89 (2007) 1–11www.elsevier.com/locate/schres

Schizophrenia Research

White matter abnormalities and brain activation in schizophrenia:A combined DTI and fMRI study

Ralf G.M. Schlösser a,⁎, Igor Nenadic a,d, Gerd Wagner a, Daniel Güllmar b,Katrin von Consbruch a, Sabine Köhler a, C. Christoph Schultz a, Kathrin Koch a,Clemens Fitzek c, Paul M. Matthews d, Jürgen R. Reichenbach b, Heinrich Sauer a

a Department of Psychiatry and Psychotherapy, Friedrich-Schiller-University, 07743 Jena, Germanyb Medical Physics Group, Institute for Diagnostic and Interventional Radiology (IDIR), Friedrich-Schiller-University, 07743 Jena, Germany

c Institute for Diagnostic and Interventional Radiology (IDIR), Friedrich-Schiller-University, 07743 Jena, Germanyd Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB), John Radcliffe Hospital, Oxford, UK

Received 28 June 2006; received in revised form 1 September 2006; accepted 7 September 2006Available online 7 November 2006

Abstract

Diffusion tensor imaging (DTI) studies of schizophrenia have revealed white matter abnormalities in several areas of the brain.The functional impact on either psychopathology or cognition remains, however, poorly understood. Here we analysed bothfunctional MRI (during a working memory task) and DTI data sets in 18 patients with schizophrenia and 18 controls. Firstly, DTIanalyses revealed reductions of fractional anisotropy (FA) in the right medial temporal lobe adjacent to the right parahippocampalgyrus, likely to contain fibres of the inferior cingulum bundle, and in the right frontal lobe. Secondly, functional MRI revealedprefrontal, superior parietal and occipital relative hypoactivation in patients with the main effect of task. This was accounted for byreduced prefrontal activation during the encoding phase of the task, but not during maintenance or retrieval phases. Thirdly, wefound a direct correlation in patients between the frontal FA reduction (but not medial temporal reductions) and fMRI activation inregions in the prefrontal and occipital cortex. Our study combining fMRI and DTI thus demonstrates altered structure-functionrelationships in schizophrenia. It highlights a potential relationship between anatomical changes in a frontal–temporal anatomicalcircuit and functional alterations in the prefrontal cortex.© 2006 Elsevier B.V. All rights reserved.

Keywords: Cingulum; Cognition; Diffusion tensor imaging (DTI); Functional magnetic resonance imaging (f MRI); Hippocampus; Medial temporallobe; Parahippocampal gyrus; Prefrontal cortex; Schizophrenia

⁎ Corresponding author. Department of Psychiatry and Psychother-apy, Friedrich-Schiller-University, Philosophenweg 3, 07743 Jena,Germany. Tel.: +49 3641 935284; fax: +49 3641 935444.

E-mail address: ralf.schloesser@uni-jena.de (R.G.M. Schlösser).

0920-9964/$ - see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.schres.2006.09.007

1. Introduction

Brain structural changes in several cortical andsubcortical structures have been repeatedly reported inschizophrenia using both post mortem techniques(Harrison, 1999) and magnetic resonance imaging(MRI) (Wright et al., 2000; Honea et al., 2005). This

2 R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

has led to the notion of disturbed connectivity or thedisconnection hypothesis of schizophrenia (Friston,1998), which posits disturbed interaction betweenseveral cortical and subcortical nodes of interconnectednetworks. While there have been several anatomicalformulations of this hypothesis (Weinberger and Lipska,1995; Buchsbaum et al., 1998; Crow, 1998; Andreasen,1999), most data rely on activation differences infunctional imaging studies and/or cognitive studies.Most current hypotheses postulate dysfunction ofprefrontal lobe areas. Evidence has been broughtforward for disturbed connectivity between prefrontalcortices and other areas, including thalamus andcerebellum (Andreasen, 1999), thalamus and striatum(Buchsbaum et al., 1998), medial temporal lobe(Weinberger and Lipska, 1995), or interhemisphericdysfunction (Crow, 1998). Recent functional imagingstudies applying structural equation modelling havesuggested a more complex pattern of disrupted func-tional connectivity including the dorsolateral prefrontalcortices, thalamus, parietal lobes and cerebellum(Schlosser et al., 2003). However, there has been nosystematic investigation combining both functional andstructural imaging to assess the disconnectionhypothesis.

Structural imaging studies using diffusion tensorimaging (DTI), as well as gene expression studies andevidence for oligodendroglial dysfunction, have sug-gested abnormalities of white matter in schizophrenia(Davis et al., 2003). Functional disconnection mighttherefore be a manifestation of structural abnormalitiesin the connecting fibre tracts.

DTI studies in schizophrenia have demonstratedreduced anisotropy in several areas (Jones et al., 2006;Kanaan et al., 2005; Kubicki et al., 2005), includingchanges in the frontal lobe (Buchsbaum et al., 1998), thecorpus callosum, and uncinate fasciculus (Kanaan et al.,2005; Kubicki et al., 2005). It remains unclear, whetherand how these structural defects might be associated withchanges in brain activity or cognitive functions (Buchs-baum et al., 1998) such as short-term memory (Nestoret al., 2004). Taking into account that structural deficits inschizophrenia are mostly assumed to be static even acrossprogression of disease (Weinberger and McClure, 2002),white matter abnormalities might constitute a trait markerof schizophrenic pathology. In this study, we combinedDTI and fMRI to relate changes in white matter structureto altered patterns of functional activation (in grey matter)with a well-defined cognitive task. We applied DTI toassess the cerebrum for changes in local anisotropy of thewhite matter using a voxel-based approach able to reporton changes in widely distributed fibre tracts connecting

