NEGATIVE FREQUENCY-DEPENDENT SELECTION BY POLLINATORS ON ARTIFICIAL FLOWERS WITHOUT REWARDS

Transcription

1 Evolution, 51(3) pp NEGATIVE FREQUENCY-DEPENDENT SELECTION BY POLLINATORS ON ARTIFICIAL FLOWERS WITHOUT REWARDS ANN SMITHSON) AND MARK R. MACNAIR Department of Biological Sciences, University of Eeter, Hatherly Laboratories, Prince of Wales Road, Eeter EX4 4PS, U.K. I Abstract.-Many species of nonmodel deceptively pollinated orchids are polymorphic for corolla color. These species are pollinatedby naive insects searchingfor nectar, and are not mimics.it has been suggestedthat the foragingbehavior of insect pollinators during the avoidance learning process results in these stable corolla color polymorphisms; for this to occur pollinators must induce negativefrequency-dependent selectionon corolla color.thereforethe hypothesis that pollinator behavior results in a preference for rare color morphs of deceptive species was tested eperimentally. Bumblebees (Bombus terrestris) foraged in the laboratory on arrays of artificial flowers with different corolla color morphs. Morphs were varied in frequency, and bumblebee preferences were recorded on arrays where morphs did and did not contain sucrose solution rewards. Bumblebees preferred the most common color morph when flowers contained sucrose solution rewards, but overvisited rare morphs when sampling flowers that contained no rewards. Bumblebees also tended to move between unlike color morphs when these were unrewarding, suggesting that a probabilistic sampling strategy was adopted. Thus eperiments demonstrated that pollinator behavior could result in a selective advantage for rare color morphs of plant species that are pollinated by deception without mimicry, which would induce negative frequency-dependent selection on corolla color. The observed pollinator behavior could allow stable corolla color polymorphisms to be maintained by selection in nonmodel deceptively pollinated species. Key words.-bombus terrestris, deceptive pollination, disassortative mating, orchids, sampling behavior. The behavior of pollinators foraging on different floral phenotypes will affect the relative fitness of the plants they are visiting. We have previously shown (Smithson and Macnair 1996) that bumblebees, typical pollinating insects, when offered artificial flowers that differed only in color, forage in a frequency-dependent way, strongly preferring the most common color morpho Since the relative fitness of floral morphs will be some positive function of the number of visits received by pollinators through both seed set (Waser and Price 1981) and pollen removal (Stanton et al. 1989), pollinator preference for the most common morph would result in differential reproductive success potentially through both male and female function on corolla color morphs and would cause positive frequency-dependent selection (FDS) through common morph advantage. Common corolla morph advantage has been found in field populations of flowers with different corolla shapes (reduced outcrossed seed set in rare corolla morphs of Phlo pollinated by Lepidoptera, Levin 1972). Positive FDS will eventually lead to monomorphism for that trait (Thomson 1984; Endler 1988). Corolla colors of insect-pollinated plants often show intraspecific variation (Kay 1978), and since this variation is heritable and pollinating insects do discriminate between the morphs (e.g., Kay 1978; Stanton 1987; Stanton et al. 1989), how may the observed corolla color polymorphisms be retained in plant populations if pollinators forage in a way that selects against the rarest forms? The simplest eplanations are that the observed corolla color variation could be due to transient polymorphism, genetic drift, balance between mutation and selection, or to heterozygote advantage. Alternatively, there may be some situations in which pollinators eert negative FDS on these traits, which will result in stable polymorphism for corolla color (Clarke and O'Donald 1964). To eert negative FDS, pollinators must prefer the rarest, as opposed to the most common, corolla color morpho 1997 The Society for the Study of Evolution. All rights reserved. Received January 12, Accepted January 6, A wide range of predatory animals have been shown to select different color morphs of prey species in a frequencydependent way, also showing preferences for common morphs (reviewed by Allen 1988). When prey are consumed, common prey. morphs are at a selective disadvantage and predators will thus induce negative FDS on their prey populations, to be contrasted with the positive FDS that pollinators would induce on floral morphs. There are, however, two situations in which predators show a preference for rare prey morphs, potentially inducing positive FDS on prey populations. At high densities of prey, predators show a preference for rare prey color morphs (reviewed Allen 1988). There is no evidence, however, that pollinators show a switch from common morph preference to rare morph preference as flower density increases (Smithson and Macnair 1997). It has been shown that predatory birds eat a higher proportion of rare prey color morphs when these are distasteful (Benson 1972; Greenwood et al. 1989; Mallet and Barton 1989). Because oversampling of rare morphs of distasteful prey would induce positive FDS, it is believed that this eplains why prey species distasteful to their predators are usually monomorphic, and why different distasteful species converge in color pattern to form MUllerian mimicry complees (Greenwood et al. 1981). Does an analogous switch from common to rare morph preference ever eist for pollinators? It has been suggested that the high levels of intraspecific floral diversity observed in some species of orchids may be related to their deceptive system of pollination (Heinrich 1975; Nilsson 1980; Dafni 1984; Ackerman and Galarza Perez 1991). Species such as Dactylorhiza sambucina (Nilsson 1980), Orchis caspia (Dafni 1983), Orchis morio (Dafni 1987), Dendrobium speciosum (Slater and Calder 1988), Thelymitra epipactoides (Cropper and Calder 1990), and Tolumnia variegata (Ackerman and Galarza-Perez 1991) offer no rewards to their bee pollinators, nor mimic any specific

