Variation in floral sex allocation and reproductive success within inflorescences of Corydalis ambigua (Fumariaceae): pollination efficiency or resource limitation?


G. Kudo (fax + 81 11 7064954;


  • 1The variation of floral sex allocation with flower position within inflorescences was investigated in the spring ephemeral, Corydalis ambigua. Investment in female function (pistil), attraction (corolla) and nectar production decreased from bottom to top flowers, whereas male investment (stamen) did not differ.
  • 2 This self-incompatible species appears to set seeds as a result of visitation by nectar robbing bumblebee queens. The tendency of bees to visit lower flowers first and then move upwards within an inflorescence should result in directional pollen flow from bottom to top flowers.
  • 3 Naturally pollinated upper flowers set fewer seeds than intermediate and lower flowers due to pollen limitation. The lack of differences in seed set and seed mass per pod following artificial outcrossing indicated that resource limitation did not explain the variation in seed production of flowers in different positions. Pollen viability also did not differ significantly between flower positions.
  • 4 A model of pollination was developed that incorporated the visitation pattern of bumblebees and observed variations in nectar distribution between flower positions. This predicted that receipt of outcross pollen would decrease from bottom to top flowers, but that pollen export to other plants would not differ between flower positions provided that the pollen exchange rate of pollinators was either small or positively correlated with nectar content of each flower position. The observed pattern of floral sex allocation would then be parallel to relative success of pollen export and import between flower positions within inflorescences.


Most flowering plants are hermaphrodite, but the pattern of sex allocation can vary both within and between individuals (Devlin & Stephenson 1987; Ross 1990); and even within inflorescences, resource allocation to male and female functions may differ between flowers. An extreme example is the occurrence of andromonoecy in which a single inflorescence may contain both male and hermaphrodite flowers (Bertin 1982; Solomon 1985). Such variation in floral sex expression can reflect differences in reproductive success through each sexual function (Solomon 1985; Devlin & Stephenson 1987).

Stereotypic behaviour of pollinators may lead to directional pollen flow within inflorescences and thus influence floral sex allocation (Brunet & Charlesworth 1995). The first flowers visited will receive more pollen from other plants than will flowers that are visited subsequently (Robertson 1992; Harder & Barrett 1996). Correspondingly, although the last flowers visited before pollinators leave the inflorescence tend to receive self pollen, they are the most successful at exporting pollen to other plants (Robertson 1992; Harder & Barrett 1996). Thus, the contribution of male and female functions to reproductive success is expected to differ between flower positions within inflorescences, and floral sex allocation might vary accordingly. Larger investment in ovule production would be expected to increase the reproductive success of earlier visited flowers, and increased pollen production to benefit those visited later. Although such variation is especially likely in self-incompatible plants (Brunet & Charlesworth 1995), few studies have reported data with which to assess this prediction.

Resource conditions also affect sex allocation patterns in hermaphrodite plants (e.g. Lloyd 1980; Charnov 1982). Resources available for reproduction may vary between flower positions leading to differences in ovule number and seed production (Lee 1988; Jennersten 1991) and pollen germination rates or growth rates of pollen tubes may also depend on the resource condition of plants (Young & Stanton 1990). Thus, to assess variation in floral sex allocation within inflorescences, it is important to discriminate between the effects of pollen limitation and of resource limitation on each sexual function.

We evaluate which of these contrasting mechanisms explains the variation in floral sex allocation within inflorescences in Corydalis ambigua Chem. et Schlecht (Fumariaceae), a spring ephemeral plant. This species is self-incompatible and is visited exclusively by bumblebee queens (Ohara & Higashi 1994). Because bumblebees tend to visit bottom flowers first and move upwards within an inflorescence (e.g. Darwin 1877; Pyke 1979; Corbet et al. 1981; Barrett et al. 1994), we predict female-biased allocation to lower flowers and male-biased allocation to those in upper positions. Accurate assessment of pollen flow is necessary, because the ratio of self to outcross pollen carried by a pollinator changes continually throughout geitonogamous pollination of an inflorescence (e.g. Harder & Barrett 1995, 1996). First, we compared patterns of seed-set for each flower position between naturally pollinated and artificially outcrossed plants: if seed set of upper flowers is limited by resources, they should set fewer seeds than lower flowers even when outcrossed. To assess the effect of resource limitation in male function, we then compared pollen viability between flower positions by comparing seed set via crossing experiment. For upper flowers to be successful as pollen donors under natural conditions, frequent pollen removal from the self plant must be accompanied by high viability. If, however, upper flowers suffer resource limitation, their pollen viability, and thus their contribution as pollen donors, may be reduced. Finally, we evaluated the success of pollen receipt and donation at each flower position by modelling the movements of self and outcross pollen within a single inflorescence based on the foraging pattern of bumblebees. The results of such experiments should allow us to discuss the ecological consequence of floral gender variation within inflorescences in C. ambigua.

