onbehavioral and cellular levels.
Acentralproblem inthestudyofbehavior is the analysis of the mechanisms underlying the various forms oflearning. What arethecellular mechanismsof associativelearningand how do they relatetothoseof nonassociativelearning?Aredifferent forms of learning governed by separatemechanisms or by variations ona commonmechanism?Answers to these questions arebeginningtoemerge.Theyhave comeprimarilyfromthe useofsimple experimentalsystems inwhichthe animals are capable of various forms of learning and are accessible to analysisonthe cellularlevel(1-7).
Becauseof therelativesimplicityof its nervous system and the
detailed
knowledge available on the biophysical, bio-chemical, and morphologicalpropertiesof its neurons(8), the marinesnail Aplysiahasbeen usefulforanalyzing the mech-anismsof twoformsof nonassociativelearning:habituationand sensitization.Each oftheseforms hasbeenshown to bedueto achangeinstrengthat aspecificsetof synaptic connections(2, 9), andineachcasethechangeinsynapticstrengthresults from an alteration in the calcium currentcontrolling transmitter releaseatthepresynapticterminals(10, 11).Alevelofcomplexity beyond habituationand sensitization isassociativeconditioning. Amajorlimitation tothefurther studyof themechanismsoflearninginAplysia has been the failure to demonstrate associative learning in this animal (12-14).Wehere describeapowerfulformofclassical condi-tioninginAplysia.
Theparadigmwehave used was based upon classical de-fensive or aversive conditioning (15). In this paradigm one stimulus, theconditioned stimulus(CS),isrepeatedly paired withanaversive unconditioned stimulus (US). Thistraining commonlygives risetoa setofconditioned responses,which
can be studiedin twodifferent ways. First, overt motor re-sponses.totheCS sometimesdevelop and theycanbe examined directly (5, 15). A second approach-which we have used here-istoexaminetheability of the conditioned stimulus to modulate other (test) behaviors not involved in the original conditioning procedures (16, 17). The test behavior we have used isescapelocomotion.
Despite itstechnical advantages and theoretical interest, this second approach, commonly used in vertebrates, has not been directly exploredininvertebrates. This approach is useful for
tworeasons. (i) Examination ofconditioned modulatory effects
Thepublication costsof thisarticle weredefrayedinpartby page
chargepayment. Thisarticlemusttherefore be hereby marked "ad-vertisement"inaccordance with 18 U. S. C. §1734solelytoindicate this fact.
6675
also for the behavioral study of fundamental psychological issues, such as the distinction betweenlearning and performance (18). (ii) This protocol permits the use of a test system that survives restraint of the animal and surgical exposure of its central nervous system, a prerequisite for cellularanalysis.
Ourbehavioral index of aversive classicalconditioningwas themodulationby the CS of escape locomotion(Fig. 1).This response offers severaladvantagesfor thestudy oflearningin Aplyswa on both
behavioral
and cellular levels:(i)Locomotion iseasy toquantify because the individual steps are discrete and readily identifiable.(ii). Locomotiondisplays short-term and long-term sensitization (ref. 19 andunpublished data). (iii)The central program for locomotion is located within the circum-esophageal gangliaand can bemeasuredreliablyinthe pat-ternedactivityofbothperipheral
nervesand identifiedmotorneurons insimplifiedpreparations(20-22). RESULTS
ConditioningProcedures. Theexperimental procedureswe used are summarized in Fig. 2. All animals underwent the pretest on day 1 toassessbaseline locomotorresponsiveness before training. The test stimulus was a train of weakpulsed electric shockappliedacrossthe tail with spanningAg/AgCl electrodes. Thelatencyof the first step andthenumber ofsteps takenwithin a 5-minperiodafter theshockwererecorded by using criteriadescribedinthelegend of Fig. 1.