2 716 A. SMITHSON AND M. R. MACNAIR reward-producing species, but for successful pollination they rely on naive pollinators making eploratory visits in the search for potential nectar sources (Nilsson 1980). Such a method of nonmodel (sensu Dafni 1986) deceptive pollination is believed to be widespread (Ackerman 1986). Considerable variation in corolla color is recorded: D. sambucina has two color morphs, red and yellow (Nilsson 1980), T. epipactoides varies in color between brown, blue, and green (Cropper and Calder 1990). The different color morphs of both species are found within the same populations. The standard hypothesis for the maintenance of such corolla color polymorphisms in populations of deceptive species is that floral variation results in each pollinator making more visits to the deceptive species before learning to avoid it, as floral variation inhibits learning of a "search image" of the deceptive species by the pollinator (Heinrich 1975; Nilsson 1980; Little 1983; Dafni 1984; Cropper and Calder 1990; Ackerman and Galarza-Perez 1991). A search image is defined as a perceptual change in the ability of a predator to detect familiar cryptic prey (Tinbergen 1960; Dawkins 1971). Since flowers are not cryptic to their pollinators, search images cannot be involved in learned avoidance by pollinators (see Lawrence and Allen 1983). Further, eperiments show that bumblebees do not take more visits to learn to avoid unrewarding flowers as the number of unrewarding flower colors increases (Dukas and Real 1993a). Differential morph fitnesses are also not predicted by this hypothesis, and for polymorphism to be maintained by selection, pollinator behavior must cause negative FDS. Therefore we have considered in more depth the mechanism behind pollinator responses to encountering flowers of deceptive species, which might lead to negative FDS on corolla color. Two alternative hypotheses are suggested: 1. Negative FDS results from pollinators taking a certain number of flower visits to learn that each color morph is unrewarding. In predator-prey systems, the mechanism believed to cause oversampling of rare distasteful prey morphs is that each ineperienced predator eats the same number of individuals of each distasteful morph before learning to associate distastefulness with morph color, thus rare morphs are proportionately overeaten (Greenwood et al. 1981; Greenwood 1984). Ifpollinators sample similar numbers of flowers of each deceptive morph before learning to avoid them, then rare morphs will be visited proportionately most often. 2. Rare morphs are at an advantage in deceptive system through the way pollinators learn to find rewarding flowers. When visiting rewarding flowers of two morphs, bees visit flowers of similar colors sequentially (Stanton 1987; Smithson and Macnair 1996). Conversely, after encountering an unrewarding morph, a pollinator may subsequently move to a morph of a different phenotype ("deceit encourages inconstancy." Plowright and Laverty 1984). If this pattern of visitation is followed, then dissimilar phenotypes will be visited sequentially as pollinators search deceptive morphs for nectar or pollen. Negative FDS will occur since rare morphs will be overvisited while the identity of rewarding morphs are being learned. We tested these hypotheses eperimentally by offering an array of artificial flowers to captive pollinators and recording their preferences. We used two colors of artificial flowers (blue and purple) to represent a flower species polymorphic for corolla color that either offered rewards to pollinators or was rewardless, and thus deceptive. In addition, we used another artificial flower color (yellow) to represent an alternative flower species for pollinators that was always rewarding and monomorphic. We tested the central prediction of the FDS hypothesis-that when the polymorphic species is rewarding, pollinators show a preference for its most common morph, but when the polymorphic species is deceptive, pollinators overvisit its rarest morph (prediction 1). Additionally, it would be epected that the proportion of the polymorphic species visited will decline when deceptive but not when rewarding (prediction 2). If negative frequency dependence is discovered, then the two mechanisms discussed above may be discriminated eperimentally. Hypothesis 2 predicts that when visiting the polymorphic species, similar morphs will be visited sequentially when rewarding, but dissimilar morphs will be visited sequentially when deceptive, whereas hypothesis 1 does not predict differences in flower visitation order (prediction 3). Hypothesis 1 predicts that the same number of visits would be made to the deceptive species before learned avoidance occurs, irrespective of the relative abundance of the deceptive species relative to the rewarding species, whereas hypothesis 2 predicts that fewer visits will be made to the deceptive species as its abundance relative to rewarding species decreases (prediction 4). MATERIALS AND METHODS In these eperiments, an array of artificial flowers with different color morphs was presented to captive colonies of bumblebees (Bombus terrestris), using a methodology identical to that given in previous eperiments (Smithson and Macnair 1996). The colonies were obtained from a commercial supplier (Bunting Biological Control, Colchester, Esse, UK), were fed sucrose solution into the colony bo prior to eperiments, and thus had no prior foraging eperience before eperiments commenced. Bees had access to pollen in the colony bo at all times, and were given sucrose solution overnight during eperiments to maintain positive energy budgets. All eperiments were carried out inside a netted cage (0.9 m X 0.9 m X 0.9 m), situated in a laboratory, to which access was provided through a gated tube from the colony bo. During eperiments, worker bumblebees foraged singly on an array of artificial flowers presented at the base of the- cage. The array consisted of a transparent perspe board (775 mm X 775 mm X 6 mm) into which a grid of 30 X 30 wells 25 mm apart had been drilled, with well-size diameter of 2.5 mm and depth of 3 mm. Colored discs of card, all 16 mm in diameter, placed under selected wells, acted as "flowers." The wells of the flowers could then be filled with a quantity of 40% (w/v) sucrose solution, dispensed with a Gilson micropipette, to act as a reward on which the bees could feed. Bees were trained to forage from the array using 33 of each of the three colors used in the eperiments to be described, with the flowers being placed randomly on the board and filled with approimately 20 tj..l sucrose solution mied with honey. Once the bees were feeding from all flower colors, honey was not included subsequently in the sucrose solution so that no olfactory cues for

