Insect Learning

views updated


Why study learning in insects? What can it contribute to a general knowledge of how learning takes place in a wide variety of animals? There are many potential answers to these questions. This review will focus on the general contribution that can be made to systematic understanding of how learning has evolved and is controlled in a wide variety of vertebrate and invertebrate species. Any systematic study must begin with a well-defined phylogenetic lineage. Insects are in the phylum Arthropoda, which contains animals that have jointed exoskeletons (e.g., insects, ticks, crabs, lobsters, spiders). With at least 2 million extant species (some estimates range as high as 30 to 50 million), the arthropod class Insecta comprises the most diverse group of multicellular organisms (Borror, Triplehorn, and Johnson, 1989). Insects have adapted to a wide array of living conditions, ranging from most terrestrial to many aquatic environments. The diverse insect species found in these environments must solve the basic problems inherent in locating resources such as food or mates and avoiding predatory or environmental threats. The learning abilities observed in any laboratory situation probably evolved to solve these problems.

Because of this species and habitat diversity, insects provide an excellent means of testing patterns of phylogenetic emergence of different learning mechanisms. The appropriateness of using cognitive explanations, the lack of generally agreed upon operational definitions, the need for learning taxonomies, and a recognition of individual differences in performance are all relevant issues in the study of invertebrate learning. One must also carefully distinguish phylogenetically homologous traits from analogous ones. Homology refers to traits that arise from a common ancestral condition. Researchers must always consider rigorous criteria for any proposed homology between behaviors of different groups of invertebrates and/or vertebrates (Wenzel, 1992). A monophyletic group of animals that possesses homologous traits is a clade, and the process of modification of those traits is cladogenesis.

Through a comparison of traits that may appear dissimilar and thus unrelated in several closely related, extant species, it is possible to obtain a picture of which traits are ancestral (plesiomorphic) and which are derived. For example, are nonassociative learning mechanisms homologous to associative mechanisms? That is, did ancestral species possess the ability to modify behavior through habituation and sensitization prior to the ability to express associative conditioning? Were changes in the evolution of learning abilities gradual, adaptive alterations or the result of rapid, discontinuous changes resulting from dramatic reorganization of neural tissue (Wyers, Peeke, and Herz, 1973)? Through a study of insect species whose phylogenies with respect to other characteristics (e.g., morphological, physiological) are known, such hypotheses can be tested.

Analogous traits are physically similar but have been derived from very different ancestral conditions. For example, the expression of operant conditioning of leg movement in an insect might be analogous to operant conditioning of leg movement in a vertebrate. Insect legs and vertebrate legs are not homologous structures, but both enable animals to move throughout their environment and thus the rules for operant conditioning of each may be similar. Analogous traits associated with complex learning abilities arise through convergent evolution, perhaps because of similar environmental problems that require one or more ways to modify behavior based on experience. This description of analogous learning abilities in terms of degrees of complexity is anagenesis (Demarest, 1983). A comparative study of potentially analogous learning mechanisms in such phylogenetically diverse groups as insects and vertebrates allows the testing of hypotheses about the conditions that give rise to analogous learning abilities. Thus learning abilities studied in insects have an important value for deriving hypotheses that are testable in vertebrates. Even if learning traits do not have a common phylogenetic origin, working out mechanisms in one species can generate conceptual advances in understanding a similar learning ability in another species.

No single study can be designed to investigate all of these issues. In the long run this approach must make extensive use of several species that are chosen to appropriately test defined phylogenetic hypotheses (Wenzel, 1992). What follows, then, is not a comprehensive review of the invertebrate learning literature but rather a highlighting of significant studies.


The order Orthoptera comprises cockroaches, grasshoppers, and locusts. Horridge (1962) published results of an experiment in which headless cockroaches and locusts learned to keep one leg raised to terminate a series of electric shocks. This experiment generated considerable interest because it was one of the first to suggest that an insect can be used to explore the physiology of learning and memory. Subsequently, leg-position learning, or the Horridge paradigm, has demonstrated learning in a wide variety of experimental situations, ranging from intact animals to a single ganglion. This latter information demonstrated that learning need not be confined to a single area of the central nervous system (e.g., the brain) but can be distributed throughout several stimulus and motor processing pathways in a nervous system.

