Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=aaas. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Association for the Advancement of Science is collaborating with JSTOR to digitize, preserve and
extend access to Science.
http://www.jstor.org
Single-unitstudiesof striatalneuronsare
The Basal Ganglia and Adaptive in accord with this general view, because
many of these neuronsexhibit highly con-
Motor Control text-dependentfiringpatterns(1 1, 16, 17).
They are active in relation to movements
triggeredby particularcues, includingmem-
Ann M. Graybiel,*ToshihikoAosaki, Alice W. Flaherty, ory-encodedcues as well as environmental
MinoruKimura ones. Most of the cells recorded in the
striatum are its projection neurons. It is
The basal ganglia are neural structures within the motor and cognitive control circuits in these projectionneurons,rather than the
the mammalian forebrain and are interconnected with the neocortex by multiple loops. rarerstriatalintemeurons,that degenerate
Dysfunction in these parallel loops caused by damage to the striatum results in major in Huntington'sdisease (18). By virtue of
defects in voluntary movement, exemplified in Parkinson's disease and Huntington's their differenttransmitterphenotypesand
disease. These parallel loops have a distributed modular architecture resembling local connectivity patterns, striatal projection
expert architectures of computational learning models. During sensorimotor learning, neurons may enhance or suppressmove-
such distributed networks may be coordinated by widely spaced striatal interneurons that ments throughbasalgangliapathwayswith
acquire response properties on the basis of experienced reward. inhibitoryand disinhibitory("release")ef-
fects on their thalamic and brainstemtar-
gets (9, 12, 19). Dysfunctionof these cir-
cuits is thus thought to produceeither hy-
During voluntary movement, large num- counterpartin coherent firing pattems re- pokinetic or hyperkineticbasalgangliadis-
bers of neurons in the forebrain and hind- latedto perceptualbinding.Relativelylittle orders(9, 12, 19).
brain become active. It is a major goal of informationis yet available about coordi- It has puzzledinvestigatorsfor yearsthat
research on the motor system to understand nated neuralfiring in motor circuits,how- there are millions of projectionneuronsin
the particular functions of individual corti- ever, and with the notable exception of the primatestriatum,but they projectto a
cal areas and subcortical sites in generating work on cerebellaroscillations (7), almost very much smaller set of neurons in the
volitional acts. Earlier views of cortical nothing is known about coherence in the basal ganglia output nuclei (20). Why
movement control emphasizing the primary subcorticalstructuresthat provideinforma- would striatalneuronshave such very spe-
motor cortex have now been augmented by tion to primaryand higher order motor cializedpropertiesif they simplyblendtheir
models that involve distributed coding corticalareas. outputsby convergingat the next step in
across multiple cortical areas, selective ac- A particularlyinterestingset of subcor- processing?Partof this puzzlewas resolved
tivation of cortical areas during the prepa- tical structuresare the basalganglia,which when it was discoveredthat there are mul-
ratory phases preceding movement, and dif- in the primatebrainforma massivecollec- tiple, parallel channels from the cortex
ferences in the degree of activation of cor- tion of neuronswith outputsmainlydirect- throughthe basal gangliaand back to the
tical areas before intemally guided and ed towardthe motorand prefrontalareasof cortex (8, 21). But this findingraisedother
stimulus-triggered movements (1). the frontal lobes, as well as towardsome questions:If informationis sent in parallel
Subcortical movement circuits, includ- brainstemmotor sites (8-11). Dysfunction throughthe basalganglia,whatdo the basal
ing the cerebellum and basal ganglia, also of the basal ganglia and the brain nuclei ganglia contributeto the processing?And
show large-scale changes in activation be- interconnectedwith them leads to distur- how are differentaspectsof the processing
fore and during motor activity (2). One of bancesof movementand cognition,includ- coordinated(the bindingproblem)?