Animals were thenmatchedonthebasis of theirpretest scores andassigned to one of three training groups. The"untrained" groupreceivednofurther trektmentduringthe2-daytraining period. The "unpaired" group received training with the shrimp CSand the head shock-USexplicitly unpaired. The "paired"group was trained withspecific temporalpairingof the CS and US(seeFig.2). Three training trials perday (in-tertrialinterval,3hr)weregivenfor 2days.The training pro-cedureineach trialwassimilar tothat ofMpitsosand Collins (5).Animals inthe pairedgroupreceivedtheCSappliedover the anteriorhead region; 60 sec after onset of the CS the US was appliedviaspanningAg/AgClelectrodes acrossthefront of the head. Theunpairedgroup was trained with the same CS and
US asthepairedanimals,but received themspecifically un-paired;oneach trialthe CSwasdelivered90minaftertheUS. Eighteen hours after the last training trial all animals were tested, usingablindprocedure (byanobserver who didnot
know the animals' individual traininghistories). In thetest
sessioneach animalreceived theCSfor60 secandthen,with
Abbreviations: CS, conditioned stimulus; US, unconditioned stim-ulus.
Proc.Natl. Acad.Sci. USA 76(1979)
Pretest (Day 1) Test
JL-period---4
5min Training (Days 2 and 3)
3 hr
Paired _FL L
-U-Tail shock
3JL 15sec E Headshock (US)
30sec -FJI-Shrimp(CS) 90sec JL 1.5 hr 1.5 hr Unpaired AW - ,- -C D E F
FIG. 1. Onecompletestepinthe locomotorsequenceof Aplysia. (1/4 life size.) (A and B) Elevation and extension of the head; (C and D) arching of the midbody; (E and F) retraction of the tail. Stepswere
countedasthe number oftail retractions because thesearediscrete
andeasily observed.
theCSstillpresent,thesameteststimulus usedinthepretest
(weak tail shock)wasdelivered. Thelatencyand number of
steps taken within 5 min were monitored for each animal
duringthetestsession.These trainingproceduresresultedin
aform ofconditioning characterized by (i) temporal specificity,
(ii)a requirementof CS presencefor thelearningtobe
ex-pressed,and(iii) rapidacquisition.Each of these characteristics will bedescribedinturn.
Temporal Specificity. A critical feature of associative
learningistemporalspecificity. Inboth classical and
instru-mentallearning the subject Displaysachangeinbehavior due
tospecific temporal relationships betweenevents.Acommon
interpretation(23)isthatthesubject changesitsbehaviorasit
learnsthatoneevent(theCSinclassicalconditioningand the
operantresponseininstrv-mentalconditioning)comestopredict theoccurrenceofanother(theUSorreinforcement).Thefirst
questionweexaminedwaswhethertrainingwithapatternof
temporally pairedCSandUSpresentationsproducesadifferent
outcomethan training witha patternofexplicitly unpaired
Test (Day4)
Test
period-l
FIG. 2. Behavioral protocol. Animals (150-350 g)werehoused
individually inperforated circular pans (28-cm diameter, 7.6-cm
depth) thatweresuspendedina750-liter tank of aeratedartificial seawater(15'C).The teststimuluswas a15-sectrainof50-mAac pulses (1.5seceach), 0.33 Hz,tothe tail. The USwas ahead shock,
30secof 400-mAacpulses (1.5 sec), 0.33 Hz. Largecurrentswere necessaryfor effective stimulation because of theseawatershunt between the electrodes and skin. The CSwas1.5mlof crudeshrimp
extract.ImmediatelypriortoCSonsetaplastic bagwasplaced around eachpantoseparateits contentsfrom thewaterinthe tank.CSoffset wasaccomplished by siphoning all thewateroutof thepanand then removing the bag. These procedureswereidentical for paired and unpairedgroups.