3 NEGATIVE FDS BY BUMBLEBEES 717 reward location were provided. Active foragers were individually marked on the thora with spot combinations of solvent-free correction fluid. Each marked individual was used in one treatment only (described below), and for each bee a treatment consisted offive sequential bouts, during each bout the bee was allowed to forage on the array and between bouts the bee returned to the colony. After completing five bouts, the bee was then removed from the colony to prevent reuse. All eperiments were carried out on an array of 150 flowers (approimately 248 m- 2 ), at various frequencies of flower colors, and replicates of workers were tested at each frequency for each treatment. Each worker bee eperienced the same flower color frequency over the five bouts, but the positions of flowers were randomized between each bout. The board was cleaned and flowers refilled between each bout. To test predictions 1-3 given above, an array of three distinct color morphs was used. One morph, yellow, did not vary in frequency, being always present at a frequency of one-third of the total number of flowers used (50 flowers of 150), and was always rewarding. The remaining two morphs only, blue and purple, were varied in frequency, and were rewardless in one series of treatments, but controls where these morphs had rewards were also performed for comparative purposes. The treatments were carried out as follows: (1) control 1, all morphs rewarding; yellow, blue, and purple morphs had 2 f.ll sucrose in all flowers; (2) control 2, all morphs rewarding; yellow, blue, and purple morphs had 5 f.ll sucrose in all flowers; and (3) deceptives, one rewarding morph, two unrewarding morphs; yellow morphs had 5 f.ll sucrose, blue and purple morphs had 5 f.ll distilled water. It was necessary to place 5 f.ll sucrose solution in yellow flowers in treatment 3 to ensure that bumblebees did not ehaust all the sucrose available on the array and then start to revisit emptied yellow flowers. In previous eperiments, we used 2 f.ll sucrose solution in flowers as this gives long bout lengths of approimately 40 flowers, while flower revisitation is minimal (10%; Smithson and Macnair 1996, 1997). For comparative purposes, it was thus necessary to include additionally a control array with 5 f.ll sucrose solution in all flowers; although with large rewards, bout lengths are short and bumblebee preferences can be reduced (Smithson and Macnair 1996). The blue and purple morphs (a total of 100 flowers of 150 on the array were either blue or purple) were presented to bumblebee workers in nine frequencies ranging from 10% to 90% purple flowers, and replicates of five to nine workers were tested at each frequency. The predictions made above were tested by recording the numbers of blue and purple morphs visited (a visit was counted only when the bee probed the well of the flower) by the bees and comparing with the frequency of blue and purple morphs available (prediction 1), and the combined numbers of blue and purple morphs visited as the total number of flowers visited increased (prediction 2). Prediction 3 was tested by recording the order in which flowers of different colors were visited. Prediction 4 was tested in a separate eperiment, where only two color morph were used, yellow and blue. Yellow flowers were always rewarding, and contained 5 f.ll sucrose solution, whereas blue flowers were unrewarding and contained 5 f.ll distilled water. One hundred fifty flowers in total were used, but Yellow Ultraviolet 2 3 FIG. 1. Trichromatic color diagram to show how the flower and background colors used in eperiments would appear to the eyes of bumblebees. Each ais represents the relative quantum flu to each of the respective photoreceptors (ultraviolet, blue, yellow) from each of the colors analyzed: (I) green background; (2) yellow; (3) purple; and (4) blue. 4 Flower Colors Blue morph frequencies were varied from 50% to 90% yellow flowers. Replicates of three to five workers were tested at each frequency, and the number of unrewarding flowers of the total visited was recorded. To carry out unbiased eperiments using three color morphs, it must be ensured that the morph that is always rewarding is not similar in color to either of the other morphs, or the eperiments could model a system where a deceptive morph mimics a specific model. Such pollination systems are known (Nilsson 1983; Ackerman 1986; Dafni 1986) and different selection pressures would act on floral traits under such a system (Schemske and Agren 1995). To ensure that the flower colors used would apply to a nonmodel deceptive system, the colors of the flowers used were quantified by obtaining the reflectance spectrum for each color and for the green background. Reflectance spectra were obtained using a Unicam SP1800 spectrophotometer fitted with a SP890 reflectance unit (Kearns and Inouye 1993, pp ). The spectrophotometer was zeroed using magnesium oide, and reflectances were recorded over 330 to 700 nm every 10 nm. To quantify how the colors would appear to the eyes of bumblebees, which do not see within the same wavelength range as do human eyes (Kevan 1978), resultant reflectance spectra were multiplied by spectral distributions of daylight (standard D 65 daylight function, Henderson 1977) and by the spectral sensitivities of each of the three bumblebee photoreceptor types, as described by Lunau (1992). Each flower color was thus quantified as the relative quantum flues to each of the three photoreceptor types (Lunau 1992). The colors of the flowers used in eperimentsand the background color as would be seen by the eyes of bumblebees are plotted in a trichromatic color diagram (Fig. 1), with the aes representing the relative quantum flu to each of the three bumblebee photoreceptors. Results show that the two