the prefrontal cortices, i.e. frontothalamic tracts (from theanterior thalamic peduncle, passing the anterior internalcapsule before fanning out in the anterior parts of thecorona radiata), the anterior corpus callosum (connectinghomologous prefrontal cortices), the fasciculus uncinatus(connecting inferior prefrontal areas with the anteriortemporal cortices (Ebeling and von Cramon, 1992)),as well as the cingulum bundle (providing connectionsbetween the prefrontal and parietal, as well as medialtemporal areas). All study participants also underwentfMRI scanning to assess brain activation differencesduring a working memory task. This cognitive functionhas consistently been reported as impaired in schizophre-nia and is possibly at the core of neuropsychologicaldysfunction in this disorder (Aleman et al., 1999; Cannonet al., 2005). This particular task was used in previousstudies and shows robust activation of dorsolateral andventrolateral prefrontal, thalamic, and parietal areas(D'Esposito et al., 1999).

2. Methods

2.1. Subjects

The study included 18 patients with a DSM-IVdiagnosis of schizophrenia (mean age 29.6 years, S.D.7.0; 4 female, 14 male) without any concurrent otherpsychiatric disorder or any neurological condition.Diagnosis was established using a symptom checkliston the basis of DSM-IV criteria (abbreviated SCIDinterview) and confirmed by two clinical psychiatrists(R.S. and I.N.). Patients were free of any concurrentpsychiatric diagnosis and had no neurological condi-tions. All patients were in remission from an acutepsychotic episode and on stable medication with eitheratypical (n=12) or typical (n=6) antipsychotics and insome cases with additional mood stabilisers. Psychopa-thology as rated with PANSS showed a mean score of69.4 (SD 23.7; n=16). A group of 18 healthy volunteers(mean age 29.0, SD 10.0; 6 female, 12 male) served as amatched control group. The two groups did not differsignificantly in age ( p=0.84; two-tailed T-Test) orgender ( p=0.457; Pearson Chi-Square test). All parti-cipants were right-handed (Edinburgh HandednessScale (Oldfield, 1971)). None of the study participantshad a concurrent or past history of neurologicalconditions, or head trauma. Controls were additionallyscreened for absence of first-degree relatives with axis Ipsychiatric disorders. All participants gave writteninformed consent to a study protocol approved by thelocal Ethics Committee (Friedrich-Schiller-University,Medical School).

3R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

2.2. Imaging protocols

Magnetic resonance imaging was performed on a 1.5Tesla whole-body scanner using a standard CP transmit/receive head coil (Magnetom Vision plus, SiemensMedical Solutions, Erlangen, Germany). Subjects firstcompleted two fMRI sessions, which were followed bythe structural protocol with DTI. Foam pads were usedfor positioning and immobilisation of the subject's headwithin the head coil and to restrict movement. A screenwas positioned on the head coil, onto which stimulusmaterial was projected using a LCD beamer (Toshiba,Japan). Earplugs were used to reduce influence ofscanner noise.

Functional studies were performed using a T2*-weighted EPI sequence (TR 2700 ms, TE 60 ms, flipangle α 90°) obtaining 24 contiguous axial slices of 3 mmthickness covering the entire brain (FOV 220mm; in-plane resolution: 64×64 voxels with 3.44×3.44 mm). Aseries of 536 whole-brain volume sets were acquired intwo sessions (268 scans in each session), with the firstthree images of each series being discarded. We used amodified Sternberg task in an event-related design toexamine working memory functions. In this delayed-response task, subjects are presented letters as visualstimulus material and are then required either to simplyretain the information in its original order (forwardcondition) or perform additional alphabetical reorderingof this sequence (alphabetise condition) during asubsequent delay period. The applied paradigm wassimilar to a paradigm used before, which has been shownto activate the dorsolateral prefrontal and parietal cortices(D'Esposito et al., 1999; Postle et al., 1999). A chain of 3letters randomly selected from the alphabet was presentedfor 2.5 s, followed by a brief 750 ms delay, an instruction(“forward” or “alphabetise”) presented for 750 ms, and amain delay of 8 s. In the subsequent letter-digit probesubjects had to indicate whether the displayed numbercorrectly gave the displayed letter's position in thealphabetised chain (two-alternative forced choice) bypressing one of two designated buttons of a fibre opticresponse device. The forward and alphabetise conditionswere balanced across the experiment. To increase sam-pling rate, we additionally introduced a temporal jitterwith temporal onsets varying by TR/8 between blocks.

DTI was performed with a twice-refocused spin echoEPI sequence (TRSE-EPI, TR 4000 ms, TE 100 ms,field-of-view 240×240 mm) (Reese et al., 2003)acquiring b0 images as well as diffusion weightedimages (b ≈940 s/mm2) in six independent directions.Diffusion gradients were applied according to thefollowing scheme: (0,0,0; −1,1,0; −1,0,1; 0,1,1; −1,

−1,0; −1,0,1; −0,1,−1). We acquired two interleavedblocks of 19 slices, each 3 mm thick with a 3 mm gap(in-plane resolution 2.5 mm×1.88 mm, matrix 96×128,interpolated to 256×256), which were acquired succes-sively, the second block being shifted along the z-axisby 3 mm. This provided a scan of almost the entire brainexcept the lower cerebellum and brainstem with 3 mmslice thickness.