Materials and methods

Study site and study species

This study was conducted in the understorey of a natural deciduous forest near Sapporo (43°25′ N, 143°32′ E), Hokkaido, northern Japan. Soon after snow melt (early April), a succession of spring ephemerals, including Adonis ramosa, Corydalis ambigua, Trillium camschatcense and Anemone flassida, emerge and bloom before the leaves of the canopy trees expand.

Corydalis ambigua is one of the most common spring ephemerals in northern Japan. Each plant generally produces a single inflorescence and each of the one to 20 flowers has a spur in which nectar collects. In the study C. ambigua population, flowering usually starts during late April and lasts until mid May. Although flowers open progressively from the bottom to the top of an inflorescence, there is considerable overlap in the anthesis period of most of the flowers. On average it is only 2 days (range 1–6 days) from the onset of first flowering to opening of the last flower within an inflorescence, but the total flowering period of an inflorescence is 18 days (range 12–25 days) (G. Kudo, unpublished data). Seeds mature during early June, when they fall from the pods and are dispersed secondarily by ants (Ohkawara et al. 1997).

Queens of the bumblebee, Bombus hypocrita sapporensis Cockerell, a major pollinator of C. ambigua (Ohara & Higashi 1994), usually emerge from hibernation during late April at the same time as C. ambigua begins flowering. Although nectar is usually collected through perforations made in the nectar spurs, fertilization and therefore seed set does occur, probably because some pollen is deposited on the bee during nectar robbing (illustrated in Higashi et al. 1988). The anthers are protected within an inner petal and unvisited flowers contain large amounts of pollen with no evidence of disturbance. We can therefore easily discriminate whether a given flower has been visited by bumblebees by checking for the presence of a robbing scar and examining anther condition.

We observed the behaviour of bumblebees on inflorescences in a 100 × 100 m area on calm days from the end of April to early May 1995. We classified flower position into lower, intermediate or upper according to total flower number (Table 1). For each inflorescence visited, we recorded the total number of flowers per inflorescence, position of the first flower visited, whether the bee then moved upward or downward, the total number of flowers visited and the position of the final flower visited.

Table 1.  Classification of flower position within inflorescences of Corydalis ambigua with reference to flower number. In a case of plants having 10 flowers, for example, the lower three flowers are classified as lower position, upper three flowers as upper position, and others as intermediate position
 Number of flowers per plant
Position class23456789101112131415161718

Floral allocation

During late April 1995, we collected 49 unvisited but fully open inflorescences with various total numbers of flowers. Each flower was classified with respect to its position within the inflorescence (Table 1), before division into stamens, pistil and corolla. Components were dried at 70 °C for 48 h and then weighed. The mean of all flowers in a position class from a given inflorescence was used for analysis.

Nectar production per flower was compared between flower-position classes. During early May 1995, we collected 25 inflorescences with unperforated corollas just before flowering and transported them to the laboratory, where we placed them in bottles filled with water. Twenty-four hours after each flower had opened, we measured its nectar volume (n = 159 flowers) with a capillary tube (0.6 mm internal diameter). The mean of all flowers in a position class from a given inflorescence was used for analysis.

Ovule fertilization and seed production

Prior to flowering in 1995, we randomly selected 90 reproductive plants, numbered them with plastic tags and recorded the number of flower buds before leaving them exposed to natural pollination. Just before seed dispersal, we harvested all pods on the marked plants and classified them into the appropriate position class.

At the same time, we selected a further 37 plants to compare the pattern of seed production when plants received abundant pollen with that of open-pollinated plants. These were chosen at random from those in flowers but which had not yet been visited by bumblebees. Each plant was hand-pollinated using a separate donor located more than 5 m distant (to remove the effect of genetic variation on seed production due to relatedness between plants). All flowers within an inflorescence had the surface of the stigma completely covered with pollen and hand-pollinated inflorescences were then bagged to prevent further pollination. All pods were subsequently harvested and all mature seeds were weighed for 26 of the inflorescences.