presentationsornotrainingatall (Fig. 3). Beforetrainingthere
were no marked differences amongthe threegroups in the
amountof locomotion elicitedby tail shock (theteststimulus). Aftertraining,a one-wayanalysis ofvarianceindicatedthere was anoverallsignificant difference
(F2,
45= 27.4; P<0.01)amongthegroupsintheirescapelocomotorresponsestothetest
stimulusinthepresenceof theshrimpCS.Pairedcomparisons
showed that thenumber ofstepstakenbytheuntrainedgroup
wasnotsignificantly different from thatof theunpairedgroup, but the animalsinthepairedgroupwalkedmorethanthosein
either controlgroup(P<0.005ineachcase;Fig.3A).Adirect
12 A B 0 4.~~~~~~~~~~~~0 6 ~ ~ ~ ~ ~ -E2 C 0
FIG.3. (A )Responsesinthetestsession.Eachscoreisthemean
(± SEM)number ofsteps taken within5minoftheteststimulus. Paired animals(n = 18)walkedsignificantlymorethanuntrained animals(n=12)orunpairedanimals(n=18),P<0.005ineachcase.
Allpairedcomparisonswerebyttestsforindependentgroups;all
probabilityvaluesaretwo-tailed.(B)Differencesbetweentestand pretestscores(fromsameexperimentasA).Pairedanimals walked
significantlymoreinthetestthan inthepretest(P <0.005,tfor correlatedmeans).Thedecrease in locomotionof theuntrainedgroup
isnotstatisticallysignificant,whereasthedecrease oftheunpaired
groupis(P<10.025). Such decreasesareoften,butnotinvariably,
observed in controlgroups.
A B JAL :0- - 7. Ar- lw 'i;_-r .*. m
FIG. 4. (A) Stimulus site control. Differences (mean+SEM) betweentestandpretestscores.USdelivery during trainingwaswith seawater-filledcapillary electrodes (0.5-cm diameter, 0.2-cm
sepa-ration)toobtainrepeatable,restricted stimulation of the anterior head. Paired animals (n=8) showed significantlygreaterlocomotion
(P<0.005) thanunpairedanimals(n=8). (B) Differences between
testandpretestscoresof animals trained with CS and USunpaired (n=8)orwithUSalone(n=8).Therewas nosignificantdifference between thesegroups.(C)Differences betweentestandpretestscores of untrained animals(n=8) and animals trained with CS alone (n
= 8). Therewasalsonosignificant difference between thesegroups.
comparisonof thedifferenceinlocomotionexhibitedby the differentgroups
before,
and aftertraining isshowninFig. 3B.The data presented inthis wayclearly show that the paired
groupisdifferentnotonlyintheamountofescapelocomotion
exhibitedinthetestsessionbut alsointhe direction of theeffect:
Bothcontrolgroupsshowedadecreaseinlocomotioninthetest sessionwhereas thepairedgroupshowedanincrease.No
sig-nificant differenceswerefoundinthemeanlatenciestothe first
stepamongthesegroupsinthetestorpretest.
Although thetesting inthisexperimentwasconducted blind,
thetrainingwasnot.Thusthe possibility existed thattrainerbias could have contributedtothe differences observed.Tocontrol for this and related artifacts in the trainingprocedure, we
carriedoutseveralcontrol experiments. We firstreplicatedthe experimentbut used ablindtraining procedure aswellasa
blindtestingprocedure. Trainingwasdonebyaperson-who
knew neither thedesignnorthepurposeof thestudy.Asinthe
previousexperiment,animalsinthepairedgroup(n =8)
ex-hibited significantlymoreescapelocomotionduring thetest session(P<0.005) thandid theunpairedcontrolgroup(n =
8).
Evenwhen thetrainingwascarriedoutblind,thedifference intraining procedures for paired and unpairedgroups may
have producedanunintentionalbiasonthepartof thetrainer
(for example,hemayhavedeliveredastrongerUStoonegroup
oranother bydirect contact with the spanningelectrodes).
