to biological signals and receivers is still in its infancy, our evidence is fragmentary. We hope, however, that it will stimulate future research to add the missing pieces of the puzzle. We shall first discuss possible causes of the diversity of floral color signals and then move on to the evolution of pollinator color-vision systems.

Pollination syndromes and flower colors

One way to explain the diversity of flower colors is to use the concept of pollination syndromes, which holds that particular classes of pollinators are specifically associated with particular floral traits, including floral color (Faegri & van der Pijl 1978). There is current debate on how tight and exclusive these associations are (Waser et al. 1996; Johnson & Steiner 2000; Thomson et al. 2000; Gegear & Laverty, this volume). One argument involves the significance of red flowers in the context of hummingbird pollination. In the classical view, red flower coloration is a strategy to kill two birds with one stone: such coloration was thought to be invisible for bees and at the same time attractive for hummingbirds (Raven 1972). Therefore, flowers that are morphologically adapted to bird pollination, and on which bees transfer pollen less efficiently than birds do, should be colored red. The premises are flawed, however. Bees do visit red hummingbird flowers, and they can be trained to distinguish red from a green, foliage-like background, as well as from yellow and orange model flowers (Chittka & Waser 1997). Researchers working on hummingbirds have not been able to find a preference for red (Lunau & Maier 1995). Thus, the association between hummingbirds and red flowers is not exclusive.

A recent study by Thomson et al. (2000), however, does indicate that the association exists. In seven lineages of the genera Penstemon and Keckiella (Scrophulariaceae), flowers frequented by hummingbirds are more often orange and red than their bee-visited close relatives. Also, red coloration is strongly associated with other floral traits linked to ornitho-phily. But what it is the significance of such coloration, if it is neither attractive for hummingbirds nor invisible for bees.? In our view, there is no necessity for exclusivity: any change in floral trait may be subject to selection if it confers a change in fitness, however small. Red coloration might be an adaptation to facilitate detection by hummingbirds, or to decrease detectability by bees, or both - even by a few percent. For flowers that are adapted to hummingbirds, bumble bees may not only transfer pollen less efficiently than birds, thereby acting as pollen thieves

(Thomson et al. 2000); they may also rob nectar, further reducing plant fitness (Irwin & Brody 1999). In such circumstances selection would favor any character that diminished visitation by bees.

In many situations, hummingbirds and most bees choose nectar flowers on the basis of their net caloric rewards (Waser et al. 1996; Healy & Hurly, this volume; Waddington, this volume). These depend not only on the nectar content of the flowers, but also on the time taken to locate the flowers. Thus, we need to evaluate the search times that hummingbirds and bees take for finding red, UV-absorbing flowers, and compare these with times taken to search for flowers of other colors. Data for hummingbirds have yet to be obtained, but results for bumble bees are now available. In a flight arena, we presented Bombus terrestris workers with a random arrangement of three identical model flowers, all of which were rewarded. We measured the time taken from entering the arena to landing on the last flower, excluding flower-handling times. Search times strongly depended on color; the larger the color contrast of the flowers with their background, the more rapidly bees would detect the flowers. Red and white (UV-reflecting) model flowers, which make the poorest contrast with their backdrop, took more than twice as long to find than did blue or yellow flowers, for example (Fig. 6.1). Thus, red coloration may indeed be a strategy to reduce visitation by bumble bees to some degree. Another (non-exclusive) possibility is that hummingbird flowers use red color to form a mimicry ring, so that each species will be identifiable as a suitable food source by hummingbirds using experience gained on flowers of different species (Healy & Hurly, this volume).

