R E V IV. Cornilleau-P6r& and C. Gielen – Processing of optic flow

AcknowledgementsThiswork was

supportedby EspritProjectMucorrc26615. We thankTjeerdDijkstra,

[acques Droulez,FranckBrernmerand WernerGraf

for valuablecommentson the

manusm”pt.

36 Buizza, A. et al. (1980) Exp. Brain Res. 39, 165-17637 Nakayama, K. (1981) VisionRes. 21, 1475-148238 Dijkstra,T.M.H. et al. (1995) VisionRes. 35, 453-46239 Van den Berg, A.V. and Collewijn, H. (1986) Vision Res. 26,

1209–122240 Ungerleider,L.G. and Mishkin, M. (1982) in Analysis of Visual

Behavior (Ingle, D.J., Goodale,M.A. and Mansfield,R.J.W., eds),pp. 549–586, MIT Press

41 Maunsell, J.H.R. and Van Essen, D.C. (1983) J. Neurosci.3,2563-2586

42 Koenderink, J.J. and Van Doom, A.J. (1975) Optica Acta 22,773-791

43 Saito, H. et al. (1986) J. Neurosci.6, 145-15744 Lagae, L. et aZ. (1994) /. NeurophysioL71, 1597-162645 Duffy, C.J. and Wurtz, R.H. (1991) ]. NeurophysioL 65,

1329-134546 Motter, B.C. and Mountcastle, V.B. (1981) 1. Neccrosci.1, 3–2647 Orban, G.A.et al. (1992) Proc.NatlAcad. Sci. USA89, 2595–259948 Allman,J., Miezin,F. and McGuinness,E. (1985) Perception14,

105–126

49 Maunsell, J.H.R. and Van Essen, D.C. (1983) J. Neurophysiol.49, 1148-1167

50 Roy,J.P. and Wurtz, R.H. (1990) Nature 348, 160-16251 Bradley, D.C., Qian, N. and Andersen,R.A. (1995) Nature 373,

609-61152 Andersen, R.A. (1987) in Handbook of Physiology (Section 1,

Vol. 5) (Brookhart, J.M. and Mountcastle, V.B., eds),PP. 483-518, AmericanPhysiologicalSociety

53 Komatsu, K.and Wurtz, R.H. (1988) ]. NeurophysioL60, 580-60354 Sakata, H., Shibutani, H. and Kawano, K. (1980)

J. Neurophysiol.43, 1654-167255 Kawano, K., Sasaki, M. and Yamashita, M. (1984)

J. Neurophysiol.51, 340-35156 Thier, P. and Erickson, R.G. (1992) Eur.J. Neurosci. 4, 539-55357 Berman, N., Blakemore, C. and Cynader, M. (1975) J. PhysioL

246, 595-61558 Hoffmann, K.P. (1982) in Functional Basis of Ocular Motility

Disorder (Lennerstrand, G., Zee, D.S. and Keller, E.L., eds),pp. 303–311, Pergamon

59 Grasse,K.L. (1994) VisionRes. 34, 1673-1689

J.G.R.Jeff2rysisat the Dept of

Physiology,TheMedicalSchool,

Unive7si@ofBirmingham,

Birmingham,UXB1527T.

MilesA.Whittbcgtonis at

the NeuronalNetworksGroup,

DeptofPhysiologyand Biophysics,

StMary’sHospitalMedicalSchool,

Imperial College,London,

UK W2 lPG.RogerD. Traub is at

the IBMResearchDivision,T.!.

WatsonResearchCenter,Yorktown

Heights,NY 10598,USA,and the Dept

of Neurology,Columbia

Universi~,NewYork,NY 10032,

USA.

Neuronal networks for induced ’40 Hz’rhythmsJohn G.R. Jefferys, Roger D. Traub and Miles A. Whittington

A fast,coherentEEG rhythm,calleda gammaor a ’40Hz’ rhythm,hasbeenimplicatedboth in

higherbrainfunctions,suchasthe‘binding’of featuresthataredetectedbysensorycorticesinto

perceivedobjects,and in lowerlevelprocesses,suchas the phasecodingof neuronalactivity.

Computer simulationsof severalparts of the brain suggestthat gamma rhythms can be

generatedby poolsof excitatoryneurones,networksof inhibitoryneurones,or networksof both

excitatoryand inhibitoryneurones.The strongestexperimentalevidencefor rhythm generators

hasbeenshownfor: (1) neocorticaland thalamicneuronesthat are intrinsic’40Hz’ oscillators,

althoughsynchronystill requiresnetwork mechanisms;and (2) hippocampaland neocortical

networksof mutually inhibitory interneuronesthat generatecollective40Hz rhythms when

excitedtonically.

Trends Neurosci. (1996) 19, 202-208

FAST, GAMMA RHYTHMS have been implicated inhigher cognitive function. They are also known as

‘40 Hz’rhythms, but actuallyrange from 30 to 100Hzand might vary in frequency during a response. The20-100 Hzrangeweconsiderhereoverlapswiththe betaband (15–30Hz)of the EEG,but wewillignorethe finerpoints of EEG classification. The natural history andfunctionalrolesof synchronousgammaoscillationshavebeen reviewedrecentlyl-3,andsowillbe consideredonlybriefly.