The Horridge paradigm also brought into focus the adequacy of the yoked-control design in separating learning from nonassociative effects. In the original Horridge experiment, learning was inferred from a difference in the number of shocks received by experimental subjects and by their yoked controls. The experimental subjects were shocked contingent on leg position; control subjects were yoked in such a way that they received a shock whenever the experimental subjects did, but independent of their own behavior. The yoked paradigm has been extensively criticized in the literature. Church (1964), for instance, pointed out that because of the nature of the yoked paradigm, random differences in inherent responsiveness will lead to artifactual learning in the population.

To answer such criticisms, a new experimental design was developed for training leg position in the locust (Forman, 1984). Rather than simply requiring the animal to raise a leg to terminate a series of shocks, Forman required his locust to maintain a particular range of leg movements arbitrarily selected by the experimenter. After a few minutes of training, the animal learns to shift its leg position to an angle that terminates aversive heat to the head or produces access to food. Locusts can also be trained to manipulate leg position to produce heat to the head in a cold environment. The task can be made more complex by narrowing or shifting the range of leg movement necessary to control the heat stimulus. The Forman experiment is important because it represents the first significant improvement in the Horridge paradigm; both the response and the reinforcer are arbitrary, and learning can be identified in an individual animal. This procedure has the additional advantage of eliminating shock as the aversive stimulus.

The rationale for developing the Forman paradigm is to obtain data on the cellular mechanisms underlying operant behavior. By using electromyograms and intracellular techniques, the motor neurons involved in learning have been found and characterized. Forman and Zill (1984), for example, identified three separate motor strategies utilized by the locusts during training ("kicking," changes in muscle tonus, and tonic slow excitor motor neuron activity). Each of these strategies can be selectively trained. An exciting application of the technique and a fine example of the comparative method is an analysis of the similarities and differences in response strategies between locusts and the weta, a primitive New Zealand insect related to the locust (Hoyle and Field, 1983).


The order Diptera comprises all flies. Fly species that have been extensively used in studies of learning include Phormia regina (blowflies) and Drosophila melanogaster (fruit flies). Research interest in flies was generated by the extreme ease of maintaining populations under controlled mating conditions over generations that cover only weeks rather than years. Controlled breeding experiments have characterized the behavioral, genetic, and biochemical bases of different learning mechanisms.

Some of the first comprehensive studies of learning behavior in flies began with the pioneering work of Dethier and colleagues (1990), which worked out in considerable detail the stimulus control of feeding reflexes in Phormia regina. They described a procedure in which the tarsal (leg) receptors that mediate sucrose taste sensation were stimulated to elicit extension of a subject's proboscis (the sucking mouthparts) through which it feeds on the sucrose-water droplet. They found that prior exposure to sucrose greatly increased the probability that a fly would extend its proboscis to the presentation of water alone. The motivational state that was modulated by the sucrose exposure was termed central excitatory state (CES), which describes a nonassociative (sensitization) modification of the probability of proboscis extension to a neutral stimulus (water). CES can be characterized by at least three factors: There is a decay over time between the sensitizing and test stimuli; increased sucrose concentrations lead to increases in CES; food deprivation leads to increased levels of CES for any given sucrose concentration.

Tully and Hirsch's studies (Tully, 1984) have documented genetic bases for CES effects in P. regina, and other studies extended the results on the genetic basis for CES effects to D. melanogaster (Vargo and Hirsch, 1986). Bidirectional selection for high and low CES lines in P. regina has shown that the response to selection is rapid and may reach asymptotic levels in one or a few generations. Hybridization of the different lines indicated that one major gene segregated in the selected lines was responsible for producing most of the variability in CES effect. Selection for CES effects in D. melanogaster has shown a slightly different genetic basis. Sometimes a low but not a high line was produced, or vice versa. These data indicate that several genes may be involved in regulating CES in fruit flies. Further studies have shown that genes reside on at least two chromosomes; heritable cytoplasmic factors may also be involved (Vargo and Hirsch, 1986).