4 718 A. SMITHSON AND M. R. MACNAIR deceptive morphs would both appear distinct from the rewarding yellow morph to a bumblebee. PREDICTION 1 Data Analysis We tested here the central prediction of the FDS hypothesis, that bumblebees show a preference for the most common of blue and purple flowers when they are rewarding, but overvisit the rarest when they are not. Counts of the numbers of visits to blue and purple flowers were totaled over the five bouts analyzed for each bumblebee. Because bout length depended on the reward provided, data from each treatment was standardizedby analysing results for the first 100 flower visits for each bee. Few visits were recordedto unrewardingmorphs after worker bees had visited 100 flowers (see below). The proportion of visits to purple flowers was calculatedand compared to the proportion of purple and blue flowers available for each treatment, as in Smithson and Macnair (1996), by fitting a sigmoidal switching curve (Greenwood and Elton 1979) using the equation: (VA1)b a ,,,, Oloi:::...~---L-':':"-----'--~-'---",--~---J o Proportion of Purple Flowers Available b. 1, n i",,, )( 0.4,,,,,,,, ;:<,, 0.2 II',, Proportion of Purple Flowers Available Results Raw data found for each treatment, together with the curves fitted by linear least-squares regression, are shown in Figure 2. Fitted values of parameters and standard errors are FIG. 2. Results of bumblebees selecting purple and blue flowers when purple and blue flowers vary in frequency on an array ofthree flower colors. The figure also shows the lines fitted to equation (1) as described in the tet (solid) and the line of random selection (dotted): (a) treatment 1; (b) treatment 2; and (c) treatment 3. C. 1 't:l, QJ...,,, 'in,, :>,', en 0.8 ' loo,, X ~., 0,,' ~ 0.6, X X QJ,' ~ 'is. X loo., :l, Il.. le 0.4 ~'..., X X 0 le,'» C, M.s X,' loo 0 X l:l. 0 loo,, II c, 0, Proportion of Purple Flowers Available

5 NEGATIVEFDS BY BUMBLEBEES 719 TABLE 1. Summary of results of fitting data from the three eperiments to the curve given in equation (1) using least-squares regression. Fitted parameters for both frequency-dependent and frequency-independent components of choice are given, and parameter standard errors were calculated from 1000 bootstrap estimates of regression parameters (Efron and Tibshirani 1986). Significant differences of parameters from the null hypothesis of no selection, calculated using the r-test methods of Greenwood and Elton (1979), are indicated (* P < 0.05, ** P < 0.01, *** P < 0.001). b V Sample size (frequency-dependent (frequency-independent Treatment (number of bees) selection) :t I SE selection) :t I SE 1. Control :!: *** :!: *** 2. Control :!: :!: * 3. Deceptives :!: ** :!: ** given in Table I, along with the significance of differences from a null hypothesis of no selection. It can be seen that weak but significant preference for the most common color was recorded for treatment I, when blue and purple flowers were rewarding. When rewards were greater in all flowers (treatment 2), a bias toward the most common morph was recorded, although preferences did not differ statistically from random choice. This was probably due to short bout length, and reduced selectivity when rewards were high. However, when blue and purple flowers were deceptive, significant preference for the rare color was observed (Table 1). Comparing fitted values of b over all three treatments, significant heterogeneity was found (F 2,70 = , P < 0.001). This shows that a change from providing rewards to being deceptive results in a concomitant change from preference for common colors to overvisitation of rare ones. "0"0 =u «I... u.~.:> call) u o ~ ~ =~ I... ol:t i II... u 0- l:l..l:l.. 0'" := ~~ ] 0! Eperiment 1 Eperiment 2 o Eperiment 3 I I I I I! I!!! T.1-! I :2: 2: II ll: I I Number of Flowers Visited FIG. 3. Proportion of blue and purple flowers visited plotted against the total number of flowers visited from 1 to 150 flowers in groups of 25 flowers. Each point represents the mean proportion :!: 1 standard error of the previous 25 flowers visited. V is significantly less than 1 in all three treatments, indicating bias to blue flowers irrespective of frequency dependence. When fitted values of V are compared across the three treatments, significant heterogeneity was found (F 2,66 = , P < 0.001). It can be seen (Table 1) that fitted values of V are similar for controls (comparing Vbetween treatments 1 and 2, z = 0.396, P > 0.05). There was, however, a tendency for an increased bias to blue in treatment 3 (comparing a - pooled estimate of for the controls with treatment 3, Z = 1.877, P = 0.062), suggesting stronger directional selection when flowers are deceptive. PREDICTION 2 To test the hypothesis that the proportion ofblue and purple flowers visited declines when they contain no rewards but remains the same when rewarding, the datasets for each bee were divided into sequential sets of 25 flowers visited to a maimum of 150 flowers, ignoring breaks between bouts. The proportion of visits to blue and purple flowers together of the total number of visits made by each individual bee (i.e., to yellow, blue, and purple flowers) was calculated. The mean proportion of blue and purple flowers visited was found for each set of 25 flowers, and the results, which show the effect of increasing eperience of individuals on the number of blue and purple morphs visited, are shown in Figure 3. The proportion of purple and blue flowers visited decreased significantly with eperience for treatment 3 (ANOVA on arcsine-transformed data, F = , P < 0.001), but this proportion did not change during either treatments 1 or 2 (ANOVA on arcsine-transformed data: treatment 1, F = 0.801, P > 0.05; treatment 2, F = 0.381, P > 0.05). After 100 flowers have been visited, the numbers of unrewarding flowers probed in treatment 3 had declined almost to zero (Fig. 2). Slightly fewer yellow flowers were visited overall during treatment 1 than during treatment 2 (Fig. 2; ANOVA on arcsine-transformed data: F 1,5 = , P < 0.001). PREDICTION 3 Data Analysis Differences in the order in which blue and purple flowers were visited was predicted by one of the behavioral mechanisms that could cause rare morph overvisitation by bumblebees, depending on whether the flowers were deceptive or rewarding. To test for nonrandomness of visitation order, for each sequence of flower visits over the first 100 flowers visited for each individual worker bee, the numbers of moves

