Stephan 2006 Plasticity

download Stephan 2006 Plasticity

of 11

description

neuro

Transcript of Stephan 2006 Plasticity

  • S oninKla n

    Cu e roma tionab articmo yscopa sparsys esis frev rainele nneinc theorto enia

    Keglu

    TschWeanlatecoidifinabschbethesugpafun

    hyuseusiimbritheneuroimaging paradigms. The aim of this approach is to establish

    molevprocatph

    AbPla

    phFrineredcoiolter(Ta

    coanassdisplacoonschneity (e.g., NRG1 or dysbindin) (Harrison and Weinberger 2005).

    Fro

    Ad

    Rec

    1Be

    000doConversely, any impairment in synaptic plasticity would affectthe way long-range connections are established in the develop-ing brain. This is because the strength of functional couplingbetween two neurons determines whether their connectionsurvives developmental pruning (Hua and Smith 2004). Further-more, functional coupling is a function of experience-dependentsynaptic plasticity (Zhang and Poo 2001). Dysconnectivity due toimpaired experience-dependent synaptic plasticity is consistentwith the observation that schizophrenia cannot be explained by

    m the Wellcome Department of Imaging Neuroscience (KES, KJF), Insti-tuteofNeurology, and InstituteofChildHealth andGreatOrmondStreetHospital (TB), University College London, London, United Kingdom.dress reprint requests to Klaas E. Stephan, Wellcome Department ofImaging Neuroscience, Institute of Neurology, University CollegeLondon, 12 Queen Square, London WC1N 3BG, United Kingdom;E-mail: [email protected] June 15, 2005; revised October 14, 2005; accepted October 29,2005.cause of the original meaning of the Latin prefix dis ( apart), theterm disconnection might suggest connectivity in schizophrenia isreduced, leading to less interaction between neural units. This is notwhat the original disconnection hypothesis implies (Friston 1998). Toavoid confusion and emphasize the notion of abnormal synapticplasticity in schizophrenia, we use the term dysconnection (the Greekprefix dys meaning bad or ill).

    2Of course, functional dysconnections due to abnormal synaptic plastic-ity must also have microstructural correlates, e.g., changes in themorphology of dendritic spines and/or the number or structure ofreceptors. Here, we restrict impaired structural connectivity to thelevel of cellular processes such as axonal fiber bundles.

    BIOL PSYCHIATRY 2006;59:9299396-3223/06/$32.00i:10.1016/j.biopsych.2005.10.005 2006 Society of Biological Psychiatryynaptic Plasticity and DyscSchizophrenia

    as E. Stephan, Torsten Baldeweg, and Karl J. Fristo

    rrent pathophysiological theories of schizophrenia highlight thnifest 1) anatomically, through structural changes of associaerrant control of synaptic plasticity at the synaptic level. In thisdulation of synaptic plasticity. In particular, we discuss how dtients, could result from abnormal modulation of N-methyl-D-atems. We focus on the implication of the dysconnection hypothiew recent advances in measuring plasticity in the human bctroencephalography (EEG) that can be used to address dyscolude perceptual and reinforcement learning. We describe howa mechanistic understanding of synaptic plasticity in schizophr

    y Words: Dynamic causal models, mismatch negativity, NMDA,tamate, acetylcholine, dopamine

    he notion that schizophrenia is not caused by focal brainabnormalities, but results from pathological connectivitybetween brain regions, has been an influential idea in

    izophrenia research. This idea was initially proposed byrnicke (1906), who postulated that psychosis arises fromatomical disruption of association fiber tracts, and reformu-d later in terms of psychopathology by Bleuler (1911), whoned the term schizophrenia to denote the splitting offerent mental domains. More recently, this theme re-emergedneurophysiologic and neuroimaging experiments showingnormal distributed activity and functional connectivity inizophrenia (Volkow et al 1988; Hoffman et al 1991; Wein-rger et al 1992; Friston and Frith 1995). In an attempt to explainse observations, the disconnection hypothesis (Friston 1998)gested that the core pathology of schizophrenia is an im-ired control of synaptic plasticity that manifests as abnormalctional integration of neural systems, i.e., dysconnectivity.1

    In this article, we review evidence for the dysconnectionpothesis and its convergence with recent genetic studies. Wethis to motivate studies of plasticity in schizophrenic patientsng noninvasive techniques like functional magnetic resonanceaging (fMRI) and electroencephalography (EEG). Following aef and selective review of synaptic plasticity, we look at someoretical models that may constrain psychopharmacologicalnection

    le of altered brain connectivity. This dysconnectivity couldfibers at the cellular level, and/or 2) functionally, throughle, we review the evidence for these theories, focusing on thennectivity, observed between brain regions in schizophrenictate (NMDA)-dependent plasticity by other neurotransmitteror functional imaging at the systems level. In particular, weusing functional magnetic resonance imaging (fMRI) andctivity in schizophrenia. Promising experimental paradigmsetical and causal models of brain responses might contribute.

    dels of synaptic plasticity that can be evaluated at the systemsel using functional imaging of individual patients. This ap-ach may eventually furnish surrogates for diagnostic classifi-ion, facilitate genetic studies, and have predictive validity forarmacotherapy.

    normal Anatomical Connections, Abnormal Synapticsticity, or Both?

    There is wide-ranging evidence for dysconnection in schizo-renia (Andreasen et al 1999; Hoffman and McGlashan 2001;ston, 2005b). For example, in terms of functional connectivity,uroimaging studies of language have shown consistentlyuced frontotemporal coupling in schizophrenia relative tontrol subjects. Similarly, patients show abnormal electrophys-ogical measures of functional connectivity, e.g., reduced in-regional gamma-band synchrony during sensory processingble 1).The question is how dysconnectivity is caused. Functionalupling between brain areas might be abnormal because theiratomical connections are altered, e.g., due to a miswiring ofociation fibers. Alternatively, functional coupling could beturbed due to impairments in synaptic transmission andsticity.2 These mechanisms are not necessarily exclusive butuld coexist, because they have a common cause or becausee causes the other. For example, several genes linked toizophrenia are involved in both establishing long-range con-ctions during development and in regulating synaptic plastic-

  • genen

    conspe(Tadismaof 200indancontioThstudiftiv

    theheareknmiind197symaresim

    intleabescrtioityelifinren

    ticnocusevstugeNMrom(20varphciagic

    dysynthaThis mrev

    3N-

    Table 1. Evidence for Abnormal Functional Connectivity in Schizophrenia

    Exp s (Sele

    Aba

    ith 1999602berg

    AbP

    (revie20042005

    AbM

    al 20 19938

    onnethe ic patsele oints

    imag

    930 BIOL PSYCHIATRY 2006;59:929939 K.E. Stephan et al

    wwetics alone but only by interactions between genes andvironment (Sullivan et al 2003).Several studies have reported abnormalities in intra-arealnectivity in schizophrenia, demonstrating area- and lamina-cific reductions in dendritic field size and dendritic spinesble 2). Although this could be interpreted as a developmentalturbance of microcircuit formation, it may also reflect abnor-l synaptic plasticity: long-term potentiation (LTP) induces growthdendrites and upregulation of dendritic spines (Monfils et al4; Engert and Bonhoeffer 1999), whereas blocking LTP orucing long-term depression (LTD) decreases dendritic lengthd spine density (Monfils and Teskey 2004). There is lessvincing evidence, so far, that long-range anatomical connec-

    ns are compromised in schizophrenia (Harrison 1999; Table 3).is may be due to a lack of sensitive postmortem methods fordying anatomical connectivity in the human brain. However,fusion weighted imaging (DWI) studies have delivered nega-e results or widely varying findings (Table 3).In contrast to the sparse evidence for abnormal connectivity at level of cellular processes, pharmacological studies ofalthy volunteers show that impairments of synaptic plasticity consistent with schizophrenic symptoms (Table 2). It is wellown that N-methyl-D-aspartate (NMDA) antagonists, dopa-ne (D2) agonists, amphetamines, and serotonin agonists canuce psychotic symptoms in healthy subjects (Allen and Young8; Javitt and Zukin 1991; Kapur 2003). Additionally, psychoticptoms and cognitive deficits induced by NMDA antagonists similar to those of schizophrenia (Domino et al 2004 listilarities and differences).3 Finally, and perhaps even more

    methyl-D-aspartate (NMDA) receptors (NMDARs) also play a role insynaptic transmission of sensory information (Fox et al 1990; Kellyand Zhang 2002). It is therefore likely that NMDA antagonists alsodiminish the strength of glutamatergic synapses directly by blockingNMDAR-dependent excitatory postsynaptic currents (EPSCs). How-ever, (in)activation of NMDARs leads to various intracellular pro-cesses that change -amino-3-hydroxy-5-methyl-4-isoxazole propi-onic acid receptor (AMPAR)-mediated EPSCs as well, e.g., throughphosphorylation of AMPAR subunits or rapid trafficking of AMPARsfrom intracellular sites into the postsynaptic density or vice versa.These processes operate at fast time scales, i.e., from seconds tominutes (Montgomery and Madison 2004; Passafaro et al 2001; Bagalet al 2005). Therefore, it is likely that the cognitive effects seen afterketamine administration to healthy volunteers result both from reduc-ing NMDAR EPSCs and from NMDA-dependent synaptic plasticity,i.e., changes in the functional state and number of AMPARs.