Gamma rhythms occur in humans and other mam-mals following sensory stimuli, often in brief runs.‘Inducedrhythms’at 50-60 Hzwerefirstdescribedin theolfactorybulbbyAdrian4,andhavesincebeenidentifiedin the olfactory cortexs,visualcortex3’&9,auditorycor-tex10’11,somatosensorycortexlz and motor cortex13-15.Gammaoscillationsalsooccur in the hippocampuslc’17,wherethe linkwithexternalsensorystimuliis lessdirect,but mightstillexistin the formof multimodalinputsre-ceivedfromhigher-ordersensorycortices.Hippocampalgamma rhythms tend to occur duringthe theta band(4-12 Hz)of the EEG,whichis a prominentfeatureof thehippocampusin vivo1618,especiallyduringexploration.

In humans the auditory response includes brief‘40Hz transientresponses’’9’20,which increasewhenthesubjectpaysattention, andwhichdisappearwith lossofconsciousnessduringanaesthesia21.Repetitiveauditorystimulation at -40 Hz generatesa large ’40 Hz steady-stateresponse’zz.Recordingsof brain magnetic activity(magnetoencephalogramsor MEGs)in humanssuggestthat gammarhythms can be verywidespread23,duringboth waking and dream states. Other MEG measure-ments in humans suggestthat gammarhythms mightbe organizedto sweepacrossthe whole brain, perhapsproviding ‘temporal binding...into a single cognitiveexperience’24.

Neuronalfiring

Single-unit recordings in vivo have revealed muchaboutthe eventsor featuresto whichneuronesrespond.Individualneuronesdonot detecttheirpreferredsensoryfeaturesin isolation,but formpartof neuronalnetworkswhoseemergentpropertiesdefinethe feature-detectionpropertiesof the corticalcolumn. In the visualsystem,itusedto be thought that successivehierarchiesof neur-ones encoded progressivelymore-complex featuresof

202 TINS Vol. 19, No. 5, 1996 Copyright @ 1996, Elsevier Science Ltd. All rights reserved. 0166- 2236/96/S15.00 PII: S0166-2236(96)1 OO23-O

REVIEW J.G.R. Jefferys et cd.– Neuronal networks for induced ‘40 Hz’ rhythms

A B c D

Recurrent Mutual Intrinsic Mutualinhibition excitation oscillator inhibition

in network

Fig. 2. Simplified representationsto illustrate the essentialfeaturesof severalmechanismsproposed ta be involvedin the generation of gamma oscillations. In each case,E (excitatory) and / (inhibitory) representnetworksofneuronesthat are mutually connected,the continuouslines indicate the key connectionsfor their respectivemecha-nisms,and the dot-dash arrow indicatesthe ffowof specificinformation through the network.(A) i//ustratesthe recur-rent inhibitory loop model proposedby Freemanet al.s Computertheoreticalanalysishas subsequentlyshown thatmutual excitationisrequired,but that mutua/ inhibition isnot. (B) showsa simi/armode/in whichthe timede/aysa/ongthe axonscouplinggroupsof excitatoryneuronesplay a keyro/e33,and whichalsoreceivescontributionsfrom recurrentinhibition. (C) proposesthat neuroneswith intrinsic-osci//atorpropertiescan imposetheir own rhythm on the synapticnetworkin which theyare embedded34-37.(D) representsour own mode/of gamma oscillationsin the hippocampus,inwhich interneuronesare tonica//yexcited(thin arrows)so that they will fire at a rate >40 Hz. Thedivergentinhibitoryconnectionsbetweentheseneuronesresult in synchronizedinhibition acrossthe population. When this decays,theneuraneswill discharaedue to the tanic excitationthat drivesthe rhvthm38imDosinoo rhvthm of about 40 Hz. Notice. ,./,that in its simplestform (D) separatesthe role of the oscillator(or clock)and the processor.

sites, or to technical differences(gammarhythms canoccur in brief bursts with a considerable jitter in thefrequency”, so any correlation could conceivably besmudgedout when measurementsareaveragedduring0.5s runs of an EEG, or 20 cycles at 40 Hz; Ref. 28).However,the reasons for these discrepanciesremainunresolved.

Coherent rhythms might have other functions. Oneideais that they providea timing referencefor a neuralcode that depends on the phase relationship of indi-vidual neurones with the reference oscillation. Thestronger the excitation to an individualneurone, theearlierin the cycle it willfire.Thusneuronesthat fireatsimilarphasesin the rhythm will havereceivedsimilarintensifies of input which might be used,for example,to lock their outputstogether for a more effectivesum-mation.Thishypothesiswasproposedforthetarhythmsin the hippocampus30,and also, more recently, forgamma rhythms31.

The roleof gammarhythmsisunknown.Theymightbe central to our cognitivefunction, be fundamentaltothe neural code, have some entirely different role, orsimplybean epiphenomenon32with no deepmeaning.We believethat one keystepto resolvingthese issuesisto understandthe cellularandnetworkmechanismsthatgenerate gamma rhythms, and to developpharmaco-logicaltools that willallowusto probetheirrolesinvivo.