In general, we expect sharper discontinuities between syndromes where classes of pollinators differ strongly in morphology (so that one type of pollinator transfers pollen substantially better than another) and in sensory system (so that, for example, a particular color is poorly detectable by one type of pollinator, but conspicuous for another). Red hummingbird flowers fit these prerequisites, but we stress that pollinator segregation achieved by red coloration is nowhere near exclusivity. We suspect that this observation extends beyond red flowers. The concept of "private channels" in sensory biology may apply to spectacular specializations such as ultrasonic hearing. However, in many cases, the ranges of sensory systems will overlap, sometimes heavily. In such cases, interactions between signals and signal receivers will not follow a simple crypsis vs. conspicuousness dichotomy. We may have to look for more subtle

Fig. 6.1. (a) Color loci of targets for bumble bees in the color hexagon. The colors appear to humans as follows: l-yellow; 2-white (UV absorbing); 3-blue; 4-turquoise; 5-red; 6-white (UV reflecting). The angular position of a color in the color hexagon informs us about the bee-subjective hue. We assume the photoreceptors adapt to the background against which the stimuli are presented; see Chittka (1996) for details. As a consequence of this adaptation process, the background lies in the center of color space. Thus, by definition, the color contrast of a model flower with its background is determined by the distance of its color locus from the center of color space. The distance between the center and each of the hexagon's corners is unity, (b) Correlation between color contrast (target vs. background) and search time (rs = - 0.83; n = 6; p <0.05). Three colored chips of 0 = 28 mm were placed in a flight arena at random positions. We measured the time elapsed from entering the arena to landing on the third chip. Note that the correlation of detectability with color contrast is good only for large color targets; for smaller ones, an increasingly strong influence of green contrast is found (see Giurfa & Lehrer, this volume).

Fig. 6.1. (a) Color loci of targets for bumble bees in the color hexagon. The colors appear to humans as follows: l-yellow; 2-white (UV absorbing); 3-blue; 4-turquoise; 5-red; 6-white (UV reflecting). The angular position of a color in the color hexagon informs us about the bee-subjective hue. We assume the photoreceptors adapt to the background against which the stimuli are presented; see Chittka (1996) for details. As a consequence of this adaptation process, the background lies in the center of color space. Thus, by definition, the color contrast of a model flower with its background is determined by the distance of its color locus from the center of color space. The distance between the center and each of the hexagon's corners is unity, (b) Correlation between color contrast (target vs. background) and search time (rs = - 0.83; n = 6; p <0.05). Three colored chips of 0 = 28 mm were placed in a flight arena at random positions. We measured the time elapsed from entering the arena to landing on the third chip. Note that the correlation of detectability with color contrast is good only for large color targets; for smaller ones, an increasingly strong influence of green contrast is found (see Giurfa & Lehrer, this volume).

differences in effectiveness of different signals for different receivers, and in their actual fitness effects.

There is also the possibility of an evolutionary "arms race". If, for example, red hummingbird flowers are so profitable that bumble bees might significantly improve their fitness by exploiting them, then bees might be selected to improve their sensory skills to detect such flowers. As we discuss below, this might have happened in Bombus occidentalis, a bee species known for extensively robbing hummingbird flowers (Irwin & Brody 1999).

How do we explain the diversity of flower colors whose major reflectance falls within the visual range of practically all pollinators, such as UV, violet, blue, pink, white (typically UV-absorbing), or yellow (with or without UV-reflectance)? Some scientists have extended the syndrome concept to such flowers as well, but if partitioning by syndromes is the major selective pressure that drove floral color diversification, why do we not see stronger segregation.? More bluntly, why are not all bumble bee flowers blue, all butterfly flowers orange, and all fly flowers white, for example? In many phylogenetic lineages, switches from one flower color to another occur without an associated morphological adaptation to a different class of pollinator (W.S. Armbruster, unpublished data). In one study on a nature reserve near Berlin, we did not find any differences among the colors of flowers visited by large and small bees, butterflies, flies, and beetles (Waser et al. 1996). In a phylogenetic analysis of the distribution of flower colors within two plant genera, Armbruster (unpublished data) found that all the variation occurred in association with bee pollination (see below). Thus, direct selection by pollinators in the sense of an innate affinity (as suggested by some adherents of pollination syndromes) surely cannot explain all the existing variation in floral color (Gegear & Laverty, this volume). In the following paragraphs, we highlight alternative explanations for why floral colors might diverge. Not all of these involve pollinators.