Theories of anagenesis predict that as metazoan life becomes physically more complex, more complex learning abilities will emerge (Demarest, 1983). Thus associative learning mechanisms may be mechanistically related to simpler nonassociative processes. Accordingly, learning studies with both P. regina and D. melanogaster have focused on developing associative conditioning paradigms and testing for genetic correlates with nonassociative processes. By associating either a saline or a water conditioned stimulus (CS) with sucrose, P. regina can learn to extend their proboscises to the CS (Nelson, 1971). Tully, Zawistowski, and Hirsch (1982) selected for high and low learning lines of blowflies. These lines showed a positive correlation between CES levels and asymptotic levels of learning performance. Therefore, CES and associative conditioning appear to have at least some common genetic bases, which might include pleiotropic genetic effects.

Aversive conditioning has been widely used to select large numbers of D. melanogaster in order to rapidly isolate mutant strains that show deficiencies in learning and/or memory (Dudai, 1983). The procedure involves exposing flies sequentially to two odors while they walk across a metal grid (Tully and Quinn, 1985). While they are exposed to one odor, they receive shocks through the grid. Flies are then presented with a sequence of new "collection" tubes that contain either the odor paired with shock (S+) or the odor that was not paired with shock (S-). The response measure is the number of flies that enter a new tube that contains the S+ odor versus the number that enter a tube containing the Sodor. Over several runs, decreased entries into the tube with the S+ odor relative to the tube with the S-odor indicate learning.


The order Hymenoptera contains a diverse group of insects commonly referred to as sawflies, ants, wasps, and bees. Hymenopterans such as ants and honeybees have been widely used to document learning abilities related directly to learning problems in the animal's natural environment. Furthermore, experiments with ants and bees in laboratory learning paradigms have demonstrated that these abilities conform to standard definitions of learning. But an ant's or a bee's learning ability may be less complex and/or less generalizable to new situations than that of animals with larger, more elaborate nervous systems (Demarest, 1983). Indeed, the crucial question is how complex these abilities are and to what natural situations they can be applied.

Learning in ants was first brought into the laboratory by Fielde (1901), who reported that ants can successfully negotiate a simple maze. Schnierla (1946) described the chemical, visual, and kinesthetic cues used by ants in solving a more complex maze. DeCarlo and Abramson (1989) used a different procedure to extend vertebrate learning paradigms to ants. They demonstrated an ant's ability to choose one compartment of a two-compartment chamber based on rates of stimulus delivery.

The honeybee (Apis mellifera) is an ideal species with which to research similar questions. On warm, sunny days worker bees regularly depart from the colony on foraging trips during which they collect resources (e.g., nectar, pollen, water) crucial for survival and reproduction of the colony. A large number of studies have documented the abilities of freely flying honeybees to learn the relationships of visual, tactile, and olfactory cues to appetitive and aversive stimuli (Menzel, 1990; Bitterman, 1996). For example, forager bees learn the association of nectar, which for most conditioning studies is replaced with a sucrose-water mixture and floral color, shape, odor, and the time of day that floral rewards are available. Other work has documented the honeybee's ability to learn compounds of stimuli. The unconditioned stimulus-preexposure effect and latent inhibition have also been studied. For studies of aversive learning, Abramson (1986) has demonstrated the ability of freely flying bees to use certain stimuli as a means of avoiding exposure to an aversive-shock stimulus. The learned avoidance ability of the honeybee may have evolved as a means to cope with bitter and even toxic nectars found in some flowers.

Proboscis-extension conditioning of honeybees is a common technique for studying learning under easily controllable stimulation variables (Menzel 1990). Honeybees restrained individually in harnesses can be readily conditioned to extend their proboscises upon presentation of a floral odor. After one or a few pairings of an odor-conditioned stimulus with a sucrose unconditioned stimulus, 40 to 90 percent of the subjects will extend their proboscises (conditioned response) to the odor alone. Enhancement of a background rate of proboscis extension to odor is specific to situations in which the CS precedes the US (forward pairing) and is sensitive to latency of onset of odor relative to sucrose. Proboscis-extension conditioning has been used to study a variety of phenomena in honeybees, such as sensory discrimination (Smith and Menzel, 1989a); control of motor systems (Smith and Menzel, 1989b); memory consolidation (Menzel, 1990). More recent work has extended to restrained bees Abramson's (1986) study of aversive conditioning in freely flying bees. The majority of subjects that received a shock contingent upon their response to sucrose in the context of a particular odor learned to withhold proboscis extension to sucrose in order to avoid shock (Smith, Abramson, and Tobin, 1991).