Whatdrivesthegammarhythm?

The original models for binding and segmentationintroduced the idea that neurones that oscillatedtogetheralsoworkedtogethe~s,reflectingthat synchro-nization is a more important factor than a narrowbandwidth327.Although single episodes of neuronalsynchronizationmight occur by chance, repeatedsyn-chronizationis much lesslikelyto do so. Atthe cellular

levelrepeatedsynchronizationcouldpromote temporal summation atactive synapses.

Severaltheories exist for the gen-eration of gamma oscillations invariouspartsof the brain, which allneedfurtherexperimentaltesting.Itis possible that gamma oscillationsariseby differentmechanismsin dif-ferent parts of the brain, and thatseveralmechanismscan combine inindividual regions. Figure2 showshighly simplifiedrepresentationsofsome of the components of thesedifferentmechanisms,whichwewillconsider below in roughly chrono-logical order.Feedback loops between excitatory andinhibitory neurones

(The main regionsimplicatedin-clude: the olfactory bulb, the piri-form cortex, the entorhinal cortexand the primary visual cortex.)Freemanandcolleaguesdevelopedamodel for induced rhythms in sev-eral olfactorystructures,which pro-posedthat synchronous oscillationis generated by a feedback loopbetween excitatory and inhibitoryneuroness.Theyproposedthat somemutual connectivity was also re-

quiredwithinthe poolsofboth excitato~ andinhibitoryneuronesto stabiiizethe oscillations.Ermentrout39hasshown that mutual excitation amongst the excitatoryneurones is necessaryfor stable oscillations to be gen-erated by a recurrent inhibitory loop. However,ourrecent simulationssuggestthat other conditions mightsufficefor stable oscillations (R.D.Traub, unpublishedobservations).

Freemanetal.s predictedthat inhibitory cells shouldlag behind the excitatory cells by a quarter of a cycle(6.5 ms at 40Hz). Experimental support came fromsingle-unitandEEGrecordingsin vivofromthe olfactorybulb, anteriorolfactorynucleus,prepiriformcortex andentorhinalcortexs.The signalsfell into two groups:oneset firedin phasewith the gammaEEG,and one eitherled or laggedthe gamma EEGby a quarter of a cycle.Unfortunatelythese measurementscannot identifythetypes of neurones in each group. In contrast, hippo-campalinterneuronesrecordedduringthe gammaEEGfirein phasewith pyramidalcells”. This is predictedbyour inhibitory networkmodel (see below),both whenisolatedfrom the excitatory network38,and when con-nected with pyramidalcells (R.D.Traub et aZ.,unpub-lished observations). Why the hippocampal and the(superficiallysimilar)olfactorycortical circuitryshoulddifferremainsunclear5.

Wilson and Bowermade similar models of the piri-form cortex33and the primaryvisualcortex32.The geo-metric structure of these models differed, but theessential idea in both wasthat the amplitudeand thefrequency of coherent 30-60Hz oscillations, elicitedby afferent volleys, were determined (or ‘tuned’)by afast-feedbackinhibitory loop (Fig. 2A). Essentially, ifthe stimulus is appropriate (not too strong), enoughactivityin the recurrentexcitatoryconnectionsbetweenpyramidalcells persists after the recurrent inhibition

204 TINS VOI. 19, NO. 5,1996

J.G.R. Jefferys et al.– Neuronal networks for induced ‘40 Hz’ rhythms REVEW

wanes in order to re-excite the pyramidal-cellpopu-lation. In the case of the piriform-cortexmodel, theyshowed that the time constant of inhibition ‘tuned’the frequency of the gamma rhythm, so that longertime spent open for the chloride channels resultedinslowerrhythms (and also a loss of power).

In their model of the primaryvisual cortex Wilsonand Bower32note, in passing, that local mutual inhi-bition betweenthe interneurones‘improvedfrequencylocking and produced auto- and cross-correlationswith more pronouncedoscillatorycharacteristics’.Thisdiffersfrom the central role of similar connections inthe generationof gammarhythmsin the hippocampus,where they wereboth necessaryand sufficient38,40.

In the visual-cortex model, horizontal pyramidalcell axons wereessentialfor long-range(>1mm) cross-correlations32.These had zerophase lag as long as theexcitatory postsynaptic potentials (EPSPS)that theygeneratedwerenot too strong. StrongerEPSPSresultinphaselagsconsistentwith delaysin axonalconduction,while weakerEPSPSwerereminiscent of other kindsoflooselycoupledoscillators.In both the visual-cortexandthe piriform-cortex versions of this model, gammarhythms arose from interactions between networksofexcitatory neurones, could dependon the conductionvelocitiesof intrinsiccorticalconnections (Fig.2B), andwere tuned by the time constants of excitatory andinhibitory synapses.We are not awareof any attemptsto dissectthesecomplexinteractionsexperimentally;inparticular,an investigationof the effectsof conductiondelayson cortical oscillations wouldbe instructive.Intrinsic oscillations in individual neurones

(The main regions implicatedinclude the thalamusandthe neocortex; seeFig.2C.) Neuronesin manypartsof the brain have the intrinsic capacity to oscillate atabout 40 Hz. Space does not permit an exhaustive re-viewof intrinsic oscillators;herewe outline one or tworelevant cases. For example, severaltypes of neuronesin the thaiamocortical system such as the reticular34and intralaminar35neurones do so. In the neocortexitself, examplesof intrinsic cellularoscillators include:sparselyspiny, layer-4 neurones3b,about 20% of long-axon projection neurones in layers 5 and 6 (Ref. 37),and ‘chattering cells’ (cells that fire brief trains ofaction potentials at 200 Hz about 40times a second),which were recently reported in vivo41.

