Inflorescence architecture affects pollinator behaviour and mating success in Spiranthes sinensis (Orchidaceae)

Authors


Author for correspondence:
Atushi Ushimaru
Tel: +81 0 78 803 7746
Email: ushimaru@kobe-u.ac.jp

Summary

  • Despite the wide inflorescence diversity among angiosperms, the effects of inflorescence architecture (three-dimensional flower arrangement) on pollinator behaviour and mating success have not been sufficiently studied in natural plant populations.
  • Here, we investigated how inflorescence architecture affected inter- and intra-plant pollinator movements and consequent mating success in a field population of Spiranthes sinensis var. amoena (S. sinensis). In this species, the flowers are helically arranged around the stem, and the degree of twisting varies greatly among individuals. The large variation in inflorescence architecture in S. sinensis results from variation in a single structural parameter, the helical angle (the angular distance between neighbour-flower directions).
  • The numbers of visits per inflorescence and successive probes per visit by leaf-cutting bees decreased with helical angle, indicating that individual flowers of tightly twisted inflorescences received less visitations. As expected from pollinator behaviour, pollinia removal and fruit set of individual flowers decreased with helical angle. Meanwhile, geitonogamy decreased in tightly twisted inflorescences.
  • Our novel findings demonstrate that natural variation in inflorescence architecture significantly affects pollinator behaviour and reproductive success, suggesting that inflorescence architecture can evolve under pollinator-mediated natural selection in plant populations. We also discuss how diverse inflorescence architectures may have been maintained in S. sinensis populations.

Introduction

The diversity of floral traits has evolved under pollinator-mediated selection, and, for most flowering plants, manipulating the inter- and intra-plant movements of pollen vectors is crucial for mating success (Harder et al., 2004; Harder & Johnson, 2009). The frequency of pollinator visitations to a plant is expected to correlate with pollen export and import (Harder & Barrett, 1996), whereas the number of successive visits within a plant is predicted to correlate with self-pollination between flowers (geitonogamy: Harder & Barrett, 1995).

Floral and inflorescence characteristics are thought to have evolved to control these pollinator behaviours. The mating success of a plant depends on the reproductive outcomes of all flowers of a plant, and inflorescences as well as floral traits are important for a complete understanding of plant pollination. Because of their reproductive importance, individual flowers have historically been the primary subject of pollination biology studies (Harder et al., 2004). During the last couple of decades, the number of flowers opening at one time (display size), which is a trait of inflorescences, has attracted a great deal of attention. For example, several studies have documented the relationship between display size and two aspects of pollinator behaviour: large floral displays attract more pollinators compared with small displays, and individual pollinators visit more flowers on large displays (e.g. Ohashi & Yahara, 2001). However, despite the extensive diversity of inflorescence architecture (three-dimensional flower arrangement within inflorescences) within angiosperms (Troll, 1964; Weberling, 1989), the effect of inflorescence architecture on inter- and intra-inflorescence pollinator movements and mating success has not been sufficiently studied (Harder et al., 2004).

A few studies have examined the effects of inflorescence architecture on pollinator behaviour. Using artificial inflorescences consisting of plastic artificial flowers or inflorescences in which natural flowers are spatially rearranged, these studies have investigated the effects of inflorescence architecture on the following three aspects of pollinator behaviour. First, attractiveness: Fishbein & Venable (1996) found that Asclepias tuberosa inflorescences with intermediate flower densities attracted more pollinators than those with the same number of flowers at higher or lower densities, and Ishii et al. (2008) further showed that the relative attractiveness of a large display size depends on the three-dimensional arrangements of flowers. Secondly, the number of flowers probed on a given inflorescence in one pollinator visit (the number of successive probes per visit): Hainsworth et al. (1983) found that hummingbirds probed fewer flowers during visits to artificial hemispheric inflorescences than during visits to one- or two-dimensional inflorescences, and Ishii et al. (2008) also demonstrated that the number of flowers visited by bumblebees varied significantly among different inflorescence architectures (i.e. racemes, panicles and umbels). Thirdly, consistency of pollinator movement within inflorescences: the three-dimensional arrangement of flowers affects the consistency of pollinator movement within inflorescences (Jordan & Harder, 2006) and can create variation in the mating environments of flowers, which may select for sex allocation gradients among flowers (Brunet & Charlesworth, 1995; Kudo et al., 2001) or even segregation of the sex role within inflorescences (Jordan & Harder, 2006).