    erimental Finding Reference

    normal Functional Connectivity Between Temporalnd Frontal Regions as Measured by PET and fMRI

    Friston and FrFriston et al 1Lawrie et al 20Meyer-Linden

    normal Gamma Synchrony During Sensoryrocessing (as measured by MEG/EEG)

    Lee et al 2003Spencer et al Symond et al

    normal Patterns of Functional Interactions aseasured by EEG

    Breakspear etKoukkou et alSaito et al 199

    This table lists some experimental findings related to abnormal functional clarge majority of studies have found the same abnormality in schizophrencted and representative subset of publications for each finding; for most pPET, positron-emission tomography; fMRI, functional magnetic resonancew.sobp.org/journalriguing, electrophysiological signatures of impaired sensoryrning are found consistently in schizophrenic patients and caninduced pharmacologically in healthy volunteers. As de-ibed below, schizophrenic patients show a significant reduc-n in an event-related potential (ERP), the mismatch negativ- (MMN). The MMN has been interpreted as an error signal,cited during sensory learning (Friston, 2005a). A replicatedding is that the NMDA antagonist ketamine reduces the MMN,dering it very similar to that of patients (Table 2).Indirect, but important, support for abnormal synaptic plas-ity in schizophrenia is additionally provided by recent ge-me-wide linkage and allelic association studies. Reviewing therrent findings, Harrison and Weinberger (2005) identifieden candidate genes for schizophrenia for which at least threedies provided positive evidence. Remarkably, six of thesenes are intimately related to glutamatergic synapses, notablyDA receptor (NMDAR)-dependent signaling, and/or to neu-odulatory transmitters (Figure 1). Harrison and Weinberger05) concluded that these candidate genes . . . predispose, inious ways but in a convergent fashion, to the central patho-ysiological process: an alteration in synaptic plasticity, espe-lly affecting NMDA receptor (NMDAR)-mediated glutamater-transmission. . . .In conclusion, even though one cannot exclude cellularsconnectivity (along with or as a consequence of impairedaptic plasticity), it seems more promising to pursue the notiont schizophrenia is caused by abnormal synaptic regulation.is is fortuitous, because in the human brain, synaptic plasticityore amenable to experimental study. Measuring the effects ofersible pharmacological manipulations of synaptic plasticityth in vivo imaging techniques is feasible in humans, whereasperimental manipulations of structural connectivity woulduire developmental perturbations.

    urotransmitter Modulation of Synaptic Plasticity

    Although the concept of impaired synaptic plasticity is, on itsn, too broad to be useful for clinical research, it provides amework for developing testable pathophysiological models inizophrenia research. While it is beyond the scope of thisicle to review synaptic plasticity in depth, this section reviewsectively some aspects that are useful to guide the rest of theicle. For details, the reader is referred to comprehensiveiews on synaptic plasticity cited below.Synaptic plasticity can be defined as a state- and history-pendent change in synaptic strength (Zucker and Regehr

    ction) Comments

    95

    et al 2005

    Replicated finding

    w) Replicated finding

    03 Agreement that abnormalities exist, but differencesin the nature of abnormalities reported

    ctivity in schizophrenia. Replicated findingmeans that, to our knowledge,ients. Note that due to editorial space constraints, we can only provide a, additional studies exist that are not cited here.ing; MEG, magnetoencephalogram; EEG, electroencephalogram.wiexreq

    Ne

    owfraschartselartrev

    de

  • 200synthehyPAtheEh(STcomtiono200hodesynintduan

    synAMmeMalatochapsystu(sytorintexatoexalarsonNMpoverDo

    agwhONM19FucanprehipFin

    itfecdeNMprasys

    Inv

    neplaimuse

    4So

    Table 2. Evidence for Abnormal Synaptic Plasticity in Schizophrenia

    Exp s (Sel

    PhaaH

    97891 (re)

    AbbB

    erma0

    Can berge

    Red

    000

    On

    lastici

    K.E. Stephan et al BIOL PSYCHIATRY 2006;59:929939 9312). For glutamatergic synapses, on which we focus here,aptic strength depends on presynaptic transmitter release andnumber and functional states of postsynaptic -amino-3-

    droxy-5-methyl-4-isoxazole propionic acid receptors (AM-Rs). Synaptic plasticity therefore results from changing any ofse factors (Montogomery and Madison 2004; Perez-Otano andlers 2005). A key distinction is between short-term plasticitySP) and long-term plasticity (LTSP). Short-term plasticityprises short-lived phenomena, e.g., rapid synaptic facilita-

    n or depression with time constants around 100 ms, that dot require a remodeling of the synapse (Zucker and Regehr2). In contrast, LTSP includes enduring changes that lasturs and possibly years (long-term potentiation and long-termpression) and depend on lasting structural alterations of theapse. Clearly, this classification is oversimplified; there areermediate processes like early LTP that have a rapid onsete to phosphorylation and rapid trafficking of AMPARs (Freyd Morris 1998; Malinow and Malenka 2002).The NMDARs play a central role in plasticity at glutamatergicapses where the number and functional states of postsynapticPARs are controlled primarily through NMDAR-dependentchanisms (Malinow and Malenka 2002; Montgomery anddison 2004; Perez-Otano and Ehlers 2005) (Figure 1). Modu-ry transmitters affect synaptic plasticity mainly throughnges in NMDAR function (Gu 2002).4 Given the induction ofchotic symptoms by NMDA antagonists and the geneticdies described above, the relationship between AMPARsnaptic strength), NMDARs (synaptic plasticity), and modula-y neurotransmitters (modulation of plasticity) is of greaterest in schizophrenia research. Here, we list just a fewmples of neurotransmitter effects on NMDARs that are relatedcandidate genes for schizophrenia described above. Formple, metabotropic glutamate receptors (mGluR), particu-ly mGluR3 (one of the candidate genes highlighted by Harri- and Weinberger 2005) and mGluR5, interact strongly withDARs. Metabotropic glutamate receptor agonists lead totentiation of NMDAR-mediated responses and can even re-se the effects of NMDAR antagonists (Moghaddam 2003).pamine receptors also strongly alter NMDAR function: D1

    me neurotransmitters like acetylcholine (Massey et al 2001) or dopa-mine (Wolf et al 2003) can affect synaptic plasticity independently ofNMDARs.

    erimental Finding Reference

    rmacologically Induced Symptoms of Psychosisnd Schizophrenia- like Cognitive Deficits inealthy Volunteers

    Allen and Young 1Javitt and Zukin 19Kapur 2003 (review

    normal ERPs of Sensory Learning in Patients cane Mimicked by Pharmacological NMDAlockage in Healthy Controls Subjects

    Kreitschmann-AndUmbricht et al 200

    didate Genes Identified in Genetic Studies Harrison and Wein

    uction in Dendritic Field Size and Spine Density Black et al 2004Garey et al 1998Glantz and Lewis 2Rosoklija et al 2000

    set in Adulthood

    This table lists some experimental findings related to abnormal synaptic pERP, event-related potentials; NMDA, N-methyl-D-aspartate.onists and D2 antagonists increase NMDAR-dependent LTP,ereas D2 agonists decrease it (Centonze et al 2004; Tseng andDonnell 2004). Cholinergic mechanisms greatly modulateDA-dependent LTP and LTD in visual cortex (Brocher et al92; Kirkwood et al 1999) and hippocampus (Yun et al 2000).rthermore, the nicotinergic 7 nicotinic receptor, a putativedidate gene for schizophrenia (Martin et al 2004), is ex-ssed abundantly on glutamatergic synapses in cortex andpocampus both presynaptically and postsynaptically (Fabian-e et al 2001).Given the bottleneck role of NMDARs for synaptic plasticity,seems likely that whatever neurotransmitter systems are af-ted in schizophrenia, the ensuing abnormal plasticity might bependent, at least partially, on dysfunctional modulation ofDAR-dependent processes. The next section considers thectical implications of this observation for research at thetems level.