the problems posed by such multilevel neu- ing disorderedtiming of movements (12, We focus here on two sets of findings
ral activity is the motor analog to the bind- 13). The anatomical circuit arrangement that suggestnovel approachesto these is-
ing problem in perception (3). How are within which the basal ganglia reside is sues. The first is evidence that the input-
brain circuits coordinated temporally and unique.Their largestinputstation,the stri- outputarchitectureof the sensorimotorstri-
spatially so as to produce coordinated motor atum, collects inputs from the entire neo- atum has a modular design that remaps
acts? cortex and sends processed information cortical inputs onto distributedlocal mod-
One interesting possibility suggested by throughother partsof the basal gangliato ules of striatalprojectionneurons.The sec-
recent work (4) is that synchronous or os- areasof frontal cortex that have been im- ond is evidence that the class of striatal
cillatory responses in motor cortical areas plicated in motor planningand execution. neuronsknown as tonicallyactive neurons
could underlie these functions in a way These striatalcircuitsare modulatedby the (TANs) may contributeto temporalbind-
analogous to that posited for synchronous dopamine-containing nigrostriatal tract, ing across such modularnetworksduring
firing and -40-Hz oscillations in the visual which degeneratesin Parkinson'sdisease. behavioral learning by undergoingdopa-
cortex and thalamus (5). Episodes of syn- The striatumalso receivesinputsfromtha- mine-sensitive changes in their response
chrony involving a broad range of frequen- lamic sites implicatedin rhythmicfiring in properties.
cies have also been observed (6), suggesting the forebrain(14) and from the amygdala,
that the distributed systems known by ana- which functions in reward leaming and Modular Remapping in the
tomical description might have a functional emotionalbehaviors,such as the expression Basal Ganglia
of fear (15).
A. M.GraybielandA.W.Flaherty areinthe Departmentof This arrangement,and the multiple in- An earlyhint that input-outputremapping
BrainandCognitiveSciences, MassachusettsInstitute
of
Technology,45 CarletonStreet, E25-618, Cambridge, temal loops of the basal have
ganglia, led to might occur in modules in the striatum
MA02139, USA.T.Aosakiis inthe Laboratory forNeural speculationthat the basal ganglia are not came from evidence that a dispersed set of
ControlResearchCenter,Institute simply related to motor execution per se.
Circuits,Bio-Mimetic neurochemically specialized patchy zones,
of Physicaland ChemicalResearch (RIKEN), 3-8-31
M.
Rokuban,Atsuta,Nagoya,Aichi456, Japan. Kimura Instead, they may in
participate motor plan- the striosomes, tend to collect inputs relat-
is at OsakaUniversity,
Facultyof Healthand SportSci- ning or predictivecontrol, motor sequenc- ed to the limbic system and to project to the
ences, Machikaneyama, Osaka560, Japan. ing, motor learning,and action repertoires dopamine-containing substantia nigra pars
*Towhomcorrespondenceshouldbe addressed. involvingmotivationaland cognitivedrive. compacta (9, 22). It is now clear, however,
1826 SCIENCE * VOL. 265 * 23 SEPTEMBER1994
ARTICLES
that the entire striatumis modular.There in orderto focusinputsinto specialmodules a given small site in the pallidum have
are input-outputmodulesin the largema- (25). Why go to the expense? about the same dimensions (26-28).
trix compartment,and these "matrisomes" One possible answer to this question Such dispersion is great enough so that
receive sensorimotorand associativeinputs comes from experiments in which ana- the individual modules in any one input-
andprojectto the outputnuclei of the basal tomical tracerswere placed into monkey output set could have different nearest
ganglia(23). brains to label simultaneouslysensorimo- neighbor relations. This design thus fa-
Studying the primate somatic sensory tor inputs to the striatum and striatal vors local spatiotemporalcoherencewithin
and motor cortex, for example, we found outputs to the pallidum, a main output individualinput-outputmodulesas well as
that the inputsderivedfromany smallcor- structureof the basal ganglia. We found diversity across them, which could allow
tical site representinga particularbodypart that labeled input fiber clusterscan over- plasticityand variabilityin striatalprocess-
(say,the contralateralhand) forma distrib- lap clusters of labeled projection neurons ing. Individualmodulesmight also be sub-
uted, partlyinterconnectedset of zones in quite precisely(Fig. 1) (26-28). This sug- ject to differentpatternsof within-module
the striatum (24). Any given matrisome gests a pattern of remappingin the basal convergence(forexample,differentdegrees
receives overlappinginputs from the same ganglia that involves divergencefrom the of submodalityconvergence)andbe subject
body-partrepresentationin differentsubar- cortex to the striatumfollowed by recon- to different sorts of local processing(for
eas of the sensorimotorcortex, so that sev- vergence from the striatum to the palli- example,reflectingdifferentneurochemical
eral sorts of informationrelevant to that dum. In effect, the information is dis- environments).This wouldfurtherincrease
body-partconverge.This suggeststhat the persed to distributedmodules in the stri- the possibilityof dynamicprocessingin ap-
cortical representationsof the body are atum,but it can be broughttogether again parentlyparallelchannels.