3. Data analysis

Behavioural data were assessed with analysis ofvariance (ANOVA) using SPSS software. We firstdetermined reaction time (RT), number of correct respo-nses (hits) and of incorrect responses (wrong) for eachof the two task conditions separately (“forward” and“alphabetise”). We then used ANOVA to test for themain effect of GROUP as well as interaction of GROUP(patient; control) with TASK (forward; alphabetise).

Imaging data were analysed using the SPM2software package (Institute of Neurology, London,UK) implemented in MATLAB (Mathworks Inc. Sher-born, MA, USA) and own software for DTI analysis.DTI data sets were first merged to one image and con-trolled for artefacts. From the diffusion images, wecomputed fractional anisotropy (FA) maps for eachsubject using in-house software. For normalisation, weapplied SPM2 routines. First, we computed a b0template applying a 12-parameter affine co-registrationto the SPM2 EPI template for each subject's b0 image.The 36 co-registered images were then resliced to3×3×3 mm3 isotropic voxels, averaged, and smoothedwith a Gaussian kernel of 8 mm FWHM. This averagedimage was then used as a template for the followingnormalisations. Each individual b0 image was normal-ised to the b0 template using 12-parameter affine trans-formation followed by a non-linear normalisation with6×8×6 basis functions, and these individual trans-formation parameters were subsequently applied to thesubject's FA map. For statistical analysis, we set up aGLM in SPM2 and contrasted the FA maps betweengroups, testing for FA reductions in patients across theentire volume (at pb0.001, uncorrected). FunctionalMRI data sets were first reconstructed and visuallyinspected for gross artefacts. Subjects exceeding amovement of more than 3 mm or 3° were excludedfrom analysis (this affected two initially recruitedpatients). We then applied the slice timing function tocorrect for the fMRI acquisition scheme. Followingmotion correction, all fMRI scans were co-registered tothe first scan of the corresponding time series, correctingfor minor head movement. None of the analysed

4 R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

subjects showed gross distorting head movement. Allseries were then spatially normalised to standard MNIspace (non-linear normalisation). We then applied aspatial smoothing filter (10 mm FWHM Gaussiankernel), as well high-pass filtering (128 sec cut-off).Selection of the smoothing filter was based on theunderlying matching filter theorem and anticipated sizeof regional effects in the range of 6–12 mm, thusreducing noise as well as taking into consideration theanatomical variability of brain structures (Worsley et al.,1996). Brain activations were then analysed for the threeseparate temporal domains of the task (i.e. encoding,delay, and retrieval), modelling each of these as separatehemodynamic responses to assess the BOLD responseassociated with each of the three domains. For thispurpose, we applied covariates that modelled theexpected BOLD signal response (relative to the ITI)occurring in each of the task periods. Evoked fMRIsignal changes for the different task sub-componentswere modelled by a covariate of variable length box-

Fig. 1. DTI analysis revealing reductions of fractional anisotropy (FA) in sctemporal lobe and right frontal lobe. Significant voxels ( p=0.002, uncorreprojection (MIP; top left) and in red superimposed on the mean b0 image of(top right, lower panel). Diagrams (bottom left and right, resp.) show meanschizophrenia patients in the medial temporal lobe and frontal cluster, resp.

cars, convolved with a canonical hemodynamic re-sponse function (HRF). These HRFs were then used asindividual regressors within the general linear modelapplied for statistical testing. Resulting parameterestimates (approximating measured response amplitude)for each event type (i.e. encoding, delay, and retrieval)and pair-wise contrasts were computed. For eachsubject, separate statistical parametric maps of T-valuesfor contrasts of interest were computed using a fixedeffect analysis. Two types of analyses were performed,for all of which a statistical threshold of pb0.001(height threshold, uncorrected) was chosen. First, wecontrasted activation during the task compared to rest,which depicts activation during the entire trial treated asan epoch. Secondly, we modelled encoding, delay, andretrieval as three distinct stages using separate hemo-dynamic response functions for each temporal phase.This decomposition was performed for closer examina-tion of components of the task, i.e. to test for differencesat distinct stages of working memory, as shown

hizophrenia patients compared to healthy controls in the right medialcted; voxel threshold k=10 voxels) are shown as maximum intensitythe group (top right, upper panel) or single subject T1-weighted imageindividual fractional anisotropy (FA) values for healthy controls and

5R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

previously (Bedwell et al., 2005). For this modelling wedefined three hemodynamic response functions (HRF),taking into account the temporal stages and durations ofeach sub-component, i.e. duration for encoding wasdefined as the temporal stage of the 2.5 s longpresentation period, for maintenance as the 8s longdelay period, and for retrieval we adapted the HRFindividually for each trial according to the individualreaction time of the subject on that trial. Assumingmental manipulation during the alphabetise condition toelicit stronger prefrontal BOLD changes (D'Espositoet al., 1999; Cannon et al., 2005), we used thealphabetise condition for the tests, except for theencoding condition, which contains the stage beforeinstruction is given. Therefore, the encoding analysesinclude twice as many trials as the maintenance orretrieval analyses. While the absolute number of trialscompared between groups was identical, there is no biastowards one group, but higher statistical power to detectgroup differences at the encoding stage (compared tomaintenance or retrieval stages). This difference is offsetby the higher number of sampling points duringmaintenance, which is longer than the encoding periodand thus contains more image frames. A threshold ofpb0.001 (uncorrected for multiple comparisons) wasused for all fMRI analyses, as well as an extent thresholdof k=10 voxels. In the case of differential contrasts (e.g.stronger activation of healthy controls vs. patients forthe encoding vs. rest condition), we applied inclusivemasking with the overall task vs. rest contrast; thismeans that we only considered areas which had previ-ously shown involvement in the task, as demonstrated