    estigating Synaptic Plasticity in a Clinical Context

    In this section, we consider how noninvasive functionaluroimaging techniques might afford clinical tests of synapticsticity in schizophrenia. The notion that synaptic plasticity ispaired in schizophrenic patients is not sufficient to specifyful tests (cf. the discussion by Harrison and Weinberger05). The dysconnection hypothesis is more specific and sayst it is not plasticity per se that is abnormal but its modulationring reinforcement and perceptual learning (Friston, 2005b).is raises the possibility that the interaction of NMDAR functiond modulatory neurotransmitter systems lies at the heat of thethophysiology. However, given the polygenetic nature ofizophrenia and evidence from gene expression studies (Mir-s et al 2000), it is likely that mechanisms of abnormaldulation of plasticity differ across patients. For example, ine patients dysfunctional modulation of NMDAR-dependentaptic plasticity might be due primarily to changes in dopami-rgic function, whereas in others it might be caused throughnormal modulation by acetylcholine. The challenge is tovelop suitable psychopharmacological paradigms and dataalyses by which one could possibly distinguish among thesessibilities.The selective dependence of different forms of learning onferent neurotransmitter systems speaks to psychopharmaco-

    ection) Comments

    view)Replicated finding

    hr et al 2001 Replicated finding

    r 2005 (review) Almost all genes are implicated in synapticplasticity (see main text)

    Replicated finding; although structural changes,these are likely consequences of impairedsynaptic plasticity (see main text)

    Speaks to learning-dependent processes andagainst abnormal structural connectivity asthe main cause of schizophrenia

    ty in schizophrenia. See legend to Table 1 for further explanations.20thaduThanpaschnicmosomsynneabdeanpo

    difwww.sobp.org/journal

  • logdifforcoudeofpeoftomsec(Fruseintanpre

    comconindproleainfhastraimrelserassulu

    moconbemolinysi

    untio

    Mo

    synap

    Dy

    debinpecospetoeUspadaallreganexDCpo

    5Se

    Table 3. Evidence for Abnormal Structural Connectivity in Schizophrenia

    Exp es (S

    AbDC

    996Harris

    AbA

    mann1

    WhW

    2

    al 2000205

    Cha 1995 199799

    onne

    932 BIOL PSYCHIATRY 2006;59:929939 K.E. Stephan et al

    wwical studies of healthy volunteers using paradigms that induceferent types of learning (e.g., perceptual, associative, rein-cement, or procedural learning) and have different timerses (i.e., STSP vs. LTSP). For example, there is good evi-

    nce that dopamine is crucial for reinforcement and other formsemotional learning, whereas acetylcholine is important forrceptual learning (Friston, 2005b). One could argue that manythe disintegrative and autistic aspects of schizophrenic symp-s can be viewed as a failure of emotional learning that isondary to a fundamental failure of reinforcement learningiston 1998); this view points to reinforcement learning as aful paradigm. On the other hand, hallucinations could beerpreted as resulting from problems with perceptual learningd inference and a role for sensory paradigms (Friston, inss).Many schizophrenic patients are unable to perform well onplex cognitive tasks. More attractive are tasks that are notfounded by patient performance or effort and are relativelyependent of strategy and attentional set. Arguably, the mostmising paradigm that meets these constraints is implicitrning.5 Implicit learning is non-episodic learning of complexormation in an incidental manner, without awareness of whats been learned (Seger 1994) and independent of the subjectstegy of learning (Chun and Jiang 1998). Some forms ofplicit learning can take place in the absence of attention to theevant stimuli and can progress quickly. This has been ob-ved particularly in the sensory domain, e.g., audio-visualociative learning (McIntosh et al 1998) and learning of stim-s probabilities (e.g., MMN paradigms; see below).In the next section, we review recent developments indeling that are especially relevant to the neurobiologicalsiderations above. These developments enable plasticity tomeasured noninvasively using fMRI and EEG. We then turn tore theoretical models of how the brain actually learns. Thesek dysconnection theories of schizophrenia to empirical anal-s and may provide a mechanistic framework in which to

    veral studies focusing on priming effects in language have shown that,behaviorally, schizophrenic patients are not impaired in implicitlearning (Danion et al 2001). Neurophysiological studies have shown,however, that even in the absence of behavioral differences, theneural dynamics during implicit learning are different in schizo-phrenic patients relative to control subjects (Mathalon et al 2002).Furthermore, implicit learning in other domains shows clear impair-ments in schizophrenia (Schwartz et al 2003).

    erimental Finding Referenc

    errantly Located Neurons in White Matter, Implyingisturbances in Neuronal Migration and Formations ofonnections

    Akbarian et al 1Eastwood and

    normal Cytoarchitecture of Entorhinal Cortex, Implyingberrant Formation of Microcircuits

    Jakob and BeckArnold et al 199

    ite Matter Abnormalities Observed with Diffusioneighted Imaging

    Foong et al 200Steel et al 2001Hulshoff Pol etKubicki et al 20Szeszko et al 20

    nges in Morphology of the Corpus Callosum Woodruff et al Woodruff et al Highley et al 19

    This table lists some experimental findings related to abnormal structural cw.sobp.org/journalderstand how abnormal plasticity might be expressed func-nally.

    deling Synaptic Plasticity

    There are many approaches to modeling learning-relatedaptic plasticity at the systems level. Here, we review twoproaches that may be particularly useful in the present context.

    namic Causal ModelingDynamic causal modeling (DCM) is a generic approach toscribing the dynamics of interacting neural systems. It com-es a model of neural population dynamics, which entailsrturbations (e.g., sensory inputs), intrinsic connections, andntextual modulations of these connections, with a modality-cific forward model (fMRI: Friston et al 2003; EEG/magne-ncephalography [MEG]: David et al 2005, in submission).ing a Bayesian approach, with priors that constrain the valuesrameters can assume, DCMs are fitted such that the predictedta are as similar as possible to the measured responses. Thisows one to quantify and make statistical inferences aboutional responses in terms of the connectivity at the neural leveld, critically, how this connectivity changes as a function ofperimental context (e.g., time or drug effect). For example,M for fMRI is based on the following bilinear model of neuralpulation dynamics

    dz

    dtAz

    j1

    m

    ujB(j)z Cu (1)

    ere z is the state vector (with one state variable per brainion), t is time, and uj is the j-th input to the system (i.e.,me experimentally controlled manipulation). This stateuation represents the strengths of direct inputs to thedeled system (sensory stimuli; C matrix), the strength ofnnections between regions (A matrix), and the modulationthese connections (B(1). . . B(m) matrices) as a function ofgnitive set (e.g., task, attention), time (e.g., learning), org (e.g., ketamine). These parameters correspond to ratenstants of the modeled neurophysiologic processes andve as an index of connection strength. The modulation ofnnections with time allows one to estimate and quantifysticity on a macroscopic scale.This state equation only describes the behavior of largeuronal populations, without reference to any specific neu-hysiologic mechanism by which connection strengths

    election) Comments

    on 2003Agreement that abnormalities exist, but differences

    in the nature of abnormalities reported

    1986 Agreement that abnormalities exist, but differencesin the nature of abnormalities reported

    4

    Inconsistent findings across studies, severalnegative results

    (meta-analysis) Inconsistent findings across studies

    ctivity in schizophrenia. See legend to Table 1 for further explanations.whregsoeqmocoofcodrucosercopla

    nerop

  • chareforonDychbinnecatlinA ddifmopreanbyvedif

    pirmape

    FigLeaposstaNM(ACaceHararroprorec

    K.E. Stephan et al BIOL PSYCHIATRY 2006;59:929939 933ange (cf. Stephan 2004). Dynamic causal models for ERPsbased on a much more realistic neural model and account,example, for the influences of different neurotransmitters connection strengths (David et al 2005, in submission).namic causal models may be very useful in studyinganges in connectivity during learning, particularly in com-ation with pharmacological paradigms that target differenturotransmitters. For example, dopamine has been impli-ed in reinforcement (emotional) learning, while acetylcho-e may be important for perceptual learning (Friston, 2005b).ichotomy among schizophrenic subgroups with regard toferential impairments in dopaminergic and cholinergicdulation of NMDA-dependent plasticity might be ex-ssed as differences in plasticity when tested with emotionald perceptual learning paradigms, respectively. Furthermore,combining learning paradigms with pharmacological inter-ntions, the neurotransmitter specificity of learning-relatedferences could be established.Having introduced causal models to assess connections em-ically, we now consider theoretical models of learning thaty constrain the search for changes in connectivity duringrceptual learning.

