And Re as en 97

8/2/2019 And Re as en 97

http://slidepdf.com/reader/full/and-re-as-en-97 1/8

Linking Mind and Brain in the Study of MentalIllnesses: A Project for a Scientific

PsychopathologyNancy C. Andreasen

Brain research on mental illnesses has made substantial advances in recent years,supported by conceptual and technological developments in cognitive neuroscience.Brain-based cognitive models of illnesses such as schizophrenia and depression havebeen tested with a variety of techniques, including the lesion method, tract tracing,neuroimaging, animal modeling, single-cell recording, electrophysiology, neuropsychol-ogy, and experimental cognitive psychology. A relatively sophisticated picture is emerg-ing that conceptualizes mental illnesses as disorders of mind arising in the brain. Con-vergent data using multiple neuroscience techniques indicate that the neural mecha-nisms of mental illnesses can be understood as dysfunctions in specific neural circuitsand that their functions and dysfunctions can be influenced or altered by a variety ofcognitive and pharmacological factors.

In 1895, a little-known Viennese neuro-

psychiatrist named Sigmund Freud wrote alargely unnoticed work entitled A Project fora Scientific Psychology, in which he proposedthat the cognitive mechanisms of normaland abnormal mental phenomena could beexplained through orderly and rigorousstudy of brain systems. Freud began his ca-reer researching pharmacology (the thera-

peutic effects of cocaine), neurology (apha-sia in children), and basic neuroscience(staining techniques for visualizing neu-rons), but he ultimately abandoned boththe project and neuropsychiatry. During the

fin de siecle 1900s, however, Freud’s projectis slowly being achieved. This fruition re-flects the maturity of the techniques of neuroscience, as well as the convergence of efforts from multiple domains: psychiatry,cognitive psychology and neuropsychology,and clinical and basic neuroscience. Modelsof illness mechanisms have been developedthrough the use of clinical observation, ex-perimental paradigms developed in psychol-ogy, animal and human lesion studies, ana-tomic studies of neural circuits, neuroimag-ing, and behavioral neuropharmacology.The long-term goal is to achieve a “scien-

tific psychopathlogy”: to identify the neuralmechanisms of normal cognitive processesand to understand how they are injured inmental illnesses.

This overview provides a summary of some of the fundamental conceptual issuesthat are being addressed in pursuit of thislong-term goal. The work of neuroscientistsstudying two common mental illnesses—

schizophrenia and depression—illustratesthe consensus that is developing amonginvestigators who begin with different strat-egies originating from different disciplinesin the broad field of cognitive neuroscience.

Fundamental Conceptual Issues

The relationship between mind and brain.Mental illnesses have historically been dis-tinguished from other medical illnesses be-cause they affect the higher cognitive pro-cesses that are referred to as “mind.” Therelationship between mind and brain has

been extensively discussed in contemporaryphilosophy and psychology, without any de-cisive resolution (1). One heuristic solu-tion, therefore, is to adopt the position thatthe mind is the expression of the activity of the brain and that these two are separablefor purposes of analysis and discussion butinseparable in actuality. That is, mentalphenomena arise from the brain, but mentalexperience also affects the brain, as is dem-

onstrated by the many examples of environ-

mental influences on brain plasticity (2).The aberrations of mental illnesses reflectabnormalities in the brain/mind’s interac-

tion with its surrounding world; they arediseases of a psyche (or mind) that residesin that region of the soma (or body) that isthe brain.

Mind and brain can be studied as if theyare separate entities, however, and this isreflected in the multiple and separate disci-plines that examine them. Each uses a dif -ferent language and methodology to studythe same quiddity. The challenge in devel-oping a scientific psychopathology in the1990s is to use the power of multiple disci-plines. The study of mind has been the prov-

ince of cognitive psychology, which has di-vided mind into component domains of in-

vestigation (such as memory, language, andattention), created theoretical systems to ex-

plain the workings of those domains (con-

structs such as memory encoding versus re-trieval), and designed experimental para-digms to test the hypotheses in human be-ings and animals (3). The study of brain hasbeen the province of several disciplines. Neuropsychology has used the lesion methodto determine localization by observing ab-sence of function after injury, whereas neu-

roanatomy and neurobiology have mappedneural development and connectivity andstudied functionality in animal models (4).The boundaries between all these disciplineshave become increasingly less distinct, how-

ever, creating the broad discipline of cogni-tive neuroscience. The term “cognitive” hasdefinitions that range from broad to narrow;its usage here is broad and refers to all activ-

ities of mind, including emotion, perception,and regulation of behavior.

Contemporary psychiatry studies mentalillnesses as diseases that manifest as mindand arise from brain. It is the disciplinewithin cognitive neuroscience that inte-

grates information from all these relateddisciplines in order to develop models thatexplain the cognitive dysfunctions of psy-chiatric patients based on knowledge of normal brain/mind function.

Using the phenomenotype to find the bio-

type. There are at present no known biolog-

ical diagnostic markers for any mental ill-nesses except dementias such as Alzhei-mer’s disease. The to-be-discovered lesionsthat define the remainder of mental illness-es are likely to be occurring at complex orsmall-scale levels that are difficult to visu-alize and measure, such as the connectivity

of neural circuits, neuronal signaling andsignal transduction, and abnormalities ingenes or gene expression. Despite their lackof a defining objective index such as glu-cosuria is for diabetes, however, these ill-nesses are very real. Not only do they pro-

duce substantial morbidity and mortality,but advances in psychiatric nosology haveproduced objective, criterion-based, assess-ment techniques that produce reliable andprecise diagnoses (5). In the absence of apathological marker, the current definitionsof mental illnesses are syndromal and are

The authoris at theMental HealthClinical Research Cen-ter, The University of Iowa Hospitals and Clinics and Col-lege of Medicine, 200 Hawkins Drive, Iowa City, IA52242, USA.

SCIENCE VOL. 275 14 MARCH 1997 http://www.sciencemag.org1586

8/2/2019 And Re as en 97


based on a convergence of signs, symptoms,outcome, and patterns of familial aggrega-tion (6).

Finding the neural mechanisms of men-

tal illnesses must be an iterative process;syndromal clinical definitions (or the phe-nomenotype) are progressively tested, re-

fined, and redefined through the measure-ment of neurobiological aspects (or the bio-

type) (7). This process is not fundamentallydifferent from that used to study other dis-eases. The diagnosis of diabetes, for exam-

ple, has evolved from the observation of glucosuria to multiple subdivisions based onage of onset, severity of symptoms and com-

plications, degree of islet cell involvement,and genetic factors. For most mental illness-es, the task is simply made more challengingby the absence of an objective criterion thatcan provide an initial clue to assist in find-ing mechanisms, as neuritic plaques havedone for Alzheimer’s disease.

