the index patient. The derivation of a similar measure for schizophreniahas been described previously.28 Separate geneticliability scales were derived for schizophrenia and bipolar disorder. To calculatethe scales, a polygenic multifactorial liability threshold model of illnesswas used29 in which liability was assumed tobe continuous in the population with a gaussian distribution. Patients wereinitially assumed to have an expected liability above a particular threshold,which was based on the population prevalence rates of the illnesses: 0.7%for schizophrenia and 0.5% for bipolar disorder.30 Giventhese assumptions, the initial imputed liabilities were 2.78 for patientswith schizophrenia and 2.89 for patients with bipolar disorder. Other subjectswith psychotic disorders who were in families with schizophrenia or bipolardisorder were assumed to express the same phenotype as the index patient andwere assigned the same initial liability. A second threshold was includedfor families with schizophrenia to categorize subjects with personality disordersrelated to schizophrenia, assumed to have a population prevalence of 3.3%,31 which produced an initial expected liability of 2.08for such individuals. Other relatives were considered unaffected and had aninitial expected liability of −0.08 in families with schizophrenia and–0.07 in families with bipolar disorder.
For each family, we derived a vector of liabilities (L),which was initiallyimputed to each family member. These scores were then adjusted for each subjectto account for family size and affection distribution. First, a correlationmatrix for each family (R) was constructed describing the genetic interrelationshipsof all individuals older than 16 years and as far as second degree from theindex patient (ie, self = 1; first-degree relatives = 0.5;second-degree relatives = 0.25; spouse = 0). Assumingthat genes are the only source of familial resemblance (as has been demonstratedby twin studies5,6), a secondcorrelation matrix of liabilities to illness in each family (V) was producedby multiplying the off-diagonal elements of R by an estimate of heritability,considered to be 0.7 for both schizophrenia and bipolar disorder. A vectorof expected genetic risks (G) for each family is then given by the formula
G = RV−1L,
with the assumptions of normal distribution theory.32 Thesecalculations produced estimates of continuously variable genetic risk (geneticliability score) for subjects in families with schizophrenia and bipolar disorder.
For each subject, a set of 1.5-mm-thick contiguous coronal T1-weightedMRI studies representing the whole brain was obtained using a 3-dimensionalspoiled gradient recalled echo sequence with a 1.5-T scanner (N/Vi Signa System;General Electric, Milwaukee, Wis) and the following protocol: time to repeat = 13.1milliseconds, inversion time = 450 milliseconds, echo time = 5.8milliseconds, number of excitations = 1, flip angle = 20°,and acquisition matrix = 256×256×128. The scanning protocolwas identical for all participants, who underwent scanning in random orderwith respect to affection status.
Optimized voxel-based morphometry33,34 wasused to segment MRI data and coregister probabilistic maps of gray matterand white matter volume density for each participant in standard anatomicalspace. This was implemented using Matlab version 6.0 (MathWorks, Natick, Mass)with SPM99 software (Statistical Parametric Mapping, Wellcome Department ofImaging Neuroscience, University College London, London, England).
Initially, customized gray, white, and cerebrospinal fluid templateimages in standard stereotactic space were created from a sample of 52 healthycontrol subjects, who had undergone scanning using identical parameters randomlythroughout the study period, to minimize any scanner-specific bias and providea template matched to the sample. These subjects were group matched to thecombined samples of patients and relatives on the basis of age (mean ± SD,39.3 ± 14.8 years; range, 19-69 years), sex (46.2% men; n = 24),and parental social class (38.5% I or II; ie, professional, managerial, ortechnical occupations; n = 20) and had no personal or family historyof a psychotic, bipolar, or schizophrenia spectrum disorder. The MRI studyof each control subject was segmented into gray, white, and cerebrospinalfluid tissue in native space. These images were smoothed using an isotropicgaussian kernel (8 mm full width at half maximum) and then spatially normalizedusing parameters derived from applying a 12-parameter affine transformationof each unsmoothed gray matter map to the standard SPM99 T1-weighted graymatter template and applying these to the smoothed segmented images. The imageswere then averaged to create customized gray, white, and cerebrospinal fluidtissue templates in standard stereotactic space.