Slice studies revealed that oscillations of 40Hz insparselyspiny neurones in the frontal cortex are gen-eratedby persistent,voltage-dependentNa+currentsanddelayed voltage-dependentrectifier currents3b.Otherfrontal-cortexneuronesuse fastpersistentNa+currents,leak and slownon-inactivating K+currentsto generateoscillations of 4–20 Hz (Ref. 42). Variousmodels sug-gest that similarmechanismscan generateoscillationsof 40 Hz(Ref.43). At least some cortical neuroneswithintrinsic oscillatormechanismsproject to contralateralareas, and to the thalamus, providingroutes for long-rangesynchronizationof theseoscillations37.The exist-ence of cells with intrinsic oscillations at -40 Hz doesnot in itself explainthe synchronizationof local popu-lations of neurones, but it is likely to pace populationrhythms when the neurones are suitably coupled bychemical or electrical synapsesor both44.Networks of inhibitory neurones

(The main regions implicated include the hippo-campus and the parietal neocortex; see Fig. 2D.) Wehaverecentlyproposeda newmodelof gammarhythms

A Isolated interneurone

B Interneurone network

/

50 mV

50 ms

Fig. 3. Inhibitory neuronal networksgenerate gamma oscillations.(A) Computersimulationof the briskexcitationof an iso/atedinhibitoryinterneurone by an injection of current to mimic the activation ofmetabotropic glutamate (mG/u) receptors.(B) The same inhibitoryinterneuroneas part of a network of inhibitory neuronescoupledbyfast, GABA,-mediatedinhibitory postsynapticpatentiak (/f2Ps), Itsresponseto mCJureceptoractivationisnow sculptedinto an osci//ationof 33 Hz by synchronized/P5Psgeneratedby the inhibitory netwark.

based on experiments and computer simulations onthe hippocampal slice (Fig. lB,C). Essentially, whennetworksof inhibitory neurones are tonically excited,they tend to entrain each other into rhythmic firingthroughtheir mutualinhibitoryconnections. Figure3Ashowsa computersimulationof the effectof a depolar-izing current in an isolated interneurone. The rapiddischargebecomes organizedinto a rhythmic patternof -40Hz whenthe interneuroneis synapticallycoupledto a networkof similarlyactivatedinterneurones (Fig.3B). In effect the rhythm is sculpted from the tonicdischarge by synchronous inhibitory postsynapticpotentials (IPSPS).The experimental evidence for thismodel is that synchronousIPSPSat frequenciesin thegamma band occur in hippocampal and neocorticalslices where all monotropicglutamate-receptor con-taining synapses are blocked. In these experiments,interneuronesare excitedby the activation of metabo-tropic glutamate (mGlu) receptors. The experimentalblockadeof fast EPSPSexcludesmodelsthat dependonthese (Fig. 2A–C). We are not awareof hippocampalneuronesthat oscillatepreferentiallyat 40 Hz,and ourcomputer simulations show that such intrinsic oscil-lators are not necessary for gamma oscillations in aneuronal network. In this model, what is necessaryistonic excitationof the interneurones(forexample,frommetabotropic or NMDAreceptors). Fast EPSPSmightbe superimposedon the tonic excitation, but they arenot required,as shown by the original experiments38.

TZNSVOL 19, No. 5,1996 205

REVIEW J.G.R. Jefferys et al. - Ne.renal networks for induced ‘40 Hz’ rhythms

Aa b1.0

co

200 PA 200 PA

~ 0.5 100 ms 100 mskoso

2-0.5 I 1 I 1 1 1

20 40 60 20 40 60t(ins) t(ms)

B 50

1,~,

a

-7&40>0 : o520- 0al4=:30.- 0z=8 b

0020

~o0

00 5 10 15 20 25 30 35

IPSC decay constant, TD(ins)

Fig. 4. The frequency of oscillation in the inhibitory neuronal network is a function of the decay constant of theinhibitory postsynaptic current (IPSC). (A) shows autocorrelationsof voltage-clamprecordingsfrom inhibitoryinterneuronesin stratum oriensmade during an applicationof glutamate in the presenceof drugs to blockianotropicglutamate receptors.(a) Priorto additian of 20~~ pentobarbital, the netwarkoscillatedat 22.7ms [44 Hz; IPSCdecayconstant (TJ was 9.1 i 0.4ms], which is faster than pyramidal cells which have a T. of 22.4 *0.8ms. (b) Afterequilibrationwith pentobarbital theperiodslowedto 44.5 ms (22 Hz; TDreached>30ms). (B) Measurementsmadeofboth networkfrequencyand T. (opencircles)during the wash-inof 2pMpentobarbital reveala closerelationship,whichmatchesthat predictedby computersimulations(filled diamonds).More recentcomputersimulationsmatch the non-linearity found at lower frequenciesand the upperand lower limits to the synchronousnetworkoscillations,fo//owing