Pods from both open and hand pollinated plants were opened and seeds categorized as mature, eaten or underdeveloped, or unfertilized ovules were counted. Unfertilized ovules represented ovules that had not enlarged since flowering, whereas undeveloped (immature) seeds had softer seed coats and were of lighter colour and smaller size than mature seeds (but were bigger than unfertilized ovules), and were likely to be the result of abortion of fertilized ovules. Dipteran larvae often eat maturing seeds of C. ambigua, leaving empty seed coats. We measured ovule fertilization rate and seed production rate as follows:

Fertilization rate = fertilized ovule number/total ovule number;

Seed production rate = mature seed number/total ovule number.

Pollen viability

Fertilization ability of pollen from the three position classes and size of sired seeds were compared in a crossing experiment. Plants were randomly selected during late April 1998. We collected pollen from two flowers in each position class (i.e. upper, intermediate or lower flowers) of each of 32 donor inflorescences and deposited the pollen from each position on the stigmas of two lower flowers of a separate recipient inflorescence (n = 96) that had not been visited by pollinators. Unpollinated flowers on the recipient inflorescences were then removed and each recipient plant was covered with a nylon mesh to avoid additional pollination until fruiting when all pods were harvested and their seeds were counted and weighed.

Pollen flow model

We modelled pollen movements between flower positions within inflorescences based on the observed patterns of bumblebee behaviour within single inflorescences and an exponential model of pollen dispersal (Robertson 1992; Harder & Barrett 1996). We assume that the average pollen load (m) on a bumblebee body is constant and that at each flower a bee picks up the same amount of pollen that it deposits with a pollen exchange rate, p (0 < p < 1). This exchange rate is likely to vary with nectar availability, as pollinators tend to stay longer and deposit more pollen on stigmas of flowers having larger nectar volumes (Thomson 1986; Thomson & Plowright 1980; Zimmerman 1983). Longer visits by bumblebees on an inflorescence also increase seed set in C. ambigua (Ohara & Higashi 1994). To assess the effects of nectar distribution within inflorescences on pollination success between flower positions, we consider first a case where pollen exchange rate is constant for all flower positions independent of nectar distribution. Then, we assume a case where pollen exchange rate of each flower position is proportional to the nectar content of the flower.

At the first flower, both pollen deposition and removal on first-visited flowers equals mp, and m(1 − p) pollen grains from other plants remain on the bumblebee’s body (as pollen carryover). The mp pollen grains removed from this first flower are transported to other flowers on the same plant (geitonogamy) or other plants (outcrossing). Pollen transport mediated by bumblebees occurs predominantly from bottom to top, and intermediate and upper flowers may therefore receive outcross pollen by pollen carryover from lower flowers as well as from bees that laid directly on them. Thus, total pollen flow between flower positions can be expressed in terms of m and p by taking accounts of the patterns of bumblebee visitation and nectar distribution.

Statistical analysis

We used repeated-measures anova and Scheffé means comparison tests for comparing stamen weight, pistil weight, corolla weight and nectar volume between flower positions within inflorescences. Furthermore, we assessed the effect of flower size on floral allocation. When stamen or pistil weight was significantly correlated with corolla weight as an index of flower size, we used single-factor ancova to test the variation in stamen or pistil weight for the different flower positions taking corolla weight as the covariate. Fertilization rate and seed production rate per inflorescence were compared between artificially and naturally pollinated plants by t-tests after arcsine-transformation. In comparisons of these values between flower positions within inflorescences, we used repeated-measures anova and Scheffé means comparison tests. Individual seed weight and seed mass per pod of outcrossed plants were compared for the different flower positions by repeated-measures anova. Differences in visitation frequency of bumblebees between flower positions were compared by G-tests. Pollen viability of flower from different positions was tested with one-way anova for comparisons of seed number, seed production rate (after arcsine-transformation) and seed weight. All analyses were performed using StatView Version 5.0 (SAS Institute Inc., Cary, NC).