6 720 A. SMITHSON AND M. R. MACNAIR TABLE 2. Results of testing bumblebee visitation sequences for deviation from random visit order. Transition matrices of counts of movements made between unrewarding flowers with respect to color were found for each treatment, position, and frequency, and assortivity indices were calculated as described in the tet. Sequences were divided into the first and second sets of 50 flowers visited. The table shows overall significances of difference from random visit orders for each eperiment and position (z) and mean assortivity indices for the combined data. Significances are indicated (* P < 0.05, ** P < 0.01, *** P < 0.001). Position 1 Position 2 Position 3 Mean assortivity Mean assortivity Mean assortivity Treatment inde inde inde flowers flowers flowers 2.647*** D flowers from blue to blue, blue to purple, purple to blue and purple to purple flowers was calculated from each flower to the net, second-net, and third-net flower in the sequence (henceforth referred to as positions 1,2, and 3). All yellow flowers visited in the sequence were ignored, such that sequential visits to blue and purple morphs were counted whether or not they were separated by visits to yellow morphs. This method of analysis would be equivalent to assuming that in nature there is no interference in the transport of pollen between deceptive morphs even if visits are interspersed with visits to rewarding species. In hummingbird pollinated flowers, it has been shown that visiting intervening flowers of a different species does reduce pollen transfer between individuals, but due to their precise methods of pollen placement by pollinia it has been speculated that such interference may not occur in the Orchidaceae (Murcia and Feinsinger 1996). The data was divided into the first and second sets of 50 flowers visited for each bee. Counts of the numbers of moves in each category were made for each bee to give a 2 X 2 transition matri, and combined transition matrices were found by pooling datasets for individual bees, giving 27 transition matrices for each of the two datasets in each treatment. The transition matrices could then be tested for differences from random visit orders using chi-squared tests. To see whether any deviations from epected counts in each transition matri represented preference for visiting similar or dissimilar colors sequentially, an inde of assortivity could be calculated for each transition matri: I = ~(B_B_X_P_P.:...)_---.:('-BP_X_P_B...:.,) (BB + BP) X (PB + PP)' where BB = observed number of blue to blue movements, BP = observed number of blue to purple movements, PB = observed number of purple to blue movements, and PP = observed number of purple to purple movements. The inde etends from +1 to -1 and will be positive when more movements are made between the same color, and negative when between different colors. To give an overall test for the significance of difference from random visit order within each dataset, the individual chi-squared tests were combined over frequencies using the methods of Everitt (1977). The square root of each chi-square test was calculated and allotted the same sign as the equivalent assortivity inde. These values are approimately normally distributed with mean zero and unit standard deviation, so can be tested against z as follows: (2) z = (± V i).g-1/2 1=1 where Vi is the square root ofa chi-square test with its allotted sign, and g is the number of independent chi-squares in each test. Significances were found using tables of the standard normal distribution. Results Table 2 shows assortivity indices calculated for each treatment and position for the first 50 flowers visited, and for the second 50 flowers for treatment 3. Results for the second 50 flowers for treatments 1 and 2 are similar to those for the first 50 flowers. The results of combined chi-square tests and the average assortivity indicesfor each positionare also given in Table 2. It can be seen that treatments 1 and 2 showed no deviation from random visitation order, although treatment 2 showed a weak, nonsignificant tendency to assortivity (Table 2). Treatment 3 showed a significant bias to disassortative visit order for position lover the first 50 flowers, which weakened over positions 2 and 3. Over the second 50 flowers in treatment 3, disassortivity was again found, but none of the overall results differed significantly from random epectations. Further, in prediction 3, it would also have been hypothesized that when yellow flowers, which always contained rewards, were visited, bumblebees would have shown a tendency to visit these sequentially irrespective of the reward status of blue and purple flowers. This was tested by combining blue and purple flowers into one class, and testing the visit orders of yellow flowers versus blue and purple flowers combined as described above for the first 100 flowers visited. Results showed that visit order to yellow flowers was indeed significantly assortative in both treatments (mean assortivity indices for position 1 were: treatment 1, I = , z = 34.41, P < 0.001; treatment 2, I = , z = 19.75, P < 0.001; treatment 3, I = , z = 22.12, P < 0.001). However, an analysis of covariance on assortivity indices using frequency of blue to purple flowers as the covariate showed that there was significant heterogeneity in assortivity indices with treatment (F = 16.27, P < 0.001), suggesting that the assortivity inde was less positive in treatments 1 and 2, where blue and purple flowers were rewarding compared to treatment 3, where they were deceptive. (3)