Flower constancy and flower similarity

An alternative view to floral syndromes is that flowers differ in color as a strategy to promote flower constancy. Such fidelity by pollinators favors an efficient and directed pollen transfer between conspecifics (Chittka et al. 1999). Conversely, pollinators straying between flowers of different species may lose pollen during interspecific flights (Feinsinger 1987) or even reduce seed set by clogging stigmas with foreign pollen (Waser 1978).

In some closely related species, hybrids may be produced that are sometimes less viable than the parental species, thereby increasing selective pressures to diverge in floral advertising (Levin & Schaal 1970).

To understand the kind of diversity that can be expected to evolve as a strategy to promote constancy, it is critical to know the range over which pollinator-subjective color difference is correlated with flower constancy. For example, if a barely distinguishable contrast between two flower colors can produce 100% constancy, then flower constancy may drive only small-scale color differences, such as between two similar, but just distinguishable, shades of blue. However, character displacement across color categories, such as from blue to yellow, would be harder to explain by pollinator constancy if this were the case. Previous work allows us to predict how color discrimination improves with color distance (Chittka et al. 1992), but flower constancy and discrimination are unlikely to increase with color difference in the same way. In measuring flower constancy as a function of floral color difference, we do not ask: "How well can bees distinguish colors.?" Instead, the appropriate question is: "How readily do bees retrieve memories for different flower types, depending on how similar they are to the one currently visited?" Discriminability sets the upper limit for constancy, but there is no a priori reason to assume that constancy is directly determined by discri-minability.

In order to measure flower constancy as a function of color distance between flower types, we tested six species of apid bees on 15 pairs of plant species or color morphs of the same species, using a paired-flower, bee-interview protocol (Thomson 1981). We did not use the traditional Bateman's Index (Bateman 1951), because this index has a number of complications: it cannot be calculated if animals are completely constant, because the denominator in the formula becomes zero. Additionally, Bateman's Index always yields maximum constancy if the frequency of inconstant transitions from one of the flower types is 0, even if pollinators are inconstant when starting from the other flower type. Therefore, we quantified constancy using a new formula which circumvents these difficulties:

where A represents the number of constant flights from X to X, B the flights from X to Y, C the flights from Y to Y, and D the flights from Y to X. Constancy calculated in this way can range from 1 (complete constancy) through 0 (random flights between species) to - 1 (complete inconstancy).

Fig. 6.2. Flower constancy in several species of bees as a function of color distance between pairs of flower types. For each pair of flower types, we recorded at least 80 choices. Flower constancy data are calculated as explained in the text.

This formula can be used only when individuals are coming to the pair of test flowers from both types of flowers.

Even though our analysis ignores differences other than color, there is a clear relationship between bee-subjective color difference and flower constancy (Fig. 6.2). Constancy does not deviate from chance at distances below 0.1 (where discrimination is already 70%; Chittka et al. 1992). At distances of about 0.2, constancy levels rise sharply in all pollinator species and above 0.4, constancy is generally above 80%. Thus, flower constancy is negligible at small color differences, even though bees can differentiate these colors well; it is at its maximum only in cases of pronounced differences.

Unfortunately, however, floral divergence due to benefits of constancy is not easily proven. Some authors have taken color diversity of sympatric flowers itself as evidence for character displacement (Menzel & Shmida 1993), but it is critical to test an observed distribution of phenotypes against a null model. Gumbert et al. (1999) examined several sets of sym-patric and simultaneously flowering plants in a nature reserve near Berlin. A color distance distribution was generated for each set of flower colors by calculating bee-subjective color distances between all floral color loci in bee-color space. To test whether the flowers differ more strongly in color than would be expected by chance, we compared the real distributions with those produced by a random generator.

When common plants were examined, there were no significant differences between random sets and actual flower color distributions. The only significant deviations were detected in rare plants, and these effects varied with habitat. In one habitat, rare flowers were more similar than expected by chance, in two they were less similar, and in two others, there was no deviation from a random distribution. Thus, flower constancy may have influenced plant community structure in some habitats, but we need more research before concluding that such influences are widespread (Chittka et al. 1999).

