Flowers are unreliable, widely distributed food sources, normally offering minute rewards. Flowers of the same kind tend to bloom in close proximity, because plants of the same species growing in patches often bloom simultaneously, or a single plant has many blossoms. Thus, a patch of flowers of the same kind has a location in space and exists for some time, perhaps longer than the lifespan of an insect pollinator. A typical habitat consists of several to many patches of flowers, some of the same species, some of others; pollinators must choose between them.
Hymenopteran pollinators visit flowers to provide food for themselves and their brood. They frequently travel long distances between the nest site and the flower patches, carrying pollen and nectar. Since they must visit many flowers per foraging bout, they need to decide between different flowers in quick succession. Both innate preferences and experience guide the decision-making process (Menzel 1985). Since most of the approach flights are either return visits to a plant or first visits to nearby ones, pollinators are guided mainly by their memories of the location of productive flowers and the particular features of the flowers (signals, manipulatory properties, reward conditions) that the insects learned during previous visits.
Insects' learning capacity and richness of memories are usually underestimated, but studies of learning and memory in honeybees (under both natural and laboratory conditions) demonstrate that learning is fast, and comprises various levels of cognitive processing, such as generalization, categorization, concept formation, configuration, and context-dependency (Menzel & Giurfa 2001). Memory is rich, highly dynamic, and long-lasting (Menzel 1999).
Here I take up the case for the decision-making process being guided by navigational memories and memories of signals from the flowers themselves. Specifically, I shall argue that the components of the pollinators' navigational memory are intimately connected with memories of the flower signals, leading to a unique neural representation of localized and qualified objects (nest site, feeding places with particular properties, landmarks passed, etc.). Patches of flowers are localized in space, and bees navigate between loci in a goal-directed fashion. They establish locus-specific memories, and thus their navigational capacities are a major component in returning to a flower, identifying it as a productive one, and handling it efficiently.
Learning all these features of a flower - location, signals, construction - establishes composite memories, whose impact on choice behavior is continuously updated, both with reference to new experience and to elapsing time. Most importantly, memory is not a unique and stable entity of information storage - not in bees nor in any other animal (Milner et al. 1998) - but rather a dynamic process establishes different and sequential forms of memory phases, which are then transformed by consolidation processes. The central argument put forward here is that the contents and dynamics of the memory phases are the major factors controlling choice behavior, and thus flower constancy.
Our case study is the honeybee. There is no reason to believe that the honeybee is in any way special in its cognitive capacities, because the main requirement, namely, goal-directed navigation between nest site and feeding places, must be met by any hymenopteran pollinator species. In this sense, honeybees can be studied as a representative species of hyme-nopteran pollinators, including both social and solitary bees.
Localization on a rough scale: the structure of navigational memories
Foraging bees embark on feeding flights and return to the hive using sun compass information (von Frisch 1965; Wehner & Menzel 1990), visual distance estimation (Esch & Burns 1995; Srinivasan et al. 1996), path integration (Wehner 1992), and visual landmarks (von Frisch 1965; Menzel et al. 1996; Collett & Zeil 1998). These sources of information are tightly interconnected: compass directions are derived from both extended landmarks (von Frisch 1965, Dyer & Gould 1981) and from home vectors associated with local landmarks (Menzel et al. 1998), establishing a memory for the flight route between the hive and a frequently visited feeding site.
Besides this specialized route memory (SRM), bees leaving the hive for the first time do not fly straight to a feeding site, but, rather, perform elaborate orientation and learning flights (von Frisch 1965) of increasing duration and distance (Becker 1958; Vollbehr 1975; Capaldi & Dyer 1999; Capaldi et al. 2000). It was recently shown (Menzel et al. 2000) that the spatial memory established during these learning flights leads to a general landscape memory (GLM), coding and storing the layout of landmarks in a geometric sense with the hive at the center.
To uncover the structure of GLM, it was necessary to perform experiments in which bees were not trained along a route, because route-trained animals apply the vector memory for the route when released at an unexpected place, thereby giving the impression that they are unable to localize the release site relative to the intended goal (Wehner & Menzel 1990). We therefore tested bees that had not established a SRM. We chose a moving feeder (called the "variable feeder") which circled around the hive at close range (5-10 m) several times a day. Bees trained under these conditions are called V-bees. These V-bees were compared to bees trained in the usual way to a constant feeder at a greater distance to the hive (C-bees). Both groups of bees were released at five sites 350 m away from the hive. For C-bees one of these sites was their familiar training site. To measure navigational performance, both vanishing bearings at the release site and flight time were recorded.
It was found that C-bees follow their compass memory at the release site as expected. They take a long time to return to the hive, particularly when the initial flight route carries them further away from the hive, but eventually all of the bees arrived at the hive, indicating that they refer to some other form of spatial memory when their active memory about the flight vector has vanished. V-bees, on the other hand, showed a weak tendency to fly into the 180° sector toward the hive from any of the five release sites and, most importantly, arrived at the hive after a brief flight time, a flight time that was not significantly longer than the flight time of C-bees along their trained route (Menzel et al. 2000).
These results indicate that bees do indeed possess a form of geometric representation of the landmark layout when they refer to GLM, but not when they refer to SRM. Since no natural feeding spots were available during the test period, bees must have established GLM during their orientation flights. GLM is suppressed by SRM, as indicated by the fact that the C-bees first follow their active flight memory; however, SRM does not erase GLM, since, if SRM did not lead back to the hive, C-bees were able to activate GLM from a remote store and use it for navigating. Otherwise we would not have observed C-bees returning to the hive.