40 Figure adapted, with permission,from Ref.38.an increasein the connectivityof the simulatednetwork .

The frequencyof the oscillation is controlled, in part,by the time constant of the fast GABAA-mediatedIPSP.Computersimulationshavepredicted,andexperimentshave confirmed, that drugsthat slowthe decayof theIPSP(forexample,barbiturates),alsoslowthe frequencyof the oscillation(Fig.4). (In 1950,Adrianmadea similarobservationon the olfactorybulb when he noted thatfrequenciesof inducedoscillationsin the olfactorybulbdiffered between urethane-induced and barbiturate-induced anesthesia, at around 50 and 15Hz respec-tively.) In a more extensiveexplorationof the controlof these synchronous, inhibitory gamma oscillations,wefindthat they can existovera rangeof 20-70 Hz,andthat they desynchronize outside of this range40.Theoscillations speed up, in both experiment and simu-lation, with an increasedexcitatory drive,a shortenedinhibitory postsynapticcurrent (IPSC)decay constant(.~A,A),or a decreasedIPSCamplitude.The recordingsalso showthat at least two classesof interneuronepar-ticipatein this activity:fast spikingcellsin strataoriensand pyramidale40.

This inhibitory networkmodel resemblesthe recur-rent inhibitory loop mechanism (above), in that bothare sensitive to IPSPdecay constants. It differsin thatit doesnot requireintact fast EPSPS,and it doesrequire

tonic excitation of the interneur-ones (which is the case experimen-tally in the hippocampus and atleast part of the neocortex). Therecurrentinhibition model appearsto be stabilizedby mutual connec-tions within the population ofinhibitory neurones32(and also bymutuallyexcitatoryneurones39),butthis is very different from the cen-tral role that such connections playin the inhibitory networkmodel.

Thisnewmodelpredictsthat bothexcitatory and inhibitory neuronesfire in phase with the gammarhythm, becauseboth typesof neur-one are clocked by the same popu-lation IPSPS.This is the case in thehippocampus in vivo16,but appar-ently not in the olfactory bulb andrelatedareaswhere a phase lag waspredictedasa resultof the reciprocalactivity in the recurrentexcitatory–inhibitory loops. Interestingly, theolfactorybulband anteriorolfactorynucleus have a peak in power atoscillation frequencies of around75Hz, comparedwith 40-50Hz forthe hippocampus,andcouldusedif-ferent mechanisms. Lateral ento-rhinalcortex and prepiriformcortexhave peaks in power at both thesefrequencies.

Inhibitory network mechanismsmight also function in the neo-cortex. Metabotropic glutamateagonistselicitedgammaoscillationswhen the monotropic glutamatereceptors were blocked pharmaco-logically, much as they did in thehippocampus38.This meansthat thecortical inhibitory networkcan sus-

tain gamma oscillations,but we cannot exclude otherparallel mechanisms. As mentioned above, the neo-cortex contains neurones that are intrinsic oscillatorsat -40 Hz. We predict that inhibitory neurones thatoscillatewilltend to stabilizethe gammarhythm of theinhibitory network. At least some intrinsic oscillatorneurones are inhibitory.

Golomb, Wang and Rinze14’made simulations thatshowthat mutualinhibitioncan entraina network,pro-videdthat the individualneurones, when uncoupled,oscillate with a short periodrelative to the inhibitorytimecourse. In the case of gamma oscillations in thehippocampus,these conditions are met when TGABAisin the range 8–13 ms, and there is sufficient tonic‘drive’to the interneurones (Figs3 and 4). This modelwas developedfor the reticular nucleus of the thala-mus, which participatesin the generation of synchro-nous 7–12Hz ‘spindle’ dischargesand 3 Hz absence-seizuredischarges4s’4G.In both cases,the low threshold(T),voltage-dependentCa2+currentplaysa crucialrole,generatingrebound excitation when the IPSPSdecay.Computer models showed that full synchronizationdependson a sufficientlyslowinhibitorypotentialcom-paredwiththe excitationcomponent(thelow-thresholdCa2+spike)4547.In the case of spindledischargesthis is

206 TINSVOL 19, NO. 5,1996

J.G.R. Jefferys et al. - Neuronal networks for induced ‘40 Hz’ rhythms REVIEW

providedby fastGABAA-mediatedIPSPS,and in the caseof 3 Hzabsence-likeseizures,by slowGABA~-mediatedIPSPS.In practice,in the diencephalicslice,the thalamicreticular nucleus alone does not produce spindle dis-charges. Coherent oscillations also require excitatorysynaptic input from thalamocortical-projection neur-one545,48< Without the projection neuronesthe networktends to generate ‘clusters’, in which only part of thepopulation participatesin the oscillation, which thenhas a frequencythat is an integermultiple (usuallyx 2or x 3) of the mean firing rate of individual neur-ones45.This is perhapseasiestto imaginein the caseoftwo mutually inhibitory neurones with rebound exci-tation; as long as the IPSPSlast long enough to de-inactivatethe T current,then stimulatingone will leadto a persistentsequenceof alternatingdischargesin thetwo neurones.Asyet there is no experimentalevidenceas to whether clusteringexists in thalamic tissue.