Fig. 2. Activation patterns of healthy controls and patients during the encodintask rendered on the cortical surface ( pb0.001, uncorrected; k=10 voxels;encoding phase; B— activation pattern during the maintenance phase (in botschizophrenia patients).

by a significant activation during the task compared torest. The rationale of this approach is to exclude thepossibility that relative deactivations in one group (e.g.patients) with little or no change in activation in theother group (e.g. controls) would induce a spuriouspositive result on the group level that might be inter-preted as a relative activation surplus (in this example,controls activating stronger than patients). For the DTI-fMRI correlation, we first extracted the FA in patientimages for those clusters that had shown differencesbetween patients and controls. These values (a mean FAfor each patient) were then entered as a covariate into theGLM of an fMRI analysis. Thus, two separatecorrelation analyses were performed: one for a medialtemporal lobe cluster and one for a prefrontal cluster;each was computed with a threshold of pb0.001(uncorrected). In both cases, we tested for correlationsacross the entire brain, as we tested for a hypothesisedeffect of the structural abnormality on several distrib-uted cortical networks, including those remote from thestructural (white matter) abnormality.

4. Results

4.1. Behavioural data

ANOVAs for behavioural data did not reveal astatistically significant effect of the factor GROUP oneither reaction time (F(1,34)=1.395; p=0.246), correctresponses (F(1,34)=1.391; p=0.246), or incorrectresponses (F(1,34)=0.852; p=0.363). There was nosignificant effect for the GROUP x TASK interaction for

g and maintenance periods of the modified Sternberg working memorymasked with main effect of task): A — activation pattern during theh cases: images on the left for healthy controls; images on the right for

Fig. 3. Differences in brain activation between schizophrenia patients and healthy controls during a Sternberg working memory task measured withfMRI ( pb0.001, uncorrected; masked with main effect of task). A shows relative hyperactivation of controls compared to patients for the overallcomparison (i.e. task vs. rest), identifying areas in the prefrontal and occipital cortex that show relatively higher activation in controls. B showsactivation pattern differences for the three subcomponents of encoding (B.1), and maintenance (B.2). For B.1, healthy controls show relativehyperactivation to patients, while for B.2 patients show relative hyperactivation in the displayed areas compared to controls.

6 R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

reaction time (F(1,34)=0.009; p=0.924) and onlytrends for correct responses (F(1,34)=4.007; p=0.053)and incorrect responses (F(1,34)=3.936; p=0.055).

4.2. DTI

Patients showed significantly reduced local FA in theright medial temporal lobe and a right dorsolateralprefrontal area. Localisation of changes, as well as plotsof mean FA values for each cluster are shown in Fig. 1.Medial temporal lobe changes affected white matteradjacent to the parahippocampal/entorhinal cortexextending towards the hippocampus. In the temporallobe, we identified two closely juxtaposed clusters(maxima at x/y/z = 36,−15,−24, maximum voxelT=4.08, and 33,−9,−33 and T=3.54, respectively;pb0.001, uncorrected), which were continuous whenlowering the threshold to p=0.002. The position andconfiguration of this FA difference between patients andcontrols suggests that it corresponds to a region mainly

containing fibres of the temporal part of the cingulumbundle (Mori et al., 2005) and short-distance fibres ofconnections between medial temporal lobe structures.The temporal (inferior) part of the cingulum bundlecontains fibres linking the medial temporal lobe withprefrontal and parietal lobe areas (Mori et al., 2005). Thesecond major region of difference between the twogroups, in the dorsolateral prefrontal cortex, was locatedat the level of the superior/middle frontal gyri (maximum:x/y/z=42,24,24; T=6.93; pb0.001, uncorrected). Thislocation does not allow a definite association with aspecific fibre tract; the region is likely to contain differentfibre tracts, which might include short-distance U-fibres,commissural fibres, and fibres from association tracts(potentially also cingulum bundle).

4.3. Functional imaging

Both patients and controls showed task-related activa-tions in the dorsolateral and ventrolateral prefrontal,

Fig. 4. DTI-fMRI correlation in schizophrenia patients: changes in activation (BOLD signal) during the encoding stage of the modified Sternbergworking memory task correlated with the mean fractional anisotropy (FA) in a prefrontal cluster previously identified to differ between schizophreniapatients and healthy controls ( pb0.001, uncorrected). Results are displayed as maximum intensity projection (left) and rendered views (right).

7R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

anterior cingulate, occipital and lateral posterior parietalcortices (see Fig. 2). For the encoding condition, activationswere more prominent in the visual cortex (includingprimary visual as well as secondary/lateral occipitalcortices, extending to the inferior temporal cortex, andposeterior parietal cortex, and the dorsolateral prefrontalcortex, bilaterally. For the maintenance condition, bothgroups revealed activations in the superior parietal lobe(bilaterally), as well as the left and right dorsolateral/ventrolateral prefrontal cortices (Fig. 2). For the overallcontrast between groups, we found increased activationfor controls relative to the patients in lateral prefrontal,superior parietal, and occipital areas (Fig. 3A), but noareas of higher activation in patients relative to controls.When analysing these patterns separately for encoding,maintenance and retrieval conditions (Fig. 3B), the largestdifferences were seen for encoding, duringwhich controlsshowed higher activation in prefrontal and occipital areas.Therewas no area inwhich patients had higher activationsthan controls with this contrast. For the maintenancecondition, patients showed higher activation in a cluster inthe inferior left precentral gyrus (Fig. 3B.2). There wereno areas of higher activation for controls in themaintenance condition. For the retrieval condition, therewere no significant differences between the two groups.

