Review article

Experimental access to associativelearning in honeybees

J Mauelshagen U Greggers

Institut für Neurobiologie, Freie Universität Berlin, Königin-Luise-Str 28/30, 1000 Berlin 33, Germany

(Received 29 October 1992; accepted 14 April 1993)

Summary — The pioneering work on the behavior and physiology of bees provides a fundamentalframework on which new experiments can be designed in order to investigate the nature of associa-tive learning in bees. Such studies require investigations not only on the behavior of free-flying bees,but also on the behavior of restrained bees as a basis for studies on the cellular level. In order tocombine the results about free-flying bees with the results obtained in the laboratory, it is first neces-

sary to test the validity of the restrained preparation. Therefore, one has to deal with the clarificationof associations formed during both classical conditioning and instrumental learning. If it is possible toidentify comparable associations, then the results could lead to mutually supportive interpretationswith respect to the mechanisms and biological meaning of learning and memory in the honeybee.

Apis mellifera / instrumental learning / classical conditioning / associative learning

INTRODUCTION

Honey bees (Apis mellifera L) are a well-known model for behavioral studies of

learning and memory (Opfinger, 1931; Lin-dauer, 1963; von Frisch, 1967; Gould,1984; Menzel, 1990). Individual bees haveto remember several parameters regard-ing effective food sites, such as their loca-tion, color, odor and shape. As social in-

sects bees are well suited for an

investigation of the mechanisms underly-ing memory formation. Despite their highlearning capabilities, insect brains also

provide access to studies at a cellular lev-

el (Menzel et al, 1991). Therefore, to in-

vestigate learning and information pro-cessing in the bee brain, different methodsare applied, such as studies on the behav-ior of free-flying bees (Lindauer, 1963; vonFrisch, 1967; Menzel et al, 1974; Gould,1984; Bitterman, 1988; Menzel, 1990;Greggers and Menzel, 1993) and on thelearning behavior of restrained bees (Er-ber and Schildberger, 1980; Bitterman etal, 1983; Menzel and Bitterman, 1983;Smith, 1991; Smith et al, 1991; Menzel etal, 1993). Furthermore, electrophysiologi-cal studies are carried out, which allow the

application of simple learning paradigmsduring extracellular recordings (Rehder,

1987; Smith and Menzel, 1989a, 1989b;De Jong and Pham-Delegue, 1991; Braunand Bicker, 1992) or intracellular record-ings (Erber, 1980, 1981, 1983; Hammer,1991, 1993 submitted; Mauelshagen,1993). Essential support is provided bymorphological (Mobbs, 1982; 1984, 1985;Rybak, 1987; Rehder, 1988, 1989; Rybakand Menzel submitted), pharmacological(Mercer and Menzel, 1982; Mercer et al,1983; Michelsen, 1988; Braun and Bicker,1992; Wittstock, 1993) and immunocyto-chemical studies (for review see Bicker,1993). The most recent progress has

come from introducing new techniques, iepatch clamp recordings on an identified

neuronal population in the bee brain

(Schäfer et al, 1993), optical recordings(Lieke, 1991, 1993) and biochemistry (Alt-felder and Müller, 1991; Altfelder et al,1991; Müller and Altfelder, 1991). Thusthere is a kind of fragile framework whichis composed of some pioneering andsome very specific studies; however, thepieces often do not fit and some funda-mental questions have only been partlyaddressed.

In this review, 2 approaches are em-phasized: behavioral studies on the forag-ing behavior of free-flying bees and theelectrophysiology of single neurons in thebee brain during application of simplelearning paradigms which is based on arestrained preparation. To focus attentionon some basic problems, the relevanceand validity of both approaches is dis-

cussed.

THE PROBLEM

Studies of learning and memory in honeybees need to analyze the kind of associa-tions which are formed during foraging-related behaviors and the mechanismswhich underlie such associations. An ex-

perimental situation has to be designed toextract the relevant parameters which

might be strong indicators of a learned be-havior. Basically, there are 2 approachesin studying the behavioral components ofhoneybee learning. The observation of

free-flying bees leaves the bee in the mostnatural conditions, with the disadvantagethat the experimental situation is less con-trolled; in the other approach, the animal isrestrained so that only single componentsof the natural behavior can be studied, butwith the advantage of a well-controlled ex-perimental situation. On the basic knowl-

edge of learning theories, however, oneruns into the problem of eventually study-ing 2 different aspects of learning (Menzeland Bitterman, 1983), ie with free-flyingbees learning based on an operant behav-ior (instrumental learning) is studied; withrestrained bees learning based on Pavlov-ian (classical) conditioning is investigated(Bitterman et al, 1983).

During instrumental learning one has toconsider 3 important parameters: a behav-ioral response, R; a positive or negativeoutcome, O, of this response (reinforcer);and a stimulus, S, which are present dur-ing this contingent event (Colwill and Res-corla, 1986). Thus associations would bepossible between the stimulus and the re-sponse (stimulus-response associations),between the response and the reinforcer

(response-reinforcer associations) or both(2-process theory). Although there are

good reasons to believe that, at least forvertebrates, the most effective associa-tions are formed between the responseand the reinforcer, there are still some un-certainties about it, in particular as regardsthe interpretation of the involvement of thestimuli.

During classical conditioning (Pavlov,1927) a conditioned stimulus (CS) be-comes predictive for an unconditionedstimulus (US). After conditioning (pairing ofCS and US) the conditioned stimulus elic-

its the conditioned response (CR) whichresembles the response to the uncondi-tioned stimulus (UR) before conditioning.Thus the conditioned response is the con-

sequence of a stimulus-stimulus associa-tion (ie stimulus-reinforcer association).Usually a conditioned response developswithin multiple conditioning trials, but thereare also a few examples which report ahigh learning rate after only a single condi-tioning trial (Menzel et al, 1974; Sahley etal, 1981; Crow and Forrester, 1990).

Therefore, the most striking differencebetween the 2 experimental approacheswith honey bees is that learning in free-

flying bees is a consequence of their deci-sions during foraging, while with restrainedbees learning is a stimulus-induced behav-ior, whereby the temporal relation and se-lection of the stimuli are determined by theexperimenter.