The question whether bees and ants have a cognitive map has generated controversy. That is, have they learned a "mental analogy of a topographic map, i.e., an internal representation of the geometric relations among noticeable points in the animal's environment" (Wehner and Menzel, 1990, p. 403)? Evidence to date points to a vector-based navigation system combined with memory matching of relative positions of landmarks (Cartwright and Collett, 1987) rather than to a more complex topographic representation.

New Developments in Invertebrate Learning

Later research in insect learning has emphasized one species of insect: the honeybee (Bitterman, 1996). This work emphasizes the type of research program into learning mechanisms that ought to be taken up on a much wider array of species.


Kamin (1968) conditioned rats to one stimulus (A) and followed that with conditioning to a second stimulus that was a compound of the first with a novel stimulus (AX). When finally tested with X, subjects typically revealed that they had learned less about X than subjects in appropriate control groups. This "blocking" phenomenon went on to spawn a tremendous volume of both theoretical and empirical research. Most research on vertebrates has emphasize blocking between cues from different stimulus modalities (e.g., compounds of one visual and one acoustic cue). Recently, blocking has been investigated in a series of studies in honeybees (Smith 1997). In contrast to vertebrates, blocking is more evident in compounds made of the same stimulus modality (mixtures of two olfactory or two visual cues), although intermodal blocking has been identified for some stimulus compounds (Couvillon, Campos, Bass, and Bitterman, 2001). Whether this stimulus specificity represents a fundamental difference between vertebrates and invertebrates must await further study.

CS Preexposure

Unreinforced exposure to a conditioned stimulus leads to retardation of learning about that CS when it is subsequently paired with reinforcement in a way that would normally result in robust conditioned responding. This mechanism has also been referred to as "latent inhibition." Abramson and Bitterman (1986) initially reported a CS-preexposure effect in free-flying honeybees. Studies of odor preexposure have revealed the effect in laboratory studies in honeybees (Chandra, Hunt, and Smith, 2001). These latter studies highlight the importance of recognition of individual difference in expression of a preexposure effect. Individual honeybees differ in the extent that they exhibit the effect, and both studies were successful in demonstrating that individual differences have a genetic basis. This latter finding indicates that individual differences are most likely not an artifact of the approach, and they may point in future studies to an ecological meaning for individual differences. Chandra, Hunt, and Smith (2001) isolated segments of chromosomal DNA that presumably house genes that influence this trait.

Risk Sensitivity and Choice Behavior

Two studies have revealed risk sensitivity in honeybees (Shafir, Wiegmann, Smith, and Real 1999; Shapiro, Couvillon, and Bitterman, 2001). When two different conditioned stimuli are reinforced at the same mean rate but the reinforcement differs in variance, animals may prefer the less variable option (risk aversion), the more variable option (risk prone), or neither (risk insensitive). Under normal conditions honeybees are risk-averse. As with CS preexposure, individuals differ in their degree of risk sensitivity. This behavior resembles risk sensitivity in vertebrates. But a model that incorporates associationist concepts can account for risk sensitivity in honeybees (Shapiro, Couvillon, and Bitterman, 2001). So more cognitive interpretations of this phenomenon need not be invoked. More recently, Shafir, Waite, and Smith (2002) have shown that honeybees violate basic properties of rational choice because the choice between two options for reward depends on available alternatives. Their relative preference between two options changes with the introduction of a third, relatively unattractive option.


Honeybees master more abstract relationships among cues, such as "sameness" or "difference" (Giurfa et al., 2001). Honeybees can be conditioned to respond to a cue based on whether or not it matched a sample cue to which animals were recently exposed. This rule can also be used in a more general sense: once the rule is learned in one stimulus modality. it can apply to cues from a different modality. Risk sensitivity and matching-to-sample may yet prove to be fruitful for pursuit of more cognitive manifestations of invertebrate learning.