Inhibitory networksynchronization differsbetweenthalamus and hippocampus.In gamma oscillations inthe hippocampus,the ‘reboundexcitation’ is not dueto a Ca2+current, but rather to the sustainedinwardcurrent that is turned on by activating mGlu recep-torsqg.This causes inhibitory neurones to dischargeassoon as the IPSPhas decayedsufficiently. More workis neededto find out whether this mechanism appliesin other regions that can generate gamma rhythms,such as olfactory structures and neocortex. Some ofthe mechanisms outlined above could well coexist inparticular parts of the brain: inhibitory network andintrinsic mechanismscould combine in the neocortex;and excitatory networksmight be essential for the re-current inhibition loop to work39.Ourmuch improvedunderstandingof networkoscillatorshasstartedto maketheir mechanisms amenable to direct experimentaltesting.

Functionalconsequencesof rhythmicinhibition

Both theorysoand experiment38r51-53show that in-hibitory neurones are very effective at determiningwhen a pyramidalcell will fire. Ourproposalis that theinhibitory networkreceivesa steadyor slowexcitatorydrivewhich makesit oscillate,providinga clock whichdetermines when pyramidalcells can fire, if they re-ceive suprathreshold, excitatory afferent inputs. Thefact that the inhibitory networkcan, by itself, sustaina rhythm in the gamma frequency range without arequirement for fast EPSPS,separatesthe synchroniz-ing control or ‘clock’ from the specific neuronal pro-cessingof information ‘the Central ProcessingUnit’. Apossibleimplication is this: if firingof pyramidalneur-ones is constrainedto occur at particulartimes imposedby a 40Hz interneuronal network clock, then brainregions expressing40 Hz might not use rate-encodingof information, but rather might encode informationthrough selection of which pyramidalcells fire at all.Furtherwork also is requiredto find out whether thisprovides the means for associating or binding attrib-utes of specificobjects, for controlling summationandpotentiation, or for a phase code for neural signaling.

If the role of gamma rhythms is indeed to mediatebinding, then mechanisms must exist for the selectivecoupling of areasinvolved in processingcommon en-tities. One ideais that reciprocalexcitatoryconnectionscan do this, although there seemsto be a constraint inthat the conduction delaymust be less than one thirdof the period of the rhythm54.Others argue that the

coherence providedby this kind of mechanism is tooweakto providefor reliablebinding, in that it appearsin simulations only when multiple trials are used32.Scaling to largernetwprksor the presence of intrinsicoscillators, or both, might circumvent this problem.Alternatively(or additionally)oscillations in differentpartsof the cortexcouldresonatethroughthe thalamo-cortical loop3s(but see Ref. 1). We have a long way togo before we can identify how these long-rangelinksare made and broken, and understandingthe mecha-nisms of local networkoscillations is a key step in thisdirection.

We startedthis reviewdescribingthe association offast rhythms with higher cortical functions. Evidencefrom MEG recordings suggests that impairments ofthese rhythms couldbe involvedin brain diseasesuchas Alzheimerlsdisease23.The relationship of gammarhythms with selective attention has prompted ideasthat disruptionof its mechanism could play a role in

“ lg. Understandingthe cellular and net-schizophremaworkmechanismsthat generategamma rhythms pro-vides a starting point for thinking how they could bemanipulatedtherapeutically.

Selectedreferences1 Gray, C.M. (1994) J. CompuL Neurosci. 1, 11-382 Engel, A.K.et aL (1992) TrendsNeurosci.1S,218-2263 Singer, W. and Gray, C.M. (1995) Annu. Rev. Neurosci. 18,

555-5864 Adrian, E.D. (1950) Electroencephalogr. Clin. Neurophysiol.2,

377-3885 Eeckman,F.H.and Freeman,W.J. (1990)BrainRes.528,238-2446 Gray, C.M. et al. (1989) Nature 338, 334-3377 Engel, A.K.et al. (1991) Science 252, 1177-1179

8 Engel, A.K., Konig, P. and Singer, W. (1991) Proc.Natl Acad.Sci. USA88, 91369140

9 Freeman,W.J. and van Dijk,B.W. (1987)BrainRes.422,267-27610 Madler, C. et aL (1991) Br. 1.Anaestlr.66, 81-8711 Keller,L et aL (1990) Clin.Hectroencephakgr. 21, 88-9212 Bouyer, J.J. et aZ. (1987) Neuroscience22, 863-86913 Ptimtscheller, G., Flotzinger, D. and Neuper, C. (1994)

Electroencephalogr.Clin. Neurophysiol.90, 456-46014 Murthy, V.N. and Fetz, E.E. (1992) Proc. Natl Acad.Sci.USA89,