Gray and white matter maps normalized to these customized tissue templateswere produced for each subject included in the study as follows. Each subject’sMRI study was segmented into gray, white, and cerebrospinal fluid tissue classesin native space. Parameters were derived from the spatial normalization ofeach subject’s gray matter map to the customized gray matter templateand iteratively applied to the original brain image to produce an image optimallynormalized for gray matter segmentation. The images were resliced at a finalvoxel size of 1.5 mm3 and resegmented using the customized tissuetemplates as prior probability maps, and the gray matter maps were retained.This procedure was repeated using parameters derived from normalizing eachwhite matter map to the white matter template and iteratively applying themto the original image to derive white matter tissue maps for each subject.The gray and white matter images were then modulated by multiplying voxelvalues by the Jacobian determinants from the spatial normalization to correctfor volume changes introduced at this step.33,34 Finallyall normalized, segmented, modulated gray and white matter tissue maps weresmoothed using an isotropic gaussian kernel (4 mm full width at half maximum).
Multiple regression models were specified to estimate the associationbetween genetic liability and brain structural variation at each intracerebralvoxel with gray or white matter volume density as dependent variables, geneticliability score as the key predictor variable, and age, sex, and affectionstatus as covariates. Analyses were performed separately for the familieswith schizophrenia and bipolar disorder. A map of the standardized regressionmodel coefficient of interest (β) coding the association between anatomicalvariation and genetic risk at each voxel was thresholded such that if β>1.96(probability of β<0.05), the voxel value was set to β−1.96;otherwise the voxel value was set to 0. This procedure generated a set ofsuprathreshold voxel clusters in 3 dimensions, each with a mass, or sum ofsuprathreshold voxel statistics. We tested the null hypotheses of no associationbetween brain structure and genetic risk by permutation at cluster level,as described in detail elsewhere.14,35,36 Stringentthresholds for statistical significance were derived from the permutationdistribution so that the expected number of false-positive test results ineach map was less than 1. Significant clusters were anatomically localized,and Brodmann areas were ascribed when relevant from the coordinates of thecentroid voxel and the 2- dimensional spatial extent of each cluster in eachaxial slice in accordance with the standard atlas of Talairach and Tournoux.37
The mass of each significant cluster for each individual was transferredto a spreadsheet and, when multiple clusters were present, principal components(PC) analysis without rotation was performed to explore the extent of correlationbetween endophenotypic regions. In general, we found that anatomical variationwas strongly correlated between brain regions associated with genetic risk;that is, the first PC always accounted for more than 70% of total variance.We therefore used individual scores for the first PC as summary measures ofanatomical variation in endophenotypic systems comprising 2 or more correlatedgray or white matter regions associated with genetic risks for schizophreniaor bipolar disorder.
We anticipated that variation in putative anatomical endophenotypesshould be associated to the same extent with variable genetic risk in bothpatients and relatives and that endophenotypic variation might be specificallyassociated with genetic risk for 1 type of psychosis or generally associatedwith genetic risks for both types of psychosis. To explore these issues, wemodeled the association between anatomical variation in endophenotypic systems(as defined by PC scores) and genetic liability using hierarchical observationmodels that accommodated the nonindependent clustering of some individualswithin the same families. Multilevel modeling was implemented using Statasoftware version 6.0 (Stata Corporation, College Station, Tex), and a 2-tailedprobability threshold for significance in these systems-level analyses wasset at P = .05.
We first explored the association between genetic liability and relatedendophenotypic systems separately for groups of patients with schizophreniaor bipolar disorder and their relatives to test the hypothesis that geneticrisk was associated with endophenotypic variation in relatives without psychosisas well as patients. Second, we explored the associations between endophenotypes,defined by prior analysis of families with that disorder, and genetic liabilityin unaffected relatives from both types of families. Disorder-specific endophenotypesare associated with genetic risk only in unaffected relatives of index patientswith a diagnosis of that disorder, whereas disorder-generic endophenotypesare associated with genetic risk in unaffected relatives of patients withboth types of disorder.