4.4. DTI-fMRI-correlations

We found significant correlations between FA in thefrontal white matter cluster and activation in severalprefrontal areas, including the gyrus adjacent to thewhite matter area of FA reduction (Fig. 4). There was nostatistically significant correlation between FA in thetemporal lobe cluster and fMRI activation patterns withthis task. In addition, analysis of the encoding phase forthe forward and the alphabetise trails separately did notsignificantly alter the results of the combined analysis.

Furthermore, there was no correlation between FAvalues in either of the two clusters and BOLD changesin healthy controls.

5. Discussion

Despite a number of both functional and structuralimaging studies in schizophrenia, the relation of brainstructure and function has not been comprehensivelystudied so far. This is the first study that we are aware ofwhich attempts to relate directly alterations in brainactivity with fMRI to changes in white matter structure inthis disorder. There are three major findings from thisstudy.

First, our DTI comparison of schizophrenia patients inremission and matched controls reveals white matterchanges in two brain regions that were hypothesised toshow changes based on previous data (Lipska andWeinberger, 2002). One of the major fibre tracts linkingthese regions is the cingulum bundle, which containsfibres in its inferior portion within the white matteradjacent to the parahippocampal cortex and providesconnections to prefrontal cortical areas (Mori et al.,2005). Although this interpretation of our data would beconsistent with the fronto–temporal hypothesis ofdisconnection between hippocampus/temporolimbic cor-tices and dorsolateral prefrontal cortex as put forward byWeinberger and colleagues (Weinberger and Lipska,1995; Lipska andWeinberger, 2002), our findings are nota definitive proof of this model, since they do not showchanges all along this fibre tract and cannot rule out thatthe two significant clusters actually might be related toindependent pathologies. Similarly, the microscopiceffects of white matter changes on adjacent corticesremain incompletely understood. It is of interest, thatour medial temporal lobe finding is adjacent to theparahippocampal cortex, which not only has connections

8 R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

to prefrontal areas, but is also functionally relevant for arange of cognitive abilities, including memory encoding(Schon et al., 2004). The perirhinal cortices are a crucialelement in interactions between the hippocampus and theneocortex (de Curtis and Pare, 2004). The linkingpathway is assumed to provide a substrate for memoryacquisition, consolidation, and retrieval functions (Fellet al., 2002). A recent computational study has foundevidence that compromised parahippocampal functionleads to memory deficits closely resembling those seen inschizophrenia (Talamini et al., 2005). Reduction of FAimplies either a difference of fibre architecture (e.g. axondensity, size, orientation, or internal structure) or inmyelination of the fibres. A developmental difference infibre architecture might be present through life, butrelative differences in myelination could be age-depen-dent and dynamic (Paus, 2005). The parahippocampalcluster is located in a part of the medial temporal lobe that(to a lesser extent) also includes local fibres connectingadjacent structures of the medial temporal lobe. Some ofthese fibres myelinate only late in adolescence or earlyadulthood, thus forming a potential substrate for theexpression of the clinical phenotype of schizophrenia(Benes et al., 1994). This neurodevelopmental hypothesisis supported by studies of displaced white matter neuronsin the frontal lobe and parahippocampal gyrus (Eastwoodand Harrison, 2005). Our finding of reduced FA in thefrontal lobe is in line with previous studies, of whichseveral, but not all, found changes in these areas,although the location of the alteration differed signifi-cantly across studies (Kanaan et al., 2005; Kubicki et al.,2005). While the voxel-based analysis of FAvalues offersadvantages in resolution and thus comparison ofanatomically variable structures, some potential limita-tions deserve consideration. Firstly, minor imprecisionsin image normalisation as well as smoothing might resultin voxels of adjacent but anatomically distinct structuresto contribute to a finding. This is particularly an issue forcortical areas, which have much smaller anisotropy thanthe white matter. In both clusters of our results, however,mean individual FA values in both groups were wellabove a threshold of 0.15, which can be applied to ruleout involvement of grey matter. It is therefore veryunlikely, that the two significant clusters reflect changesin nearby cortical structures, but are rather attributable towhite matter change (see Fig. 1). Secondly, despite theincreased resolution, the clusters likely contain certainamounts of crossing white matter fibre tracts. In the caseof the medial temporal lobe finding this would meanthat fibres other than the cingulum might be affected aswell. Since ROI-based methods also suffer this problem,we believe our voxel-based approach to offer more

advantages (including independence of user bias inoutlining ROIs) and thus to justify this approach. ROImethods, on the other hand, avoid errors related to reg-istration or normalisation, which justifies their use as analternative method. Further studies are necessary to fullyevaluate the relative advantages and disadvantages ofthese two approaches. Ultimately, the DTI methodologyis limited in spatially resolving the composition of fibretracts within a given voxel. In the case of our frontalcluster, affected fibres could include inter-and intra-hemispheric tracts, e.g. callosal fibres, as well as fibres ofthe cingulum bundle, local U-fibres etc. Increasing fieldstrength and optimised sequences will be helpful in im-proving the methodology.