Together, these findings demonstrated that inflorescence architecture may affect mating outcomes by influencing pollinator behaviour. However, these studies did not fully reveal the functions of inflorescence architecture, in that the effects of natural variation in floral arrangement on pollinator behaviour have not yet been examined. Furthermore, although Fishbein & Venable (1996) assessed the effects of floral arrangement on the rate of pollinia removal and insertion per flower, they did not consider intra-inflorescence movements, which could enhance geitonogamy. To our knowledge, no other work has examined the effect of floral arrangement on mating success.

In the present study, we investigated how natural variation in inflorescence architecture affects pollinator behaviour and consequent reproductive success in a field population of Spiranthes sinensis var. amoena (S. sinensis). This species is an ideal test subject because its flowers are helically arranged around the stem, and the degree of twisting varies greatly among individuals (Fig. 1). This large variation in inflorescence architecture in S. sinensis results from variation in a single structural parameter, the helical angle (the angular distance between neighbour-flower directions; Fig. 2). This distinguishing character enabled us to easily quantify inter-individual differences in inflorescence architecture.

Figure 1.

Diversity of Spiranthes sinensis inflorescence architecture. Flowers are spirally placed around the stems, and the degree of twisting varies substantially among individuals. Helical angles of the respective inflorescences are 0° (a), 25.7° (b), 51.4° (c) and 90° (d).

Figure 2.

Scheme of the inflorescence architecture of Spiranthes sinensis. The dihedral angle between neighbouring flowers on the stem is defined as the ‘helical angle’ of the inflorescence. The movement of a bee towards a neighbouring flower is defined as ‘normal’, whereas movements towards spatially adjacent, but not neighbouring, flowers are defined as ‘skips’.

We hypothesized that the helical angle of S. sinensis inflorescences would affect pollinator behaviours in a variety of ways. First, attractiveness may either increase or decrease with helical angle. Loosely twisted inflorescences (i.e. inflorescences with a smaller helical angle; Fig. 1a) may attract more pollinators than tightly twisted inflorescences (i.e. inflorescences with a larger helical angle; Fig. 1d) if the series of continuous flowers visible from a certain direction acts as a single attraction unit. However, extremely loosely twisted inflorescences on flat ground, as is the case of S. sinensis habitats, may only attract pollinators from a limited direction (Ushimaru et al., 2006). Secondly, twisting increases the distance between neighbouring flowers within an inflorescence (Supporting information Fig. S1), potentially decreasing the number of successive probes per visit. This hypothesis is based on the idea that plant traits that increase the probing cost per flower, such as flower complexity and inter-flower distance within a plant, probably reduce the number of successive probes in a plant (Hainsworth et al., 1983; Ishii & Sakai, 2001; Ohashi, 2002). Thirdly, a tightly twisted arrangement may induce pollinators to skip neighbouring flowers within an inflorescence, as tightly twisted arrangements offer many spatially closed, non-neighbouring flowers. Given the nature of the upward movement habit of bees in a vertical inflorescence (Waddington & Heinrich, 1979), skip behaviour may also reduce successive probes within an inflorescence, as it would hasten a visiting bee to reach top flowers. Furthermore, skip behaviour would affect the distribution of empty flowers within inflorescences because the distribution of empty flowers reflects the visitation history by bees. The dispersion pattern of empty flowers within an inflorescence would further influence the behaviour of subsequent visitors.