Comparative Analysis of Learning in Africanized Honeybees

Another recent development in learning of honeybees is the study of the Africanized honeybee. Studies have examined conditioning to various stimuli, extinction (both unpaired and CS only), conditioned inhibition, color and odor discrimination, and learning in day-old bees. The results suggestion a subspecies difference between European and Africanized honeybees. In addition to work on basic phenomena, experiments on practical applications of conditioning methodology are illustrated with studies demonstrating the effects of insecticides on learning and the reaction of Africanized bees to consumer products (Abramson, Aquino, Silva, and Price, 1997; Abramson, Aquino, Ramalho, and Price, 1999; Abramson, Aquino, and Stone, 1999).

The Development of a Social Insect Model of Alcoholism

Honeybees are also a model for studies of alcoholism in humans. Honeybees have much to recommend them for such studies, including a language and social structure. Studies have shown that bees will self-administer ethanol and that locomotion and learning is impaired in a dose-dependent manner (Abramson et al., 2000).



Abramson, C. I. (1986). Aversive conditioning in honeybees (Apis mellifera). Journal of Comparative Psychology 100, 108-116.

Abramson, C. I., Aquino, I. S., Ramalho, F. S., and Price, J. M. (1999). Effect of insecticides on learning in the Africanized honey bee (Apis mellifera L.). Archives of Environmental Contamination and Toxicology 37, 529-535.

Abramson, C. I., Aquino, I. S., Silva, M. C., and Price, J. M. (1997). Learning in the Africanized honey bee: Apis mellifera L. Physiology and Behavior 62, 657-674.

Abramson, C. I., Aquino, I. S. and Stone, S. M. (1999). Failure to find proboscis conditioning in one-day old Africanized honeybees (Apis mellifera L.) and in adult Uruçu honeybees (Melipona scutellaris). International Journal of Comparative Psychology 12, 242-262

Abramson, C. I., and Bitterman, M. E. (1986). Latent inhibition in honeybees. Animal Learning and Behavior 14, 184-189.

Abramson, C. I, Stone, S. M., Ortez, R. A., Luccardi, A., Vann, K. L., Hanig, K. D. and Rice, J. (2000). The development of an ethanol model using social insects I: Behavior studies of the Honey Bee (Apis mellifera L.). Alcoholism: Clinical and Experimental Research 24, 1,153-1,166.

Bitterman, M. E. (1996). Comparative analysis of learning in honeybees. Animal Learning and Behavior 24, 123-141.

Borror, D. J., Triplehorn, C. A., and Johnson, N. F. (1989). An introduction to the study of insects, 6th edition. Philadelphia: Saunders College Publishing.

Cartwright, B. A., and Collett, T. S. (1987). Landmark maps for honeybees. Biological Cybernetics 57, 85-93.

Chandra, S. B. C., Hunt, G., and Smith, B. H. (2001) Quantitative trait loci associated with reversal learning and latent inhibition in honeybees (Apis mellifera) Behavior Genetics 31, 275-285.

Church, R. M. (1964). Systematic effect of random error in the yoked control design. Psychological Bulletin 62, 122-131.

Couvillon, P. A., Arakaki, L., and Bitterman, M. E. (1997). Intramodal blocking in honeybees. Animal Learning and Behavior 25, 277-282.

Couvillon, P. A., Campos, A. C., Bass, T. D., and Bitterman, M. E. (2001). Intermodal blocking in honeybees. Quarterly Journal of Experimental Psychology B. 54, 369-381.

DeCarlo, L. T., and Abramson, C. I. (1989). Time allocation in the carpenter ant (Camponotus herculeanus). Journal of Comparative Psychology 103, 389-400.

DeJianne, D., McGuire, T. R., and Pruzan-Hotchkiss, A. (1985). Conditioned suppression of proboscis extension in Drosophila melanogaster. Journal of Comparative Physiology and Psychology 99, 74-80.

Demarest, J. (1983). The ideas of change, progress, and continuity in the comparative psychology of learning. In D. W. Rajecki, ed., Comparing behavior: Studying man studying animals. Hillsdale, NJ: Erlbaum.

Dethier, V. G. (1990). Chemosensory physiology in an age of transition. Annual Review of Neuroscience 13, 1-13.

Dudai, Y. (1983). Mutations affect storage and use of memory differentially in Drosophila. Proceedings of the National Academy of Sciences of the United States of America 80, 5,445-5,448.

Fielde, A. (1901). A further study of an ant. Proceedings of the National Academy of Sciences of the United States of America 53, 521-544.