    ure 1. Strongly simplified schema of the hierarchical relation among synapving presynaptic mechanisms aside, the efficacy or strength of a glutatsynaptic AMPA receptors (AMPARs). The AMPARs are rapidly inserted intotes (Montgomery and Madison 2004; Malinow and Malenka 2002). Both thDA receptors (NMDARs). The function of NMDARs is influencedbymGluRs ah), norepinephrine (NE), serotonin (5HT), and dopamine (DA) (Gu 2002). Notylcholine (Massey et al 2001), can also affect AMPAR function independenrison and Weinberger (2005), six genes (right part of figure) encode protws). The majority of these genes affect NMDAR function directly. A few excesses as well. AMPA, -amino-3-hydroxy-5-methyl-4-isoxazole propioniceptors; mGluRs, metabotropic glutamate receptors; ACh, acetylcholine; NErarchical Empirical BayesianModels ofrceptual InferenceOver the last years, perceptual inference has been increas-ly understood from the perspective of hierarchically orga-ed cortical systems that implement predictive coding based Empirical Bayes (Rao and Ballard 1999; Friston 2003). Here,h level of the hierarchy strives to attain a compromise betweenttom-up information about sensory inputs provided by theel below and top-down predictions (or priors) provided by theel above. We describe briefly how this perspective may proveful for understanding aberrant learning and hallucinations inizophrenia (see Friston 2005a for details).Put simply, the challenge of perceptual inference is to deter-ne the most likely cause, in the external world, of sensory data.e can frame the problem of representing causes in terms of aterministic nonlinear function

    u g (v, ) (2)

    ere v is a vector of causes (e.g., properties of a physicalject), u represents sensory input, and are parameters. g(v,)generative model, i.e., a function that generates data from the

    ength, synaptic plasticity, and its modulation at glutamatergic synapses.rgic synapse depends largely on the number and functional state ofemoved from the cell membrane and have discrete electrophysiologicalfficking and state-dynamics of AMPARs are mainly under the control ofl as by a variety of neuromodulatory transmitters, including acetylcholineat some of these neuromodulators, e.g., dopamine (Wolf et al 2003) and NMDARs. Out of seven candidate genes for schizophrenia identified bynown to be involved in synaptic plasticity and its modulation (dashedles of their functions are listed; note that all genes are involved in otherAMPAR, AMPA receptors; NMDA, N-methyl-D-aspartate ; NMDAR, NMDApinephrine; 5HT, serotonin; DA, dopamine.HiePe

    ingnizoneacbolevlevusesch

    miOnde

    whobis a

    tic strmateand re traswelte thtly ofeins kampacid;, norewww.sobp.org/journal

  • causes. The problem the brain has to contend with is to find afunction of the inputs that recognizes the underlying causes, i.e.,tocormopro

    tiolikinstruconpridebedethe

    wigiv

    whthe

    moeastiois taprecmothiof

    thetostadeinpenquzatcaubuchaconma

    This can be implemented in a simple neuronal architecture asshown in Figure 2.

    whmamoanbeWh(cobaabetetopweco

    leamenealsbu19theleastrmespesynMMlea

    MMDu

    oumohawhby(devaltharelpo(Amieliliga pusepa

    attthrdaseasigbriweet theris

    934 BIOL PSYCHIATRY 2006;59:929939 K.E. Stephan et al

    wwinvert g in Equation 2 and to learn the parameters (whichrespond to connection strengths in the brains generativedel of how inputs are caused). In general, this is a difficultblem because of nonlinear interactions among the causes.Generative models are specified in terms of a prior distribu-n over the causes p(v;) and the generative distribution orelihood of the inputs given the causes p uv;. Importantly,the framework of Empirical Bayes, cortical hierarchies con-ct their own priors. At each level of the hierarchy, theditional density of the causes, i.e., the compromise betweenors from the level above and the likelihood (sensory evi-nce) from the level below, represents the prior for the levellow in the hierarchy. Thus, throughout the hierarchy, priorspend on the sensory input at the lowest level and the priors athighest level:

    u g1(v2, 1)1

    v2 g2(v3, 2)2

    v3 . . .

    (3)

    th u v1. At each level, the conditional density of the causes,en the inputs, is given by the recognition model

    p (v|u;)p (u|v;) p (v;)

    p (u;)(4)

    ere the numerator corresponds to the marginal distribution ofsensory inputs

    p (u;) p (u|v;) p (v;)dv (5)Recognition corresponds to inversion of the generativedel. However, the generative model may not be invertedily, e.g., it may not be possible to parameterize this recogni-n distribution. This is crucial because the end point of learninghe acquisition of a useful recognition model that the brain canply to sensory inputs. One solution is to posit an approximateognition model q(v) that is consistent with the generativedel and that can be parameterized. Estimating the moments ofs density corresponds to inference. Estimating the parametersthe underlying generative model corresponds to learning.Let us take a brief look at how this could be implemented inbrain. Mathematically, neuronal dynamics can be consideredminimize the so-called free energy (F), a concept fromtistical physics (Neal and Hinton 1998). The free energypends on the conditional density of the causes given theuts, or its approximation q(v), and some hyperparameters coding the uncertainty of this approximation. These twoantities can be updated iteratively using expectation maximi-ion (EM). The E-step decreases F with respect to the expectedse, ensuring a good approximation to the recognition distri-tion implied by the parameters . This is inference. The M-stepnges (the synaptic efficacies of forward and backwardnections between levels), enabling the generative model totch the input density. This corresponds to learning:

    Inference E-step: q (v)minq

    F (6)

    Learning M-step:min

    F

    min

    F (7)w.sobp.org/journalPerceptual inference in such a hierarchical system would failen either the parameters or hyperparameters were not esti-ted correctly in the M-step. This would happen with abnormaldulation of synaptic plasticity, because adjusting parametersd hyperparameters corresponds to adjusting synaptic efficaciestween and within neural levels, respectively (see Figure 2).at would be the consequences? In both cases, prediction errornveyed by forward projections) and priors (provided byckward projections) would be wrong. This would result inerrant learning. An interesting case is where the hyperparam-rs, encoding the relative uncertainty about bottom-up and-down information, are learned improperly. If too muchight is afforded to the prior expectation from supraordinatertical levels, hallucinations may result (Friston, in press).In this section, we have focused on models of perceptualrning. Similar themes arise in theoretical models of reinforce-nt learning that were the original motivation for the dyscon-ction hypothesis. These models of value-dependent learningo rely on prediction error, pertaining not to the sensory inputt to the prospective value of a particular action (Friston et al94). The strength of hierarchical Empirical Bayes models is thaty provide a mechanistic understanding of perception andrning. They furnish the principles behind the neuronal infra-ucture that encodes statistical properties of complex environ-ntal input, while accounting for learned regularities. Thisaks to one of the most attractive paradigms for studyingaptic plasticity, i.e., implicit learning, as described above. TheN is a particularly promising example of implicit perceptualrning and is reviewed in the next section.

    N as a Paradigmatic Example of Synaptic Plasticityring Perceptual Learning

    Although MMN can be observed in various sensory domains,r review will be restricted to the auditory system where it isst pronounced and where abnormalities in schizophreniave been described. Classically, a MMN potential is eliciteden a sequence of repeated stimuli (standards) is interrupteda stimulus that differs in intensity, frequency, or durationviant). It is also elicited when temporal (interstimulus inter-s) or higher-order features (patterns) are changed, suggestingt memory traces about regularities of stimulus-to-stimulusationships are encoded (Ntnen et al 2001). Although MMNtentials can be affected by attention under some conditionsrnott and Alain 2002), attentional effects on MMN are usuallynor or absent (Ntnen et al 1993a, 2001). Moreover, MMN iscited in the absence of attention (and even during sleep andht coma) (Atienza et al 2002a; Fischer et al 1999), implying thatreattentive echoic memory trace of the preceding standards isd as a template against which incoming sounds are com-red.Being able to obtain neural signals not confounded byention, cooperation, and performance, except perceptualesholds (Todd et al 2003), makes MMN an interesting candi-te for diagnostically useful paradigms in schizophrenia re-rch. A meta-analysis showed that more than 30 studies foundnificant reductions of MMN amplitude in schizophrenia (Um-cht and Krljes 2005). Moreover, individual MMN correlatesll with disease severity and cognitive dysfunction (Baldewegal 2004) and functional status (Light and Braff 2005). However,re are conflicting reports about its association with genetick for schizophrenia (Michie et al 2002; Bramon et al 2004).