The demographic characteristics of the subjects are listed in Table 1. There was a significant age differencebetween patients with schizophrenia and their relatives because the relativegroups included parents as well as siblings, and there was a greater proportionof men in the schizophrenia group. All subjects were of white ethnicity. Allpatients with schizophrenia were taking antipsychotic medication. Of the patientswith bipolar disorder, 31 were taking mood stabilizers, 1 was taking olanzapine,and 5 were receiving no medication. Unaffected relatives had never experienceda psychotic illness, but 10 relatives of patients with schizophrenia and 9relatives of patients with bipolar disorder had experienced another DSM-IV Axis 1 disorder at some point in their lives, mostlymajor depressive disorder. Four relatives of patients with schizophrenia alsofulfilled the criteria for schizotypal personality disorder.
Genetic risk for schizophrenia was associated with distributed graymatter volume deficits in the orbital, prefrontal, and premotor parts of thefrontal cortex, caudate nucleus, and bilateral thalamus as well as the leftinsula and lateral temporal cortex (Figure 1Aand Table 2). The PC analysis showedthat these gray matter deficits were highly correlated across regions, implyinggenetically determined effects on the volume of a cortical-subcortical network.All regions of gray matter volume deficit loaded positively for the firstPC (Table 2), which explained 73.5%of the total variance in the group of patients with schizophrenia and theirrelatives. Scores for the first PC were strongly associated with genetic riskin patients with schizophrenia and their relatives without psychosis (Table 3). There was no significant interactionbetween subject group (patient vs relative) and genetic liability score, indicatingthat this pattern of gray matter deficit was not determined solely by abnormalitiesin the patients (Table 3 and Figure 1B). The relationship between increasedgenetic risk and greater gray matter volume deficits in this cortical-subcorticalsystem remained significant when the analysis was confined to the 20 familiesin which the patient’s family history consisted specifically of schizophrenia(β = −1.49; P = .02;95% confidence interval, −2.63 to −0.31).
In contrast, genetic risk for bipolar disorder was associated with graymatter deficits in an almost completely separate and relatively circumscribedset of regions, principally the right anterior cingulate gyrus and ventralstriatum (Figure 1A and Table 2). Regional analysis confirmed that genetic risk was associatedwith reduced gray matter volume of the anterior cingulate gyrus and striatumin patients with bipolar disorder and their relatives (Table 3); there was no significant interaction between subject groupand genetic liability score, again indicating that this association was notdetermined solely by abnormalities in the patients (Table 3 and Figure 1C).
We also found strong associations between genetic risk for each typeof psychosis and anatomical variation in white matter. However, the whitematter endophenotypes associated with genetic risk in the 2 groups were anatomicallyoverlapping, in contrast to their anatomically distinct gray matter endophenotypes.Risk for schizophrenia was associated with white matter deficits in the posteriorcorpus callosum and left frontal and temporoparietal regions (Figure 2A and Table 2). Deficitsin these regions were highly correlated, and all regions of white matter volumedeficit loaded positively for the first PC (Table2), which explains 82.5% of the total variance in the group of patientswith schizophrenia and their relatives. First PC scores were significantlyassociated with genetic liability, and the interaction between subject group(patients vs relatives) and genetic liability score was not significant (Table 3 and Figure2B).
Genetic risk for bipolar disorder was associated with white matter deficitsin the anterior corpus callosum and bilateral frontal, left temporoparietal,and right parietal regions (Figure 2Aand Table 2). All regions of white mattervolume deficit loaded positively for the first PC, which explains 80.7% ofthe total variance. First PC scores were strongly associated with geneticliability, and there was no significant interaction between subject group(patients vs relatives) and genetic liability score (Table 3).
Genetic risk for bipolar disorder was not significantly associated withvolume deficits in the gray matter endophenotype for schizophrenia, and therewas a significant interaction between the 2 relative groups (relatives ofpatients with schizophrenia vs relatives of patients with bipolar disorder)and genetic liability on PC scores (Table 3 and Figure 1B). These results indicate that graymatter variation in this distributed frontostriatal and temporal system isan endophenotypic marker specifically associated with genetic risk for schizophrenia.
Likewise, genetic risk for schizophrenia was not significantly associatedwith volume deficits in the gray matter endophenotype for bipolar disorder,and there was a significant interaction between the 2 relative groups (relativesof patients with schizophrenia vs relatives of patients with bipolar disorder)and genetic liability on PC scores (Table 3 and Figure 1C). These results indicate that graymatter variation in this relatively circumscribed cingulate and striatal systemis an endophenotypic marker specifically associated with genetic risk forbipolar disorder.