Defining the boundary between normal and

abnormal cognitive processes: continua versuscategories. Defining a boundary betweennormal and abnormal cognitive processes,which demarcates the phenomenotype of specific disorders, is often considered to bethe first step in identifying the neural sub-strates of mental illnesses. This task hasbeen difficult. Many of the symptoms of mental illness are on a continuum withnormality. The dysphoric mood of depres-sive illness shares many features with thenormal sadness experienced as a conse-quence of a personal loss such as the deathof a loved one or the termination of amarriage. At what point does such normal

sadness become a form of psychopathology?Most children are rambunctious and haveshort attention spans. At what point doesthis pattern become sufficiently severe todiagnose attention deficit hyperactivity dis-order? Thresholds based on duration (suchas dysphoria that persists for more than amonth after a personal loss) or on severity(such as inattentiveness that interferes sig-nificantly with school performance) areusually applied to resolve this problem (8).These are boundaries of convenience thatpermit reliable definition, not boundarieswith any inherent biological meaning.

This approach to defining the boundariesbetween mental illnesses and normality mayseem imprecise. Traditional medical think-ing teaches that disease processes are discon-

tinuous from normality. One either has can-

cer or one does not. Categorical disease mod-els are being challenged, however, by therecent data indicating that individuals maycarry a genetic risk factor to develop a dis-order that can be measured premorbidly(such as the BRCA1 and ApoE lipoproteingenes) and that may or may not ultimatelybe expressed as the full form of the disorder,

depending on the occurrence of a variety of cofactors (9). Such observations raise ques-tions about the discontinuity requirementfor a definition of a disease process. Currentmodels of the etiology of mental illnesses infact share many features with cancer. In bothgroups of disorders, genetic factors may playa role in inducing vulnerability, but addi-tional “hits” appear to be needed to produce

a full-blown catastrophic condition that is

recognized as an illness. Before the develop-

ment of a full-blown catastrophic condition,however, the boundary between health anddisease may be blurry.

Focus on diseases versus symptoms. Asscientists studying mental illnesses work togive shape and focus to these apparentlyamorphous disease processes, they must alsodetermine whether the problem might bemade more soluble by examination of symp-

toms rather than disease categories. Mostdisease categories are constellations of symptoms. The nature of the constellation

may provide a clue to mechanism (as thecombination of dysphoria with diurnal dys-regulation and appetitive changes suggestsendocrine dysfunction in depression). Analternate strategy in the search for neuralmechanisms is to reduce the field of viewand examine specific components of diseaseprocesses, thereby potentially improvingthe focus of the investigation.

For example, hallucinations are a com-

mon symptom of severe mental illnesses.Understanding their neural mechanismscould potentially tell us something aboutbroader aspects of brain dysfunction in thesedisorders. Hallucinations are defined as sen-

sory experiences (such as seeing objects orpeople, or having the sensation that insectsare crawling under the skin) that occur with-

out any external stimulation. The most com-

mon form of hallucination is auditory. Au-ditory hallucinations are the subjective ex-perience of hearing a voice that is clearlydistinct from the person’s own thoughts andhas the qualities of an auditory perception(such as volume, pitch, and intonation). Theauditory hallucinations experienced during apsychotic illness such as schizophrenia arealmost invariably experienced as humanvoices that speak in sentences or fragments

of sentences. Because the brain mechanismsthat control auditory perception are wellunderstood and those governing languageprocessing are at least partially understood,identifying the brain disturbances that pro-duce this particular symptom could providesome leverage for understanding the natureof the brain injury in schizophrenia. Shiftingthe focus of investigation from disease cate-gory to specific symptoms offers the possibil-ity of making the search for mechanismsmore tractable.

This strategy also introduces other chal-

lenges, however. First, hallucinations occurin a variety of sensory modalities. Auditoryhallucinations are the most common, buthallucinations may also be visual, tactileolfactory, or (very rarely) gustatory. Doesthe same basic process underly all halluci-natory experiences, with some more specificprocess leading them to be expressed in agiven sensory domain? Or is each a different

discrete process? The search for the neuralmechanisms of hallucinations will be drivendifferently, depending on which of thesehypothetical alternatives is chosen. A sin-

gle basic process suggests (but does not re-

quire) a subcortical locus, whereas a discreteprocess suggests specific cortical loci.

Second, hallucinations occur in thesame modality in different disease processes.For example, auditory hallucinations arevery common in mania as well as schizo-

phrenia. Might one more powerfully seekthe common basic mechanism of hallucina-tions by pooling patients who are tradition-

ally studied separately? On the other hand,hallucinations also occur in different mo-dalities in patterns that are highly charac-teristic of different disease processes. Forexample, auditory hallucinations are verycommon in schizophrenia and may co-occur with tactile hallucinations, whereasvisual and olfactory hallucinations aremuch less common. Olfactory hallucina-tions are common in temporal lobe epilep-sy, whereas visual hallucinations are com-

mon in a variety of drug-induced psychoticstates (such as delirium tremens or LSD orPCP psychosis) but not in all (for exampleamphetamine psychosis). The variability in

patterning across disease states arguesagainst the likelihood that hallucinationshare the same mechanism in differentmental illnesses, and it may argue againstpooling patients for study who experiencehallucinations in the same modality inde-pendently of disease process.

Finally, hallucinations have differentand characteristic courses in different dis-ease processes. The auditory hallucinationsof schizophrenia tend to be relativelychronic and lifelong, whereas those of ma-nia are brief and only occur during theeuphoric mood state that defines a manic

episode. This observation suggests that neu-

roanatomic mechanisms exert a more pow-

erful effect in schizophrenia, whereas themechanism driving hallucinations in maniais more rapidly reversible and plastic, andtherefore neurochemical. Both types canhowever, be powerfully affected by the sameantipsychotic drugs and may therefore sharesome common neurochemical mechanism.

Characteristics of heuristic cognitive modelsof mental illnesses. Creating a scientific psy-chopathology requires the development ofheuristic and testable models. Cognitive

http://www.sciencemag.org SCIENCE VOL. 275 14 MARCH 1997 1587

ARTICLES

8/2/2019 And Re as en 97


models of mental illnesses may developfrom a variety of different vantage points,such as neurobiology, neuropsychology,cognitive psychology, or psychiatry. What-ever the point of origin, models convergeon a final common pathway that leads to ashared set of characteristics. The modelsmay begin at the level of systems, but theymust also be adaptable to smaller biological

scales (such as cells, membranes, and mol-

ecules), at which the primary lesions of mental illness probably occur.

Cognitive models of mental illnessesthat can be heuristically applied on multi-ple levels share three characteristics. (i)They provide a general theory of the diseasethat is consistent with our current level of clinical knowledge, based on observation of signs and symptoms, response to treatment,course of illness, and familial/genetic data.(ii) They provide a theory that can betested experimentally in human beings,preferably with several types of techniques

so that comparative confirmatory or discon-

firmatory data can be obtained. (iii) Theyprovide a theory that can also be modeledand tested in animals, because animal mod-els offer the most flexible and powerfultechniques for rapidly screening new treat-ments and for identifying molecular mech-

anisms of illness.