  • stimfornei.e.thoCotheet incinctrape200

    bestrobo200mimereg

    Fig(prencpretex

    K.E. Stephan et al BIOL PSYCHIATRY 2006;59:929939 935The classical view of the MMN as a fixed cortical response toulus change has been challenged by considerable evidencea role of cortical plasticity in MMN generation. Mismatch

    gativity amplitude depends on STSP (immediate stimulus past,, at a time scale of seconds) and LTSP (over hundreds andusands of stimuli, i.e., at a time scale of minutes to hours).ncerning the former, the MMN increases progressively with number of standard repetitions in the stimulus train (Samsal 1983; Imada et al 1993; Javitt et al 1998), leading to anreasing strength of the echoic memory trace. Long-termreases in MMN have been related to emergence of memoryces for complex sounds and are correlated to improvements ofrceptual performance (Ntnen et al 1993b; Atienza et al2b).The role of synaptic plasticity in MMN generation has alsoen established by neuropharmacological studies. There isng evidence for the role of NMDAR in MMN generation inth monkeys (Javitt et al 1996) and humans (Umbricht et al0). Further studies have looked at the effects of neurotrans-tters that modulate activity-dependent synaptic modificationsdiated by NMDARs. Whereas there is still limited informationarding the modulatory effects of serotonin and dopamine

    ure 2. Upper panel: Schematic of a hierarchical neural architecture for imiors) to representations in the level below. Theupper circles represent neuraoding the conditional expectationof causes (). At all levels, these expectatdiction from the level above and the input from the level below. These twot). Lower panel: Details of the influences on representational and error unitviewed in Umbricht and Krljes 2005), converging evidencews that cholinergic stimulation increases and cholinergicckage decreases the MMN amplitude (Baldeweg et al, inss; Harkrider and Hedrick 2005; Pekkonen et al 2001).In terms of the neural mechanisms underlying MMN, somedies have suggested that it could result from stimulus-specificaptation of neurons in primary auditory cortex (Ulanovsky et2003; Jskelinen et al 2004), possibly due to neuronalractoriness and lateral inhibition (May et al 1999). This adap-ion theory, however, is challenged by a variety of findings.st of all, there is strong consensus across studies and tech-ues that MMN responses are not only expressed in primaryditory but also in secondary auditory and prefrontal areasrazdil et al 2005; Doeller et al 2003; Liasis et al 2001; Pincze et2001). This is compatible with strong and reciprocal anatom-l connectivity between auditory and prefrontal areas (Roman- et al 1999). Prefrontal effects on MMN expression are alsogested by studies of patients with prefrontal lesions whohibit diminished temporal MMN amplitudes (Alain et al 1998).ondly, invasive recordings from cat auditory cortex demon-ate that MMN amplitude was inversely related to deviantbability (Csepe et al 1987; Pincze et al 2002). Similar findings,

    enting predictive coding. Each level in this model provides predictionsencodingprediction error () and the lower circles represent neural unitshange tominimize theprediction error, i.e., the discrepancybetween thestraints correspond to prior and likelihood terms, respectively (see main single level in the hierarchy. See Friston (2003, 2005a) for details.(reshoblopre

    stuadal reftatFirniqau(Bal icaskisugexSecstrpro

    pleml unitsions ccon

    s at awww.sobp.org/journal

  • using EEG, have been made in humans (Javitt et al 1998; Winkleret al 1996; Sato et al 2000; Sabri and Campbell 2001). Together,theenplaha200tiogenDeinpretholevprefincalrolchahieanstru

    regpropreconexadigoprellogres

    Co

    paHaapwhunbinprodybinuasufthebraexpsysthacomstacespaindia200(Peturma

    has provided some heuristics for combining functional imagingtechniques with computational models to characterize abnormal-itie

    firsMMMEexuseseclogchthiatresgastual peab

    diabiodis

    chcolres

    Akb

    Ala

    Alle

    An

    Arn

    Arn

    Ati

    Ati

    Bag

    Bak

    Bal

    Bal

    Bla

    936 BIOL PSYCHIATRY 2006;59:929939 K.E. Stephan et al

    wwse findings speak to MMN as an example of how the braincodes the probabilistic structure of sensory input and thusces MMN in the context of perceptual learning. This notions been formalized by predictive coding models (Friston,5a), which, as described above, implicate reciprocal interac-ns between hierarchical processing levels, where higher levelserate predictions about the sensory input to lower levels.viations from that prediction elicit an error (mismatch) signallower levels, which then drives modifications of top-downdiction. The learning of relative stimulus probabilities isught to involve changes in connection strengths betweenels (as in the M-step above). This model makes severaldictions, all of which are supported by the experimentaldings described above: 1) the existence of several, hierarchi-ly arranged, areas that show MMN responses; 2) the centrale of synaptic plasticity, expressed by NMDAR-dependentnges in connection strengths between the levels of thisrarchy; 3) the modulation of this plasticity by acetylcholine;d 4) the dependency of MMN amplitude on the probabilisticcture of the stimulus sequence.Clearly, more work is needed to test detailed predictions withard to how connection strengths change as a function of thebabilistic structure of the input sequence. However, thesedictions can now be tested explicitly in terms of changes innectivity, using DCM (see David et al, in press, for anmple). The key point here is that the combination of para-ms like MMN with causal modeling like DCM provide anportunity to study, noninvasively and quantitatively, learning-ated synaptic plasticity. Combining this with neuropharmaco-ical manipulations may be of great interest for schizophreniaearchers.

    nclusions: Synaptic Plasticity and Schizophrenia

    Many questions about the role of synaptic plasticity in thethophysiology of schizophrenia remain. As pointed out byrrison and Weinberger (2005, p 56), . . . it will not be syn-ses per se but the neural circuits in which they participateich will prove to be the appropriate explanatory level toderstand how the genetic influences operate . . . various com-ations of susceptibility genes can converge on synapticcessing in these microcircuits to effect a common pattern ofsfunction and emergent symptoms, though the specific com-ation of genes and possibly alleles can vary across ill individ-ls. There are two important points here. First, it is notficient to identify susceptibility genes. We need to understandcontributions these genes make to the workings of particularin systems and how changes in the structure and/or theression of these genes impact mechanistically on thesetems. The dysconnection theory is a mechanistic hypothesist can be formulated in terms of (and indeed arose from)putational models of learning. These models help under-

    nd the consequences and loci of abnormal plasticity pro-ses. Second, we need combinations of causal models andradigms (e.g., DCM and MMN) that can characterize plasticityindividual patients and can help clinical decisions, e.g.,gnostic classification and optimal pharmacotherapy (Stephan4). Given some encouraging findings in depression researchzawas et al 2005), one may hope that connectivity estimatesn out to be much more sensitive and specific biologicalrkers of disease than local response amplitudes. This articlew.sobp.org/journals in the modulation of learning-related plasticity.Our own research program is planned in three phases. In thet, we are investigating a variety of perceptual (includingN) and reinforcement learning paradigms, using fMRI andG/EEG in healthy volunteers to establish 1) whether theyhibit robust changes of connectivity and 2) whether they areful candidates for practical use in a clinical setting. In aond step, we will combine these paradigms with pharmaco-ical manipulations and ensure we can measure the ensuinganges in synaptic plasticity in a dose-dependent fashion. Therd step will translate these models into patient studies, lookingpredictive validity in relation to diagnosis and treatmentponse. Hopefully, programs of this nature will provide surro-te markers, based on neuroimaging that can be used in geneticdies (cf. Gottesman and Gould 2003; Egan et al 2004; Baker et2005). As noted above, this is important, not only for thera-utics and drug targeting but also for an understanding ofnormal synaptic regulation at the molecular level.In short, we hope that this sort of work will eventually lead tognostic tests for psychiatric diseases that are as useful asphysical and biochemical tests established in other medicalciplines.