Thus, afunctionallystronger orweaker US delivered toone
group might have contributed to thedifferential effectwe
observed. To test thispossibility weexamined the two
con-ceivable US intensity differencesbygivingunpairedtraining
with(i)afunctionallystrongerUSand(ii)afunctionally weaker
USthan thatpreviouslyused. The weak USwas ashock with
half thecurrent(200 mA) normallyused. ThestrongUS
con-sisted of thesamecurrentasnormal(400 mA),butthe
elec-trodes, instead of being heldneartheskin,werebroughtinto
directcontactwith theskin. Separate pilot experimentshad indicated that suchcontactcausesconsiderablymore
sensiti-zationofavariety ofresponsesthan does shock from
noncon-contactto the middle of the anterior head(centered between theoral tentacles and therhinophores)evenwhen the animal's head became withdrawn. With this different method of VS deliverythepairedgroup still showedsignificantly more escape locomotion in the presence of the CS than theunpaired group (P < 0.005; Fig. 4A). These experiments demonstrate that trainingwith the CSand USspecifically pairedendows the CS withproperties not observed with unpaired presentation of the samestimuli.
We alsofound that training with either the US alone(Fig. 4B)orwith the CS alone(Fig. 4C)failed toproducethe
con-Unpaired Paired lO1-a, 0 el 2 (A -4 -6 "-6 MI _8 Test Test withoutCS with CS -10
FIG. 5. Differences betweentestandpretestscoresintheabsence
and(3hrlater)inthepresenceof theCS. Pairedanimals(n= 8)
walkedsignificantly morethan unpaired animals (n = 8) in the
presenceof the CS(P<0.025).Thepairedanimalsalso walked sig-nificantlymorethantheyhad in the absence of theCS(P<0.01,t
forcorrelatedmeans).
a 6 4 B t I A
14- 12- 10- 8- 6-Q 4-0 2-E 0 8B B 6- 4- 2-A I I I I I IU-- I 0 2 4 6 9 Trials
FIG. 6. Acquisition. Differentgroups weregiven fromzerotonine
trainingtrials and thentested 18 hr after the last trial. Eachpoint is
themean(+SEM) number ofstepstaken in thetestsession. Three conditionswereexamined:paired training (A, 0), unpairedtraining
(B, 0), andnotraining(e& in A andB,eachrepresenting thesame
data).Numbers of animalspergroupwere:untrained,n=27;paired
andunpaired(respectively),onetrial,n= 16,n=16;twotrialsn =
24,n=24;three trials,n=24,n=24;four trials,n=16,n=16;five
trials,n=16,n=15;sixtrials,n=26,n=26;ninetrials,n =16,n
=16.Thehorizontal broken lines indicate baseline (untrained)
per-formance.Significant differences between paired and unpairedscores werefoundontrials5,6,and9(indicated byarrows,P<0.025,P<
0.005, and P<0.005,respectively).
ditioned facilitatory effect of theCSonescape.Thesefindings
furthersupporttheconclusion that theability of theCSto
en-hanceescapelocomotionisdependentuponthespecific tem-poral relationship betweenCSandUSduring conditioning.
RequirementofCSPresence. Associativelearningisusually
characterizedbymarked stimulus specificity (24). Thepresence
of theCSisnecessaryforexpressionof the conditionedresponse:
it isnotelicited by other stimuli,exceptbythose thatarequite
similar.Wehavebeguntoexaminethese features bytestingfor
the requirementof the CS during thetestsession. If paired traininghadmerelyproduceda generalincrease in
respon-siveness,onecould predict thattheteststimulus alone might
elicit theconditioned effectinthe absenee of theCS. To test this possibility we first trained animals with the
standardpaired andunpaired protocols (Fig. 2). Thesegroups
werethentestedtwice,firstintheabsenceoftheCS,and then,
3hrlater,inthepresenceof theCS.Aftertrainingneithergroup
exhibitedmuch locomotioninresponsetotail shockinthe
ab-sence oftheCS (Fig. 5). However, in the presence of the CS, the paired group exhibited significantly more escape locomotion in response to the test stimulus than did the unpaired group (P < 0.025), and, in addition, significantly more locomotion than it had shown in response to the same test stimulus in the absence of the CS (P < 0.01, t for correlated means). In the absence of the CS both groups showed a decrease in locomotion relative to the pretest scores. In the presenceof the CS only the paired group showed an increase in locomotion relative to the pretest.