Floral allocation

Pistil (female function) and corolla (attraction function) mass per flower decreased significantly from lower to upper flowers (F2,48 = 106.41, P < 0.0001 and F2,48 = 17.19, P < 0.0001, respectively; Fig. 1) but stamen mass (male function) did not vary significantly between flower positions (F2,48 = 0.18, P > 0.8). The total mass of individual flowers differed significantly between flower positions (lower, 6.0 ± 0.1 (SE) mg; middle, 5.7 ± 0.1 mg; upper, 5.4 ± 0.1 mg; F2,48 = 42.10, P < 0.0001), largely reflecting variation in corolla mass with comprised 81% of the total. Pistil mass was positively correlated but stamen mass was not correlated with corolla mass (R2 = 0.11, P < 0.0001, R2 = 0.01, P > 0.1, respectively, n = 154). Results of the ancova in which corolla mass was used as the covariate revealed that pistil mass differed significantly between flower positions (F2,151 = 13.76, P < 0.0001). Thus, relative allocation to female function decreased upwards within inflorescences. Ovule number per flower obtained from open pollinated plants varied in parallel with pistil weight differing significantly between flower positions (F2,76 = 58.78, P < 0.0001, repeated-measures anova; 9.5 ± 0.1, 8.8 ± 0.1 and 7.9 ± 0.1 for lower, intermediate and upper flowers).

Figure 1.

Dry weight allocation and nectar volume per flower with reference to flower position in Corydalis ambigua. Mean value of each flower position per inflorescence was used for analysis. Vertical bar shows SE. Data were collected from 49 inflorescences for dry weight allocation and 25 for nectar volume. Values with different letters are significantly different at P = 0.05 by Scheffé tests.

Mean nectar production per flower decreased from bottom (5.1 µL) to top flowers (2.7 µL) within inflorescences (F2,24 = 21.94, P < 0.0001; Fig. 1). Nectar volume per corolla weight, an index of nectar productivity, was 1.1 µL mg−1 in lower flowers, 0.9 in intermediate flowers, and 0.6 in upper flowers.

Seed production

Ovule fertilization rate of hand-pollinated inflorescences (0.85 ± 0.03 SE) was significantly higher than that of naturally pollinated inflorescences (0.58 ± 0.02; d.f. = 125, t = 6.31, P < 0.0001). Similarly, total seed set of hand-pollinated inflorescences (0.63 ± 0.03) was significantly higher than that of naturally pollinated inflorescences (0.30 ± 0.02; d.f. = 125, t = 9.43, P < 0.0001). These results indicate that pollen limitation normally restricts seed production.

Under natural conditions, upper flowers had significantly lower fertilization rate and smaller proportion of mature seeds than intermediate and lower flowers (F2,89 = 6.72, P = 0.0015 and F2,89 = 11.45, P < 0.0001, respectively; Fig. 2). Mature seed number per flower was 3.2 ± 0.3 (SE) in lower, 3.1 ± 0.2 in intermediate, and 1.7 ± 0.2 in upper flowers. In contrast, there were no significant differences in either fertilization rate (F2,36 = 0.29, P > 0.7) and seed production rate (F2,36 = 0.09, P > 0.9) between flower positions in hand-pollinated inflorescences (i.e. when there is no pollen limitation) (Fig. 2).

Figure 2.

Patterns of ovule fertilization and seed production for flower positions under open pollination (●, n = 90) and artificial cross-pollination (○, n = 37) in Corydalis ambigua. Vertical bar shows SE. Values with different letters are significantly different at P = 0.05 by Scheffé tests. Further details in text.

In hand-pollinated plants, individual seeds of upper flowers were significantly lighter than those from other flower positions (F2,25 = 6.25, P = 0.004; Table 2) but the lack of any significant differences in total seed mass per pod between flower positions (F2,25 = 2.02, P > 0.10; Table 2) suggests that seed production in upper flowers is not limited by resources.

Table 2.  Individual seed weight (mg) and seed mass per pod (mg) of artificially outcrossed plants of Corydalis ambigua. Significance levels by repeated-measures anova are shown. Mean values with different superscript letters are significantly different (P < 0.05, Scheffé tests) from other means (n = 26, mean ± SE)
Flower positionLowerIntermediateUpperP
Individual seed wt.0.67 ± 0.02a0.66 ± 0.03a0.60 ± 0.03b   0.004
Seed mass per pod4.35 ± 0.334.36 ± 0.363.72 ± 0.39> 0.10

Foraging behaviour of bumblebees

During 7 days of observation (about 2 hours per day) we observed the behaviour of 324 queens of Bombus hypocrita sapporensis. Most bees visited lower flowers within an inflorescence first (67% of total first visits; Fig. 3 and Table 3), with first visits to upper flowers being rare (7%). For bumblebees that moved between flower positions on an inflorescence, 80% moved upwards. About half of the bumblebees left inflorescences from upper flowers (54%), with 26% and 20% leaving from intermediate and lower flowers, respectively. Although we did not record all movements within inflorescences, we observed that bees rarely skipped from lower to upper position (less than 5% of movements; G. Kudo, unpublished data). We can therefore postulate a stepwise movement up an inflorescence, with the probability of a bee leaving the lower positions of 30% (= 20/67) and thus the probability of moving to an intermediate position is 70% (Fig. 3). Similarly, the probabilities of a bumblebee on an intermediate flower leaving the inflorescence or moving to upper position are 36% and 64%, respectively (Fig. 3).