Genetic liability for bipolar disorder was associated with anatomicaldeficits in the white matter endophenotype defined by univariate analysisof the schizophrenia group; similarly, genetic liability for schizophreniawas associated with anatomical deficits in the white matter endophenotypedefined by analysis of the bipolar disorder group (Table 3). A finer-grained analysis of genetic risk and endophenotypicassociation for the white matter systems showed that genetic liability scorewas generally associated with variation in the left hemispheric parts of bothschizophrenia and bipolar disorder endophenotypes (Figure 2C) but that genetic risk for bipolar disorder was specificallyassociated with the right hemispheric parts of the bipolar disorder whitematter endophenoype (further details are available from us on request). Therewas no material change in the results of the analyses of the combined relativesgroup after excluding the 24 relatives who had a previous diagnosis of anyAxis 1 disorder, schizotypal personality disorder, or alcohol or substancedependence or were taking any psychotropic medications. These results indicatethat white matter variation in the left frontal and temporoparietal regionsis an endophenotypic marker generically associated with genetic risk for bothschizophrenia and bipolar disorder.
These results provide support for the Kraepelinian dichotomy of psychosisto the extent that we have demonstrated markedly different gray matter endophenotypesassociated with the genetic risks for schizophrenia and psychotic bipolardisorder. Genetic risk for schizophrenia was associated with a relativelyextensive system of frontal, temporal, and subcortical gray matter deficits.These regions are compatible with regions of structural deficit identifiedby prior case-control studies of patients with schizophrenia.12- 14 However,interregionally correlated anatomical variation in this gray matter systemwas associated with variable genetic risk in relatives without psychosisas well as patients with schizophrenia (suggesting that anatomical variationin this system is a marker for genetic risk rather than for caseness) butwas not significantly associated with genetic risk among unaffected relativesof patients with bipolar disorder (suggesting that this endophenotypic brainsystem is indicative of genetic risk specifically for schizophrenia).
In contrast, risk for bipolar disorder was associated with more localgray matter deficits in the right anterior cingulate gyrus and ventral striatum,both of which are components of brain circuits for emotional processing38 and have been identified as exhibiting abnormalitiesin previous case-control studies of patients with familial bipolar disorderusing structural and functional neuroimaging.39 However,in this article we have clarified that anatomical variation in these regionsis a marker for genetic risk even among relatives without psychosis, not merelya marker for the presence of bipolar disorder in patients, and we have shownthat this endophenotypic brain system is indicative of genetic risk specificallyfor bipolar disorder.
Studies examining unaffected relatives or discordant twins of patientswith schizophrenia have previously linked genetic risk to volumetric reductionof the thalamus40,41 and prefrontaland temporal cortical gray matter,42,43 especiallythe dorsolateral prefrontal cortex,43 but therewas no evidence of gray matter reduction with genetic risk in a recent twinstudy of bipolar disorder.44 Some studies comparingunaffected relatives of patients with schizophrenia with controls have reportedthat genetic risk is related to volume reduction of the hippocampus,41,45,46 which did not emergein this study. However, the effect of genetic risk on this structure remainsto be fully elucidated because other studies failed to find hippocampal volumereduction in unaffected relatives47,48;evidence also suggests that hypoxic birth complications and the transitionto psychosis influence medial temporal lobe volume deficits in schizophrenia.46,48- 50 Inrelation to these prior data, the distinctive value of our results is thatthey provide a more comprehensive map of the gray matter endophenotype inschizophrenia throughout the brain, and they allow an unprecedented directcomparison with the gray matter endophenotype in bipolar disorder.
The unique comparative design of this study also draws attention toaspects of the brain phenotype that are expressed in common between the 2forms of psychosis. Genetic risk for both disorders was associated with distributedwhite matter volume deficits that were anatomically coincident in the leftprefrontal and temporoparietal regions. White matter abnormalities have beenreported in case-control studies of both schizophrenia14,51,52 andbipolar disorder.16,17,53 Studiesof discordant twins have reported a genetic effect on global white mattervolume reduction in schizophrenia54 and lefthemispheric white matter volume reduction in bipolar disorder,44 althoughother studies assessing unaffected relatives of subjects with schizophreniahave failed to find a genetic effect on global white matter volume.42,55,56 Our data map thewhite matter endophenotype for psychosis more precisely to territories normallyoccupied by major intrahemispheric tracts: the left superior longitudinalfasciculus, which connects the frontal lobe to the temporal, parietal, andoccipital lobes; and the left inferior longitudinal fasciculus, which connectsthe temporal pole to the occipital lobe.