    This work was supported by the Wellcome Trust.We thank our colleagues, particularly Chris Frith and Mi-ael Breakspear, for helpful discussions and apologize to thoseleagues whose work we have been unable to cite due to spacetrictions.

    arian S, Kim JJ, Potkin SG, Hetrick WP, Bunney WE Jr, Jones EG (1996):Maldistribution of interstitial neurons in prefrontal white matter of thebrains of schizophrenic patients. Arch Gen Psychiatry 53:425436.in C, Woods DL, Knight RT (1998): A distributed cortical network forauditory sensory memory in humans. Brain Res 812:2337.nRM,YoungSJ (1978): Phencyclidine-inducedpsychosis.AmJPsychiatry135:10811083.dreasen NC, Nopoulos P, OLeary DS, Miller DD, Wassink T, Flaum M(1999): Defining the phenotype of schizophrenia: Cognitive dysmetriaand its neural mechanisms. Biol Psychiatry 46:908920.old SE, Hyman BT, Van Hoesen GW, Damasio AR (1991): Some cytoarchi-tectural abnormalities of the entorhinal cortex in schizophrenia. ArchGen Psychiatry 48:625632.ott SR, Alain C (2002): Stepping out of the spotlight: MMNattenuation asa function of distance from the attended location.Neuroreport 13:22092212.enza M, Cantero JL, Dominguez-Marin E (2002a): Mismatch negativity(MMN): An objective measure of sensory memory and long-lastingmemories during sleep. Int J Psychophysiol 46:215225.enza M, Cantero JL, Dominguez-Marin E (2002b): The time course ofneural changes underlying auditory perceptual learning. Learn Mem9:138150.al AA, Kao JP, TangCM, Thompson SM (2005): Long-termpotentiation ofexogenous glutamate responses at single dendritic spines. Proc NatlAcad Sci U S A 102:1443414439.er K, Baldeweg T, Sivagnanasundaram S, Scambler P, Skuse D (2005):COMT Val 108/158 Met modifies mismatch negativity and cognitivefunction in 22q11 deletion syndrome. Biol Psychiatry 58(1):2331.deweg T, Klugman A, Gruzelier J, Hirsch SR (2004): Mismatch negativitypotentials and cognitive impairment in schizophrenia. Schizophr Res69:203217.deweg T, Wong D, Stephan KE (in press): Nicotinic modulation of humanauditory sensory memory: Implications for mismatch negativity deficitsin schizophrenia. Int J Psychophysiol.ck JE, Kodish IM, Grossman AW, Klintsova AY, Orlovskaya D, Vostrikov V,et al (2004): Pathology of layer V pyramidal neurons in the prefrontalcortex of patients with schizophrenia. Am J Psychiatry 161:742744.

  • Bleuler E (1911): Dementia Praecox or the Group of Schizophrenias. EnglishTranslation 1961. New York: 9International Universities Press.

    Bra

    Bra

    Bre

    Bro

    Cen

    Chu

    Cse

    Dan

    Dav

    Dav

    Do

    Do

    Eas

    Ega

    Eng

    Fab

    Fisc

    Foo

    Fox

    Fre

    Fris

    FrisFris

    Fris

    FrisFris

    Fris

    Friston KJ, Harrison L, Penny W (2003): Dynamic causal modeling. Neuroim-age 19:12731302.

    Fris

    Ga

    Gla

    Go

    Gu

    Ha

    Ha

    Ha

    Hig

    Ho

    Ho

    Hu

    Hu

    Ima

    J

    Jak

    Jav

    Jav

    Jav

    Kap

    Kel

    Kirk

    Kou

    Kre

    Kub

    K.E. Stephan et al BIOL PSYCHIATRY 2006;59:929939 937mon E, Croft RJ, McDonald C, Virdi GK, Gruzelier JG, Baldeweg T, et al(2004): Mismatch negativity in schizophrenia: A family study. SchizophrRes 67:110.zdil M, Dobsik M, Mikl M, Hlustik P, Daniel P, Pazourkova M, et al (2005):Combined event-related fMRI and intracerebral ERP study of an auditoryoddball task. Neuroimage 26:285293.akspear M, Terry JR, Friston KJ, Harris AW, Williams LM, Brown K, et al(2003): A disturbance of nonlinear interdependence in scalp EEG ofsubjects with first episode schizophrenia. Neuroimage 20:466478.cher S, Artola A, Singer W (1992): Agonists of cholinergic and noradren-ergic receptors facilitate synergistically the induction of long-term po-tentiation in slices of rat visual cortex. Brain Res 573:2736.tonze D, Usiello A, Costa C, Picconi B, Erbs E, Bernardi G, et al (2004):Chronic haloperidol promotes corticostriatal long-term potentiation bytargeting dopamine D2L receptors. J Neurosci 24:82148222.n MM, Jiang Y (1998): Contextual cueing: Implicit learning andmemoryof visual context guides spatial attention. Cognit Psychol 36:2871.pe V, KarmosG,MolnarM (1987): Evokedpotential correlates of stimulusdeviance during wakefulness and sleep in cat-animal model of mis-match negativity. Electroencephalogr Clin Neurophysiol 66:571578.ion JM, Muelemans T, Kauffmann-Muller F, Vermaat H (2001): Intactimplicit learning in schizophrenia. Am J Psychiatry 158:944948.id O, Harrison L, Friston KJ (2005): Modeling event-related responses inthe brain. Neuroimage 25:756770.id O, Kiebel SJ, Harrison LM, Mattout J, Kilner JM, Friston KJ (in submis-sion): Dynamic causal modeling of evoked responses in EEG and MEG.Neuroimage.eller CF, Opitz B, Mecklinger A, Krick C, Reith W, Schroger E (2003): Pre-frontal cortex involvement in preattentive auditory deviance detection:Neuroimaging and electrophysiological evidence. Neuroimage 20:12701282.mino EF, Mirzoyan D, Tsukada H (2004): N-methyl-D-aspartate antago-nists as drugmodels of schizophrenia: A surprising link to tobacco smok-ing. Prog Neuropsychopharmacol Biol Psychiatry 28:801811.twood SL, Harrison PJ (2003): Interstitial white matter neurons expressless reelin and are abnormally distributed in schizophrenia: Towards anintegration ofmolecular andmorphologic aspects of the neurodevelop-mental hypothesis.Mol Psychiatry 8:821831.nMF, Straub RE, Goldberg TE, Yakub I, Callicott JH, Hariri AR, et al (2004):Variation in GRM3 affects cognition, prefrontal glutamate, and risk forschizophrenia. Proc Natl Acad Sci U S A 101:1260412609.ert F, Bonhoeffer T (1999): Dendritic