THE "FREE-FLYING" APPROACH

Considering an individual bee which hasbeen trained to collect sucrose solutionfrom a feeding place, what are the associa-tions the bee needs in order to reliablyidentify this feeding place, to distinguish it

from other food sources and to evaluate its

efficiency? It is well known that free-flyingbees can learn the location, color, odorand shape of a food source and the time ofday at which an efficient reward can be ob-tained (Opfinger, 1931; Lindauer, 1963;Gould, 1984; Bitterman, 1988; Menzel,1990). But this does not automatically im-ply that the presented stimuli are directlyassociated with the reward itself or with the

behavior which can be composed of sever-al parameters such as flying to or awayfrom the food source, sitting down, han-dling, extending the proboscis etc.

Opfinger (1931) first addressed the

question of when, during visiting a feeding

place (during arrival, sucking or departure),bees learn the cues that characterize the

stimuli and/or landmarks of the feedingplace. She found that colors and land-marks were learned best during the arrivalflight and hardly at all on departure, whileGould (1986) claimed that colors werelearnt only on arrival and close land-marksonly during the departure flight. The learn-ing of colors at a feeding place was furtherstudied by Menzel (1968), who more quan-titatively determined the time dependen-cies of colors presented during arrival flightor during sucking. He suggested that beescan learn colors both applied during arrivaland during sucking. Those studies indicat-ed that learning about a food source couldbe phase-specific. This was supported bythe experiments of Grossmann (1970,1971), who concluded that the color duringarrival significantly dominates the color

during sucking. He found that learningabout the sucking color had an influenceonly on the number of approach flights, butnot on the decision to land on the suckingcolor during the test phase. Furthermore,he reported that the arrival color waslearned faster, if both arrival color and

sucking color were the same, suggestingat least some learning about the suckingcolor itself. If an odor was added, learningwas significantly increased, even if arrivaland sucking were delayed by more than 8s. Most importantly, Grossmann (1971)showed that despite perfect contiguity col-ors could not be learned passively, ie byjust seeing the color a sufficient amount oftime before sucking rather than during thearrival flight. Learning about colors and/orlandmarks present only during departurewas reported by Hannes (1930) and wasstudied systematically by Lehrer (1991),who interpreted her results as excitatorybackward conditioning. Couvillon et al

(1991) studied learning of stimuli on arrivaland departure by emphasizing the ques-tion of when, during a visit at a feeding

place, bees learn about its location. Theyconcluded that bees learn both during arri-val and departure, and that there was nophase-specific preference for learning col-ors or landmarks. However, as they cor-rectly concluded, the learning on departurecannot be readily interpreted as backwardconditioning, but that other parameterssuch as associations with interceptivestimuli have to be considered. Thus the

question of the nature of the associationsremains unsolved by the studies describedabove.

Other investigations tried to unravel thisproblem by closely relating the learning ex-periments with bees to those performed onvertebrates (Grossmann, 1973; Sigurdson1981 a, 1981b; Bitterman, 1988). Manytheoretical paradigms developed for thestudies on vertebrates have also been

successfully applied to the bee. The appli-cation of fixed ratio and fixed interval

schedules as developed by Skinner

(1938), for example, were shown to havethe same effects as training vertebrates

(Grossmann, 1973, Sigurdson, 1981 a,1981b; Bitterman, 1988). Thus, phenome-na such as resistance to extinction, theoverlearning-extinction effect, successive

negative incentive contrast, conditioned in-hibition and within-compound associationswere found to parallel the results in the

vertebrate literature (Bitterman, 1988).Further experiments will be challenged

by controlling an experimental situation sothat extracting the relevant parameters be-comes possible. One recent study whichallows controlled access to the informa-

tional components may offer at least thenecessary conditions to analyze the natureof associations during foraging by a singlebee (Greggers, 1989; Greggers and Men-zel, 1993). In this study single bees weretrained to collect sucrose solution in a

patch of 4 electronic flower dummies

(feeders). As an input function for analyz-ing the system the learning ability of the

bee was challenged by using a dynamicreward situation. In contrast to traditional

experiments using a static reward systemby offering unitary rewards in dual choice

experiments (Staddon, 1983), differentlyrewarded feeders with a constant flow of

sucrose solution were applied in a multiplechoice situation in this study. As a conse-quence of the bee’s visiting all of the 4feeders the amount of sucrose solution ob-

tained depends on the sequence and fre-quency of visits at each feeder. Therefore,the bee experiences graded rewards dur-ing the moment-to-moment decisions

(fig 1). In different experimental arrange-ments concerning reward rates and colormarks of the feeders, the behavior of thetest bee was monitored by a computer inreal time with several photodetectors in-

stalled in each feeder. This is the first suc-

cessful technique for analyzing a numberof very relevant parameters such as thesequence and frequency of visits to eachparticular feeder, the handling time and licktime during each visit and the flight timebetween successive visits. Thus one hasthe powerful tool to manipulate the experi-mental parameters in order to determinewhich stimuli and which of the choice- and

time-related parameters may be relevantfor the learning process.

To test how the bee’s choice behavior

corresponds to the average amount of re-ward, the data from patches with differentreward rates were compared to data fromexperiments with equal reward rates. Nodifference was found between the choice

frequencies of 4 equally rewarding feed-ers. These results demonstrate that feed-ers with the same sucrose solution flowrates are chosen equally frequently. This istrue irrespective of whether the feeders aremarked with the same or different colors or

odors, and thus are distinguishable eitherboth by color or odor and location, or by lo-cation alone. Therefore, if there are associ-ations between the stimuli and the ob-

tained reward, then those stimuli are

equally effective for the formation of an as-sociative memory. For groups with differ-ent reward rates, the test bee optimizes bypartially matching its choice behavior to

the reward rates of the feeders ie the beechooses a higher reward feeder more fre-quently than a lower reward feeder (fig 2).This clearly demonstrates that the bee ad-justs its behavior to the specific reward sit-uation in the artificial patch.