5670-567415 Sanes,J.N. and Donoghue,J.P. (1993) Proc. Natl Acad.Sci. USA

90, 4470-447416 Bragin, A. et al. (1995)]. Neurosci.15, 47-6017 Stumpf, C. (1965) Electroencephalogr. CZin. NeurophysioL 18,

477-48618 Soltesz,L and Deschi2nes,M. (1993) J NeurophysioL70, 97-11619 Tiitinen, H. et al. (1993) Nature 364, 59-6020 Pantev, C. et al. (1991) Proc.Natl Acad. Sci. USA88, 8996-900021 Kulli,J. and Koch, C. (1991) TrendsNeurosci. 14, 6-1022 Galambos, R., Makeig, S. and Talmachoff, P.J. (1981) Proc.

Natl Acad. Sci. USA78, 2643-264723 Ribary,U. et aL (1991) Proc.NatZAcad.Sci. USA88,11037-1104124 LlinAs,R. and Ribary, U. (1993) Proc. NatZAcad. Sci. USA90,

2078-208125 von der Malsburg, C. and Schneider, W. (1986) BioZ.Cybem.

54, 29-4026 Gray, C.M. et al. (1992) VisualNeurosci. 8, 337-34727 Engel, A.K., Konig, P. and Schillen, T.B. (1992) Cur’r.BioL2,

33%33428 Young, M.P., Tanaka, K. and Yamane, S. (1992)

f. Neurophysiol.67, 1464-147429 Tovee, M.J. and Rolls, E.T. (1992) NeuroReport3, 369-37230 O’Keefe,J. and Recce, M.L. (1993) Hippocampus 3, 317-33031 Hopfield,J.J. (1995) Nature 376, 33-3632 Wilson, M.A.and Bower,J.M. (1991) NeuralComput.3,498-50933 Wilson, M. and Bower,J.M. (1992) J. NeurophysioL67, 981-99534 Pinault, D. and Desch@nes,M. (1992) Neuroscience51,245-25835 Steriade, M., Curro Dossi, R. and Contreras, D. (1993)

Neuroscience56, 1-936 Llinfis,R.R., Grace, A.A.and Yarom, Y. (1991) Proc. Natl Acad.

Sci. USA88, 897–90137 Nuilez,A., Amzica, F. and Steriade,M. (1992) Neuroscience51,

7-1o38 Whittington, M.A., Traub, R.D. and Jefferys, J.G.R. (1995)

Nature 373, 612-615

TINS Vol. 19, ~0, S, 1996

AcknowledgementsThis workwassupportedby theWellcome TrustandIBM.We thankGyorgyBuzsaki,Bard Errrrentrout,Charles GrayandJohn Rinzelforhelpful discussionsduringthepreparationof thismanuscript.

207

REVIEW J.G.R. Jefferys et al.– Neuronal networks for induced ‘40 Hz’ rhythms

39 Ermentrout, G.B. (1995) in The Handbook of Brain Theory andNeuraZNetworks (Arbib,M.A., cd.), pp. 732–738, MIT Press

40 Traub, R.D. et aL 1. I%ysiol. (in press)41 McCormick, D.A., Gray, C.M. and Wang, Z. (1993) Soc.

Neurosci.Abstr. 19, 35942 Gutfreund, Y., Yarom, Y. and Segev, I. (1995) ~.Physiol. 483,

621-64043 Wang, X-J. (1993) NeuroReport5, 221-22444 Llinfis,R.R. (1988) Science 242, 1654-166445 Golomb, D., Wang, X-J. and Rinzel,J. (1994) J. NeurophysioL

72, 1109-112646 Wang, X-J. and Rinzel,J. (1993) Neuroscience53, 899-90447 Destexhe, A. et d. (1994) J. NeurophysioL72, 803-818

BOOK REVIEW

48 Von Krosigk,M., Bal, T. and McCormick, D.A. (1993) Science261, 361-364

49 McBain, C.J., DiChiara, T.J. and Kauer,J.A. (1994) J. Neurosci.14, 4433-4445

50 Lytton, W.W. and Sejnowski,T.J. (1991) J. Neuro@rysioL66,1059-1078

51 Buhl, E.H., Halasy, K. and Somogyi, P. (1994) Nature 368,823-828

52 Lacaille,J.C. et aL (1987) J. Neurosci. 7, 1979-199353 Knowles,W.D. and Schwartzkroin, P.A. (1981) J. Neurosci. 1,

318-32254 Konig, P., Engel, A.K. and Singer, W. (1995) Proc. Natl Acad.

Sci. USA92, 29&294

Methods in Enzymology.Vol. 255- SmallGTPasesand Their Regulators,Part A: RasFamily;

Vol. 256- SmallGTPasesand Their Regulators,Part B: Rho Family

editedby W.E. Balch,CJ Der andA. Hall,AcademicPress,1995.$99.00 (xxxi + 548 pages)ISBFJO /2 /82/56 O(Vol.255); $80.00 (xxix+ 40/ pages)ISBNO 12 182/57 9

(Vo/.256)

The Rasand Rhoproteinsare membersofa large superfamilyof smallGTPasesthatare activateduponGTP bindingandreturnto an ‘off’ state when GTP is cleavedtoGDP + Pi as a result of their intrinsicGTPase property. The Ras proteins en-coded by viral Rasgenesdifferfrom thoseencodedby cellulargenesby a few aminoacids.These point mutationsimpair theirGTPaseactivityandthereforeinterferewith’their normal shut-off mechanism,makingthem constitutivelyactive.