spine changes associated withhippocampal long-term synaptic plasticity. Nature 399:6670.ian-Fine R, Skehel P, Errington ML, Davies HA, Sher E, Stewart MG, et al(2001): Ultrastructural distribution of the alpha7 nicotinic acetylcholinereceptor subunit in rat hippocampus. J Neurosci 21:79938003.her C, Morlet D, Bouchet P, Luaute J, Jourdan C, Salord F (1999): Mis-match negativity and late auditory evoked potentials in comatose pa-tients. Clin Neurophysiol 110:16011610.ng J, Symms MR, Barker GJ, Maier M, Miller DH, Ron MA (2002): Investi-gating regional white matter in schizophrenia using diffusion tensorimaging. Neuroreport 13:333336.K, Sato H, Daw N (1990): The effect of varying stimulus intensity onNMDA receptor activity in cat visual cortex. J Neurophysiol 64:14131428.y U, Morris RGM (1998): Synaptic tagging: Implications for late mainte-nance of hippocampal long-term potentiation. Trends Neurosci 21:181188.ton K (2003): Learning and inference in the brain. Neural Netw 16:13251352.ton KJ (1998): The disconnection hypothesis. Schizophr Res 30:115125.ton KJ (2005a): A theory of cortical responses. Proc R Soc Lond B Biol Sci360:815836.ton KJ (2005b): Dysfunctional connectivity in schizophrenia. AmJ Psychiatry 162:429432.ton KJ (in press): Hallucinations and perceptual inference. Behav Brain Sci.ton KJ, Frith CD (1995): Schizophrenia: A disconnection syndrome? ClinNeurosci 3:8997.ton KJ, Frith CD, Fletcher P, Liddle PF, Frackowiak RS (1996): Functionaltopography: Multidimensional scaling and functional connectivity inthe brain. Cereb Cortex 6:156164.ton KJ, Tononi G, Reeke GH, Sporns O, Edelman GE (1994): Value-depen-dent selection in the brain: Simulation in a synthetic neural model.Neuroscience 39:229243.rey LJ, Ong WY, Patel TS, Kanani M, Davis A, Mortimer AM, et al (1998):Reduced dendritic spine density on cerebral cortical pyramidal neuronsin schizophrenia. J Neurol Neurosurg Psychiatry 65:446453.ntz LA, Lewis DA (2000): Decreased dendritic spine density on prefrontalcortical pyramidal neurons in schizophrenia. Arch Gen Psychiatry 57:6573.ttesman II, Gould TD (2003): The endophenotype concept in psychiatry:Etymology and strategic intentions. Am J Psychiatry 160:636645.Q (2002): Neuromodulatory transmitter systems in the cortex and theirrole in cortical plasticity. Neuroscience 111:815835.rkrider AW,HedrickMS (2005): Acute effect of nicotine on auditory gatingin smokers and non-smokers. Hear Res 202:114128.rrisonPJ (1999): Theneuropathologyof schizophrenia. A critical reviewofthe data and their interpretation. Brain 122:593624.rrison PJ, Weinberger DR (2005): Schizophrenia genes, gene expressionandneuropathology:On thematter of their convergence.Mol Psychiatry10:4068.hley JR, Esiri MM, McDonald B, Cortina-Borja M, Herron BM, Crow TJ(1999): The size and fibre composition of the corpus callosum withrespect to gender and schizophrenia: A post-mortem study. Brain 122:99110.ffman RE, Buchsbaum MS, Escobar MD, Makuch RW, Nuechterlein KH,Guich SM (1991): EEG coherence of prefrontal areas in normal andschizophrenicmales duringperceptual activation. J Neuropsychiatry ClinNeurosci 3:169175.ffmanRE,McGlashanTH (2001): Neural networkmodels of schizophrenia.Neuroscientist 7:441454.a JY, Smith SJ (2004): Neural activity and the dynamics of central nervoussystem development. Nat Neurosci 7:327332.lshoff Pol HE, Schnack HG, Mandl RC, Cahn W, Collins DL, Evans AC, et al(2004): Focal white matter density changes in schizophrenia: Reducedinter-hemispheric connectivity. Neuroimage 21:2735.da T, Hari R, Loveless N, McEvoy L, Sams M (1993): Determinants of theauditory mismatch response. Electroencephalogr Clin Neurophysiol 87:144153.skelinen IP, Ahveninen J, Bonmassar G, Dale AM, Ilmoniemi RJ, LevanenS, et al (2004): Human posterior auditory cortex gates novel sounds toconsciousness. Proc Natl Acad Sci U S A 101:68096814.ob H, Beckmann H (1986): Prenatal developmental disturbances in thelimbic allocortex in schizophrenics. J Neural Transm 65:303326.itt DC, Grochowski S, Shelley AM, Ritter W (1998): Impaired mismatchnegativity (MMN) generation in schizophrenia as a function of stimulusdeviance, probability, and interstimulus/interdeviant interval. Electro-encephalogr Clin Neurophysiol 108:143153.itt DC, Steinschneider M, Schroeder CE, Arezzo JC (1996): Role of corticalN-methyl-D-aspartate receptors in auditory sensory memory and mis-match negativity generation: Implications for schizophrenia. Proc NatlAcad Sci U S A 93:1196211967.itt DC, Zukin SR (1991): Recent advances in the phencyclidine model ofschizophrenia. Am J Psychiatry 148:13011308.ur S (2003): Psychosis as a state of aberrant salience: A framework linkingbiology, phenomenology, and pharmacology in schizophrenia. Am JPsychiatry 160:1323.ly JB, Zhang H (2002): Contribution of AMPA and NMDA receptors toexcitatory responses in the inferior colliculus. Hear Res 168:3542.wood A, Rozas C, Kirkwood J, Perez F, Bear MF (1999): Modulation oflong-term synaptic depression in visual cortex by acetylcholine andnorepinephrine. J Neurosci 19:15991609.kkou M, Lehmann D, Wackermann J, Dvorak I, Henggeler B (1993): Di-mensional complexity of EEG brain mechanisms in untreated schizo-phrenia. Biol Psychiatry 33:397407.itschmann-Andermahr I, Rosburg T, Demme U, Gaser E, Nowak H, SauerH (2001): Effect of ketamine on the neuromagnetic mismatch field inhealthy humans. Brain Res Cogn Brain Res 12:109116.icki M, Westin CF, Maier SE, Frumin M, Nestor PG, Salisbury DF, et al(2002): Uncinate fasciculus findings in schizophrenia: A magnetic reso-nance diffusion tensor imaging study. Am J Psychiatry 159:813820.www.sobp.org/journal