remappedin the striatumby being broken at the next stage of processing (Fig. 2). There is a provocative similarity be-
up into local moduleswith convergentin- This divergence-reconvergence pat- tween this biological architectureand the
puts (Fig. 1). tern suggestsone way to have an overall network architectureproposed by Jacobs
It is biologicallyexpensiveto set up such parallel processing scheme for cortico- and colleagues for a supervised learning
patchy distributedsystems.This is not be- basal ganglia interactionsand at the same system (29). The input-output modules
cause they necessitateextra axonal length, time to maximize computational power resemblethe "local experts"of their mod-
but because,in additionto implyinga gen- within channels. The estimated amount el, which are subsets of elements that
eral topographyto organize the connec- of divergence and convergence is impres- perform distinct subtasks of a training
tions, they must also follow local con- sive. The modules labeled from a roughly routine and which therefore split up the
straintsthat breakthe rules of topography 1-mm-widesite in the sensorimotorcor- computational problem. The outputs of
tex stretch over as much as 7 mm of the these local experts can be independently
length and width of the putamen and fill adjusted and gated and then summed at
a volume three to five times the volume an output stage. This architectureshould
|L F.'
of the cortical site from which they were have advantagesfor dealing with the de-
labeled. The striatalmoduleslabeled from grees-of-freedomproblem in motor plan-
3a 4 ning (30). In biological systems, we and
Neocortex others find input and output modules
throughout the striatum, not just in the
sensorimotorstriatum (23). For example,
in the head of the caudatenucleus, which
has been implicated in higher order cog-
nitive functions and has abnormalactiv-
Striatum { ity in some neuropsychiatric disorders
(31 ), this local clusteringphenomenon is
very pronounced,and the limbic system-
related striosomalmodules are most con-
DA
spicuouslyrepresentedthere. Thus, there
GP
I may be many sets of local expert modules,
with some, probablyincluding striosomes,
serving nonsensorimotor functions, and
there may be different sorts of nearest
Fig. 1. Modularity of striatalinputsand outputs
neighbor relations among the modules as
documentedinan axontransportexperiment.(A)
SN well. These input-output modules in the
An injectionof anterogradetracerwas made in a striatum may be one alternative to the
smallsite in.the motorcortex(area4) representing multilayerprocessing typical of the neo-
the foot (\). Inthe same hemisphere,a smallsite in Fig. 2. Model of divergent-reconvergent process- cortex (32).
the pallidumwas injectedwith retrogradetracer ing in basal ganglia pathways. Experimental evi- These computational considerations
(7). Boththe labeledaxon projectionsfromcortex dence favors the divergence of cortical inputs to suggestthat local processingorganizedwith
to terminalsites inthe striatum(\) and the labeled modules in the striatum. Any given module can respectto matrisomaland striosomalmod-
striatopallidal output cells (7) are organizedas receive somatotopically matched inputs, symbol- ules may be a key featureof input-output
sets of patches in the putamen (B). The input ized by F (foot), from differentSI areas (areas 3a, remappingin the striatum.Buthow can the
clusters and outputclusters overlapextensively 3b, and 1) and from the motor cortex (area 4). This
[cross-hatchingin(B)].Corticalrecordingsites (tri- divergence can be followed by reconvergence
activity in differentconstellationsof mod-
angles)andstimulation sites (circles)areshown in onto sets of basal ganglia output cells in the palli- ules be coordinated?Studyingthe physio-
the insetabove (A)(L,leg; F, foot;open symbols, dum (GP). Inputs from the midbrain substantia logicalchangesthat occurin a particularset
no response)for areas 3a and 4. ON, caudate nigra(SN) using the neurotransmitter
dopamine of striatal neurons during behavior learning
nucleus;P, putamen;GPi,internal segment of the (DA)modulatethis processing,as do local inter- in the monkey, we found evidence for one
pallidum. neurons(smalldots). potential solution to this problem.