The second main finding is related to relativelydecreased brain activation in the schizophrenia patientsduring the encoding phase of the working memory task.Impairment in working memory has been repeatedlyshown to be accompanied by changes in frontal activation(Manoach, 2003; Cannon et al., 2005; Glahn et al., 2005).Our fMRI study extends these findings by assessing notonly the overall activation pattern, but also the differenceswith neuropsychologically distinct stages of processing(encoding, maintenance, and retrieval). As previouslyshown in healthy subjects (Bedwell et al., 2005),hemodynamic modelling allows a sufficiently reliableseparation of the different stages in matching-to-sampletasks. While the overall activation map implied a frontaland occipital relative activation deficit in patients, theadditional analyses show that this is mostly related todeficient activation during the encoding phase. Animalstudies showing sustained activity of prefrontal neuronsduring the short-term maintenance of information havesuggested that this phase might be particularly vulnerablein schizophrenic patients (Goldman-Rakic and Selemon,1997). Our findings rather suggest that the majorfunctional neural substrate is earlier, during task encoding.Cognitive studies have shown that patients are impairedon both a delayedmatching-to-sample task, aswell as on asimilar task without delay (Lencz et al., 2003). Similarly,other experimental procedures have suggested encodingof information to be slowed or inefficient in schizophrenia(Hartman et al., 2003). The prefrontal activation patternsare in line with the notion that schizophrenia patients donot show global hypofrontality, but relative frontalhypoactivity that varies as a function of parameters suchas task difficulty. Extending previous findings andhypotheses on cortical inefficiency (Callicott et al.,2003, Manoach, 2003), we can infer that deficits inprefrontal activation might not only depend on taskdifficulty or design, but are expressed differentially overthe temporal course of the task.Aswe appliedmasking for

9R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

overall activation, we can also infer that the detecteddifferences are specific to the task and do not reflect signalchanges that are actually not related to task performance.The choice of different temporally segregated hemody-namic response functions has been demonstrated to be amethodologically promising approach for modellingdifferent temporal domains of a complex workingmemory task (Postle et al., 1999; Bedwell et al., 2005).We were able to consistently demonstrate the presence ofrelative hypoactivation in frontal areas during encoding inpatients. We suggest that modelling of different temporalstages of a task would significantly enhance analyses ofcognitive tasks in schizophrenia to refine the existinghypotheses and minimise inconsistencies across studies(Manoach, 2003; Glahn et al., 2005). Interestingly, therewas no main effect of group on behavioural data,indicating that patients performed equally well on thetask. Such dissociation of differences in activation, butlack of (statistically significant) difference in performanceis not unusual in functional neuroimaging studiesassessing the neural basis of cognitive functions (Wilk-inson and Halligan, 2004). A possible explanation for ourdata is that dysfunction (and thus altered activation)occurs only at the encoding stage. Since our behaviouraldata only reflect overall task performance, they give noinformation on the different encoding stages or sub-components of the task. A recent meta-analysis of work-ing memory in schizophrenia, however, supports the ass-umption that the performance deficit is independent frommaintenance (for delaysN1 s duration), and thus likely tobe related to encoding deficits (Lee and Park, 2005).

The third, and most important finding of our study is,however, the correlation of FA reductions with fMRIactivation patterns. In the two correlation analyses, wecould show that FA in the medial temporal lobe cluster isnot related to the activation deficits, while the frontal FAreduction shows correlations to frontal and occipitalactivation. While previous studies have providedpreliminary evidence of anisotropy changes to be relatedto glucose metabolism (Buchsbaum et al., 1998) andcognitive performance (Nestor et al., 2004), this study isthe first direct demonstration that we are aware of forthis disease in which specific white matter changes arerelated to altered fMRI activation patterns. A correlationwas seen only for frontal FA reduction, suggesting thatwhile the frontal white matter changes were relevant tothis task, those in the temporal lobe may not have been.Other tasks may need to be used to find functionalcorrelates of the temporal lobe white matter changes. Ina most recent study (Schon et al., 2004), it was shownthat modifications of a delayed matching to sample taskcan demonstrate parahippocampal cortical involvement

in stimulus encoding. Thus, changes in task designmight facilitate the detection of activations in that area.In the case of our working memory task, the frontal FAabnormality was correlated with several cortical areas ofBOLD changes, even though these did not completelyoverlap with the activation differences seen in theprevious fMRI analysis. For example, the differences inlateral prefrontal and occipital activations were notsubstantiated by white matter changes in fronto–occipital fibre tracts. This might be for several reasons.The structural deficit might affect connectivity within anetwork that is activated or relevant only for specifictasks or at a temporal stage of the task. The latter casewould also explain a lack of effect of DTI findings onoverall task performance, which did not differ signifi-cantly between groups.

Our results are the first findings showing white matterchanges in schizophrenia to affect the hemodynamicBOLD response. This demonstrates that a structural def-icit might have functional effects in remote areas. Thisunderlines that functional activation deficits might reflecteffects secondary to structural lesions and that interfer-ence on a “primary lesion” needs consideration of bothstructural and functional modalities. One further aspectthat needs to be considered is the temporal stability ofDTI findings. Both psychopathology and cognitivedeficits show considerable variation in patients overtime, while major structural changes (e.g. white matterarchitecture) could be relatively stable. It thus needs to beconsidered that dynamic white matter changes, forexample related to transient changes in myelination ordysfunction of oligodendrocytes (Davis et al., 2003),might be reversible and thus exert an effect on cognition orpsychopathology only for a limited duration. Thecorrelation between white matter structure and brainfunctional activation therefore may be heterogeneousbetween patients and variable in a patient over the courseof disease, allowing only the most consistent correlationsto be identified. Currently, there is little longitudinal DTIdata that could support or disprove these considerations.Since, however, some studies suggest whitematter volumeto change during psychotic episodes (Christensen et al.,2004), this factor needs consideration in future studies.