7 NEGATIVE FDS BY BUMBLEBEES 721 PREDICTION 4 One of the behavioral mechanisms that was proposed to eplain rare morph overvisitation of deceptive species predicted a negative correlation between the number of visits made to deceptive flowers and their abundance relative to rewarding ones. To test this, the number of blue unrewarding flowers visited in total over the first 100 flowers was counted for each individual bumblebee, and correlated with the relative abundance of blue unrewarding to yellow rewarding flowers. A significant negative correlation was found between the two variables (r = -0.44, P = 0.05), indicating that more visits were taken to learn a preference for the rewarding flowers as the proportion of unrewarding flowers increased. DISCUSSION In these eperiments, it is shown that bumblebees switch from a common morph preference when flowers were rewarding, to a rare morph preference when flowers were deceptive. The results were not due to any specific mimicry between flowers-all flowers were discriminable by the eyes of a bumblebee. The effects of the observed bumblebee behavior on plant populations would be that the most common corolla color morphs will be overvisited when a plant species provides rewards for its pollinators, but rare morphs will be overvisited when a plant species is deceptive, assuming no specific mimicry of another rewarding species occurs. If morph fitness is positively correlated with pollinator visitation, then oversampling of rare morphs by pollinators will result in negative FDS being eerted on deceptive species of plants that rely on pollinator sampling to effect pollination. This will result in the potential to maintain a stable polymorphism for corolla color (Clarke and O'Donald 1964). Although bumblebees also showed directional selectionin favor of blue flowers, purple flowers were still overvisited above a frequency of 30% purple flowers, and stable polymorphism can still be maintained under such directional selection although the equilibrium morph frequencies will be biased toward the more favored morph (Clarke and O'Donald 1964). The equilibrium morph frequency will also be influenced by the genetic control of the corolla color traits (Clarke and O'Donald 1964). Two hypotheses were proposed regarding the mechanisms that could cause bumblebees to prefer rare deceptive morphs, and these may be restated in terms of the behavioral rules that bumblebees use when learning to avoid unrewarding flowers. Hypothesis 1 suggests "visit morphs in proportion to their encounter rate. Once a certain minimum number of flowers of a color morph have been sampled without finding a reward then learn to avoid this morph, but continue to sample remaining morphs in a similar way, until only the rewarding morph is visited." Hypothesis 2 proposes "if the previous morph visited was unrewarding, increase the probability of sampling a different morph on the net visit, and continue the process until a rewarding morph is encountered, after which increase the probability of sampling a similar morph." Hypothesis 2 is supported by the analysis of visitation orders. There is a tendency for pollinators to visit dissimilar morphs sequentially when foraging on deceptive flowers in these eperiments, at least over the first 50 flowers visited. These results also suggest that as eperience with the rewarding flowers increased, these patterns of disassortative movement between deceptive morphs became less directed. The observed nonrandom visit orders would result in disassortative mating between corolla color phenotypes of nonmodel deceptively pollinated species. Theoretical models of disassortative mating between genotypes result in a stable equilibrium for alleles at that locus (Crow and Kimura 1970). This disassortative behavior on unrewarding flowers contrasts to the pattern of behavior on rewarding flowers, both in this paper and elsewhere (Smithson and Macnair 1996). In previous eperiments (Smithson and Macnair 1996), we found large and highly significant biases to visiting flowers of similar colors sequentially when rewards were present and where considerably greater positive FDS was found than observed in these eperiments (mean assortivity inde for position 1 for an array of blue and yellow flowers containing rewards). In this study, we showed that movements from rewarding yellow flowers were assortative in all eperiments, and there was a greater tendency to move assortatively when blue and purple flowers were rewarding. However, there was a comparatively weak tendency to move assortatively between blue and purple flowers when these were rewarding. Overall, these results would be in agreement with a probabilistic bumblebee behavioral sampling rule, which is sensitive to encountering rewardless flowers.. If hypothesis 1 had been correct and bees sampled unrewarding flowers according to their encounter rate up to some threshold number, at which point that flower color was then avoided, then the total number of unrewarding flowers sampled until unrewarding flowers were no longer visited would be independent of their encounter rate. The results of testing this prediction showed, however, that the number of unrewarding flowers sampled decreased as their abundance relative to rewarding flowers decreased. Such a relationship would be found if more frequent encounters with rewarding morphs increased the likelihood that they would be visited in the future, as was predicted by the probabilistic sampling rule. In this eperiment, it should be noted that blue unrewarding flowers and yellow rewarding flowers differ both in color and reward. Although the eperiment indicated that the bees' preference for yellow versus blue flowers is proportional to their frequency, the results and consequences of this must not be confused with the first eperiment described above, which tested for frequency-dependent preferences amongst blue and purple flowers: the latter flowers differed in color only and not in reward. Overall, our results are thus consistent with the second probabilistic behavioral rule-bumblebees overvisited rare morphs when deceptive probably because of short-term avoidance of morphs that have been found to be unrewarding and longer-term learning of the rewarding morph, and this visitation pattern may be interpreted as active sampling behavior on the part of the bumblebees. Further evidence that supports this hypothesis as found by Dukas and Real (l993a), who showed that increasing the numbers of unrewarding color morphs in arrays of Abelia jloribunda did not increase the rate taken for bumblebees to learn the rewarding morph, but increasing the numbers of rewarding color morphs did in-