Finally, pollen flow between populations at different sites may prevent local adaptations to conditions at those sites (Stanton & Galen 1997). In such a situation, plants may gain fitness if gene flow between populations is depressed. Thus, if (and only if) there is a genetic correlation between a trait favoring constancy (such as flower color) and a trait involved in, say, resource acquisition under different ecological conditions, floral signal divergence may indeed be favored (Jones, this volume).

Pleiotropy, exaptation, constraint, and chance in the evolution of flower color

Biologists interested in the evolution of plant and animal signals tend to attribute signal diversity to selective pressures exerted by the signals' receivers (but see Newbigin 1898; Lutz 1924, for early attacks on this view). There are alternative explanations. One is pleiotropy, or indirect selection through genetically correlated characters (Armbruster et al. 1997). Carotenoids, responsible for yellow to orange coloration, are essential accessory pigments to chlorophyll in all plants (Scogin 1983). Many other pigment classes involved in floral coloration, or the biochemical pathways leading to the production of such pigments, may also protect against herbivores, UV radiation, and frost, or have unspecified effects on plant vigor (Onslow 1920; Levin & Brack 1995; Armbruster et al. 1997; Fineblum & Rausher 1997).

For example, Osche (1979) suggested that the yellow flavonoid coloration of pollen was already present in wind-pollinated ancestors of extant anthophilous plants and primarily served as protection against mutagenic UV radiation. He suggested that in the early stages of insect pollination, many pollinators might have formed an innate preference for yellow floral signals, and that many plants later evolved large yellow nectar guides as supernormal stimuli to cater to this preference (Osche 1979). This hypothesis remains to be tested by phylogenetic tests, however.

To examine the possibility of pleiotropic effects in floral color evolution, two plant genera with great flower-color variation, Dalechampia and Acer, were examined using phylogenetically informed analyses (Armbruster et al. 1997; Armbruster, unpublished data). In both, flowers shared pigments that were also found in leaves and stems. In Dalechampia, similar changes in flower color occurred several times independently in evolutionary history, but these changes were not associated with pollination mode. Instead, in all species with pink or purple flowers, anthocya-nins were also expressed in stems and leaves, where they possibly affect plant survival in ways not related to pollination. In Acer, the evolution of autumn leaf color actually predates changes in flower color. Again, this suggests that evolutionary changes in flower color may have occurred without any relation to pollination: rather, selective pressures operating on vegetative traits may have first favored the expression of different chemicals (see also Newbigin 1898; Onslow 1920). Then selection to enhance floral detectability may have favored expression of the same compounds in petals. In such cases, the use of particular pigments in the flowers is an exaptation with respect to pollination (Armbruster et al. 1997; Armbruster, unpublished data).

Pleiotropy is not the only constraint on flower color. If the flowers of two related species (or populations of the same species) have the same colors, this may not reflect similar selective pressures, be they on floral or vegetative traits. In fact, even if optimality arguments predict different coloration offlowers blooming at two different sites (for example because ofthe particular competing species in each habitat), they might still have the same color. One type ofconstraint is ongoing gene flow between populations, which might prevent flowers from local adaptation (Stanton & Galen 1997). Positive frequency-dependent selection by pollinators might also keep floral colors from reaching a local optimum (see Smithson, this volume). In addition, there are phylogenetic constraints on flower color in several plant taxa (Chittka 1997). In many species, a change from one floral color to another may simply require an improbable sequence of mutation events. Finally, genetic drift can act as a kind of constraint, too: evolutionary chance processes will, with some probability and depending on the size of the population, eliminate intraspe-cific variance, unless it is continuously added by new mutations or immigration (Adkison 1995).

Conversely, some plants show pronounced variation in flower color among populations (e.g., Beerling & Perrins 1993). These might reflect adaptations to local pollinator preferences, character displacement driven by different competing plants, or, through pleiotropy, adaptations to local selective pressures on vegetative traits. The possibility that simple genetic drift might account for these differences has been left largely unconsidered. To our knowledge, the only exception are the flowers of the Nigella arvensis complex, which occur not only in mainland Turkey and Greece, but also on several Aegean islands. There are strong differences in color, pattern, and floral shape among island populations; genetic drift is a likely explanation (Strid 1970). Because these island populations are small, the idea of drift is particularly palatable, but there is no a priori reason to suspect that mainland populations of plants, whose effective population sizes may be as small as those on islands, are immune to chance evolutionary processes.