Linking Mind and Brain: TheExamples of Schizophrenia and

Depression

Advances that have been made in the studyof schizophrenia and depression illustrate

the power of developing cognitive modelsthat derive from different perspectives andapply techniques from multiple domains.

Finding the common thread in schizophrenia.The name “schizophrenia” (“fragmentedmind”) was coined by Eugen Bleuler, whowished to emphasize that it was a cognitivedisorder in which the “fabric of thought andemotion” was torn or fragmented, and nor-mal connections or associations were nolonger present (10). Schizophrenia posesspecial challenges to the development of cognitive models because of the breadth anddiversity of its symptoms. The symptoms in-

clude nearly all domains of function: percep-

tion (hallucinations), inferential thinking(delusions), fluency of thought and speech(alogia), clarity and organization of thoughtand speech (“formal thought disorder”), mo-

tor activity (catatonia), emotional expres-sion (affective blunting), ability to initiateand complete goal-directed behavior (avoli-tion), and ability to seek out and experienceemotional gratification (anhedonia). Not allthese symptoms are present in any givenpatient, however, and none is pathogno-

monic of the illness. An initial survey of the

diversity of symptoms might suggest thatmultiple brain regions are involved, in aspotty pattern much as once occurred inneurosyphilis. In the absence of visible le-sions and known pathogens, however, inves-tigators have turned to the exploration of models that could explain the diversity of symptoms by a single cognitive mechanism.Exemplifying this strategy are four different

models that illustrate the melding of cogni-

tive neuroscience and psychiatry, beginningat four different points of departure. Theconvergent conclusions of these differentmodels are striking.

Cognitive psychology. Approachingschizophrenia from the background of cog-nitive psychology, Frith has divided thesymptoms of schizophrenia into three broadgroups or dimensions: disorders of willedaction (which lead to symptoms such asalogia and avolition), disorders of self -mon-

itoring (which lead to symptoms such asauditory hallucinations and delusions of

alien control), and disorders in monitoringthe intentions of others (“mentalizing”)(which lead to symptoms such as “formalthought disorder” and delusions of persecu-tion). Frith believes that all these are spe-cial cases of a more general underlyingmechanism: a disorder of consciousness orself -awareness that impairs the ability tothink with “metarepresentations” (higherorder abstract concepts that are representa-

tions of mental states) (11). Frith and hiscollaborators are currently testing this con-

ceptual framework using positron emissiontomography (PET).

Their earliest efforts focused on willed ac-

tion, tested by giving people tasks for whichthe correct response is not evident from con-

text, such as verbal fluency or choosing afinger movement. Normal people activate afrontal circuit during such tasks, whereas pa-tients with schizophrenia show relative de-creases in frontal regions and increases intemporal regions in comparison with normals(12). If the pace of the verbal fluency task isslowed, however, frontal function is similar tothat of normals and only the temporal abnor-mality remains (13). Molecular manipulationwith dopamine agonists such as apomorphinealso enhances the temporal increases (14).

Examination of the correlations betweenblood flow in frontal and temporal regionssuggests that the normal relationship betweenthem has broken down and that there is ab-

normal functional connectivity (15). Figure 1illustrates the pattern of temporal lobe hyper-activity observed in these studies.

More recently, this group has completeda systematic study of hallucinations. Theyhypothesize that hallucinations are due toan erroneous attribution of the person’sown inner speech to another person, re-flecting a defect in self -monitoring. Starting

first with normals, they developed a taskthat could potentially mimic this mecha-nism of hallucinations; people were asked toperform a sentence completion task andimagine that the response was spoken inanother person’s voice; they found that thistask led to activation of speech productionand perception regions, such as Broca’sarea, the supplementary motor area, and the

left superior and middle temporal regions(16) Applying the same task to people withschizophenia and comparing hallucinatorsto nonhallucinators, they found the hallu-cinators to have decreased flow in the areasused to monitor speech, such as the leftmiddle temporal gyrus and supplementarymotor area (17). In other studies, they haveexamined flow in patients while the pa-tients were experiencing auditory halluci-nations, and they found activations primar-ily in subcortical regions (the thalamus andstriatum), limbic and paralimbic regions(the anterior cingulate and parahippocam-

pal gyrus), and cerebellum. They speculatethat activity in subcortical regions may gen-

erate or moderate hallucinations, whereasthe content (auditory or tactile, for exam-

ple) may be determined by the specific neo-cortical regions that are engaged (18).

Fig. 1. The area in the left superior temporal gyrus

where there is overactivity during the performance

of a word generation task is shown in yellow. The

data came from 12 drug-free schizophrenic pa-

tients (9 of whom were drug-naive) scanned dur-

ing an acute phase of their illness ( 26 ). The hori-

zontal slice shown is 8 mm above the anterior

commissure–posterior commissure line. Signifi-

cant differences derived from a PET study are

shown superimposed on a standard MRI slice in

the space of the atlas of Talairach and Tournoux.

The location of peak activity is at Talairach coor-

dinates 38, 28, 8.


8/2/2019 And Re as en 97


Recently they have also developed anexperimental model for their third symptomdimension: a defect in “mentalizing,” ormonitoring the intentions of others. Peoplewere given a story comprehension task thatrequired the attribution of mental states toothers; during this task, flow was increased inthe left medial frontal gyrus and the posteriorcingulate (19); at present, no reports have

been published concerning performance of this task by schizophrenic patients.

Neurobiology. Approaching schizophre-nia from a background that blends lesionstudies and single-cell recordings in nonhu-man primates for the study of cognition,Goldman-Rakic has proposed a model sug-gesting that the fundamental impairment inschizophrenia is an inability to guide behav-ior by representations, often referred to as adefect in working memory (20). Workingmemory, or the ability to hold a represen-

tation “online” and perform cognitive oper-ations using it, permits individuals to re-

spond in a flexible manner, to formulateand modify plans, and to base behavior oninternally held ideas and thoughts ratherthan being driven by external stimuli (21).A defect in this ability can explain a varietyof symptoms of schizophrenia. For example,the inability to hold a discourse plan inmind and monitor speech output leads todisorganized speech and thought disorder;the inability to maintain a plan for behav-ioral activities could lead to negative symp-

toms such as avolition or alogia; and theinability to reference a specific external orinternal experience against associativememories (mediated by cortical and subcor-tical circuitry involving frontal/parietal/temporal regions and the thalamus) couldlead to an altered consciousness of sensoryexperience that would be expressed as de-

lusions or hallucinations. The model also

explains the perseverative behavior ob-

served in studies using the Wisconsin CardSorting Test and is consistent with thecompromised blood flow to the prefrontalcortex seen in these patients (22).