The map-like organization of GLM proves a hitherto unexpected dimension of navigational capacity in a pollinating insect. Using the harmonic radar tracking technique (Riley et al. 1996), we have recently shown that bees referring to GLM do not only return to the hive on direct flights over distances of several hundreds of meters, but may also choose to fly to a feeding site first. This indicates that the structure of GLM is not confined to spatial relationships between the central spot (hive) and landmarks, but, rather, any location within GLM can be chosen as a goal from any other location. The neural structure of GLM might be that of a maplike representation of the landscape and thus indicative for a "cognitive or mental map". Such a claim has been made by Gould (1986) on the basis of vanishing bearings of C-bees taken at the release site. Gould's observations could not be verified in any of our studies or those of other researchers (Wehner & Menzel 1990): route-trained animals always applied their SRM and flew in the wrong direction. The vanishing bearings at the release sites were the only data available to Gould, and it is still unclear how he arrived at the conclusion that bees refer to geometric structure of spatial memory.
Bees forage in a known landscape whose geometric structure is stored in their spatial memory. The locations of rewarding sites are characterized by their particular features and are memorized accordingly. Bees learn the local features (signals, localization relative to landmarks, reward conditions) of two to four feeding sites, and behave accordingly: they choose the correct color at the correct time and place (Menzel et al. 1999; Lehrer 1999) or the correct color pattern at the correct step in a sequence (Collett 1992); they choose the correct odor at a particular time (Koltermann 1971); they indicate the correct direction and distance to one of two feeding sites according to time of the day (von Frisch 1965, table 37); and, they match the frequency of their visits to the reward quantities of at least four feeding sites (Greggers & Menzel 1993).
Furthermore, bees have the capacity to switch their motivation according to recent experience and activate remote memory according to the motivational change. Take the following experiment as an example of the flexible use of location-related information. Bees were trained to two sites, one in the morning and one in the afternoon. When captured in the morning at the hive heading out to the feeding site and released at the afternoon site, or captured in the afternoon heading out to the afternoon feeding site and released at the morning site, they flew back directly to the hive from either site, indicating that the landmarks characterizing each site are able to retrieve a remote memory (here, the homeward flight vector; Menzel et al. 1996). Furthermore, when bees were released halfway between the morning and afternoon sites, at a site that resembled landmark constellations characteristic of both the morning and the afternoon sites, 50% of them flew directly toward the hive, a flight direction that they did not show at any other site and that must have resulted from the retrieval of both site-specific memories.
When these data were published, the complexity of the integration process was enigmatic, and we argued that it might be explained as an automatic process of path integration on a large scale, or as a sensorimotor routine of fast sequential reference to the landmark constellations, or as an integration process at the level of two separate memories. On the basis of the results reported above on the use of SRM and GLM we can now interpret these results more specifically. Since the bees were tested at a moment when GLM should still have been depressed by the dominant SRMs established at the two feeding sites, the results also indicate a flexible use of SRMs, and an integration of such memories if more than one is activated. Under such conditions it may also be possible that rivaling SRMs decrease their control over flight behavior, so that GLM is no longer depressed by the dominant SRM. In such a case, the novel flight direction of the bees may indicate a reference to GLM, and in that case they would have localized their release site and steered toward the intended goal (the hive) along the shorter route.
The role of a flower's location for finding and choosing it again in a foraging trip could be a function of the structure of the landscape, the kinds oflandmarks, and the vegetation density. Plants flowering in a landscape with dense subtropical and temperate vegetation tend to appear in closer patches, and these patches are thus less well-characterized by their different location in the general landscape memory (GLM). Such flowers may need to provide signals that allow spotting a patch over greater distances. The achromatic signal produced by the green receptors of the compound eye allows detection over further distances than that produced by the color vision system (Giurfa & Lehrer this volume). We thus asked how achromatic and color signals of flowers are correlated with habitat structure. In a comparison of visual signals provided by Mediterranean and desert flowers in Israel, Menzel et al. (1997) found that the achromatic signal is more pronounced in the densely grouped Mediterranean flowers than in the sparsely distributed desert flowers, whereas the color signal does not differ between the species in these two habitats. It is possible that bees mainly use their spatial memory to spot sparsely growing desert plants. Desert plants may thus rely less on their own green-contrast signals for the intermediate range of detection than densely blooming plants in the Mediterranean habitat do. The color signals of both kinds of plants should depend less on habitat features, because this signal may be needed for the flying insect's proper posture when approaching the flower for fast and effective handling, irrespective of how the plant was spotted. And indeed the color signals are not different in these two habitats. The color signal, together with the shape and pattern, may also more reliably indicate the nutritional status of the flower, a feature that should also be independent of the habitat.
This interpretation is based on two arguments. (1) Plants are the evolu-tionarily adaptive units, whereas the habitat's features are the constraints. Flowers are selected to be repeatedly located, identified, and recognized within the conditions provided by the habitat. If the habitat allows easy localization (e.g., in the desert because of the low growth density), the pollinator's navigational system may need less support from further-ranging flower signals (e.g., the green signal). (2) Spatial memory is intimately connected with associative learning processes at the feeding site. It is, therefore, likely that the different sets of external stimuli to which these navigational tasks refer are elements of a rich spatial memory with "qualified" and localized components. The "qualification" relates to the localization in the GLM and the goal's specific features (e.g., visual and olfactory stimuli, flower mechanics, reward properties). The concept of a rich and unique navigational memory composed of interrelated memory items underlying the task of navigation between nest site and feeding sites supports the view that flower signal evolution should depend on all the components guiding pollinator navigation. This is a testable hypothesis for further ecophysiological studies.
Localization on a small scale: choice sequences and memory dynamics
A foraging bout is structured in time (Fig. 2.1). Because flowers mostly occur in patches, intrapatch choices follow each other quickly and are more likely to hit on the same kind of flower. Interpatch choices are more spread out in time, and are likely to expose bees to flowers of other
Was this article helpful?