  • Lawrie SM, Buechel C, Whalley HC, Frith CD, Friston KJ, Johnstone EC(2002): Reduced frontotemporal functional connectivity in schizo-

    Lee

    Lia

    Lig

    Ma

    Ma

    Ma

    Ma

    Ma

    Mc

    Me

    Mic

    Mir

    Mo

    Mo

    Mo

    Mo

    N

    N

    N

    Nea

    Pas

    Pek

    Per

    Pez

    Pincze Z, Lakatos P, Rajkai C, Ulbert I, Karmos G (2001): Separation of mis-match negativity and the N1 wave in the auditory cortex of the cat: A

    Pin

    Rao

    Rom

    Ros

    Sab

    Sai

    Sam

    Sat

    Sch

    SegSpe

    Ste

    Ste

    Sul

    Sym

    Sze

    Tod

    Tse

    Ula

    Um

    Um

    Vo

    We

    938 BIOL PSYCHIATRY 2006;59:929939 K.E. Stephan et al

    wwphrenia associated with auditory hallucinations. Biol Psychiatry 51:10081011.KH, Williams LM, Breakspear M, Gordon E (2003): Synchronous gammaactivity: A reviewandcontribution to an integrativeneurosciencemodelof schizophrenia. Brain Res Brain Res Rev 41:5778.sis A, Towell A, Alho K, Boyd S (2001): Intracranial identification of anelectric frontal-cortex response to auditory stimulus change: A casestudy. Brain Res Cogn Brain Res 11:227233.ht GA, Braff DL (2005): Mismatch negativity deficits are associatedwith poor functioning in schizophrenia patients. Arch Gen Psychiatry62:127136.linow R, Malenka RC (2002): AMPA receptor trafficking and synapticplasticity. Annu Rev Neurosci 25:103126.rtin LF, KemWR, Freedman R (2004): Alpha-7 nicotinic receptor agonists:Potential new candidates for the treatment of schizophrenia. Psychop-harmacology (Berl) 174:5464.ssey PV, Bhabra G, Cho K, Brown MW, Bashir ZI (2001): Activation ofmuscarinic receptors induces protein synthesis-dependent long-lastingdepression in the perirhinal cortex. Eur J Neurosci 14:145152.thalon DH, FaustmanWO, Ford JM (2002): N400 and automatic semanticprocessing abnormalities in patients with schizophrenia. Arch Gen Psy-chiatry 59:641648.y P, Tiitinen H, Ilmoniemi RJ, Nyman G, Taylor JG, Naatanen R (1999):Frequency change detection in human auditory cortex. J Comput Neuro-sci 6:99120.Intosh AR, Lobaugh NJ, Cabeza R, Bookstein FL, Houle S (1998): Conver-gence of neural systems processing stimulus associations and coordi-nating motor responses. Cereb Cortex 8:648659.yer-Lindenberg AS, Olsen RK, Kohn PD, Brown T, Egan MF, WeinbergerDR, et al (2005): Regionally specific disturbance of dorsolateral prefron-tal-hippocampal functional connectivity in schizophrenia. Arch Gen Psy-chiatry 62:379386.hie PT, Innes-Brown H, Todd J, Jablensky AV (2002): Duration mismatchnegativity in biological relatives of patients with schizophrenia spec-trum disorders. Biol Psychiatry 52:749758.nics K, Middleton FA, Marquez A, Lewis DA, Levitt P (2000): Molecularcharacterization of schizophrenia viewedbymicroarray analysis of geneexpression in prefrontal cortex. Neuron 28:5367.ghaddamB (2003): Bringing order to the glutamate chaos in schizophre-nia. Neuron 40:881884.nfils MH, Teskey GC (2004): Induction of long-term depression is associ-atedwith decreased dendritic length and spine density in layers III and Vof sensorimotor neocortex. Synapse 53:114121.nfils MH, VandenBerg PM, Kleim JA, Teskey GC (2004): Long-term potentia-tion inducesexpandedmovement representations anddendritic hypertro-phy in layer V of rat sensorimotor neocortex. Cereb Cortex 14:586593.ntgomery JM,MadisonDV (2004): Discrete synaptic states define amajormechanism of synapse plasticity. Trends Neurosci 27:744750.tnen R, Paavilainen P, Tiitinen H, Jiang D, Alho K (1993a): Attention andmismatch negativity. Psychophysiology 30:436450.tnen R, Schroger E, Karakas S, Tervaniemi M, Paavilainen P (1993b):Development of a memory trace for a complex sound in the humanbrain. Neuroreport 4:503506.tnen R, TervaniemiM, Sussman E, Paavilainen P,Winkler I (2001): Prim-itive intelligence in the auditory cortex. Trends Neurosci 24:283288.l RM, Hinton GE (1998): A view of the EM algorithm that justifies incre-mental, sparse and other variants. In: Jordan MI, editor. Learning inGraphical Models. San Diego: Academic Press, 355368.safaroM, Piech V, ShengM (2001): Subunit-specific temporal and spatialpatterns of AMPA receptor exocytosis in hippocampal neurons. NatNeurosci 4:917926.konen E, Hirvonen J, Jaaskelainen IP, Kaakkola S, Huttunen J (2001):Auditory sensory memory and the cholinergic system: Implications forAlzheimers disease. Neuroimage 14:376382.ez-Otano I, EhlersMD (2005): Homeostatic plasticity andNMDA receptortrafficking. Trends Neurosci 28:229238.awas L, Meyer-Lindenberg A, Drabant EM, Verchinski BA, Munoz KE,Kolachana BS, et al (2005): 5-HTTLPR polymorphism impacts humancingulate-amygdala interactions: A genetic susceptibility mechanismfor depression. Nat Neurosci 8:828834.w.sobp.org/journaltopographic study. Clin Neurophysiol 112:778784.cze Z, Lakatos P, Rajkai C, Ulbert I, Karmos G (2002): Effect of deviantprobability and interstimulus/interdeviant interval on the auditory N1andmismatch negativity in the cat auditory cortex. Brain Res Cogn BrainRes 13:249253.RP, BallardDH (1999): Predictive coding in the visual cortex: A functionalinterpretation of some extra-classical receptive-field effects. Nat Neuro-sci 2:7987.anski LM, Tian B, Fritz J, Mishkin M, Goldman-Rakic PS, Rauschecker JP(1999): Dual streams of auditory afferents targetmultiple domains in theprimate prefrontal cortex. Nat Neurosci 2:11311136.oklija G, Toomayan G, Ellis SP, Keilp J, Mann JJ, Latov N, et al (2000):Structural abnormalities of subicular dendrites in subjects with schizo-phrenia and mood disorders: Preliminary findings. Arch Gen Psychiatry57:349356.ri M, Campbell KB (2001): Effects of sequential and temporal probabilityof deviant occurrence on mismatch negativity. Brain Res Cogn Brain Res12:171180.to N, Kuginuki T, Yagyu T, Kinoshita T, Koenig T, Pascual-Marqui RD, et al(1998): Global, regional, and local measures of complexity of multichan-nel electroencephalography in acute, neuroleptic-naive, first-breakschizophrenics. Biol Psychiatry 43:794802.sM, Alho K, Ntnen R (1983): Sequential effects on the ERP in discrim-inating two stimuli. Biol Psychol 17:4158.o Y, Yabe H, Hiruma T, Sutoh T, Shinozaki N, Nashida T, et al (2000): Theeffect of deviant stimulus probability on the human mismatch process.Neuroreport 11:37033708.wartz BL, Howard DV, Howard JH Jr, Hovaguimian A, Deutsch SI (2003):Implicit learning of visuospatial sequences in schizophrenia. Neuropsy-chology 17:517533.er CA (1994): Implicit learning. Psychol Bull 115:163196.ncer KM, Nestor PG, Perlmutter R, Niznikiewicz MA, Klump MC, Fru-min M, et al (2004): Neural synchrony indexes disordered perceptionand cognition in schizophrenia. Proc Natl Acad Sci U S A 101:1728817293.el RM, Bastin ME, McConnell S, Marshall I, Cunningham-Owens DG, Law-rie SM, et al (2001): Diffusion tensor imaging (DTI) and proton magneticresonance spectroscopy (1HMRS) in schizophrenic subjects and normalcontrols. Psychiatry Res 106:161170.phan KE (2004): On the role of general systems theory for functionalneuroimaging. J Anat 205:443470.livan PF, Kendler KS, Neale MC (2003): Schizophrenia as a complex trait:Evidence from a meta-analysis of twin studies. Arch Gen Psychiatry 60:11871192.ond MP, Harris AW, Gordon E, Williams LM (2005): Gamma synchronyin first-episode schizophrenia: A disorder of temporal connectivity? AmJ Psychiatry 162:459465.szko PR, Ardekani BA, Ashtari M, Kumra S, Robinson DG, Sevy S, et al(2000): White matter abnormalities in first-episode schizophrenia orschizoaffective disorder: A diffusion tensor imaging study. Am J Psychia-try 162:602605.d J, Michie PT, Jablensky AV (2003): Association between reduced dura-tion mismatch negativity (MMN) and raised temporal discriminationthresholds in schizophrenia. Clin Neurophysiol 114:20612070.ng KY,ODonnell P (2004): Dopamine-glutamate interactions controllingprefrontal cortical pyramidal cell excitability involve multiple signalingmechanisms. J Neurosci 24:51315139.novsky N, Las L, Nelken I (2003): Processing of low-probability sounds bycortical neurons. Nat Neurosci 6:391398.bricht D, Krljes S (2005): Mismatch negativity in schizophrenia: A meta-analysis. Schizophr Res 76:123.bricht D, Schmid L, Koller R, Vollenweider FX, Hell D, Javitt DC (2000):Ketamine-induced deficits in auditory and visual context-dependentprocessing in healthy volunteers: Implications for models of cognitivedeficits in schizophrenia. Arch Gen Psychiatry 57:11391147.lkowND,Wolf AP, Brodie JD, Cancro R, Overall JE, Rhoades H, et al (1988):Brain interactions in chronic schizophrenics under restingandactivationconditions. Schizophr Res 1:4753.inberger DR, Berman KF, Suddath R, Torrey EF (1992): Evidence of dys-function of a prefrontal-limbic network in schizophrenia: A magnetic

  • resonance imaging and regional cerebral bloodflowstudyof discordantmonozygotic twins. Am J Psychiatry 149:890897.

    Wernicke C (1906): Grundrisse der Psychiatrie. Leipzig, Germany: Thieme.Winkler I, Karmos G, Naatanen R (1996): Adaptive modeling of the unat-

    tended acoustic environment reflected in the mismatch negativityevent-related potential. Brain Res 742:239252.

    WolfME,MangiavacchiS, SunX (2003):Mechanismsbywhichdopamine recep-torsmay influence synaptic plasticity.AnnNYAcad Sci 1003:241249.

    Woodruff PW, McManus IC, David AS (1995): Meta-analysis of corpus callo-sum size in schizophrenia. J Neurol Neurosurg Psychiatry 58:457461.

    Woodruff PW, Phillips ML, Rushe T, Wright IC, Murray RM, David AS (1997):Corpus callosum size and inter-hemispheric function in schizophrenia.Schizophr Res 23:189196.

    YunSH, CheongMY,Mook-Jung I, HuhK, LeeC, JungMW(2000): Cholinergicmodulation of synaptic transmission and plasticity in entorhinal cortexand hippocampus of the rat. Neuroscience 97:671676.

    Zhang LI, Poo MM (2001): Electrical activity and development of neuralcircuits. Nat Neurosci 4(suppl):12071214.

    Zucker RS, Regehr WG (2002): Short-term synaptic plasticity. Annu RevPhysiol 64:355405.

    K.E. Stephan et al BIOL PSYCHIATRY 2006;59:929939 939www.sobp.org/journal

    Synaptic Plasticity and Dysconnection in SchizophreniaAbnormal Anatomical Connections, Abnormal Synaptic Plasticity, or Both?Neurotransmitter Modulation of Synaptic PlasticityInvestigating Synaptic Plasticity in a Clinical ContextModeling Synaptic PlasticityDynamic Causal ModelingHierarchical Empirical Bayesian Models of Perceptual Inference

    MMN as a Paradigmatic Example of Synaptic Plasticity During Perceptual LearningConclusions: Synaptic Plasticity and SchizophreniaAcknowledgmentsReferences