This model receives strong support frombasic studies of cognition in nonhuman pri-mates, which has permitted Goldman-Ra-kic to map the circuitry engaged by workingmemory tasks. Using a combination of le-sion studies, tract tracing, studies of cerebralmetabolism, and single-cell recordings inawake behaving monkeys, she and her col-leagues have demonstrated that the dorso-

lateral prefrontal cortex (the principal sul-

cus, or Walker’s area 46) plays a key role indelayed response tasks that require the an-

imals to hold the memory of the spatiallocation of a reward online. The dorsolat-eral prefrontal cortex is anatomically direct-ly connected with the posterior parietal cor-tex, which processes and stores visuo-spatialinformation relayed to it from the “dorsal”visual pathway. Other dissociable prefrontalregions appear to be specialized for object

features and to have connections with in-

ferior temporal lobe regions that code suchfeatures (23). Overall, her work supports amodel that suggests a major role for prefron-

tal regions and their multiple distributedcortical, thalamic, and striatal connectionsin a fundamental cognitive function—rep-resentationally guided behavior—that per-mits organisms to adapt flexibly to a chang-

ing environment and to achieve temporaland spatial continuity between past experi-ences and present and future actions.

The applicability of this model to schizo-phrenia has been shown through neuropa-

thology studies that demonstrate increasedcell packing density in the prefrontal cortexin schizophrenic patients, which is consis-tent with a loss of neuropil (24). The phar-macologic agents used to reduce the symp-

toms of schizophrenia have been shown toexert therapeutic efficacy through dopami-nergic blockade, with newer “atypical” neu-

roleptics such as clozapine having more po-

tent effects on D1 and D4 dopamine recep-

tors. Using iontophoresis of a selective D1antagonist onto principal sulcus neurons inmonkeys performing working memory tasks,Williams and Goldman-Rakic have shownthat D1 blockade potentiates their activityperhaps by disinhibiting their excitatory N-

methyl-D-aspartate (NMDA) receptor in-

put (25). Adding support to this hypothesis,the team has also used immunocytochemis-try to show that a D4 receptor antibody

6.0

5.0

4.0

3.0

2.0

1.0

0.0

-1.0

-2.0

-3.0

-4.0

-5.0

-6.0

ControlsPatientsFig. 2. The functional cir-

cuits used by healthy

volunteers (two columnsat right) and by schizo-

phrenic patients (two

columns at left) during

practiced recall of com-

plex narrative material are

shown as visualized with

PET. The images show

the components engaged

in coordinating cognitive

processes through pre-

frontal-thalamic-cerebel-

lar circuitry. Statistical

maps (t maps) of the PET

data, showing significantly

activated regions, are su-

perimposed on a com-posite MR image derived

by averaging the MR

scans from the individu-

als. Regions in red/yellow

tones have higher blood

flow during the memory

task. Two types of statisti-

cal maps are provided.The “peak map” (left column in each pair) shows the

small areas where all contiguous voxels exceed the predefined threshold

for statistical significance (3.61). The “t map” (right column in each pair)

shows the value of t for all voxels in the image and provides a general

overview of the landscape of increases in blood flow during the task. The

value of t is shown on the color scale at right. The controls have activity in

the bilateral frontal opercula, left thalamus, and left planum (transaxial view,

top); left thalamus and cerebellum (sagittal view, middle); and bilateral thalami

(right greater than left) and plana (left greater than right) (coronal view, bottom).

The activity in the right frontal operculum and bilateral thalami is absent in the

patients, and the cerebellar activity is much diminished. These deficits are

presumed to be the basis for their cognitive dysmetria.

ARTICLES


8/2/2019 And Re as en 97


labeled -aminobutyric acid–containing in-

hibitory interneurons in the prefrontal cor-tex, hippocampus, reticular thalamus, glo-

bus pallidus, and substantia nigra; these re-sults are also consistent with the possibilitythat atypical neuroleptics improve schizo-phrenia by disinhibiting excitatory trans-mission in cortical-striatal-thalamic path-

ways (26).

Psychobiology and neurophysiology. Usingtechniques originally derived from neuro-physiology, Braff and colleagues have devel-oped another complementary model. Thismodel begins from the perspective of tech-

niques used to measure brain electrical ac-tivity, particularly various types of evokedpotentials, and hypothesizes that the coreunderlying deficit in schizophrenia involvesinformation processing and attention (27).This model derives from the empirical clin-ical observation that patients with schizo-phrenia frequently complain that they arebombarded with more stimuli than they can

interpret (28). Consequently, they misin-

terpret (that is, have delusions), confuseinternal with external stimuli (hallucina-tions), or retreat to safety (“negative symp-

toms” such as alogia, anhedonia, or avoli-tion). Early interpretations of this observa-tion drew on the Broadbent f ilter theoryand postulated that patients had problemswith early stages in serial order processingthat led to downstream effects such as psy-

chotic or negative symptoms (29). As serialmodels have been supplanted by distributedmodels, the deficit may be better conceptu-alized in terms of resource allocation: Pa-tients cannot mobilize attentional resources

and allocate them to relevant tasks.This group has explored this model using

neurophysiological paradigms to examinesensory gating. Prepulse inhibition is a tech-

nique that measures gating of the startleresponse, which can be triggered by a brightlight or loud auditory stimulus; in normalsthe startle response can be diminished (in-

hibited) if a weak prepulse stimulus is de-livered a short time (milliseconds) before astrong startle-eliciting stimulus. Patientswith schizophrenia have impaired prepulseinhibition, which occurs across a broadrange of stimulus intensities (30). A related

paradigm, gating of the P50 evoked poten-

tial by a prepulse stimulus, has also beenshown by this group and others to be im-

paired in schizophrenic patients (31).Waldo et al. have also shown impairment inP50 gating in unaffected family members of patients with schizophrenia, raising the pos-sibility that this could be used as a trait orvulnerability measure (32).

Prepulse inhibition can be readily mod-eled in animals and used to study the effectsof medications, because all mammals dis-play the startle response. Using lesion

methods and single-cell recordings, thisgroup has demonstrated that cortical-stria-tal-pallidal-thalamic circuitry plays a keyrole in modulating the startle response inrats. They have found that dopamine ago-

nists such as apomorphine can also lead toloss of sensory gating and that both typicaland atypical antipsychotic medications re-store prepulse inhibition in apomorphine-

treated rats in a profile that correlates withtheir clinical efficacy, which suggests thatthis model could serve as a screen for neu-roleptic medications, using a cognitive mea-sure that may be related to the underlyingneural defect in schizophrenia (33).

Clinical psychiatry. Our group has usedthe clinical presentation of schizophrenia asa point of departure, initially attempting tolocalize the various symptoms in brain re-gions through the use of structural andfunctional neuroimaging techniques (34).This strategy has led to a search for abnor-malities in specific brain regions and theo-

ries about symptom-

region relationships(such as negative symptoms in the frontalcortex or hallucinations in the superiortemporal gyrus), which have been exam-

ined by a variety of investigators (35). Thisapproach is oversimplified, however, andwe are currently testing an integrated modelthat explains clinical symptoms as a conse-quence of disruptions in anatomically iden-tified circuits that mediate a fundamentalcognitive process.