8 722 A. SMITHSON AND M. R. MACNAIR crease the rate taken to learn the rewarding flowers. Dukas and Real suggested that their results might be eplained by bumblebees learning only flowers that contain a reward, a result consistent with our hypothesis. The only evidence con.trary to the hypothesis that pollinators move disassortatively between morphs of deceptively pollinated species was found by Nilsson (1980), where reanalysis of his data (Table 2, Nilsson 1980) shows that of 33 sequential visits to the two morphs of the deceptive D. sambucina by 13 pollinators, 25 were between like morphs and only eight between unlike morphs. However, morph frequencies in the areas being visited by the pollinators were not given. In these eperiments, we found an overall bias to blue flowers irrespective of frequency dependence, but that this frequency-independent preference for blue flowers increased significantly when flowers were deceptive. It was also found that the overall number of yellow flowers visited decreased when rewards in flowers were decreased. Similar changes in biases toward flower colors due to overall reward levels have been found in previous eperiments (Smithson and Macnair 1996, 1997). Quantification of the colors of the flowers that would be seen by the eyes of the bumblebees suggested that the preference for blue flowers could be due to their greater conspicuousness relative to the background, and that changes in overall reward levels caused a change in pollinator flight distance, flight path, and flight speed which resulted in different probabilities of recognizing different flower colors that differ in conspicuousness (Smithson and Macnair 1997). In conclusion, these laboratory studies show that the initial sampling behavior of pollinators visiting deceptive species that do not mimic any other rewarding species could induce negative FDS and disassortative mating on corolla color phenotypes within a plant population. This would result in maintenance of corolla color polymorphisms by selection and high phenotypic diversity for corolla color traits. In addition, if pollinators choose other floral traits like scent or flower size in a similar frequency-dependent way (Ackerman and Galarza-Perez 1991; Cresswell and Galen 1991; Moya and Ackerman 1993), high diversity of other floral traits would also be epected. When bumblebees visit unrewarding flowers, they are also induced to etend the distance that they fly before visiting another flower (Dukas and Real 1993b), and bumblebees also travel longer distances between flowers when visiting deceptive orchids (Dafni 1987). Pollinator flight distances have substantial effects on the levels of inbreeding and population subdivision within a plant population (Levin 1978), and this behavior would also result in reduced inbreeding in populations of deceptive species. The predicted patterns of floral diversity and breeding structure may be contrasted both with (1) rewarding species, where positive FDS, assortative mating, and nearest-neighbour visitation will result in monomorphism for floral traits, increased homozygosity and substructuring of plant populations (Levin 1972, 1978; Kay 1978; Levin and Watkins 1984; Stanton 1987; Smithson and Macnair 1997); and (2) intra- or interspecific mimicry, where either another species or one se within the same species acts as a reward-providing model for the deceptive species/se (Little 1983), pollinator discrimination will result in stabilizing selection acting on the mimic that increases the resemblance of model and mimic (Schemske and Agren 1995). The observed pollinator behavior potentially eplains the observed patterns of phenotypic variation observed in nonmodel deceptively pollinated orchids. By comparing floral traits, patterns of phenotype frequencies, and mating systems for species of plants with different pollination strategies, these hypotheses could be tested quantitatively. The Orchidaceae would appear to be the ideal family for such a study. ACKNOWLEDGMENTS We thank Dr. Michael Proctor for suggestions and comments on this study. Prof. Wotjek Krzanowski provided statistical advice and Prof. Roy Samb1es helped with the flower color analysis. Comments from two anonymous referees greatly improved the manuscript. AS received funds from N.E.R.C. studentship GT4/92/TLSI18 to J. E. Cresswell and MRM. LITERATURE CITED ACKERMAN, J. D Mechanisms and evolution of food-deceptive pollination systems in orchids. Lindleyana 1: ACKERMAN, J. D., AND M. GALARZA-PEREZ Patterns and maintenance of etraordinary variation in the Caribbean orchid, Tolumnia (Oncidium) variegata. Syst. Bot. 16: ALLEN, J. A Frequency-dependent selection by predators. Philos. Trans. R. Soc. Lond. B 319: BENSON, W. W Natural selection for MUllerian mimicry in Heliconius erato in Costa Rica. Science 176: CLARKE, B. AND P. O'DONALD Frequency-dependent selection. Heredity 19: CRESSWELL, J. E., AND C. GALEN Frequency-dependent selection and the adaptive surface for floral character combinations. Am. Nat. 138: CROPPER, S. C., AND D. M. CALDER The floral biology of Thelymitra epipactoides (Orchidaceae), and the implications of pollination by deceit on the survival of this rare orchid. Plant Syst. Evol. 170: CROW, J. E, AND M. KIMURA An introduction to population genetics theory. Harper and Row, New York. DAFNI, A Pollination of Orchis caspia-a nectarless plant which deceived the pollinators of nectariferous species from other plant families. J. Ecol. 71: Mimicry and deception in pollination. Annu. Rev. Ecol. Syst, 15: Floral mimicry-mutualism and unidirectional eploitation of insects by plants. Pp in B. Juniper and R. Southwood, eds. Insects and the plant surface, Edward Arnold, London Pollination in Orchis and related genera: evolution from reward to deception. Orchid Biol. Rev. Persp. 4: DAWKINS, M Perceptual changes in chicks: another look at the "search image" concept. Anim. Behav. 19: DUKAS R., AND L. A. REAL. 1993a. Learning constraints and floral choice behaviour in bumble bees. Anim. Behav. 46: b. Effects of recent eperience on foraging decisions by bumble bees. Oecologia 94: EFRON, B., AND R. TIBSHIRANI Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy. Stat. Sci. 1: ELTON, R. A., AND J. J. D. GREENWOOD Frequency-dependent selection by predators: comparison of parameter estimates. Oikos 48: ENDLER, J. A Frequency-dependent selection, cypsis and aposematic coloration. Philos. Trans. R. Soc. Lond. B 319: EVERITT, B. S The analysis of contingency tables. Chapman and Hall, London.