Forman, R. R. (1984). Leg position learning by an insect. I. A heat avoidance learning paradigm. Journal of Neurobiology 15, 127-140.

Forman, R. R., and Zill, S. N. (1984). Leg position learning by an insect. II. Motor strategies underlying learned leg extension. Journal of Neurobiology 15, 221-237.

Giurfa, M., Zhang, S., Jenett, A., Menzel, R., and Srinivasan, M. V. (2001). The concepts of "sameness" and "difference" in an insect. Nature 410, 930-933.

Horridge, G. A. (1962). Learning of leg position by headless insects. Nature 193, 697-698.

Hoyle, G., and Field, L. H. (1983). Elicitation and abrupt termination of behaviorally significant catchlike tension in a primitive insect. Journal of Neurobiology 14, 299-312.

Kamin, L. J. (1968). Attention-like processes in classical conditioning. In Jones, M. R., ed., Miami symposium on predictability of behavior: Aversive stimulation. Coral Gables, FL: Miami University Press.

Menzel, R. (1990). Learning, memory, and "cognition" in honeybees. In R. P. Kesner and D. S. Olton, eds., Neurobiology of comparative cognition. Hillsdale, NJ: Erlbaum.

Nelson, M. C. (1971). Classical conditioning in the blowfly (Phormia regina): Associative and excitatory factors. Journal of Comparative Physiology and Psychology 77, 353-368.

Schnierla, T. C. (1946). Ant learning as a problem in comparative psychology. In P. Harriman, ed., Twentieth-century psychology. New York: Philosophical Library.

Shafir, S., Waite, T. A., and Smith, B. H. (2002). Context-dependent violations of rational choice in honeybees (Apis mellifera) and gray jays (Perisoreus canadensis). Behavioral Ecology Sociobiology 51, 180-187.

Shafir, S., Wiegmann, D. D., Smith, B. H., and Real, L. A. (1999). Risk-sensitivity of harnessed honeybees to variability in volume of reward. Animal Behaviour, 57, 1,055-1,061.

Shapiro, M. S., Couvillon, P. A., and Bitterman, M. E. (2001). Quantitative tests of an associative theory of risk-sensitivity in honeybees. Journal of Experimental Biology 204, 565-573.

Smith, B. H. (1997). An analysis of blocking in binary odorant mixtures: An increase but not a decrease in intensity of reinforcement produces unblocking. Behavioral Neuroscience 11, 57-69.

Smith, B. H., Abramson, C. I., and Tobin, T. R. (1992). Conditional withholding of proboscis extension in honeybees (Apis mellifera) during discriminative punishment. Journal of Comparative Psychology 105, 345-356.

Smith, B. H., and Menzel, R. (1989a). The use of electromyogram recordings to quantify odor discrimination in the honeybee. Journal of Insect Physiology 35, 369-375.

—— (1989b). An analysis of variability in the feeding motor program of the honey bee: The role of learning in releasing a modal action pattern. Ethology 82, 68-81.

Tully, T. (1984). Drosophila learning: Behavior and biochemistry. Behavior Genetics 14, 527-557.

Tully, T., and Quinn, W. G. (1985). Classical conditioning and retention in normal and mutant Drosophila melanogaster. Journal of Comparative Physiology 157, 263-277.

Tully, T., Zawistowski, S., and Hirsch, J. (1982). Behavior-genetic analysis of Phormia regina: III. A phenotypic correlation between the central excitatory state (CES) and conditioning remains in replicate F2 generations of hybrid crosses. Behavior Genetics 12, 181-191.

Vargo, M., and Hirsch, J (1986). Biometrical and chromosome analyses of lines of Drosophila melanogaster selected for central excitation. Heredity 56, 19-24.

Wehner, R., and Menzel, R. (1990). Do insects have cognitive maps? Annual Review of Neuroscience 13, 403-404.

Wenzel, J. W. (1992). Behavioral homology and phylogeny. Annual Review Entomology 23, 361-381.

Wyers, E. J., Peeke, H. V. S., and Herz, M. J. (1973). Behavioral habituation in invertebrates. In H. V. S. Peeke and M. J. Herz, eds., Habituation, Vol. 1. New York: Academic Press.

Brian H.Smith

Charles I.Abramson