Based on relatively consistent observa-tions of abnormalities in frontal, thalamic,and cerebellar regions in schizophrenia, us-ing both magnetic resonance imaging

(MRI) and PET, we have postulated thatthe symptoms arise from impaired connec-tivity between these regions as a conse-quence of a neurodevelopmental defect orperhaps a series of them (36). Motor dys-metria has been observed in schizophreniasince its original description by Kraepelin,and “soft signs” of poor coordination arereported in more contemporary studies(37). More injurious, however, is the relat-ed “cognitive dysmetria,” which produces“poor coordination” of mental activities(37). The word “metron” literally means“measure”: A person with schizophrenia has

a fundamental deficit in taking measure of time and space and in making inferencesabout interrelationships between self andothers, or among past, present, and future.He or she cannot accurately time input andoutput, and therefore cannot coordinate theperception, prioritization, retrieval, and ex-pression of experiences and ideas. This hy-pothesis has received strong support to datefrom our work with PET, which has re-vealed abnormalities in frontal-thalamic-cerebellar circuitry across a broad range of cognitive tasks. Figure 2 shows an image

from a PET study that illustrates these ab-normalities in patients who display cogni-tive dysmetria (38).

The common thread. The common threadin all these observations, spun from fourdifferent starting points, is that schizophre-nia reflects a disruption in a fundamentalcognitive process that affects specific cir-cuitry in the brain and that may be im-

proved through medications that affect thatcircuitry. The various teams use differingterminology and somewhat different con-cepts—metarepresentations, representa-tionally guided behavior, information pro-cessing/attention, cognitive dysmetria—but they convey a common theme. Thecognitive dysfunction in schizophrenia is aninefficient temporal and spatial referencingof information and experience as the personattempts to determine boundaries betweenself and not-self and to formulate effectivedecisions or plans that will guide him or herthrough the small-scale (speaking a sen-

tence) or large-

scale (finding a job) maneu-

vers of daily living. This capacity is some-times referred to as consciousness.

Using diverse technologies and tech-

niques—PET scanning, animal models, lesionmethods, single-cell recordings, evoked po-tentials—the investigators also converge onsimilar conclusions about the neuroanatomicsubstrates of the cognitive dysfunction. Allconcur that it must involve distributed cir-cuits rather than a single specific “localiza-tion,” and all suggest a key role for interrela-tionships among the prefrontal cortex, otherinterconnected cortical regions, and subcorti-cal regions, particularly the thalamus and stri-

atum. Animal and molecular models are beingdeveloped that are based on knowledge of thiscircuitry and the fundamental cognitive pro-

cess, which can be applied to understandingthe mechanism of drug actions and develop-

ing new medications.Dissecting the anatomy of melancholy. De-

pression is the most common of the mooddisorders. It is characterized by an over-whelming sense of sadness, sometimes with-

out any obvious precipitant, accompaniedby slowing of thoughts and actions (psy-chomotor retardation), insomnia, anorexia,alterations in normal diurnal rhythms, dif -

ficulty in concentrating, and feelings of hopelessness and guilt. Because symptomssuch as insomnia or altered diurnal rhythmcould be explained by a disruption in endo-

crine regulation, particularly involving cor-tisol, during the past several decades manyinvestigators have focused on the neuroen-

docrinology of depression (39). There havebeen fewer cognitive models proposed fordepression than for schizophrenia. Howev-er, most investigators agree on the corepsychopathological feature of depression: asevere alteration in emotional tone that


8/2/2019 And Re as en 97


gives a negative coloring to many aspects of the person’s thoughts and behavior.

The lesion method and neuroimaging. Oneapproach to developing a cognitive modelof depression derives from the use of thelesion method. Robinson and others haveexamined the effects on mood of the lesionsproduced by stroke in humans and haveexplored neural mechanisms by studying

the effects of comparable lesions in rats,thereby combining human and animalmodels. They have observed that mood dis-orders occur in 30 to 50% of stroke patientsand that the course and severity of themood disturbance are not related to theseverity of the language, physical, or intel-lectual impairment produced by the stroke(40). Rather, lesion location is a crucialfactor in determining the occurrence andtype of mood disturbance; location explains50 to 70% of the variance in severity (41).Depression is strongly associated with leftanterior frontal lesions, whereas more pos-

terior left frontal lesions do not producesignificant mood disturbance. Further, le-sions in the right hemisphere produce amanic rather than a depressive syndrome(42). Robinson hypothesizes that thechanges in emotion produced by stroke area “frontal syndrome” that varies in emo-tional tone according to the hemisphereinjured by the lesion. Studies of secondarymania suggest that the amygdala may alsobe involved (43).

These changes in mood were modeled inrats by producing left or right hemisphereinfarctions through ligation of the middlecerebral artery. Infarction produces a gen-

eral reduction in biogenic amines, includingboth norepinephine and serotonin, but italso leads to focal decreases in frontal re-gions (44). This neurochemical imbalancecan be reversed by treatment with antide-

pressant medications, which affect norepi-nephine and serotonin systems, indicatingthat an anatomical lesion can be modulatedby chemical mechanisms and that the dis-tinction between “systems” and “molecules”is arbitrary (45).

Imaging studies provide independentconfirmation of many of these observationsin patients suffering from clinical depres-

sion. When compared to healthy controls,patients suffering from a full depressive syn-drome display decreased perfusion in leftdorsolateral and bilateral medial prefrontalregions and the left posterior parietal region(46). Bench et al. have shown that aftertreatment and recovery, patients at leastpartially “normalize” and have relative in-

creases in flow in frontal regions, althoughnot in parietal regions (47). Because thesefrontal regions have also been implicated in“psychomotor poverty dimension” that isrelated to the alogia and avolition of schizo-

phrenia (12), and because psychomotor re-

tardation is an important symptom of de-pression, these findings support the possi-bility that similar symptoms observed indifferent disease conditions may share acommon neural substrate (47). Further,Parkinsonian patients, who also display psy-chomotor retardation, show similar decreas-es in frontal flow in imaging studies (48).

Patients suffering from depression andschizophrenia also have similar impair-ments in sustained attention and similarpatterns of “hypofrontality” in PET studies(49).

Classical conditioning in animals and cog -nitive models of mood. Classical condition-

ing, once treated as a “black box” by behav-iorists, is now being studied by neuroscien-

tists who are identifying how the brainlearns to link a response to a stimulus.LeDoux and Davis have developed a modelthat uses conditioning to explore the rela-tions among emotion, memory, and the

brain (50), and their models provide addi-

tional insights into the neural mechanismsof mood disorders. Combining “fear condi-tioning” in rats with the lesion method andneurophysiological techniques, they haveexplored the neuroanatomy of emotion.This work indicates that exposure to anaversive experience lays down powerfulmemory traces that are difficult to extin-

guish, stored in a circuit in which the amyg-dala plays a central role, but that has tha-

lamic, hippocampal, and cortical compo-nents. There appear to be two paralleltracks for processing emotional stimuli; a

fast track engages subcortical regions (thethalamus and amygdala) and permits rapidbut relatively crude responses, whereas theslower track permits more refined identifi-cation of features and formulation of plansand includes the prefrontal cortex (51).The hippocampus also plays a role by assess-ing contextual cues (52). The molecularmechanism for the formation of emotional

memories appears to involve glutamatergicactivation of NMDA receptors (53).