Theseresults indicate that the CS is required for the conditioned effect to be exhibited and suggest that the conditioning does not take the form of a nonspecific increase in responsiveness.
Acquisition. We next examined one aspect of the acquisition
ofthe learned response by training different groups of animals with different numbers of trials. Each group was tested the morning after its last training trial. Thus animals receiving one to three trials trials were tested on day 3, while animals
re-ceivingfour to six trials, as well as animals given zero trials (no
training),were tested on day 4 (Fig. 2). Animals receiving nine
trials weretested on day 5. In each case paired and unpaired groups were run and each point on the composite acquisition curve (Fig. 6) consisted of at least two experiments for each group. The composite curve for the paired groups (Fig. 6A) was sigmoid. Intrials one through four there were no significant
differences betweenthe groups. However, after five training
trials thepaired groupexhibitedsignificantly more locomotion in response to the test stimulus than did the unpaired group (P
<0.025), in part due to an unexplained decrease in the control
group.By trials six and nine there were consistently large and
significant differencesbetween paired and unpaired groups
(P < 0.005 in each case).
Two aspects ofthe acquisition curve (Fig. 6) deserve mention.
First, althoughthe data were variable, unpaired groups tended
to walk more than-paired groupsafter receiving one to three
trials. We are notsure how reliable this small trend is. Second,
acquisitionappears tobequiterapid in thepairedgroupsafter the fourthtrial. Rapid acquisition is a common characteristic
ofaversive classical conditioning invertebrates (seebelow). DISCUSSION
These experiments provide direct evidence thatAplysia can be classically conditioned, forming a powerful temporally
specific association between a chemical CS and a noxious
electrical US. The association is rapidly acquired and is
de-pendent for its expression upon the presence of the CS. The
features of temporal specificity, the requirement of the CS
during testing, and rapid acquisition are characteristic of
aversive classical conditioningin vertebrates (18). Moreover,
inAplysia as in vertebrates, aversiveconditioning has the
in-teresting featurethat conditioning often occurswithoutovert
changes in the externalbehavioral response to the CS. Rather, conditioningleads to a change in internalstate that is manifest
in theacquired ability of the CS to modulate a varietyof
be-haviors (25, 26). For example, aversiveclassical conditioning
can endow a CS with the capability of suppressing appetitively reinforced bar pressing (16), enhancinginstrumental avoidance
responding (27), facilitating an unconditionedstartle response
(17), and accelerating heart rate (28). These conditioned
properties of the CS are specific to the temporal relationships of the CS and US during training as wellas to the particular CS
used. Theacquisition of conditioned aversiveresponses as
re-vealed by such tests is typically quite rapid in vertebrates, often peaking within 5-10 training trials (17, 29). The similarity of the effects described forAplysiatothose of aversive classical conditioning in vertebrates encourages the search for further similarities and suggests that mechanisms found in the study
A -
powerfulchangeininternal statethatbecomes manifest in the modulation of a response not directly elicitedby the CS. That aspecificmotorresponse is notconditionedis consistentwith theoretical interpretationsof aversive classicalconditioningas a conditioned motivational state rather than a conditioned motor response(18, 25, 26).That Aplysia displaywell-defined appetitiveand defensivemotivational states (14, 30) further supportsthis possibility.
The hypothesisthat Aplysia canlearn conditioned motiva-tional states has several interesting implications. One prediction isthata rangeofbehaviorswill be affected by the CS after conditioning and that these effects will be motivationally consistent;defensiveresponsesshouldbeenhancedand appe-titiveresponsessuppressed by the CS (25). A second implication isrelevanttohypotheses about the relationshipbetween sen-sitizationand classicalconditioning(8,31,32).Because sensi-tization isthoughtto be a component of motivational states(30), the conditioningof amotivationalstatewould suggest that this formof associativelearningislikelytoinvolve,inpart,aspects ofsensitization.Thisnotion is attractivebecauseitsuggeststhat differentforms oflearningmayutilize variouscombinations ofarestrictedsetoffundamentalplastic mechanisms.