To examine the effect of flower arrangement on pollinator behaviours and mating success, we addressed the following questions. Does the helical angle (i.e. floral density) of inflorescences affect pollinator attraction? Does an increased helical angle of inflorescences (i.e. increased distance between adjacent flowers within inflorescences) decrease the number of successive probes? Does an increased helical angle (i.e. increased complexity of inflorescences) lead to irregular pollinator movements and flower skipping (i.e. reduce the consistency of the foraging path within inflorescences)? Does variation in inflorescence architecture influence reproductive success in the field? In the light of our results, we discuss the adaptive significance of the inflorescence architecture for real inflorescences.

Materials and Methods

Study site and species

We studied Spiranthes sinensis (Pers.) Ames var. amoena (M. Bieb.) H. Hara inflorescences in a private grass yard in Nishinomiya, Hyogo, Japan (34°44′N, 135°20′E). Spiranthes sinensis is a summer-flowering orchid that grows in open habitats, moist-dry grassy hillsides or fields. At the study site, each plant usually had a single inflorescence composed of 8–45 flowers. The flowers were spirally arranged around the flowering stem, and the helical angle was approximately the same within a plant but differed substantially among individuals (Fig. 1), ranging from 0 to 120° (Fig. S2). The flowers bloomed sequentially from the bottom to the top of inflorescences (acropetal blooming), and the flowering periods of individual flowers were 3–4 d. The display size (number of open flowers) ranged from 1 to 45.

The flowers of this species are herkogamous and weakly protandrous: a bisexual stage is found in the field. Cross-pollination may be promoted by protandry with sequential flowering, which keeps female (older) flowers at lower positions and male (younger) flowers at upper positions. Because bees move upwards on inflorescences, they import pollen to the lower female flowers and export pollen from the higher male flowers, which may consequently enhance outcrossing. Flowers are fully self-compatible but never produce seeds without pollinator visits (T. Iwata, unpublished data). Each flower has paired pollinia, which are usually transferred as a single unit. The pollinia are granular, such that pieces of an incomplete pollinium are often transferred to the stigma (C. Fukada, M. Yokogawa, O. Nagasaki, S. Kaneko, Y. Isagi and A. Ushimaru, unpublished data). Leaf-cutting bees (Megachile nipponica and Megachile japonica) were the primary pollinators of S. sinensis at the study site during the study period, although sweat bees (Halictidae spp.) visited occasionally. Leaf-cutting bees moved upwards in the inflorescences, similar to bumble bees.

Data collection

Data were collected during July 2007 and 2008, July being the flowering peak for S. sinensis at the study site. The study population of S. sinensis was composed of c. 200–250 individuals. The population was divided into two site groups (sites A and B), as their habitats were spatially separated by a building and the timing of peak flowering differed by c. 2 wk between groups, probably as a result of differences in the light environment.

We observed pollinator behaviours from 08:00 to 17:00 h on three sunny days in July 2008 at site A to determine the attractiveness of each inflorescence, the number of successive probes within the inflorescence per visit and skips of neighbouring flowers per visit (see Fig. 2). During the 3-d observation periods, we recorded a total of 56 leaf-cutting bees entering the site (712 visits to 89 inflorescences). The display size of each observed inflorescence was recorded as the number of open flowers on each observation day. Inflorescence size was recorded as the total number of flowers of each inflorescence (including open and withered flowers and flower buds). To assess the attractiveness of each inflorescence, we recorded visited/unvisited data (i.e. binary data) for each of the 89 inflorescences by each of the 56 visitors to the sites. The number of successive probes per visit and skipping movements were measured to estimate geitonogamy and the consistency of intra-inflorescence movements of pollinators, respectively. Observations began when a bee was observed entering the study site; the bee was then followed, and behaviours were recorded until it left the site.