Irrespective of the reward situation, ageneral phenomenon is the occurrence of2 fundamentally different behaviors: anytime the experimental bee leaves a feeder(actual feeder) it may either return to thatfeeder (stay flight), or choose one of the 3

alternative feeders (shift flight). The timedependencies of the 2 choice performanc-es differ considerably in that the seconds

immediately after departing from the actualfeeder are dominated by fast stay flights.During these fast stay flights, the bee ob-tains only minute amounts of sucrose solu-tion which makes the occurrence of stayflights surprising, because one would ex-pect the bee to learn that this kind ofchoice is not successful. The most straight-forward interpretation of this and othercharacteristics of the stay behavior wouldbe that stay flights are not the conse-

quence of an associative memory but rath-er reflect nonassociative components ofthe choice process. A further indication for

the existence of nonassociative memorycomponents is taken from the analysis ofthe reward sequence during the moment-to-moment decisions which shows that the

actually experienced reward at a singlefeeder influences the information process-

ing at the following feeder. For example,the lick time was shown to depend not

only on the amount of reward offered bythe actual feeder, but in addition, on thereward experienced during the precedingvisits (table I). The associative compo-nents of the choice process can be ana-

lyzed by focussing on feeder-specificmemories, ie components which can beisolated as behavior that is unique foreach individual feeder. In this context the

analysis of revisits to the same individualfeeder after visiting one or more alterna-tive feeders in between is especially suita-ble. Several parameters at the actual feed-er (eg the lick time) can be shown to

depend on the experience during the lastvisit to this same feeder, independent ofthe experience at the feeders visited in be-tween. This suggests that a foraging hon-

eybee learns the properties of a food

source (its signals and rewards) so effec-tively that specific expectations guide thechoice behavior. The results indicate that

feeder-specific and feeder-unspecific mem-ories are developed during the continuous

learning process in the patch, which maybe interpreted as associative and nonasso-ciative components in the choice process,respectively.On the basis of these data, further ex-

periments have been designed to clarifythe nature of the associations (see below).

THE "RESTRAINED" APPROACH

To control learning-relevant parameters,numerous studies have been performed onrestrained bees (Kuwabara, 1957; Erberand Schildberger, 1980; Bitterman et al,1983; Menzel and Bitterman, 1983; Menzelet al, 1991, 1993). Such studies have uti-lized the most biologically meaningful stim-ulus combinations, intending to comparethe results obtained in this controlled situa-tion with those obtained on free-flying bees(Menzel and Bitterman, 1983; Bitterman

1988). The best known paradigm is theclassical odor conditioning of the proboscisextension reflex (PER, Kuwabara, 1957;Vareschi, 1971; Menzel et al, 1974). Anodor serving as a conditioned stimulus

(CS) is readily associated with a sucrosereward (unconditioned stimulus, US) deliv-ered to the antennae and proboscis. Aftera single learning trial (CS/US pairing) thereis a high probability of proboscis extensionin response to the conditioned odor (condi-tioned response, CR). Experiments strong-ly suggest that the association formed byPER conditioning is classical (Bitterman etal, 1983). It has been demonstrated that

concerning the rate of acquisition, generali-zation and differentiation and CS-US con-

tiguity the results of proboscis extensionconditioning have common characteristicswith those obtained with free-flying bees(Menzel and Bitterman, 1983; Bitterman etal, 1983). Also the time dependencies ofthe conditioned response after 1-trial odor

conditioning resembles the time dependen-

cies of choice behavior of free-flying beesafter a single training for color (Menzel,1987). Despite all similarities, it is not clearwhether the situations during which associ-ations are formed are comparable andwhether comparable neuronal changes oc-cur. For example the classical conditioningof colors fails in restrained bees. The con-

ditioning to mechanical stimulation or

placement (Menzel, 1990), however, indi-cate that odors are not the only stimuliwhich can serve as CS in the restrained

preparation.The behavioral experiments on free-

flying bees described above show that onecan make quite specific predictions aboutthe kind of memories applied by the beeduring foraging. However, data on the ac-quisition, storage and retrieval of learnedinformation can only be obtained by reduc-ing the level of investigation. Therefore, itis especially important to study the exacttime-dependencies of nonassociative andassociative memories and the relative con-tribution of nonassociative components toassociative memory consolidation (Menzelet al, 1993). Further, the brain structureswhich are involved in memory formationhave to be determined, the cells which areinvolved and finally the cellular mecha-nisms by which such complex behaviorcan be realized.

The successfully applied PER-

conditioning to odors serves as a basis forstudies on reduced preparations. Usingthis paradigm, it is concluded that learning-evoked changes in the response behaviorof single cells must be due to the modula-tion of the odor processing pathway. Thisis possible at sites of convergence of theCS and US processing pathways (fig 3).Odors are processed via the antenno-

glomerularis tracts (agt) which relay infor-mation from the antennal lobes to themushroom bodies (Mobbs, 1985). So far,only 1 component of the sucrose process-ing pathway is known, the VUMmx1-

neuron (Hammer, 1991; submitted). It wasshown that evoked spike activity in thisneuron is sufficient to substitute for the su-crose reward during olfactory conditioning(Hammer, 1991, submitted). The projec-tion areas of the VUMmx1 neuron overlapwith neurons of the odor processing path-way (eg agts) in the antennal lobes, thelateral protocerebrum and the calyces of

the mushroom bodies. Thus the neuralsubstrate for learning and memory in thebee brain appears to be distributed amongseveral neuropilar regions. Therefore, neu-rons which are connected to these brainstructures seem to be most interesting forthe study of the cellular basis of learningand memory.

In bees, the involvement of the mush-room bodies in associative memory con-solidation has been demonstrated usingamnestic treatments. Thus the conditioned

response probability was greatly reducedwhen the mushroom bodies had beentreated shortly after a single learning trial

(Menzel et al, 1974; Masuhr, 1976; Erberet al, 1980; Sugawa, 1986). Since an intra-cellular analysis of the mushroom body in-trinsic Kenyon cells is not possible, singlecell recordings have concentrated on thecharacterization of extrinsic elements (Er-ber, 1978, 1980, 1981, 1983; Hombergand Erber, 1979; Gronenberg, 1984, 1987;Homberg, 1984). A common feature ofthese neurons is their multimodality, iemost of them respond to more than onestimulus modality. Furthermore, in these

pioneering studies adaptive changes of theresponse behavior of mushroom body out-put neurons were already observed duringthe application of different stimulus config-urations.

The PE1 neuron is connected to mush-room body intrinsic Kenyon cells (Rybak,personal communication) and probablyprocesses the information to the medianand lateral protocerebrum. Despite its in-

teresting projection areas, there are sever-al advantages which make this neuron es-pecially suited for a cellular analysis ofinformation processing during learning(Mauelshagen, 1993): there is only onepair of PE1 neurons in the bee brain; theneuron seems to have a high integrativefunction, summing up the information of

many Kenyon cells, which might possiblyundergo changes in their response behav-

ior during memory formation; it has a largedendrite, which provides the prerequisitefor stable and repeated recordings; it has awell characterized electrophysiological sig-nature, which allows identification beforethe experiment.