unilaterallyby local infusionof the neuro- ly nearlysilent neuronswill occur (44). pauseplus a reboundin TAN firing could
toxin 1-methyl-4-phenyl-1,2,3,6-tetrahydro- Interestingly,muscariniccholinergicre- serve as a temporallycoordinatedsignal to
pyridine(MPTP) into the striataof mon- ceptor agoniststend to stabilizethe up or reset the activity states of projectionneu-
keysthat had learnedthe conditioningtask. down states of projectionneuronsby mod- ronsoverwidespreadregionsof the striatum
When the dopamine-containing fiberswere ulationof A-currents(45). If the TANs are (35). Over time, throughthe gradualacqui-
destroyedin one hemisphere,the monkeys cholinergic intemeurons, a coordinated sition of time-lockedconditionedresponses
could still perform (albeit clumsily) the
lickingtask,which requiredmuscleactivity
on both sides.However,mostTANs on the Spikes/s Spikes/s
-30 30
side of the infusion lost their pause re-
sponse, even though their tonic activity 0~ ~~~~~~~~~3
seemed normal.The TANs in the control ILLE -1&,,
i ~-5 w E15
hemisphere,by contrast, continued to re-
spond. These experimentsstronglysuggest
that the expressionof the acquiredresponses
of the TANs requirestonic dopaminergic
input.Moreover,systemicinjectionsof apo-
morphinecould reinstatethe responseslost -'CN/
on the side of the infusion,which suggests
that the dopaminesignalneededforexpres- Spikes/s A15 Spikes/s
sion of the responsemay be spatiotempo- -30 ,< 'A 730
rallypermissive(42).
w&1t5 "'A17 wh/A5
Striatal Activity and 0 .
Sensorimotor Learning
The experimentsdescribedabove suggest
that duringconditioning,striatalTANs de-
velop a responsethat is spatiallydistributed, /<H \S~~~~~
temporallycoordinated,predictive of re- - /> g 9 ~~~~~P
Ah21
ward,and dependenton dopamine.What
Spikes/s \> Spikes/s
could such a neuralsignal accomplish?An r30 * 30
impedimentto answeringthis question is
that the cellularphenotypeof the TANs is Li 491 11L
-~15 j| | 4 f i 15
not known. However, their electrophysi-
ologicalpropertiespoint to their being stri-
atal intemeurons, and in particulartheir
broadaction potentials,prolongedafter-hy-
perpolarizations, and tonic firingcloselyre- 1 ms
semble propertiesof cholinergic interneu-
Fig. 4. Similarityof pause responses acquired by TANs in widely spaced regions of the striatumas a result
rons and do not resemblethose of projec-
of conditioning with clicks and light-emittingdiode (LED)lights as conditioned stimuli.The responses of
tion neurons or noncholinergic interneu- six representative TANs recorded at particularsites (indicated by dots or squares) are shown in individual
rons, as identifiedin studiesin the rat (38, raster plots and spike histograms. Despite their widespread distribution,all of the cells show a pause
43). Our finding that TANs are differen- response after the presentation of one or both of the stimuli used as conditioning stimuli (0, cell that
tially distributedat striosomalbordersand responded to both; V, responsive cell that was tested only with a click).