9 NEGATIVE FDS BY BUMBLEBEES 723 GREENWOOD, J. J. D The functional basis of frequencydependent food selection. BioI. J. Linn. Soc. 23: GREENWOOD, J. J. D., AND R. A. ELTON Analysing eperiments on frequency-dependent selection by predators. J. Anim. Ecol. 48: GREENWOOD, J. J. D., E. M. WOOD, AND S. BATCHELOR Apostatic selection of distasteful prey. Heredity 47: GREENWOOD, J. J. D., P. A. COTTON, AND D. M. WILSON Frequency-dependent selection on aposematic prey-some eperiments. BioI. J. Linn. Soc. 36: HEINRICH, B Bee flowers: a hypothesis on flower variety and blooming times. Evolution 29: HENDERSON, S. T Daylight and its spectrum. Adam Hilger, Bristol. KAy, Q. O. N The role of preferential and assortative pollination in the maintenance of flower colour polymorphisms. Pp in A. J. Richards, ed. The pollination of flowers by insects. Academic Press, London. KEARNS, C. A., AND D. W. INOUYE Techniques for pollination biologists. Univ. Press of Colorado, Niwot. KEVAN, P. G Floral coloration, its colorimetric analysis and significance in anthecology. Pp in A. J. Richards, ed. The pollination of flowers by insects. Academic Press, London. LAWRENCE, S., AND J. A. ALLEN On the term "search image." Oikos 40: LEVIN, D. A Low frequency disadvantage in the eploitation of pollinators by corolla variants in Phlo. Am. Nat. 106: Pollinator behaviour and breeding structure ofplant populations. Pp in A. J. Richards, ed. The pollination of flowers by insects. Academic Press, London. LEVIN, D. A., AND L. WATKINS Assortative mating in Phlo. Heredity 53: LITTLE, R. J A review of floral food deception mimicries with comments on floral mutalism. Pp in C. E. Jones and R. J. Little, eds. Handbook of eperimental pollination biology, Van Nostrand Reinhold, New York. LUNAU, K A new interpretation of flower guide coloration: absorption of ultraviolet light enhances color saturation. Plant Syst. Evol. 183: MALLET, J., AND N. BARTON Strong natural selection in a warning-color hybrid zone. Evolution 43: MOYA, S., AND J. D. ACKERMAN Variation in the floral fragrance of Epidendrum ciliare (Orchidaceae). Nord. J. Bot. 13: MURCIA, C., AND P. FEINSINGER Interspecific pollen loss by hummingbirds visiting flower mitures: effects of floral architecture. Ecology 77: NILSSON, L. A The pollination ecology of Dactylorhiza sambucina (Orchidaceae). Bot. Notis. 133: Mimesis of bellflower (Campanula) by the red helleborine orchid Cephalanthera rubra. Nature 305: PLOWRIGHT, R. C., AND T. M. LAVERTY The ecology and sociobiology of bumble bees. Annu. Rev. Entomol. 29: SCHEMSKE, D. W., AND J. AGREN Deceit pollination and selection on female flower size in Begonia involucrata: an eperimental approach. Evolution 49: SLATER, A. T., AND D. M. CALDER The pollination biology of Dendrobium speciosum Smith: a case of false advertising? Aust, J. Bot. 36: SMITHSON, A., AND M. R. MACNAIR Frequency-dependent selection by pollinators: mechanisms and consequences with regard to behaviour of bumblebees Bombus terrestris (L.) (Hymenoptera: Apidae). J. Evol. BioI. 9: Density-dependent and frequency-dependent selection by bumblebees Bombus terrestris (L.) (Hymenoptera: Apidae). BioI. J. Linn. Soc. 60: STANTON, M. L Reproductive biology of petal color variants in wild populations of Raphanus sativus: I. Pollinator responses to color morphs. Am. J. Bot. 74: STANTON, M. L., A. A. SNOW, S. N. HANDEL, AND J. BERECZKY The impact of a flower-color polymorphism on mating patterns in eperimental populations of wild radish (Raphanus raphanistrum L.). Evolution 43: THOMPSON, V Polymorphism under apostatic and aposematic selection. Heredity 53: TINBERGEN, L The natural control of insects in pinewoods. I. Factors influencing the intensity of predation by songbirds. Arch. Neerl. Zool. 13: WASER, N. M., AND M. V. PRICE Pollinator choice and stabilizing selection for flower color in Delphinium nelsonii. Evolution 35: Corresponding Editor: V. L. Sork

10 本文献由 学霸图书馆 - 文献云下载 收集自网络, 仅供学习交流使用 学霸图书馆 ( 是一个 整合众多图书馆数据库资源, 提供一站式文献检索和下载服务 的 24 小时在线不限 IP 图书馆 图书馆致力于便利 促进学习与科研, 提供最强文献下载服务 图书馆导航 : 图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具