Reproducible high-quality recordingsfrom a single neuron in the protocerebrumof the bee brain requires a reduced prepar-ation (single head preparation) which leadsto distortion of the proboscis extension re-flex. Single heads in a less reduced statecan show PER after a single learning trial,which suggests that the neuronal activity insingle heads may be comparable to thosein whole insects usually used for the be-havioral analysis of PER conditioning.Therefore, the missing control over the be-havioral indicator of the conditioned re-

sponse has to be compensated by a suffi-cient amount of recordings duringapplication of one experimental paradigmin order to statistically evaluate the re-

sponse changes between groups. Never-theless, the causality between neuronal

changes and learning has to be subjectedto future experiments.

In order to find correlates for non-

associative and associative memory pro-cessing, response changes due to stimu-lus paradigms for sensitization, 1-trial con-ditioning and differential conditioning wereexamined during repeated intracellular re-cordings from the PE1 neuron.

In a first series of experiments, para-digms which usually lead to proboscis ex-tension in response to odors were applied:i) a sensitization procedure with stimulationof the antennae and/or proboscis with su-crose solution; ii) a 1-trial conditioning pro-cedure by pairing an odor (carnation) withsucrose solution (to antennae and probos-cis); iii) a control group without sucrosestimulation. In a second series a differen-tial stimulus paradigm was applied, in

which 1 odor (carnation) was paired with

sucrose and a second odor (orange) waspresented unpaired. The odor responsesof the PE1 neuron were evaluated as rela-tive spike frequency changes between re-sponses during test or training and the pre-training responses.

The results (table II) could be character-ized by several points. In the same cell,there were measurable response changesdue to both nonassociative (sensitization)and associative (conditioning) stimulus

paradigms. Sensitization by antennal andproboscis sucrose stimulation had con-

trasting effects on the subsequent odor re-sponse in the PE1 neuron. This suggeststhat the processing of sucrose stimuli viaantennae and proboscis is manifested in 2qualitatively different pathways. The effectof compound sucrose stimulation to bothantenna and proboscis (sensitization) onthe odor response was different from the

response change after 1-trial conditioningwhich uses the same sucrose stimulus.This indicates that the pairing-specific con-ditioning effect is due to a different evalua-tion of the compound sucrose stimulus.The fact that the effect of 1-trial condition-

ing on the odor response is very similar tothat observed after proboscis sensitizationsuggests that the proboscis component ofcompound sucrose stimulation during con-ditioning might be dominant over the an-tennal component. The responsibility of theproboscis pathway for an evaluation of thesucrose reward has also been discussedfor behavioral studies on restrained bees

(Bitterman et al, 1983). During differentialconditioning, there is a differential repre-sentation of the 2 odors, suggesting a

learning-specific coding of odors in themushroom bodies. Furthermore, this dem-onstrates that in the mushroom bodies,there is a differential information process-ing during nonassociative and associativememory formation. The up-and-down shiftsof spike frequencies during differential con-ditioning are highly dynamic which might

be representative of different states of

memory consolidation. So far, all observedresponse changes have been transient (inthe minute range), suggesting that the PE1neuron is not the site of long-lasting mem-ory storage but rather might function as amonitor comprising the actual changes anddynamics during successive events of acontinuous learning process.

THE CONSEQUENCES

The aim of the behavioral and the electro-

physiological approach is to analyze theprincipal associations, their temporal com-ponents and mechanisms. However, sev-eral constraints should be noted: i) with

respect to the theories on instrumental

learning the interaction of the response,the reinforcer and the stimuli has not been

clarified; ii) the learning phenomena ob-served with free-flying bees have only part-ly been addressed with restrained bees; iii)the results regarding the electrophysiologyof identified neurons are limited by a short-lived preparation (lacking behavioral con-trol), by the difficulty in recording repeated-ly from small and single neurons and bythe short recording time which does not al-low access to long-term effects.

Thus all the results of experimentswhich are designed, performed and inter-preted on the basis of the current knowl-edge are limited in their validity. In particu-lar, it is of major importance to ascertainthe role of the restrained preparation as areliable source for studying the mecha-nisms which underlie learning phenomena.

The most successful strategy for solvingthis problem would be to identify those as-sociations of free-flying bees which are

most likely stimulus-reinforcer associa-tions and which can be interpreted in anal-ogy to the associative learning during clas-sical conditioning. A second requirement isto isolate these associations from the re-

sponse-reinforcer associations which obvi-ously depend on the operant behavior ofthe free-flying bee. This leads to the hy-pothesis that restrained bees should onlybe able to form the stimulus-reinforcer as-

sociations, but fail to learn stimuli whichare closely related to response-reinforcerassociations (this could mean additional

stimulus-response associations) or whichare not related to the reinforcer itself (thiscould mean stimulus-stimulus associa-

tions, where the second stimulus is not thereinforcer).

It is obvious that the identification ofthese associations constitutes the most

challenging task. The parameters whichcan be manipulated with free-flying beesare the specific selection of stimuli, the

quality of the reward and, most important-ly, the temporal relation between the stimu-li and the reward.

For restrained bees it is well known thatodors are readily associated with sucrosesolution; colors, however, are not. With

free-flying bees good learning for bothodors and colors is observed. According tothe above-mentioned hypothesis, thiscould indicate that with odors the bee

forms stimulus-reinforcer associations,whereas with colors the bee forms stimu-

lus-response associations. However, the

question remains open of whether the pro-cess of restraining the bees simply sup-presses the ability to learn colors. The as-sumption that in free-flying bees colors arenot directly associated with the reward is,however, supported by the findings ofGrossmann (1971) which show that pas-sive learning of colors is not possible. Thishe managed by temporally isolating thestimulus from the response, but not the

stimulus from the reinforcer. If this is true,the failure of color learning in restrainedbees is no artefact. This could be the first

hint that the failure of restrained bees tolearn certain stimuli as specific predictorsfor food is a true consequence of the na-

ture of associations in free-flying bees.Thus knowledge of associations in re-

strained bees might also possibly be trans-ferable to the interpretation of results on

free-flying bees.To clarify the nature of associations the

experiments on the optimization of a singlebee in a patch of artificially arranged feed-ers (Greggers and Menzel, 1993) weremodified. In this context one makes use ofthe specific associations between certainstimuli and the quality of the reward. Asensitive measure of such associations isthe optimization by matching described

above. All stimuli and close landmarkswere removed by installing the setup onthe roof of the institute and by arrangingthe 4 feeders in an open box facing thesky. In this situation optimization failed, iethe bee could not form differential memo-ries depending on the different rewardrates of the feeders. This opens the possi-bility for future experiments to introducethe stimuli separately. If the different stim-uli are directly associated with the differentreward rates and not only with their loca-tion, then the ability for optimization shouldbe restored, even if the location of thestimuli together with their correspondingreward were rotated from visit to visit. Fur-

thermore, the results should demonstratewhether the stimuli characterizing the loca-tion of the 4 feeders are learned only in aposition relative to each other, ie learningof one location would depend on the ar-rangement of the other alternatives. Con-sequently, optimization would fail if thestimuli were not only rotated in the ab-sence of other landmarks but also shiftedin their relative position.