We recorded pollinia removal from each flower to estimate the male success of individual plants. These data were binary (remained or removed), as individual flowers have paired pollinia that are dislodged together. In 2007, we observed pollinia removal for open flowers at site A (554 flowers on 39 inflorescences) and site B (945 flowers on 117 inflorescences) on a single day. In 2008, we checked pollinia removal of almost all flowers on inflorescences (site A: 1104 flowers on 68 inflorescences). Fruit maturation of each flower was also recorded to estimate female success. Fruits were considered to be mature when they released mature seeds. We monitored fruit maturation at site A (720 flowers on 33 inflorescences) and site B (2409 flowers on 112 inflorescences) in 2007 and at site A (1342 flowers on 77 inflorescences) in 2008. In 2008, we were unable to collect data for pollinia removal and fruit maturation at site B because of widespread damage by wild boars during the preceding winter.

To determine the helical angle of each inflorescence (Figs 1, 2), we measured the number of flowers per single helical period in both years. Helical angle was then calculated as 360° divided by the number of flowers per single helical period (i.e. the average helical angle per inflorescence). When the angle was very small, and flowers within an inflorescence did not round the stem, we measured the cumulative angle among open flowers and divided this by the number of open flowers. We also recorded several variables that could potentially affect male and female success of individual flowers: the display size at the peak of anthesis, the size of each inflorescence and the position of each flower from the bottom of the inflorescence.

Data analyses

We used generalized linear mixed models (GLMMs; with binomial errors and a logit-link function) to examine the effects of three-dimensional inflorescence architecture and display size on pollinator attraction. In the full model, we chose helical angle and display size (the number of open flowers per inflorescence) as the explanatory variables, which did not correlate with each other during the three observation days (day 1: = 91, = −0.024, = 0.820; day 2: = 87, = 0.080, = 0.463; day 3: = 80, = 0.113, = 0.317). The response variable was the presence/absence (0/1) of a visit to each inflorescence by each visitor entering the site. Inflorescence identity and visitor identity (1–56) were considered random terms. Because we did not mark individual bees during our observations, the 56 bees that entered the site were considered different individuals in the analyses, even though some may have repeatedly visited the site. We applied these analyses only to data for leaf-cutting bees, as other pollinator taxa were so rare (< 2% of visits). Because the full models could include some explanatory variables that could potentially be excluded, model selection was conducted for this and the following models. We also constructed a set of alternative models excluding either or both helical angle and display size as explanatory variables. Models were compared using Akaike’s information criterion (AIC) values. The model with the lowest AIC was presented as the best model approximating the data.

We also examined the effects of three-dimensional inflorescence architecture and display size on intra-inflorescence pollinator movements using GLMMs (with Poisson errors and a log-link function). In the full models, helical angle and display size were considered explanatory variables, and inflorescence identity and visitor identity (1–56) were random terms. The response variable was the number of successive probes per visit or the number of skips per visit. The skip model did not include the number of successive probes per visit as an explanatory variable, potentially affecting the number of skips per visit for two reasons. First, our preliminary analysis using the GLMM, which included the number of successive probes per visit as a covariate of helical angle and display size, revealed that this variable had no significant effect on the number of skips: the 227.9 AIC value was higher than that of the full model (see Results), and the 95% confidence interval (CI) of its partial regression coefficients (− 0.080 to 0.164) included zero. Secondly, the variable was significantly influenced by both helical angle and display size (see Results). In the model for successive probes, we used data for inflorescences that received one or more visits, whereas data for inflorescences that received two or more successive probes were used in the skipping behaviour model. Model selection using AIC was also conducted for these models.