Figure 3.

Visit patterns of bumblebees within inflorescences of Corydalis ambigua, showing observed frequencies of first and last visits at each position and (in parentheses) estimated movements within inflorescences.

Table 3.  Positions of first and last flowers visited by bumblebees on inflorescences of Corydalis ambigua. Observed patterns are significantly different from expected ones. *** P < 0.0001, G-tests
Flower positionLowerIntermediateUpperTotald.f.G2-value
First visit (observed)216 85 233242104.97***
First visit (expected) 97130 97324  
Last visit (observed) 61 811653072 37.13***
Last visit (expected) 91125 91307  

Bumblebees visited a greater number, but a smaller proportion, of flowers on large inflorescences. On average, a bee visited 2.5 ± 0.1 (SE) flowers on inflorescences with fewer than six flowers (n = 73), 3.6 ± 0.1 for 6–10 flowers (n = 159), 4.0 ± 0.2 for 11–15 flowers (n = 82), and 5.3 ± 0.9 for more than 15 flowers (n = 10).

Pollen viability

In the crossing experiment, total seed number (F2,89 = 0.13, P > 0.7), seed production rate (F2,89 = 0.51, P > 0.5), and individual seed weight (F2,89 = 0.35, P > 0.5) did not vary significantly between flower positions on the donor plants. Pollen from upper flowers therefore has a similar ability to fertilize ovules and give rise to seeds as that from intermediate and lower flowers, suggesting that male function, like the female (see Seed Production), is not restricted by resources and upper flowers can be successful pollen donors.

Simulation of pollen flow

Based on the observed pattern of bumblebee visitation (Fig. 3), pollen flow between all flower positions is expressed in terms of m and p (see Methods) as shown in Fig. 4. For example, intermediate flowers receive 0.26 mp outcross pollen directly (arrow b) and 0.47 m(1 − p) outcross pollen by pollen carryover from lower flowers (arrow d). Similarly, intermediate flowers export 0.26 mp self pollen directly (arrow k) and 0.47 mp(1 − p) self pollen via upper flowers (arrow m).

Figure 4.

Flow of outcross and self pollen within inflorescences of Corydalis ambigua estimated from visit pattern of bumblebees (Fig. 3). Movements of self pollen derived from lower, intermediate, and upper flower positions are indicated. Outcross pollen exported from the inflorescence as a result of pollen carryover is not included. p, pollen exchange rate; m, pollen load on a pollinator.

Results of our simulations of the effects of pollen exchange rate on pollen export and import at each flower position are shown in Fig. 5. When pollen exchange rate is constant for all flower positions, the patterns of pollen export differ between flower positions, depending on the value p (Fig. 5a). When p is large, the amount of pollen export increases from lower to upper flowers, but when p is less than 0.4, the amount of pollen export is similar for all flower positions. When pollen exchange rate depends on nectar volume, and it is reduced in proportion to the differences (5.1 µL at lower compared with 4.2 µL (82%) at intermediate, and 2.7 µL (53%) at upper flowers), pollen export is similar for all flower positions irrespective of p (Fig. 5b). This indicates that each flower position has a similar potential for pollen export to other plants. The amounts of outcross pollen receipt decrease from bottom to top in both cases (Fig. 5a,b), and the difference increase with increasing p.

Figure 5.

Estimated patterns of self-pollen export (above) and outcross-pollen receipt (below) at each flower position with reference to pollen exchange rate at lowest position (p). m, pollen load on a pollinator. (a) Pollen exchange rate is assumed to be constant for all flower positions. (b) Pollen exchange rate is assumed to depend on nectar volume, i.e. 0.82 and 0.53 times the rate of lower position-flowers for intermediate and upper flowers, respectively.