With restrained bees the correlation be-tween performance and amount of rewardhas not been studied systematically. De-pendencies of the conditioned responseon modulations in US strength could be aprincipal rule underlying direct associa-tions between stimuli and reinforcer. By

testing this hypothesis with both free-flyingand restrained bees using odors or otherpossible conditioned stimuli, one could ob-tain a sensitive measure of the validity ofthe restrained preparation. Preliminaryelectrophysiological results obtained by re-cording from the PE1 neuron indicate thaton the neuronal level, there is a differentialrepresentation of 2 odors during differentialreinforcement (1% and 33% sucrose solu-tion, respectively) which would again sug-gest direct associations between odorsand sucrose solution.

Another approach to investigating thenature of the associations would be to per-form an experiment analogous to that inthe vertebrate literature. Colwill and Res-corla (1986) were successful in demon-

strating response-reinforcer associations

by applying the ’reinforcer-devaluationmethod’. Such evidence is based on differ-ential training of 2 responses with 2 differ-ent reinforcers which otherwise share thesame stimuli and location. After devalua-tion of 1 of the reinforcers in an indepen-dent situation, the rate of performance of 1

corresponding response should subse-

quently decrease in the case of a re-

sponse-reinforcer association. However,the difficulties encountered in designingsuch experiments for the bee are obvious:one has to find 2 reinforcers which excitedifferent reinforcing pathways in order tobe selectively devaluated. It may well bethat the only accessible reinforcing path-way is represented by the sucrose pro-cessing neurons via the proboscis. Fur-

thermore, the 2 reinforcers should havethe same subjective outcome for the beewhich has to perform both behavioral taskswith the same effort. Thus before design-ing the critical experiments, elaborate testswould be necessary concerning the kind ofresponses and reinforcers and the ade-

quate devaluation procedure. However,even control experiments could providesome insight into the simplicity or complex-ity of honey bee learning.

If the above-suggested experiments aresuccessful, and if it is possible to show thatin free-flying bees at least the odors are di-rectly associated with the sucrose rewardand not (or not only) with the response, thelocation or even other stimuli which char-acterize the feeding place, then it would be

possible to directly correlate results fromrestrained bees and free-flying bees. Inthis context, single cell recordings wouldbe of major importance, since they mightbe a much more sensitive measure of the

dynamics of memory processes than be-havioral studies alone, reflecting not onlythe time course of the probability of pro-boscis extension in a yes/no fashion, butcould perhaps also be correlated withsome more finely-tuned behavioral compo-nents related to stimulus evaluation and

expectation during memory consolidationor choice. Of course, to understand therole of single neurons such as the PE1neuron, recordings of other cells are abso-lutely necessary, parallel to an extensive

study on the connectivity of the complexnetwork. Only then could cellular changes,such as the described response modula-tions of the PE1 neuron, which are diver-

gent with respect to the stimulus configura-tions but convergent with respect to theirtime dependencies, be related to the be-havioral components in a meaningful man-ner.

Therefore, it should be possible to de-termine the nature of the associations andtheir mechanisms, in order to acquire aprecise knowledge of the temporal dynam-ics of memory formation by directly corre-lating experimental design and data ob-tained with the different but mutuallysupportive preparations.

ACKNOWLEDGMENTS

This review should provide encouragement tothose individuals who consider their work on

honeybee learning isolated from the natural life-

history of bees. We would also like to acknowl-edge those people in this laboratory who sup-ported our view that for an understanding of

learning it is necessary to remember that learn-

ing has a biological meaning. We thank P Maherfor improving the English manuscript.

Résumé — Approche expérimentale del’apprentissage associatif chez l’abeilledomestique, Apis mellifera L. L’approchemultiniveaux présentée ici et appliquée àl’étude de l’apprentissage et de la mémoiredes abeilles s’appuie sur 2 aspects fonda-mentaux du comportement d’apprentissa-ge : l’apprentissage instrumental d’abeillesen vol libre et le conditionnement classiquedu réflexe d’extension du proboscisd’abeilles en contention. Puisque les 2 ni-veaux sont nécessaires pour comprendrel’apprentissage dans le contexte naturel, ilest très important de s’assurer que c’est lemême type d’apprentissage qui est étudiédans les 2 approches. En termes de théo-ries de l’apprentissage, cela veut dire qu’ilfaut clarifier la nature des associations quise forment au cours des comportementsliés au butinage et pendant le conditionne-ment des abeilles en contention. Dans unesituation de conditionnement classique,des associations directes se forment entreles divers stimuli, tandis que, chez les

abeilles en vol libre, l’isolement des asso-ciations stimulus-renforcement nécessiteune analyse du rôle des différents stimuliqui caractérisent un lieu de nourrissement,de leur relation les uns aux autres, au

comportement opérant et à la qualité de larécompense obtenue.