Finally, the effects of helical angle and inflorescence size on male and female success of each flower were tested using a model selection procedure based on AIC with GLMM (with binomial errors and logit-link function). Inflorescence identity was considered a random term. In the full models, helical angle, inflorescence size (total number of flowers per inflorescence) and flower position (e.g. the positions of the lowest and highest flowers were 0 and 10, respectively, in an inflorescence with 10 flowers) were considered explanatory variables. The response variable was the presence/absence (1/0) of pollinia removal or fruit maturation in each flower. Spiranthes sinensis flowers bloom sequentially from the bottom of each inflorescence; thus, the display size experienced by each flower should vary day by day. Hence, the average display size of a flower is the most appropriate parameter for evaluating the effect of display size on pollination of the flower. However, because our data were insufficient to calculate this parameter, we used inflorescence size as an indicator of the display size of each flower, as it was strongly correlated with display size when we assessed them at the peak of the flowering season (site A in 2007: = 38, = 0.458; site B in 2007: = 108, = 0.449; site A in 2008: = 80, = 0.520). Helical angle was not correlated with inflorescence size at either site in 2007 (site A: = 40, = 0.299, = 0.061; site B: = 108, = 0.115, = 0.237), but these variables showed a significant positive correlation at site A in 2008 (= 85, = 0.241, = 0.026). Our data showed flower position to be significantly correlated with inflorescence size, perhaps because more flower positions are always found in larger inflorescences (site A in 2007: = 0.394, df = 2416, < 0.001; site B in 2007: = 0.427, df = 2416, < 0.001; site A in 2008; = 0.553, df = 1498, < 0.001). Data from different sites and years were analysed separately.

Results

Pollinator behaviour

In all models, the effects of both helical angle and display size on pollinator behaviour were included in the best-fit models with the lowest AICs (Table 1; probability of visit, AIC = 3495; successive probes, AIC = 630.4; skips, AIC = 226.4). Furthermore, the 95% CIs of the partial regression coefficients for both explanatory variables did not include zero, with the exception of the coefficient for helical angle in the pollinator attraction model, which marginally included zero. The results indicated that display size had strong and positive effects on attractiveness, successive probes within an inflorescence and skips (Tables 1, S1). Once display size was accounted for statistically, a larger helical angle significantly reduced the number of flowers probed per visit and increased flower skips and tended to reduce the attractiveness of the inflorescence to bees (Table 1).

Table 1.   Estimated coefficients of explanatory variables in the best generalized mixed model for the presence of visitation to Spiranthes sinensis inflorescences, the number of successive probes per visit and the number of skips per visit
Response variableExplanatory variableCoefficient (95% CI)
  1. Boldface indicates that the 95% confidence interval (CI) for the partial regression coefficient did not include zero.

  2. A positive value for the coefficient indicates that the explanatory variable had a positive effect on the behaviour.

Probability of visitHelical angle0.006 (0.012, 0.000)
Display size0.125 (0.090, 0.161)
Intercept−2.901 (−3.341, −2.461)
Successive probesHelical angle−0.003 (−0.005, −0.001)
Display size0.071 (0.055, 0.087)
Intercept0.208 (0.040, 0.376)
SkipHelical angle0.016 (0.005, 0.026)
Display size0.169 (0.102, 0.236)
Intercept−4.107 (−4.912, −3.302)

Mating success

The rates of pollinia removal (total number of flowers with pollinia removal/total number of flowers examined) at sites A and B in 2007 and at site A in 2008 were 0.30, 0.13 and 0.43, respectively. In all of the best models for pollinia removal, all explanatory variables (helical angle, inflorescence size and flower position) were included. The estimated coefficients for helical angle and flower position were negative in all models, and the 95% CIs for the partial regression coefficients did not include or only marginally included zero, indicating that these two factors had a significantly negative effect on pollinia removal at site A in 2008 and marginally significant effects at both sites in 2007 (Table 2). Thus, pollinia removal from each flower decreased with increased helical angle and position of flower from the bottom of the inflorescence. Meanwhile, the estimated coefficients for inflorescence size and their 95% CIs were positive in almost all cases, indicating that larger inflorescences experienced more pollinia removal (Table 2). The relative impact of the three variables varied between sites, with that of helical angle greater than those of inflorescence size and flower position at site A (Table S1).

Table 2.   Estimated coefficients of explanatory variables in the best generalized mixed model for the pollen removal and fruit maturation of Spiranthes sinensis at sites A and B in 2007 and at site A in 2008
Response variableSite and yearExplanatory variableCoefficient (95% CI)
  1. The numbers of sampling inflorescences (i) and flowers (f) are shown. Boldface indicates that the 95% confidence interval (CI) for the partial regression coefficient did not include zero. AIC, Akaike’s information criterion.