This work provides a cognitive model forexamining the neural substrates of humanillnesses such as depression or anxiety. Mostcognitive theories of depression posit that itresults from the effects of aversive experi-ences that are stored as memory traces andthat, if reactivated by appropriate cues, willproduce the characteristic negative emo-tional tone of depression. “Learned help-lessness,” the experience of being repeatedlyexposed to physical or mental trauma fromwhich there is no opportunity to escape, is

an example of experimentally induced neg-

ative memories that lead to depressive syn-

dromes in both animals and humans (54).The impact of the aversive memories maybe accessible to consciousness, giving theperson an insight as to why depression istriggered at a given time or in a given placebut the impact may also be unconscioustriggering a depression that appears to be“endogenous.” Because the amygdala hasstrong input to the hypothalamus, the neu-

roendocrinologic aspects of depression arealso explicable: Increasing activity in theamygdala will also trigger adrenocorticotrop-

Medialorbital

Amygdala

t value

0 2.25 4.50

x = 17

Fig. 3. Areas of in-

creased blood flow in the

left amygdala and the

medial orbital cortex of

people with familial, ma-

jor depressive disorder

are shown. The sagittal

image slice shown is

from an image of t val-

ues, produced by a

voxel-by-voxel compu-

tation of the t statistic,

and indicates areas

where flow is increased

in depressives relative to

controls. The statisticalsignificance of abnormal

activity in these regions

has been confirmed with

both regional blood flow

and glucose metabolic

image data from multi-

ple, independent sam-

ples of individuals. The area of increased flow in the left prefrontal cortex extends from the medial orbita

cortex to also involve areas of the lateral orbital, ventrolateral,and ventromedial portions of the prefronta

cortex. The x coordinate locates the image section in millimeters to the left of the midline. Anterior is

toward the reader’s left. [Reprinted from J. L. Price, S. T. Carmichael, W. C. Drevets, Progr. Brain Res.107, 523 (1996), with kind permission of Elsevier Science-NL, Sara Burgerhartstraat 25, 1055 KV

Amsterdam, Netherlands]

ARTICLES


8/2/2019 And Re as en 97


ic hormone release in the hypothalamus,leading in turn to corticosteroid release.

Recent neuroimaging studies in humanshave provided support for a neural model of depression that involves a thalamic-amyg-dala-frontal loop similar to that observed inconditioning studies in animals. Drevets etal. used PET to compare active and remitteddepressives with a strong family history of

mood disorder with normal volunteers inorder to explore state versus trait differences(55). Hypothetically, negative emotionalmemories are a “trait” that is always present;their activation leads to the clinical stateknown as depressive disorder, whereas theirsuppression or modulation (either by phar-macotherapy or psychotherapy) leads to re-mission of symptoms. Drevets et al. foundincreased activity in the left amygdala inboth active and remitted depressives, butonly the active depressives had increasedleft frontal activity, which suggests that theprefrontal-amygdala pathway may be acti-

vated during the depressed state. This isillustrated in Fig. 3. In addition to the fron-

tal regions and amygdala, increased activitywas also observed in the thalamus, puta-men, and hippocampus.

The basic structure. Studies of mooddisorders concur on the general nature of the underlying cognitive process of depres-sion: It is a pathological alteration in emo-

tion. Studies using the lesion method, be-havioral conditioning, and PET have be-gun to dissect the anatomy of emotion:Portions of the prefrontal cortex (probablyanterior and inferior) and the amygdalaare key nodes in the circuit, and the pari-

etal lobes and other regions may play arole as well. Although the anatomical cir-cuits clearly implicate frontal regions andrelated circuitry, the precise cognitivemechanisms are currently an area of activeinvestigation. The data from conditioningstudies and PET studies of active versusremitted patients suggest an interestinghypothesis: Memories of past pain are re-tained in regions such as the amygdala orparietal cortex and may lie dormant (“thetrait”), predisposing an individual to de-veloping a clinical depression (“the state”)if additional factors arise. A key point in

this work, however, is that the anatomy of melancholy can be modified by both psy-chological and chemical/molecular expe-riences. The depressed state can often bereversed through treatment with drugsthat affect the biogenic amine systems of the brain, but it can also be treated withcognitive therapies that attempt to reverse“negative sets,” and combination therapiesare perhaps the most effective of all. De-pression may arise as a consequence of theplastic response of mind/brain to experi-ence, and it may also remit because of

either pharmacologic or psychotherapeu-

tic manipulations of brain plasticity.

Summary and Conclusion

Examples of work applying diverse tech-

niques of cognitive neuroscience to thestudy of depression and schizophrenia indi-cate that increasingly sophisticated strate-

gies and conceptualizations are emerging aspowerful new technologies are being ap-plied. Focal regions have been replaced bycircuits and static changes by plasticity andmolecular mechanisms. The power of mod-els is enhanced by efforts to design experi-ments that can be used in nonhuman spe-cies, in order to obtain in vivo measuresthat will illuminate mechanisms. The pow-

er of neuroimaging is also permitting invivo measures of circuits and mechanismsin the human brain. These advances havecreated an era in which a scientific psycho-pathology that links mind and brain has

become a reality.

REFERENCES AND NOTES___________________________

1. J. R.Searle, The Rediscovery of the Mind (MITPress,Cambridge, MA, 1992); D. C. Dennett, Kinds of Minds: Toward an Understanding of Consciousness(Basic Books, New York, 1996).

2. M. M. Merzenich and K. Sameshima, Curr. Opin.Neurobiol. 3, 187 (1993); W. Singer, Science 270,758 (1995).

3. M. I. Posner and S. E. Peterson, Annu. Rev. Neuro- sci. 13, 25 (1990); E. Tulving, Elements of EpisodicMemory (Oxford Univ. Press, New York and Oxford,1983); J. L. McGaugh, N. M. Weinberger, G. Lynch,Eds., Brain Organization and Memory: Cells, Sys-tems and Circuits (Oxford Univ. Press, New York,1990); L. G. Ungerleider, Science 270, 769 (1995).

4. L. R. Squire and S. Zola-Morgan, The Physiological

Basis of Memory (Academic Press, New York,1983); P. Goldman-Rakic, Trends Neurosci. 7, 425(1984); B. Milner and M. Petrides, ibid., p. 403; A. L.Benton, Contributions to Clinical Neuropsychology (Aldine, Chicago, 1969); N. Geschwind and A. M.Galaburda, Eds., Cerebral Dominance: The Biologi-cal Foundations (Harvard Univ. Press, Cambridge,MA, 1984).