Has bee color vision adapted to flower color?

The discoveries that bees see ultraviolet and that flowers reflect it were made several decades ago (Kühn 1923; Lutz 1924, and references therein). Ever since, scientists have speculated that UV receptors in bees developed in a coevolutionary process with floral coloration (e.g., Menzel & Backhaus 1991). This notion received recent impetus from computer models showing that bee color vision is indeed the optimal system for detecting and identifying flowers (Chittka 1996). However, to prove that flower signals truly drove the evolution of bee color vision, it must be shown that the ancestors ofbees possessed different sets of color receptors prior to the advent of the flowering plants. One must evaluate arthropods whose evolutionary lineages diverged from those of bees before there were flowers. If the color vision of such animals is indistinguishable from that ofbees, this implies that it was present in an ancestor ofbees that predated the evolution of flower color - and this is exactly what was found (Chittka 1997). The Amax values (wavelength values of maximum sensitivity) of the Crustacea and Insecta fall into three distinct clusters in the UV (around 350 nm), blue (—440 nm), and green (—520 nm). Red receptors show up irregularly both in the Crustacea and Insecta; they have evolved several times independently.

Thus, we can infer that insects were well pre-adapted for flower-color coding more than 500 million years ago, about 400 million years before the extensive radiation of the angiosperm plants that started in the middle Cretaceous (100 million years ago). Recent data on the molecular structure of photopigments support the interpretation that the basic types of arthropod visual pigments must be placed at the very roots of arthropod evolution (Chittka & Briscoe 2001). Thus, bee color vision is an exaptation with respect to flower color.

Measured peak sensitivities of receptors vary up to 30 nm across insect species, however (Chittka & Briscoe 2001). Some of this must be measurement error, but we can not exclude the possibility of actual fine-tuning of pigments to particular visuo-ecological tasks. To examine such fine tuning, it is necessary to look at closely related species with known phy-logeny and distinct ecological conditions. We mapped the positions of maximum sensitivity of the color receptors of 11 species of bumble bees from five subgenera onto their phylogeny (Fig. 6.3). These species span habitats from European alpine (e.g., Bombus montícola) and North American temperate (e.g., B. ímpatíens) to subtropical and tropical South America (B. mono), but the Amax positions are very similar across species. Peitsch et al. (1992) claimed that bee species flying in UV-rich environments might have short-wave-shifted UV receptors, while tropical forest-dwelling bees might have long-wave-shifted UV receptors. Our analysis does not support this claim: the alpine B. montícola (whose altitude range is 900-2700 m; Hagen 1994) does not differ from B. terrestrís and B. lapídar-íus (both lowland species that are not found above 1400 m; Hagen 1994). Although the tropical B. morío has slightly long-wave-shifted UV receptors compared to the above two, it does not differ from several temperate species.

Several types of molecular constraints, and possibly pleiotropies, that might affect the evolution of color vision have been reviewed in detail elsewhere (Chittka & Briscoe 2001). One source of inertia that is often overlooked in investigations of sensory ecology is simply chance. Physiologists often assume that any superior genotype will inevitably be able to spread through a population. Because this assumption is so common, we shall elaborate in some depth why this may not necessarily happen. Imagine that a bumble bee colony produces 100 new queens, one of which carries a new mutation that has a beneficial effect on foraging

Apis mellifera

Bombus hortorum (3) fervidus (2) distinguendis morio (4) y impatiens (2) jonellus (4) monticola (4) hypnorum (4) lapidarius (4) terrestris (4) affinis (2)

SUBGENUS Megabombus Megabombus (1) Megabombus Fervidobombus Pyrobombus Pyrobombus « Pyrobombus « Pyrobombus Melanobombus • Bombus s. str. • Bombus s. str.

Was this article helpful?

0 0

Post a comment