Dans ce contexte, une approche récen-te utilisant des abeilles en vol libre semble

prometteuse (Greggers et Menzel, 1993).Le comportement de butinage d’abeilles in-dividuelles a été analysé à l’aide d’un dis-positif formé de 4 nourrisseurs artificiels in-stallés dans une prairie et présentant unesituation dynamique de récompense (fig

1). Tous les paramètres du comportementqui caractérisent une séquence de visitesont été contrôlés. L’abeille adapte son

comportement de choix à la situation de

récompense progressive (fig 2), ce qui in-dique la participation des processus d’ap-prentissage. Puisqu’il y a à la fois les para-mètres du comportement, qui ne

dépendent que de l’expérience à chaquenourrisseur artificiel, et les autres, qui ensont totalement indépendants (tableau I),on peut en conclure que les paramètresde la mémoire spécifiques des nourris-seurs et ceux qui sont non spécifiques gui-dent le comportement de choix. En utili-sant le comportement d’adaptation commeindicateur sensible de la mémoire associa-

tive, ce dispositif expérimental offre la pos-sibilité d’étudier directement le rôle des sti-muli et leur influence sur la formation desassociations. Les essais préliminaires indi-quent que si tous les stimuli et les repèrestopographiques proches sont enlevés, onn’observe pas d’adaptation, c’est-à-dire

pas d’apprentissage associatif. Cela per-met d’introduire séparément les stimuliétudiés.

Au niveau cellulaire, des enregistre-ments intracellulaires répétés de neuronesidentifiés permettent d’étudier le comporte-ment de réponse de neurones individuelspendant les paradigmes de l’apprentissa-ge olfactif. L’activité des potentiels d’actionévoqués dans le neurone VUMmx1 peutremplacer la récompense de saccharosependant le conditionnement olfactif classi-que (Hammer, 1991). Ce neurone a sonorigine dans le ganglion sous-œsophagienet se projette dans différentes régions duneuropile du cerveau, qui contiennent éga-lement des neurones des voies du traite-ment des odeurs. Le neurone PE1 assuredes connections entre les structures ducerveau qui sont des sites éventuels de laformation de la mémoire olfactive (Mauels-hagen, 1993; fig 3). L’application de para-digmes d’apprentissage simples, qui dans

les expériences comportementales provo-quent l’extension du proboscis en réponseà des odeurs, conduit à des modificationsde la réponse qui sont spécifiques du para-digme d’apprentissage appliqué (tableauII). Jusqu’à présent, toutes les modifica-tions observées étaient passagères, sug-gérant que le neurone PE1 n’est pas le sitede stockage de la mémoire à long terme,mais qu’il est impliqué dans une phase dy-namique brève au cours de la consolida-tion de la mémoire. L’électrophysiologiepourrait donc être un outil puissant pourdécrire la dynamique temporelle et les mé-canismes d’apprentissage à condition queles résultats soient inclus dans un contexte

de données issues d’études comportemen-tales sur des abeilles en vol libre.

apprentissage instrumental / condition-nement classique / apprentissage asso-ciatif

Zusammenfassung — ExperimentellerZugang zu assoziativem Lernen. Der ge-genwärtige vielschichtige Ansatz zur Erfor-schung von Lernen und Gedächtnis beider Honigbiene stützt sich auf 2 fundamen-tell verschiedene Aspekte des Lernverhal-tens: auf das instrumentelle Lernen freiflie-

gender Bienen und auf die klassische Kon-ditionierung des Rüsselreflexes bei fixier-ten Bienen. Da beide Zugänge notwendigsind, um Lernen im natürlichen Kontext zuverstehen, muß sichergestellt werden, daßin beiden Fällen vergleichbare Arten desLernens untersucht werden. Im lerntheore-

tischen Sinn bedeutet das, die Art der As-soziationen während des Sammelverhal-

tens und während der Konditionierung zubestimmen. In einer klassischen Konditio-

nierungssituation können Assoziationennur zwischen verschiedenen Stimuli auftre-

ten, während bei freifliegenden Bienen dieIsolierung solcher Assoziationen eine Ana-lyse der Stimuli erfordert, die den Futter-

platz charakterisieren, um ihren Zusam-

menhang untereinander, mit dem operan-ten Verhalten und mit der Qualität der Be-

lohnung zu erfassen.In diesem Zusammenhang scheint ein

neuerer Versuchsansatz vielversprechend(Greggers und Menzel, 1993). Auf einerWiese mit vier künstlichen Blüten wurde

das Sammelverhalten einzelner Bienen in

einer dynamischen Belohnungssituation(Abb 1) aufgezeichnet. Dies ermöglichtedie Analyse aller Verhaltensparameter, diewährend einer Abfolge von Blütenbesu-chen relevant sind. Die lerninduzierte An-

passung der Biene an die Belohnungssitu-ation äußert sich in einem "Matching"-Verhalten (Abb 2). Da es sowohl Verhal-tensparameter gibt, die nur von der jeweili-gen individuellen Blüte abhängen, als auchsolche, die von den individuellen Blüten

völlig unbhängig sind (Tabelle I), kann mandaraus schließen, daß das Entscheidungs-verhalten der Biene von blütenabhängigenund blütenunabhängigen Gedächtniskom-ponenten bestimmt wird. Da das "Mat-

ching"-Verhalten als empfindlicher Ge-dächtnis-Indikator benutzt werden kann,ermöglicht diese experimentelle Apparaturdie gezielte Analyse der relevanten Stimuliund deren Rolle bei der Gedächtnisbil-

dung. Vorläufige Ergebnisse zeigen, daßkein «Matching»-Verhalten auftritt, wennalle Stimuli und nahen Landmarken ent-

fernt werden. Das ermöglicht, die zu unter-suchenden Stimuli getrennt einzuführen.

Auf der zellulären Ebene erlauben ge-zielte intrazelluläre Ableitungen aus identi-fizierten Neuronen die Analyse von Reak-tionsveränderungen als Folge olfaktori-

scher Lernparadigmen. Evozierte Aktivitätim VUMmx1-Neuron kann die Zuckerwas-

ser-Belohnung während einer klassischenDuftkonditionierung ersetzen (Hammer,1991). Dieses Neuron hat seinen Ursprungim Unterschlundganglion und verzweigt inmehreren Neuropilbereichen des Gehirns,die außerdem Duft-prozessierende Neuro-

nen enthalten. Das PE1-Neuron verknüpftGehirnbereiche, die als Orte der Gedächt-nisbildung in Frage kommen (Mauelsha-gen, 1993; Abb 3). Die Anwendung voneinfachen Lernparadigmen, die in Verhal-tensversuchen zum Rüsselreflex auf Duft-stimuli führen, bewirkt eine Veränderungder Duftantwort dieses Neurons, die für diejeweilige Stimuluskonfiguration spezifischist (Tabelle II). Da die bisher gemessenenReaktionsveränderungen nur vorüberge-hend waren, ist das PE1-Neuron wahr-

scheinlich kein Ort lang anhaltender Ge-dächtnisspeicherung, sondern ist eher an

einem kurzfristigen und dynamischen Pro-zess während der Gedächtniskonsolidie-

rung beteiligt.Somit könnte besonders die Elektrophysio-logie im Hinblick auf die Beschreibung vonzeitlichen Dynamiken und die der Gedächt-nisbildung zugrundeliegenden Mechani-men sehr erfolgreich sein, vorausgesetztdie Ergebnisse können im Rahmen der

Gedächtnisphaenomene frei-fliegenderBienen interpretiert werden.