5. American Psychiatric Association, Diagnostic and StatisticalManual of Mental Disorders, Fourth Edition(DSM-IV) (American Psychiatric Press, Washington,DC, 1994);T. A.Widigeretal ., Eds., DSM-IV Source-

book, Volume 1 (American Psychiatric Press, Wash-ington, DC, 1994).

6. E. Robins and S. B. Guze, Am. J. Psychiatry 126,983 (1970).

7. N. C. Andreasen, Schizophr. Bull. 13, 9 (1987).8. J. P. Feighner et al ., Arch. Gen. Psychiatry 26, 57

(1972).

9. J. M. Hall et al ., Science 250, 1684 (1990); A. D.Roses, Annu. Rev. Med. 47, 387 (1996).

10. E. Bleuler, DementiaPraecoxof theGroup of Schizo- phrenias (1911), J. Zinkin, Transl. (International Uni-versities Press, New York, 1950).

11. C. D. Frith, The Cognitive Neuropsychology of Schizophrenia (Erlbaum, East Sussex, United King-dom, 1992).

12. P. F. Liddle et al., Br. J. Psychiatry 160, 179 (1992);C. D. Frith, K. Friston, P. F. Liddle, R. S. Frackowiak,Proc. R. Soc. London Ser. B. 244, 241 (1991).

13. C. Frith, Lancet 346, 615 (1995).14. P. C.Fletcher, C.D. Frith,P. M.Grasby,K. J.Friston,

R. J. Dolan, J. Neurosci. 16, 7055 (1996).15. P. K.McGuireand C.D. Frith,Psychol. Med.26, 663

(1996).

16. P. K. McGuire et al ., ibid., p. 29.17. P. K. McGuire et al ., Br. J. Psychiatry 169, 148

(1996).18. D. A. Silbersweig et al ., Nature 378, 176 (1995).19. P. C. Fletcher et al ., Cognition 57, 109 (1995).20. P. S. Goldman-Rakic, J. Neuropsychiatry Clin. Neu-

rosci. 6, 348 (1994).21. , in Handbook of Physiology F. Plum and V.

Mountcastle, Eds. (American Physiological Society,Bethesda, MD, 1987), pp. 373–417.

22. D. R. Weinberger, K. F. Berman, R. F. Zec, Arch.Gen. Psychiatry 43, 114 (1986).

23. P. S. Goldman-Rakic, Trends Neurosci. 7, 419(1984); in Neurobiology of Neocortex , P. Rakic andW.Singer,Eds.( Wiley, NewYork,1988); Neuron14,477 (1995); H. R. Friedman, J. D. Janas, P. S. Gold-man-Rakic, J. Cogn. Neurosci. 2, 18 (1990); P. S.Goldman-Rakic and L. J. Porrino, J. Comp. Neurol.242, 535 (1985).

24. L. D. Selemon, G.Rajkowska, P. S. Goldman-Rakic, Arch. Gen. Psychiatry 52, 805 (1995).

25. G. V. Williamsand P. S. Goldman-Rakic,Nature 376,572 (1995).

26. L. Mrzljak et al ., ibid. 381, 345 (1996).27. D. L. Braff, Schizophr. Bull. 19, 233 (1993).28. A.McGhieand J.Chapman,Br.J. Med.Psychol. 34,

103 (1961).29. D. E. Broadbent, Perception and Communication

(Pergamon, London, 1958).30. D. L. Braff, C. Grillon, M. A. Geyher, Arch. Gen.

Psychiatry 49, 206 (1992).31. L. L. Judd, L. A. McAdams, B. Budnick, D. L. Braff,

Am. J. Psychiatry 149, 448 (1992).32. M. Waldo et al., Schizophr. Res. 12, 93 (1994).33. N. R. Swerdlow, D. L. Braff, N. Taaid, M. A. Geyer,

Arch. Gen. Psychiatry 51, 139 (1994).34. N. C. Andreasen, Can Schizophrenia be Localized in

the Brain? (American Psychiatric Press, Washington,DC,1986);L. E.DeLisi et al.,Biol. Psychiatry 29, 159(1991); B. Bogerts et al., ibid. 33, 236 (1993); R. B.Zipursky et al., ibid. 35, 501 (1994); R. W. Buchananet al., Am. J. Psychiatry 150, 59 (1993); D. R. Wein-berger, K. F. Berman, R. Suddath, E. F. Torrey, ibid.149, 890 (1992); B. Turetsky et al., Arch. Gen. Psy-chiatry 52, 1061 (1995).

35. E. Barta, G. D. Pearlson, R. E. Powers, S. S. Rich-ards,L. E.Tune, Am.J. Psychiatry 147, 1457(1990);M. E. Shenton et al ., N. Engl. J. Med. 327, 604(1992); N. C. Andreasen et al.,Arch. Gen.Psychiatry 49, 943 (1992); D. H. Ingvar and G. Franzen, Lancet

ii, 1484 (1974); M. S. Buchsbaum et al., Arch. Gen.Psychiatry 39, 251 (1982).

36. N. C. Andreasen, Neuron 16, 697 (1996); et al., J. Am. Med. Assoc. 272, 1763 (1994); N. C.

Andreasen et al ., Science 266, 294 (1994); V. W.Swayze II,N. C. Andreasen, J. C. Ehrhardt,R. Alliger,G. Cohen, Arch. Neurol.47, 805(1990);P. Nopouloset al ., Biol. Psychiatry , in press.

37. E. Kraepelin, R. M. Barclay, G. M. Robertson, De- mentia Praecox and Paraphrenia (E&S Livingstone,Edinburgh, United Kingdom, 1919); S. Gupta, W. C.Hubbard, S. Arndt, M. A. Flaum, N. C. Andreasen,Schizophr. Res. 11, 195 (1994).

38. N. C. Andreasen et al ., Proc. Natl. Acad. Sci. U.S.A.93, 9985 (1996).

39. F. Holsboer, in Handbook of Affective Disorders, E.S. Paykel, Ed. (Churchill Livingstone, Edinburgh,United Kingdom, 1992).

40. S.E. Starkstein andR. E.Robinson, Br.J. Psychiatry

154, 170 (1989).41. R. G. Robinson, K. L. Kubos, L. B. Starr, K. Rao, T.

R. Price, Brain 107, 81 (1984).42. S. E. Starkstein, G. D. Pearlson, J. Boston, R. G.

Robinson, Arch. Neurol. 44, 1069 (1987).43. R. G. Robinson, J. D. Boston, S. E. Starkstein, T. R.

Price, Am. J. Psychiatry 145, 172 (1988).44. R. G. Robinson, Science 205, 707 (1979).45. J.R. Lipsey, R.G. Robinson,G. D.Pearlson, K.Rao,

T. R. Price, Lancet i, 297 (1984).46. L. R. Baxter Jr. et al., Arch. Gen. Psychiatry 49, 681

(1989); J.-L. Martinot et al., Am. J. Psychiatry 147,1313 (1990).

47. C. J. Bench, R. S. J. Frackowiak, R. J. Dolan, Psy-chol. Med. 25, 247 (1995); C. J. Bench et al ., ibid.22, 607 (1992).