Learning inAplysia persistsafter the animal has been re-strainedand the head ganglia have beensurgicallyexposed. Moreover, we have recently observed neural correlates of aversiveconditioning inidentifiedpedalmotorneurons (un-published data).Three other forms of associative learning have also been demonstrated in relatedgastropodmollusks: avoid-anceconditioninginPleurobranchaea (4, 5),bait-shyness in Limax (6),and intersensory associations inHermissenda (7). In each of these animals neuronal correlates of associative learninghavealsobeenreported(33-35). Thus, in addition to itsbeing possibleto examinethe relationship between nonas-sociativeand associativelearning withinthesame species, it maybe possibletoexamine onthe cellular level the relationships amongdifferentforms of associative learning across related species.
Wethank F. Adeshinaand J. Sliney for technical assistance; R. Hawkins, I. Kupfermann, V. Castellucci, and J. Koester for helpful comments;K.HiltonandL.Katzfor artwork; and P. Kandel for pho-tographicassistance.This work was supported by NationalInstitutes of Health Fellowships to E.T.W. (5T32MH1574), T.J.C.
12. Kupfermann,I.&Pinsker,H.(1968)Commun.Behav. Biol.2,
PartA, 13-17.
13. Pinsker, H., Hening,W.,Carew,T. J. &Kandel,E. R. (1973)
Science 182,1039-1042.
14. Kupfermann,I.(1974)Behav. Biol. 10,1-26.
15. Pavlov,I. P.(1927)ConditionedReflexes,translatedbyAnrep, G. V.(OxfordUniv.Press,London).
16. Estes, W.K. &Skinner,B.F.(1941) J.Exp.Pyschol. 29, 390-400.
17. Brown,J.,Kalish,H. & Farber, I. (1951)J. Exp.Pyschol. 41, 317-328.
18. Mackintosh,N.J.(1974) ThePsychologyofAnimalLearning
(Academic,London).
19. Walters,E.T.,Carew,T.J.&Kandel,E.R.(1978)Soc. Neurosci. 4, 209(abstr.).
20. Jahan-Parwar,B. &Fredman,S.(1978)Comp.Biochem.Physiol. 60A,459-465.
21. Hening, W.,Carew,J. J.&Kandel,E. R.(1977)Soc.Neurosci. 3,1219(abstr.).
22. Hening,W.A.,Walters,E.T.,Carew,T.J.&Kandel,E. R.(1979)
BrainRes. 179,231-253.
23. Rescorla,R.A.(1967)Psychol.Rev. 74,71-80.
24. Miller,N. E.(1967)inTheNeurosciences: AStudyProgram,eds. Quarton,G., Melnechuk, T. & Schmitt,F. (RockefellerUniv. Press,NewYork),pp.643-652.
25. Rescorla,R. A. &Solomon,R. L.(1967) Psychol.Rev.74, 151-182.
26. Hammond,E.J.(1970)inPhysiologicalCorrelatesofEmotion, ed.Black,P.(Academic,NewYork),pp.245-259.
27. Rescorla,R.A. &Lolordo,V. M.(1965)J. Comp.Physiol.
Psy-chol. 59,406-412.
28. Black,A. H.(1965)inClassicalConditioning:ASymposium, ed.Prokasy,W.F.(Appleton,NewYork),pp.20-47.
29. Annau,Z.&Kamin,L.J.(1961)J. Comp.Physiol. Psychol. 54, 428-432.
30. Hawkins,R. D.&Advokat,C.(1978)EighthSymposiumofthe Societyfor Neuroscience, ed. Ferrendeli, J. A..(Society for Neuroscience,Bethesda, MD),pp.16-32.
31. Wells, M. (1968)Lower Animals (Weidenfeld & Nichollson, London).
32. Razran,G.(1971) MindinEvolution:AnEast-WestSynthesis
of Learned Behavior and Cognition (Houghton Mifflin, Boston).
33. Davis,W.J. &Gillette,R.(1978)Science199, 801-804. 34. Chang, J. J. & Gelperin, A. (1978) Soc. Neurosci. 4, 189
(abstr.).