Pollen removalSite A in 2007Helical angle0.012 (0.025, 0.002)
(i = 39, f = 554)Inflorescence size0.056 (0.005, 0.107)
(AIC = 619.3)Flower position−0.058 (−0.099, −0.017)
 Intercept1.230 (2.313, 0.148)
Site B in 2007Helical angle0.011 (0.024, 0.003)
(i = 117, f = 945)Inflorescence size0.029 (0.032, 0.091)
(AIC = 679.8)Flower position−0.109 (−0.168, −0.051)
 Intercept1.336 (2.757, 0.085)
Site A in 2008Helical angle−0.010 (−0.018, −0.002)
(i = 68, f = 1104)Inflorescence size0.053 (0.020, 0.087)
(AIC = 1420)Flower position−0.041 (−0.062, −0.020)
 Intercept0.526 (1.145, 0.093)
Fruit maturationSite A in 2007Helical angle0.014 (0.051, 0.023)
(i = 33, f = 720)Inflorescence size0.266 (0.123, 0.410)
(AIC = 523.5)Flower position−0.388 (−0.452, −0.323)
 Intercept0.511 (3.486, 2.464)
Site B in 2007Helical angle0.012 (0.024, 0.000)
(i = 112, f = 2409)Inflorescence size0.100 (0.045, 0.155)
(AIC = 1962)Flower position−0.128 (−0.147, −0.110)
 Intercept1.691 (3.069, 0.312)
Site A in 2008Helical angle0.002 (0.011, 0.016)
(i = 77, f = 1342)Inflorescence size0.201 (0.139, 0.262)
(AIC = 1316)Flower position−0.260 (−0.296, −0.225)
 Intercept0.770 (1.806, 0.266)

Fruit set ratios (total number of fruits/total number of flowers examined) at sites A and B in 2007 and at site A in 2008 were 0.54, 0.25 and 0.57, respectively. In all of the best models for fruit maturation, all explanatory variables (helical angle, inflorescence size and flower position) were included. However, the effect of helical angle was not consistent between years. Fruit maturation of each flower decreased with helical angle at both sites in 2007, whereas the reverse trend occurred at site A in 2008 (Table 2). At site B in 2007, the 95% CI for the partial regression coefficients of helical angle marginally included zero, indicating a substantial negative effect on fruit maturation at site B in 2008. At site A in 2007 and 2008, the 95% CIs for the partial regression coefficients of helical angle included zero, indicating that the effects of helical angle on fruit maturation were ambiguous at that site. By contrast, inflorescence size and flower position had consistent effects on fruit maturation at both sites A and B in 2007 and at site A in 2008, such that larger inflorescences had a higher probability of fruit maturation and upper flowers produced fewer fruits (Table 2). The relative impact of flower position on fruit maturation was larger than those of helical angle and inflorescence size (Table S1).

Discussion

In the present study, we successfully demonstrated that helical angle affected pollinator behaviour and consequent reproductive success in a natural population of S. sinensis. This study is the first to document that natural variation in inflorescence architecture influences pollinator behaviour and reproductive success, suggesting that inflorescence architecture can evolve under pollinator-mediated natural selection in plant populations. In general, the responses of leaf-cutting bees to the inflorescence architecture in S. sinensis were consistent with the predicted patterns: an increase in helical angle reduced the number of visits per inflorescence, the number of successive probes within inflorescences and the consistency of foraging path within inflorescences. We now consider the implications of these effects for plant reproduction and evolution.

Pollinator attraction

We found that pollinator attraction decreased with helical angle (Table 1). This result implies that the higher floral density in loosely twisted S. sinensis inflorescences poses an advantage in terms of reproductive success, as frequent visits promote pollen import and export and mate diversity (Harder & Barrett, 1996). Floral density correlates with two intrinsic properties of inflorescences, that is, colour contrast to the background and display area visible from a certain direction, both of which may affect the visibility of inflorescences and consequently the probability of a pollinator finding inflorescences. In S. sinensis, these visual traits probably function collectively to induce a variety of visitation rates to inflorescences with different helical angles.