We surmise that risk for psychosis in general is associated with a patternof white matter abnormality that is likely to compromise intrahemisphericanatomical connectivity between the left prefrontal and temporoparietal cortex.This conjecture is compatible with a substantial body of case-control dataand theory implicating disintegration or disconnectivity of large-scale neurocognitivenetworks, especially frontotemporal disconnectivity, as a critical substratefor the generation of psychotic symptoms.57- 60 Weacknowledge that the neuropathological substrate of these white matter changesis incompletely determined by the magnetic resonance signal changes reportedin this article. For example, it is possible that the white matter changeswe have described as deficits could reflect changes in the magnetic resonancesignal owing to abnormal myelination rather than reduction in the number ofaxons. There is prior evidence from case-control studies of gene expressionin the frontal cortex for the down-regulation of genes related to myelinationand oligodendrocyte function in both schizophrenia and bipolar disorder.61,62 In future studies, we will directlyinvestigate associations between allelic variation in candidate genes andstructural variation in the gray and white matter endophenotypes defined inthis article. Such studies are expected to improve the power to detect pathogeneticallyrelevant genes for psychotic disorders and to enhance understanding of thecellular substrates of MRI endophenotypes.
Some methodological aspects of our study deserve comment. The patientsparticipating in this study were carefully diagnosed according to operationalizedcriteria and were drawn exclusively from multiply affected families. We treatedgenetic risk as continuously variable among relatives without psychosis ratherthan assuming that all relatives shared the same level of risk. We suggestthat this is a more realistic assumption, in light of the likely variationbetween families in their exposure to multiple susceptibility genes, thatmay have conferred greater statistical power to detect brain endophenotypeswith our regression analysis of anatomical variation and continuous geneticliability scores than would have been attainable by, for example, an analysisof variance treating patients and relatives as 2 discrete levels of geneticrisk. We also used a customized, computerized “pipeline” for computationalmorphometry of the whole brain structure that incorporated software sourcedfrom several laboratories for optimized nonlinear image registration and nonparametrichypothesis testing of spatially informed cluster-level statistics. All imageswere registered to a single template image constructed for this purpose fromMRI studies acquired using the same scanner and pulse sequence of a groupof healthy comparison subjects demographically matched to the patient andrelative groups.
Enduring controversy often indicates that more than one view is reasonablytenable. We suggest that the long-standing dialectic between categorical anddimensional accounts of major mental illness is related to the main implicationof these data: genetic risks for schizophrenia and bipolar disorder are associatedwith both specific and generic brain structural endophenotypes. The anatomicallysegregated expression of specific and generic genetic effects that, to ourknowledge, we have demonstrated for the first time is consistent with morphometricdeviations linked to the clinical phenotypes of schizophrenia and bipolardisorder. These results also provide an important basis for future studiesseeking to more powerfully identify susceptibility genes for psychosis byassociation with neuroimaging endophenotypes. We conclude that Kraepelin’spivotal distinction was neither wholly right nor wholly wrong.It is more apt, perhaps, to think of psychosis as a sibling pair of neurogeneticsyndromes than as 1 or 2 discrete disease entities.
Correspondence: Colm McDonald, MB, MRCPsych,Division of Psychological Medicine, Institute of Psychiatry, de CrespignyPark, London SE5 8AF, England (email@example.com).
Submitted for Publication: September 26, 2003;final revision received March 12, 2004; accepted April 5, 2004.
Funding/Support: This study was supported bythe Wellcome Trust (Drs McDonald, Bramon, and Bullmore) and the Medical ResearchCouncil (Dr Wickham), London, England; and by the Stanley Medical ResearchInstitute, Bethesda, Md.
Acknowledgments: We are grateful to all ofthe families for participating in this study and to the National SchizophreniaFellowship (Rethink), London, England, and the Manic Depressive Fellowship,London, for help with recruitment.