8/2/2019 And Re as en 97


48. H. S. Mayberg et al., Ann. Neurol. 28, 57 (1990).49. R. M. Cohen et al., Neuropsychopharmacology 2,

241 (1989).50. J. E. LeDoux, Sci. Am. 6, 50 (June 1994); M. Davis,

Annu. Rev. Neurosci. 15, 353 (1992).51. J. E. LeDoux, Curr. Opin. Neurobiol. 2, 191 (1992); L.

M. Romanski andJ. E. LeDoux, J. Neurosci.12, 4501(1992); J. L. Armony, J. D. Cohen, D. Servan-Schre-iber, J. E. LeDoux, Behav. Neurosci. 109, 246 (1995).

52. K. P. Corodimas andJ. E. LeDoux, Behav. Neurosci.109, 613 (1995).

53. M. J. D. Miserendino, C. B. Sananes, K. R. Melia,M. Davis, Nature 345, 716 (1990); C. Farb et al.,

Brain Res.593, 145 (1992);M. Davis, D.Rainne, M.Cassell, Trends Neurosci. 17, 208 (1994).

54. M. E. P. Seligman, J. Abnorm. Psychol.74, 1 (1976);F. Schneider et al., Am. J. Psychiatry 153, 206(1996).

55. W. C. Drevets et al ., J. Neuroscience 12, 3628(1992).

56. Ssupported in part by National Institiute of MentalHealth grants MH31593, MH40856, and MH-CRC43271; by a Research Scientist Award,MH00625;and by an Established InvestigatorAwardfrom the National Association for Research inSchizophrenia and Affective Disorders.

A Neural Substrate ofPrediction and Reward

Wolfram Schultz, Peter Dayan, P. Read Montague*

The capacity to predict future events permits a creature to detect, model,and manipulatethe causal structure of its interactions with its environment. Behavioral experimentssuggest that learning is driven by changes in the expectations about future salient events

such as rewards and punishments. Physiological work has recently complemented thesestudies by identifying dopaminergic neurons in the primate whose fluctuating outputapparently signals changes or errors in the predictions of future salient and rewardingevents. Taken together, these findings can be understood through quantitative theoriesof adaptive optimizing control.

An adaptive organism must be able topredict future events such as the presence of mates, food, and danger. For any creature,the features of its niche strongly constrainthe time scales for prediction that are likelyto be useful for its survival. Predictions givean animal time to prepare behavioral reac-

tions and can be used to improve the choic-

es an animal makes in the future. Thisanticipatory capacity is crucial for decidingbetween alternative courses of action be-cause some choices may lead to food where-as others may result in injury or loss of resources.

Experiments show that animals can pre-dict many different aspects of their environ-

ments, including complex properties such asthe spatial locations and physical character-istics of stimuli (1). One simple, yet usefulprediction that animals make is the proba-ble time and magnitude of future rewarding

events. “Reward” is an operational conceptfor describing the positive value that a crea-ture ascribes to an object, a behavioral act,

or an internal physical state. The functionof reward can be described according to thebehavior elicited (2). For example, appeti-tive or rewarding stimuli induce approachbehavior that permits an animal to con-sume. Rewards may also play the role of positive reinforcers where they increase the

frequency of behavioral reactions duringlearning and maintain well-established ap-

petitive behaviors after learning. The re-ward value associated with a stimulus is nota static, intrinsic property of the stimulus.Animals can assign different appetitive val-ues to a stimulus as a function of theirinternal states at the time the stimulus isencountered and as a function of their ex-

perience with the stimulus.One clear connection between reward

and prediction derives from a wide varietyof conditioning experiments (1). In theseexperiments, arbitrary stimuli with no in-

trinsic reward value will function as reward-

ing stimuli after being repeatedly associatedin time with rewarding objects—these ob-jects are one form of unconditioned stimu-lus (US). After such associations develop,the neutral stimuli are called conditionedstimuli (CS). In the descriptions that fol-low, we call the appetitive CS the sensorycue and the US the reward. It should bekept in mind, however, that learning thatdepends on CS-US pairing takes many dif -ferent forms and is not always dependent onreward (for example, learning associated

with aversive stimuli). In standard condi-tioning paradigms, the sensory cue mustconsistently precede the reward in order foran association to develop. After condition-

ing, the animal’s behavior indicates that thesensory cue induces a prediction about thelikely time and magnitude of the rewardand tends to elicit approach behavior. Itappears that this form of learning is associ-

ated with a transfer of an appetitive orapproach-eliciting component of the re-ward back to the sensory cue.

Some theories of reward-dependentlearning suggest that learning is driven bythe unpredictability of the reward by thesensory cue (3, 4). One of the main ideas isthat no further learning takes place whenthe reward is entirely predicted by a sensorycue (or cues). For example, if presentationof a light is consistently followed by food, arat will learn that the light predicts thefuture arrival of food. If, after such trainingthe light is paired with a sound and this pair

is consistently followed by food, then some-

thing unusual happens—the rat’s behaviorindicates that the light continues to predictfood, but the sound predicts nothing. Thisphenomenon is called “blocking.” The pre-diction-based explanation is that the lightfully predicts the food that arrives and thepresence of the sound adds no new predic-tive (useful) information; therefore, no as-sociation developed to the sound (5). Itappears therefore that learning is driven bydeviations or “errors” between the predictedtime and amount of rewards and their ac-tual experienced times and magnitudes [butsee (4)].

Engineered systems that are designed tooptimize their actions in complex environ-ments face the same challenges as animals,except that the equivalent of rewards andpunishments are determined by designgoals. One established method by whichartificial systems can learn to predict iscalled the temporal difference (TD) algo-rithm (6). This algorithm was originallyinspired by behavioral data on how animalsactually learn predictions (7). Real-worldapplications of TD models abound. Thepredictions learned by TD methods can alsobe used to implement a technique called

dynamic programming, which specifies howa system can come to choose appropriateactions. In this article, we review how thesecomputational methods provide an inter-pretation of the activity of dopamine neu-

rons thought to mediate reward-processingand reward-dependent learning. The con-

nection between the computational theoryand the experimental results is striking andprovides a quantitative framework for futureexperiments and theories on the computa-tional roles of ascending monoaminergicsystems (8–13).

W. Schultz is at the Institute of Physiology, University of Fribourg, CH-1700 Fribourg, Switzerland. E-mail:[email protected] P. Dayan is in the Depart-ment of Brain and Cognitive Sciences, Center for Bio-logical and Computational Learning, E-25 MIT, Cam-bridge, MA 02139, USA. E-mail: [email protected] P.R. Montague is in the Division of Neuroscience, Centerfor Theoretical Neuroscience, Baylor College of Medi-cine, 1 Baylor Plaza, Houston, TX 77030, USA. E-mail:[email protected]

*To whom correspondence should be addressed.

ARTICLES


And Re as en 97

Documents

Transcript of And Re as en 97