arestronglyrepresentedin the matrixis also
compatiblewith the view that TANs could
be cholinergicinterneurons(35). Fig. 5. Preferential distribution of
In sharpcontrastto the TANs, striatal TANs in the matrixand at the edges
of striosomes. Shown are strio-
projection neurons are peculiar in being somes (irregularshaded forms) in
almost completelysilent in the absenceof nine serial sections from a squirrel
movement or equivalent activation. But monkey striatum, reconstructed in
once they are broughtto firing threshold, three dimensions and plotted to-
they exhibit burstsof firing (38, 39). They gether with the sites of TANs (red
tend to be in "up"or "down"states and to dots) marked in acute recording ex-
flip between them. It is likely that the periments in this monkey. A dispro-
burstlike projection neuron firing, rather portionate number of TAN sites lie
than the smallmodulationsof TAN firing, at striosome-matrix borders; others
lie in the matrixand may be at ma-
contributesthe immediatetransferof neural trisome-matrisome borders.
informationto striataloutputtargetsin the
pallidum and substantianigra. However,
the TANs could have a considerableinflu-
ence on the subthresholdpotentialsof pro-
jection neurons, potentials that are of great
importance in determining whether and
when the burstlike activity of these normal-
shown after systemic treatment with MPTP [M. Fil- cortex or thalamus or could have maximum effect 50. 0. Hornykiewicz,in Biochemistry and Pharmacology
ion, L. Tremblay, P. I. Bedard, Brain Res. 444, 165 when the inputs to TANs and to TAN targets are of the Basal Ganglia, E. Costa, L. J. Cot6, M. D. Yahr,
(1988); D. S. Rothblat and J. S. Schneider, J. Neu- contemporaneous. Eds. (Raven, New York, 1966), pp. 171-185; 0.
rosci. 13, 4372 (1993)]. 48. H. V. S. Peeke and M. J. Herz, Science 173, 80 Lindvallet al., Science 247, 574 (1990).
43. Y. Kawaguchi, J. Neurophysiol. 67, 1669 (1992); J. (1971); M. Mishkin, B. Malamut, J. Bachevalier, in 51. H. Klawans, C. Goetz, C. Tanner, Eds., Textbook of
Neurosci. 13, 4908 (1993). Neurobiology of Human Leaming and Memory, G. Clinical Neuropharmacology and Therapeutics
44. Because most projection neurons are nearly silent Lynch, J. L. McGaugh, N. M. Weinberger, Eds. (Guil- (Raven, New York, 1992).
except in highly context dependent circumstances ford, New York, 1984), pp. 65-77; D. Cook and R. P. 52. P. Calabresi, R. Maj, A. Pisani, N. B. Mercuri, G.
(16, 17), it is extremely difficultto test directly the Kesner, Behav. Neural Biol. 49, 332 (1988); J. A. Bernardi,J. Neurosci. 12, 4224 (1992); P. Calabresi,
effects of TAN activity on projection neurons during Saint-Cyr, A. E. Taylor, A. E. Lang, Brain 111, 941 R. Maj, N, B. Mercuri,G. Bernardi, Neurosci. Lett.
behavioraltasks. (1988); R. J. McDonald and N. M. White, Behav. 142, 95 (1992).
45. P. T. Akins, D. J. Surmeier, S. T. Kitai,Nature 344, Neurosci. 107, 3 (1993). 53. Supported by Javits Award, NIH grants NS25529
240 (1990). 49. The dopaminergic substantia nigra pars compacta and NEI 02866, the Human FrontierScience Pro-
46. J. Hertz, A. Krogh, R. G. Palmer, Introductionto the has also been implicated in memory and learning in gram, and Japanese ESC grant 03NP01 01. We
Theory of Neural Computation (Addison-Wesley, behavioralstudies [A. Routtenberg and N. Holzman, thank M. Anderson and S. Nelson for comments on
Redwood City, CA, 1991). Science 181, 83 (1973); A. R. Cools, Behav. Brain the manuscript, C. Shneider for typing of the manu-
47. The change in TAN activity could sharpen already Res. 1, 361 (1980); M. G. Packard and N. M. White, script, G. Holm and H. Hallfor technical support, and
fairly well synchronized activity in inputs from the Behav. Neurosci. 105, 295 (1991)]. M. Johnson, who produced the image in Fig. 5.