Pattern of seed production within inflorescences

Previous studies have reported that, under conditions of natural pollination, C. ambigua receives insufficient pollen of adequate quality to result in full seed set (Ohara & Higashi 1994; Yasaka et al. 1994). This study shows, in addition, that upper flowers suffer greater pollen limitation than intermediate and lower flowers. The pattern of seed production is consistent with the consequences of geitonogamy caused by bumblebees tending to visit lower flowers first before moving upwards, which for self-compatible plants will increase the fraction of self pollen delivered from bottom to top flowers (Barrett et al. 1994). In self-incompatible species such as C. ambigua, the reduction in cross-pollen deposited on stigmas will reduce seed production from bottom to top flowers. As expected, ovule production was enhanced in lower flowers, leading to increased success as a female parent, but this pattern is not a consequence of resource limitation at upper positions because all flowers of hand-pollinated plants showed similar seed production.

Pattern of pollen flow within inflorescences

Although we expected that loss of pollen due to geitonogamy should enhance pollen export from upper flowers, which would thus benefit from greater male investment, stamen mass did not vary significantly with flower position. Simulations of pollen flow (Fig. 5) indicated that differences in pollen export betwen flower positions might sometimes be small. Naturally pollinated plants showed similar seed production rates for intermediate and lower flowers, and the simulations predict that this will be the case when p is small, i.e. pollinators have small pollen exchange abilities. Pollen exchange rates for bumblebees are known to range from 0.08 to 0.39 (references in Robertson 1992), and such values would give similar pollen deposition between intermediate and lower flowers (Fig. 5). Unfortunately, we do not have information about pollen exchange rates of bumblebees visiting C. ambigua. However, 85% of pollinated C. ambigua inflorescences are subject only to robber-like visitation of B. hypocrita sapporensis queens (Higashi et al. 1988), and it is likely that both pollen carryover and exchange rates will be lower than for legitimate visitation. If p is indeed low, models in which pollen exchange rate is assumed to be either independent (Fig. 5a) or dependent (Fig. 5b) on nectar distribution can successfully explain the similar male allocation between flower positions within inflorescences of C. ambigua.

Implications for geitonogamy

Variation in nectar production has been considered as a strategy whereby plants can control pollen flow by influencing the foraging behaviour of pollinators (e.g. Zimmerman 1988). In our simulations, changing the pollen exchange rate depending on nectar volume at each flower position resulted in changes in the pattern of pollen export but not in patterns of pollen receipt. This indicates that variation in nectar production may be a strategy that controls the patterns of pollen donation (male success), rather than the deposition of pollen on stigmas (female success) as predicted elsewhere by Klinkhamer & de Jong (1993).

In self-compatible plants, selfing rates increase and outcrossed siring success decreases with increasing geitonogamous pollination (Harder & Barrett 1995), and both male and female reproductive success are reduced through decrease in pollen dispersal, dilution of outcross pollen or interference by self-pollen grains (Galen et al. 1989; Snow et al. 1996). The negative effects of geitonogamy are thus considered to be strongest in self-incompatible plants (Klinkhamer & de Jong 1993). Floral strategies that reduce geitonogamous pollination include the spatial and temporal separation of pollen and stigmas (i.e. herkogamy and dichogamy) or the presentation of only some fraction of flowers within an inflorescence as a result of sequential flower opening (Lloyd & Webb 1986; Galen et al. 1989; Harder & Barrett 1996). However, in C. ambigua anthers are located close to the stigma and dehisce soon after the onset of flowering, leading to exposed pollen being deposited around the stigma. Although flowering progresses from bottom to top, there is much overlap of anthesis between flower positions. This species therefore does not appear to use the mechanisms available to it to reduce geitonogamy and pollen discounting. Pollen deposition over large and variable areas of the pollinator’s body may, however, result in low pollen exchange rates (Waser & Price 1984; Robertson 1992), thus reducing the negative effects of geitonogamy. Therefore, the evolutionary consequences of the high frequency of robber-like bumblebee pollination, which have previously been questioned in C. ambigua (Higashi et al. 1988; Ohara & Higashi 1994), may result in illegitimate pollination which may contribute to reducing the negative effects of geitonogamy in this species.


We thank M. Ohara for comments on the earlier version of the manuscript; N. Wada and H. Fukuda for their valuable discussion on this study; and S. Suzuki, Y. Nishikawa, Y. Ejima, T. Kasagi and N. Akashi for assistance in data collection. We are grateful to S. Barrett and two anonymous referees for their useful comments and revision of the manuscript. This study was partly supported by a grant from the Ministry of Education, Science and Culture, Japan (no. 11440223).

Received 3 January 2000 revision accepted 19 June 2000