In conclusion, our analyses provide a first fMRIdemonstration of a brain functional impact of whitematter abnormalities assessed with DTI. They suggestthat the reductions of white matter anisotropy found inthis and previous studies are relevant to brain functionand thus contribute to the substrate for disease genesis.This structure-function approach will enhance ourunderstanding of the neurobiological mechanisms andits effects on cognition. Further studies, including a

10 R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

range of pathologically more specific MR imagingmethodologies and longitudinal studies of patientsthrough their disease courses, should prove an effectivemeans of addressing these questions.

Acknowledgements

This work was supported by an IZKF programmegrant of the Friedrich-Schiller-University of Jena (to R.S.and H.S.)/TMWFK (B307-04004) and the Bundesmi-nisterium für Bildung und Forschung BMBF (Core-UnitMRT-Methodik, 01ZZ0405). I.N. was supported by aJunior Scientist Grant (Nachwuchsförderung) of theIZKF Jena.

References

Aleman, A., Hijman, R., de Haan, E.H., Kahn, R.S., 1999. Memoryimpairment in schizophrenia: a meta-analysis. Am. J. Psychiatry156, 1358–1366.

Andreasen, N.C., 1999. A unitary model of schizophrenia: Bleuler's"fragmented phrene" as schizencephaly. Arch. Gen. Psychiatry 56,781–787.

Bedwell, J.S., Horner, M.D., Yamanaka, K., Li, X., Myrick, H., Nahas,Z., George, M.S., 2005. Functional neuroanatomy of subcompo-nent cognitive processes involved in verbal working memory. Int.J. Neurosci. 115, 1017–1032.

Benes, F.M., Turtle, M., Khan, Y., Farol, P., 1994. Myelination of a keyrelay zone in the hippocampal formation occurs in the human brainduring childhood, adolescence, and adulthood. Arch. Gen.Psychiatry 51, 477–484.

Buchsbaum, M.S., Tang, C.Y., Peled, S., Gudbjartsson, H., Lu, D.,Hazlett, E.A., Downhill, J., Haznedar, M., Fallon, J.H., Atlas, S.W.,1998.MRI white matter diffusion anisotropy and PETmetabolic ratein schizophrenia. NeuroReport 9, 425–430.

Callicott, J.H., Mattay, V.S., Verchinski, B.A., Marenco, S., Egan, M.F.,Weinberger, D.R., 2003. Complexity of prefrontal cortical dysfunc-tion in schizophrenia: more than up or down. Am. J. Psychiatry 160,2209–2215.

Cannon, T.D., Glahn, D.C., Kim, J., Van Erp, T.G., Karlsgodt, K.,Cohen, M.S., Nuechterlein, K.H., Bava, S., Shirinyan, D., 2005.Dorsolateral prefrontal cortex activity during maintenance andmanipulation of information in working memory in patients withschizophrenia. Arch. Gen. Psychiatry 62, 1071–1080.

Christensen, J., Holcomb, J., Garver, D.L., 2004. State-related changesin cerebral white matter may underlie psychosis exacerbation.Psychiatry Res. 130, 71–78.

Crow, T.J., 1998. Schizophrenia as a transcallosal misconnectionsyndrome. Schizophr. Res. 30, 111–114.

D'Esposito, M., Postle, B.R., Ballard, D., Lease, J., 1999. Maintenanceversus manipulation of information held in working memory: anevent-related fMRI study. Brain Cogn. 41, 66–86.

Davis, K.L., Stewart, D.G., Friedman, J.I., Buchsbaum,M.,Harvey, P.D.,Hof, P.R., Buxbaum, J., Haroutunian, V., 2003.White matter changesin schizophrenia: evidence for myelin-related dysfunction. Arch.Gen. Psychiatry 60, 443–456.

de Curtis, M., Pare, D., 2004. The rhinal cortices: a wall of inhibitionbetween the neocortex and the hippocampus. Prog. Neurobiol. 74,101–110.

Eastwood, S.L., Harrison, P.J., 2005. Interstitial white matter neurondensity in the dorsolateral prefrontal cortex and parahippocampalgyrus in schizophrenia. Schizophr. Res. 79, 181–188.

Ebeling, U., von Cramon, D., 1992. Topography of the uncinatefascicle and adjacent temporal fiber tracts. Acta Neurochir. (Wien)115, 143–148.

Fell, J., Klaver, P., Elger, C.E., Fernandez, G., 2002. The interaction ofrhinal cortex and hippocampus in human declarative memoryformation. Rev. Neurosci. 13, 299–312.

Friston, K.J., 1998. The disconnection hypothesis. Schizophr. Res. 30,115–125.

Glahn, D.C., Ragland, J.D., Abramoff, A., Barrett, J., Laird, A.R.,Bearden, C.E., Velligan, D.I., 2005. Beyond hypofrontality: aquantitative meta-analysis of functional neuroimaging studies ofworking memory in schizophrenia. Hum. Brain Mapp. 25,60–69.

Goldman-Rakic, P.S., Selemon, L.D., 1997. Functional and anatomicalaspects of prefrontal pathology in schizophrenia. Schizophr. Bull.23, 437–458.

Harrison, P.J., 1999. The neuropathology of schizophrenia. A criticalreview of the data and their interpretation. Brain 122 (Pt 4),593–624.