Instrumentelles Lernen / klassische

Konditionierung / assoziatives Lernen

REFERENCES

Altfelder K, Müller U (1991) Cyclic nucleotide-dependent protein kinases in the neural tis-sue of the honeybee, Apis mellifera. InsectBiochem 21 (5), 487-494

Altfelder K, Müller U, Menzel R (1991) Ca/calmodulin and Ca/phospholipid-dependentprotein kinases in the neural tissue of the

honeybee, Apis mellifera. Insect Biochem 21(5), 479-486

Bicker G (1993) Chemical architecture of anten-nal apthways mediating proboscis extensionlearning in the honeybee. Apidologie 24 (3)

Bitterman ME (1988) Vertebrate-invertebrate

comparisons. NATO ASI Ser Intell Evol Biol17, 251-275

Bitterman ME, Menzel R, Fietz A, Schäfer S(1983) Classical conditioning of proboscisextension in honeybees (Apis mellifera). JComp Psychol 97, 107-119

Braun G, Bicker G (1992) Habituation of an ap-petitive reflex in the honeybee. J Neurophysi-ol 67(3), 588-598

Colwill RM, Rescorla RA (1986) Associativestructures in instrumental learning. In: The

Psychology of Learning and Motivation (Bow-er GH, ed) Academic Press, New York, 55-104

Couvillon PA, Leiato TG, Bitterman ME (1991)Learning by honeybees (Apis mellifera) onarrival at and departure from a feeding place.J Comp Psychol 105(2), 177-184

Crow T, Forrester J (1990) Inhibition of proteinsynthesis blocks long-term enhancement ofgenerator potentials produced by one-trial invivo conditioning in Hermissenda. Proc NatlAcad Sci USA 87, 4490-4494

De Jong R, Pham-Delègue M-H (1991) Electro-antennogram responses related to olfactoryconditioning in the honey bee (Apis melliferaligustica). J Insect Physiol 37(4), 319-324

Erber J (1978) Response characteristics and af-ter effects of multimodal neurons in themushroom body area of the honeybee. Phys-iol Entomol 3, 77-89

Erber J (1980) Neural correlates of non-

associative and associative learning in the

honey bee. Verh Dtsch Zool Ges 250Erber J (1981) Neural correlates of learning in

the honeybee. Trends Neurosci 4, 270-272

Erber J (1983) The search for neural correlatesof learning in the honey bee. In: Neuroetholo-gy and Behavioral Physiology (Huber F,Markl H, eds) Springer, Berlin 216-227

Erber J, Masuhr T, Menzel R (1980) Localiza-tion of short-term memory in the brain of thebee, Apis mellifera. Physiol Entomol 5, 343-358

Erber J, Schildberger K (1980) Conditioning ofan antennal reflex to visual stimuli in bees

(Apis mellifera L). J Comp Physiol 135, 217-225

Frisch K von (1967) The Dance Language andOrientation of Bees. Harvard Univ Press,Cambridge, MA

Gould JL (1984) The natural history of honeybee learning. In: The Biology of Learning

(Marler P, Terrace H, eds) Springer, Berlin,89-114

Gould JL (1986) The biology of learning. AnnuRev Psychol 37, 163-192

Greggers U (1989) Optimizing strategies inchoice behaviour in the honeybee. In: NeuralMechanisms of Behavior: Proc 2nd Int CongrNeuroethology (Erber J, Menzel R, PflügerHJ, Todt D, eds) Georg Thieme, Stuttgart,219

Greggers U, Menzel R (1993) Memory dynamicsand foraging strategies of honeybees. BehavEcol Sociobiol 32, 17-29

Gronenberg W (1984) Das Protocerebrum derHonigbiene im Bereich des Pilzkörpers - eineneurophysiologisch-anatomische Charakteri-sierung. Dissertation, FU Berlin

Gronenberg W (1987) Anatomical and physio-logical properties of feedback neurons of themushroom bodies in the bee brain. Exp Biol46, 115-125

Grossmann K (1970) Erlernen von Farbreizenan der Futterquelle durch Honigbienenwährend des Anfluges und während des

Saugens. Z Tierpsychol 27, 553-562Grossmann KE (1971) Belohnungsverzögerung

beim Erlernen einer Farbe an einerkünstlichen Futterstelle durch Honigbienen.Z Tierpsychol 29, 28-41

Grossmann KE (1973) Continuous, fixed-ratio,and fixed-interval reinforcement in honeybees. J Exp Anal Behav 20, 105-109

Hammer M (1991) Analyse der funtionellenRolle des Neurons VUMmx1 bei der klassis-chen Konditionierung des Rüsselreflexes derBiene. Dissertation, FU Berlin

Hannes F (1930) Über die verschiedenen Artendes "Lernens" der Honigbiene und der Insek-ten überhaupt. Zool Jahrb Abt Allg Zool

Physiol Tiere 47, 89-150

Homberg U (1984) Processing of antennal infor-mation in extrinsic mushroom body neuronsof the bee brain. J Comp Physiol A 154, 825-836

Homberg U and Erber J (1979) Response char-acteristics and identification of extrinsicmushroom body neurons of the bee. Z Natur-forsch 34, 612-615

Kuwabara M (1957) Bildung des bedingten Re-flexes von Pavlovs Typus bei der Honig-biene, Apis mellifica. J Fac Sci HokkaidoUniv Ser VI Zool 13, 458-464

Lehrer M (1991) Bees which turn back and look.Naturwissenschaften 78, 274-276

Lieke E (1991) Asymmetric inhibitory interac-

tions in the antennal lobes of honeybees re-vealed by real-time optical imaging. Soc Neu-rosci Abstr 17, 640

Lieke E (1993) Optical recording of neuronal ac-tivity in the insect CNS: odorant coding bythe antennal lobes of honeybees. Eur J Neu-rosci 5, 49-55