The different interconversionstatesofthe GTPases(whichhasmadethem popu-larly known as ‘molecularswitches’)regu-late manyintracellularsignalingpathways.Althoughboth the Rasand Rho familyofGTPasesbelongto a classof proteinsthatend with the sequenceCXXXX, they arefunctionallydistinct.Rasproteins regulatecell growth and differentiationby provid-inga linkbetweengrowth factor receptorsand gene expression’,whereas Rho pro-teins regulate the assemblyof focal ad-hesionsandcell movement.AlthoughRhoproteins belong to the GTPase super-family,they are mainlyinvolvedinthe regu-lationof actincytoskeletalorganization2’3.

Volumes 255 and 256 of Methods inEnzymologyare dedicatedto the biologyand biochemistryof Ras-and Rho-relatedproteins, respectively, and are dividedinto four sections.Both volumesare cat-egorized in a similar fashion. The firstsectiondescribesthe methodsusedfor thecloning and purification of recombinantRasor Rho proteinsfrom bacteria,yeastand the baculovirus–insect-cellsystem.Inthis section, emphasisis placed on thepost-translationalmodificationsof the Rasproteins,which makesthem hydrophobic

and allows them to become attachedtothe plasmamembrane.The importanceofthis modificationiswell addressed.

The second section deals mainlywiththe cyclicprocessof interconversionbe-tween the GTP (’on’) state and the GDP(’off’) state of Rasproteins.The technicaldetails required to monitor the GTP-bindingproperty both at an in vivo(in situ)level and at an in vitro level havebeen ad-equatelycovered.

The third section describes the various

approachestaken to identifythe protein-protein interactionsbetween componentsof the Ras-relatedsignaltransductionpath-way, for which a rangeof techniqueshasbeenwidelyused,from classicaltechniques(such as metabolic labelingand immuno-precipitation)to more recentmolecularap-proaches(suchasthe two-hybridsystem).

The final sectiondescribesthe variousfascinatingapproachesthat havebeenusedto monitor the biologicalactivity of Rasgenes, which include oocyte and mam-malian microinjection assays,fibroblastcomplementationassaysandthe screeningof phagepeptidelibrariesfor SH3 Iigands.

Both of these volumes have severalstrengths:the readableandconcisecollec-tion of chaptersare written by some ofthe leadersin the field of signaltransduc-tion, the logical organization of infor-mation makes it quick to access infor-mation related to specific Ras or Rhogenes,and manyof the figuresare gener-ally convincingand well done, particularlythe photomicrographsin the section‘therole of Rho proteins in cellularfunction’,whichare of excellentquality.

However, there are a few areaswherethese books are not as strong as they

mightbe.The lackof detailedintroductionin some of the chapters might make itdiftlcult to follow for students or thoseenteringthe area of signaltransductionforthe first time, and a few articleshave notbeen appropriately assignedto the rel-evant chapters. In some articles theauthors have failed to emphasize theimportance of post-translational differ-ences (suchas glycosylation,phosphoryl-ation andfarnesylation)for the functionofRas-related proteins when expressingthem in different systems (for example,E. coli,yeastand insectcells).This distinc-tion isimportantandcouldhavebeenindi-cated. The importance of selectingthecorrect system or cell line has beenignored in the chapters that describeprotein–protein interactions,which couldlead to bona fide protein-protein inter-actions being missed due to weak ex-pressionor lackof interactingproteins.Inaddition, misfolding of proteins (forexample, in E. coli, yeast, Sf9 cells ormammaliancell lines)or failure of certainpost-translationaleventsin the target pro-teins might lead to incorrect conclusions.The antisenseapproachthat hasbeenusedto inhibit Ras function is convincing,butcontrol experimentsare missingor havenot beendescribedhere4.

Overall, these books are a most usefuland valuable resource to everyoneinvolvedin the field of protein research.Theywill certainlyserveasguidancebooksand many of the techniques describedmight remain central to the field of signaltransductionin the future.

Tilat A.RJzviLlept of Cc//Biology,Neurobiologyand

Anatomy, Universityof Cincinnati MedicalCenter,231 8ethesdaAvenue,Cincinnati,

OH 45267-0521, USA.

References1 Barbacid, M. (1987) Annu.Rev.Biochern.

56, 779-8272 Burridge, K. et aL (1988) Annu.Rev. Cell

BioL 4, 487-5253 Vincent, S., Jeanteur, P. and Fort, P.

(1992) Mol.CeKBioZ.12, 3138-31484 Gura, T. (1995) Science 270, 575-577

208 TINSVoL 19, No. S, 1996 Copyright 01996, Elsevier Science Ltd. All rights reserved. 0166- 2236/96/$15.00 PII: S0166-2236(96)60012-5