Hartman, M., Steketee, M.C., Silva, S., Lanning, K., McCann, H.,2003. Working memory and schizophrenia: evidence for slowedencoding. Schizophr. Res. 59, 99–113.

Honea, R., Crow, T.J., Passingham, D., Mackay, C.E., 2005. Regionaldeficits in brain volume in schizophrenia: a meta-analysis of voxel-based morphometry studies. Am. J. Psychiatry 162, 2233–2245.

Jones, D.K., Catani, M., Pierpaoli, C., Reeves, S.J., Shergill, S.S.,O'Sullivan, M., Golesworthy, P., McGuire, P., Horsfield, M.A.,Simmons, A., Williams, S.C., Howard, R.J., 2006. Age effects ondiffusion tensor magnetic resonance imaging tractography measuresof frontal cortex connections in schizophrenia. Hum. Brain Mapp.27, 230–238.

Kanaan, R.A., Kim, J.S., Kaufmann, W.E., Pearlson, G.D., Barker, G.J.,McGuire, P.K., 2005. Diffusion tensor imaging in schizophrenia.Biol. Psychiatry 58, 921–929.

Kubicki, M., McCarley, R., Westin, C.F., Park, H.J., Maier, S., Kikinis,R., Jolesz, F.A., Shenton, M.E., 2005. A review of diffusion tensorimaging studies in schizophrenia. J. Psychiatr. Res.

Lee, J., Park, S., 2005. Working memory impairments in schizophre-nia: a meta-analysis. J. Abnorm. Psychology 114, 599–611.

Lencz, T., Bilder, R.M., Turkel, E., Goldman, R.S., Robinson, D., Kane,J.M., Lieberman, J.A., 2003. Impairments in perceptual competencyand maintenance on a visual delayed match-to-sample test in first-episode schizophrenia. Arch. Gen. Psychiatry 60, 238–243.

Lipska, B.K., Weinberger, D.R., 2002. A neurodevelopmental modelof schizophrenia: neonatal disconnection of the hippocampus.Neurotox. Res. 4, 469–475.

Manoach, D.S., 2003. Prefrontal cortex dysfunction during workingmemory performance in schizophrenia: reconciling discrepantfindings. Schizophr. Res. 60, 285–298.

Mori, S., Wakana, S., Nagae-Poetscher, L.M., van Zijl, P.C.M., 2005.MRI Atlas of Human White Matter. Elsevier, Amsterdam.

Nestor, P.G., Kubicki, M., Gurrera, R.J., Niznikiewicz, M., Frumin,M., McCarley, R.W., Shenton, M.E., 2004. Neuropsychologicalcorrelates of diffusion tensor imaging in schizophrenia. Neuro-psychology 18, 629–637.

Oldfield, R.C., 1971. The assessment and analysis of handedness: theEdinburgh inventory. Neuropsychologia 9, 97–113.

Paus, T., 2005. Mapping brain maturation and cognitive developmentduring adolescence. Trends Cogn. Sci. 9, 60–68.

11R.G.M. Schlösser et al. / Schizophrenia Research 89 (2007) 1–11

Postle, B.R., Berger, J.S., D'Esposito, M., 1999. Functional neuroan-atomical double dissociation of mnemonic and executive controlprocesses contributing to working memory performance. Proc.Natl. Acad. Sci. U. S. A. 96, 12959–12964.

Reese, T.G., Heid, O., Weisskoff, R.M., Wedeen, V.J., 2003. Reductionof eddy-current-induced distortion in diffusion MRI using a twice-refocused spin echo. Magn. Reson. Med. 49, 177–182.

Schlosser, R., Gesierich, T., Kaufmann, B., Vucurevic, G., Hunsche,S., Gawehn, J., Stoeter, P., 2003. Altered effective connectivityduring working memory performance in schizophrenia: a studywith fMRI and structural equation modeling. NeuroImage 19,751–763.

Schon, K., Hasselmo,M.E., Lopresti, M.L., Tricarico, M.D., Stern, C.E.,2004. Persistence of parahippocampal representation in the absenceof stimulus input enhances long-term encoding: a functionalmagnetic resonance imaging study of subsequent memory after adelayed match-to-sample task. J. Neurosci. 24, 11088–11097.

Talamini, L.M., Meeter, M., Elvevag, B., Murre, J.M., Goldberg, T.E.,2005. Reduced parahippocampal connectivity produces schizo-

phrenia-like memory deficits in simulated neural circuits withreduced parahippocampal connectivity. Arch. Gen. Psychiatry 62,485–493.

Weinberger, D.R., Lipska, B.K., 1995. Cortical maldevelopment, anti-psychotic drugs, and schizophrenia: a search for common ground.Schizophr. Res. 16, 87–110.

Weinberger, D.R., McClure, R.K., 2002. Neurotoxicity, neuroplasti-city, and magnetic resonance imaging morphometry: what ishappening in the schizophrenic brain? Arch. Gen. Psychiatry 59,553–558.

Wilkinson, D., Halligan, P., 2004. The relevance of behaviouralmeasures for functional-imaging studies of cognition. Nat. Rev.Neurosci. 5, 67–73.

Worsley, K.J., Marrett, S., Neelin, P., Evans, A.C., 1996. Searchingscale space for activation in PET images. Hum. Brain Mapp. 4,74–90.

Wright, I.C., Rabe-Hesketh, S., Woodruff, P.W., David, A.S., Murray,R.M., Bullmore, E.T., 2000. Meta-analysis of regional brainvolumes in schizophrenia. Am. J. Psychiatry 157, 16–25.