Lindauer M (1963) Allgemeine Sinnesphysiolo-gie. Orientierung im Raum. Fortschr Zool 16,58-140

Masuhr T (1976) Lokalisation und Funktion desKurzzeitgedächtnisses der Honigbiene, Apismellifica L. Dissertation, TH Darmstadt

Mauelshagen J (1993) Neural correlates of ol-

factory learning paradigms in an identifiedneuron in the honey bee brain. J Neurophysi-ol 69 (2), 609-625

Menzel R (1968) Das Gedächtnis der Honig-biene für Spektralfarben. I. Kurzzeitiges undlangzeitiges Behalten. Z Vgl Physiol 60, 82-102

Menzel R (1987) Memory traces in honeybees.In: Neurobiology and Behavior of Honeybees(Menzel R, Mercer A, eds) Springer, Berlin,310-325

Menzel R (1990) Learning, memory, and cogni-tion in honey bees. In : Neurobiology of Com-parative Cognition (Kesner RP, Olten DS,eds) Erlbaum Inc, Hillsdale, NJ, 237-292

Menzel R, Bitterman ME (1983) Learning byhoney bees in an unnatural situation. In: Be-havioral Physiology and Neuro-Ethology:Roots and Growing Points (Huber F, Markl H,eds) Springer, Berlin, 206-215

Menzel R, Erber J, Masuhr T (1974) Learningand memory in the honeybee. In: Experimen-tal Analysis of Insect Behaviour (Barton-Browne L, ed) Springer, Berlin, 195-217

Menzel R, Hammer M, Braun G, MauelshagenJ, Sugawa M (1991) Neurobiology of learningand memory in honeybees. In: The Behaviorand Physiology of Bees (Goodman LJ, FisherRC, eds) CAB Int, Wallingford, UK, 323-251

Menzel R, Greggers U, Hammer M (1993) Func-tional organization of appetitive learning andmemory in a generalist pollinator, the honeybee. In: Insect Learning - Ecology and Evolu-tionary Perspectives (Papaj DR, Lewis AC,eds) Chapman and Hall, New York, 79-125

Mercer AR, Menzel R (1982) The effects of bio-genic amines on conditioned and uncondi-tioned responses to olfactory stimuli in the

honeybee, Apis mellifera. J Comp Physiol145, 363-368

Mercer A, Mobbs PG, Evans PD, Davenport A(1983) Biogenic amines in the brain of the

honey bee, Apis mellifera. Cell Tissue Res

234, 655-677

Michelsen DB (1988) Catecholamines affect

storage and retrieval of conditioned odourstimuli in honey bees. Comp Biochem Phys-iol 91C, 479-482

Mobbs PG (1982) The brain of the honeybeeApis mellifera I.The connections and spatialorganization of the mushroom bodies. Phil

Trans R Soc Lond B 298, 309-354

Mobbs PG (1984) Neural networks in the mush-room bodies of the honeybee. J Insect Physi-ol 30, 43-58

Mobbs PG (1985) Brain structure. In: Compre-hensive Insect Physiology Biochemistry andPharmacology (Kerkut GA, Gilbert LI, eds)Pergamon Press, Oxford, 299-370

Müller U, Altfelder K (1991) The Ca-dependentproteolytic system —calpain-calpastatin— inthe neural tissue of the honeybee Apis mellif-era. Insect Neurochem 21(5), 473-477

Opfinger E (1931) Über die Orientierung derBiene an der Futterquelle. Z vergl Physiol 15,432-487

Pavlov IP (1927) Conditioned Reflexes. OxfordUniv Press, London

Rehder V (1987) Quantification of the honey-bee’s proboscis reflex by electromyographicrecordings. J Insect Physiol 33, 501-507

Rehder V (1988) A neuroanatomical map of thesuboesophageal and prothoracic ganglia ofthe honey bee (Apis mellifera). Proc R SocLond 235, 179-202

Rehder V (1989) Sensory pathways and moto-neurons of the proboscis reflex in the suboe-sophageal ganglion of the honey bee.J Comp Neurol 279, 499-513

Rybak J (1987) Die morphologische Charakteri-sierung von alpha-Lobus extrinsischen Neu-ronen. Diplomarbeit, FU Berlin

Sahley CL, Gelperin A, Rudy JW (1981) One-trial associative learning modifies food odorpreferences of a terrestrial mollusc. Proc NatlAcad Sci USA 78, 640-642

Schäfer S, Rosenboom H, Menzel R (1993) lon-ic currents of identified neurons from themushroom bodies of the honeybee. In:

Beiträge zur 21. Göttinger Neurobiologenta-gung. Thieme, Stuttgart (abstr)

Sigurdson JE (1981a) Methods and designs -automated discrete techniques of appetitiveconditioning in honey bees. Behav ResMethods Instrum 13, 1-10

Sigurdson JE (1981b) Measurement of consum-matory behavior in honey bees. Behav ResMethods Instrum 13, 308-310

Skinner BF (1938) The Behavior of Organisms.Appleton-Century-Crofts, New York

Smith BH, Menzel R (1989a) The use of electro-mygram recordings to quantify odorant dis-crimination in the honey bee, Apis mellifera.J Insect Physiol 35, 369-375

Smith B, Menzel R (1989b) An analysis of vari-ability in the feeding motor program of the

honey bee: the role of learning in releasing amodal action pattern. Ethology 82, 68-81

Smith BH (1991) The olfactory memory of the hon-eybee Apis mellifera. I. Odorant modulation ofshort- and intermediate-term memory after sin-

gle-trial conditioning. J Exp Biol 161, 367-382Smith BH, Abramson CI, Tobin TR (1991) Con-

ditional withholding of proboscis extension inhoneybees (Apis mellifera) during discrimina-tive punishment. J Comp Psychol 105(4),345-356

Staddon JE (1983) Adaptive Behavior and

Learning. Cambridge Univ Press, New York

Sugawa M (1986) Zeitliche Dynamik und Lokali-sation des Engraumes im Honigbienengehirn(Apis mellifera). Diploma Thesis, TU Berlin

Vareschi E (1971) Duftunterscheidung bei der

Honigbiene - Einzelzell-Ableitungen und Ve-rhaltensreaktionen. Z vergl Physiol 75, 143-173

Wittstock S, Kaatz HH (1993) Inhibition of pro-tein synthesis by cyclohemimide does not af-fect formation of long-term memory in honeybees after olfactory conditioning. J Neurosci(in press)