By removing and tying flowers, Fishbein & Venable (1996) manipulated the inflorescences of A. tuberosa and found that floral density affected the visitation rate of pollinators, which is consistent with our results. However, the two studies differ in terms of the optimum design for pollinator attraction: the highest density of flowers attracted the most pollinators in S. sinensis, whereas an intermediate floral density was most attractive in A. tuberosa. One possible cause for this discrepancy is the difference in the range of floral density offered in the two studies. The range of floral density of S. sinensis inflorescences was relatively small compared with that of the experimentally prepared A. tuberosa inflorescences, as the former included only natural variants and the latter were artificially rearranged. If the natural variation of S. sinensis inflorescences offers only a portion of the floral density presented by manipulated A. tuberosa inflorescences, the response of pollinators to the two plants would probably differ. Alternatively, the optimum design for attractiveness may vary with conditions such as plant–pollinator pairing.

Successive probes

We found that successive probes on inflorescences decreased with helical angle, suggesting that inflorescences with larger inter-flower distances received fewer probes per visit (Table 1). This result implies that inflorescence architecture affects opportunities for geitonogamous pollination in S. sinensis. Generally, an increase in intra-inflorescence movements of a pollinator reduces the amount of pollen export to other plants (pollen discounting: Harder & Barrett, 1996), the number of outcrossed seeds (seed discounting: Lloyd, 1992) and the average offspring performance as a result of inbreeding depression (Charlesworth & Charlesworth, 1987). These disadvantages are potentially eliminated by spatial and temporal sexual segregation (e.g. dioecy, monoecy, synchronous dichogamy and protandry with acropetal blooming) when pollinators visit female flowers before male flowers. However, even though S. sinensis has protandrous flowers with an acropetal-blooming schedule, these negative effects of successive probes would not be eliminated because the protandry of each flower is attributable to the temporal change in the position of the column relative to the upper tepals and lip of flowers, such that, in the male phase, the space between the column and the upper petal and sepals is larger than that between the column and the lip, and vice versa in the subsequent female phase (Coleman, 1933). Because these structural changes progress gradually over days, individual flowers inevitably experience a bisexual period, allowing facilitated and geitonogamous self-pollination. Furthermore, even if all flowers blooming in an inflorescence are functionally male at an early developmental stage, pollen discounting could occur as a result of pollinia loss during successive probes. These facts suggest that, while protandry in S. sinensis has a limited effect on reducing pollinator-mediated self-pollination, a larger helical angle should be effective for lowering the cost of geitonogamous pollination, as it decreases successive probes within an inflorescence per visit.

Our findings support the energy-based hypothesis of Pyke (1982), which predicts that bees leave inflorescences when their instantaneous rate of energy gain falls below the average rates of energy gain for the habitat. The increased inter-flower distance in tightly twisted inflorescences would decrease the relative instantaneous rate of energy gain on the inflorescences, which may promote the early departure of bees from the inflorescences. Similar observations have been reported for an artificial inflorescence–hummingbird system (Hainsworth et al., 1983) and an Iris–syrphid fly system (Ishii & Sakai, 2001).

Skip behaviour (number of skips of neighbouring flowers per visit) increased with helical angle (Table 2), indicating that bees used relatively diverse foraging paths on inflorescences with larger helical angles. As stated above, the distance between neighbouring flowers was relatively large in tightly twisted inflorescences, in which neighbouring flowers were not located vertically. By contrast, tightly twisted arrangements offer several spatially close, non-neighbouring flowers, some of which are located vertically (Fig. 1). Relatively diverse foraging paths in tightly twisted designs may be attributable to these structural alternations. The consistency of foraging routes would affect the distribution of rewarding flowers within the inflorescences, which could affect the intra-inflorescence behaviours of subsequent visitors, as bees are thought to determine whether an inflorescence is currently rewarding based on experience probing only a few flowers, especially on vertical inflorescences on which bees routinely follow an upward path (Pyke, 1978; Waddington, 1981). Diverse foraging paths in tightly twisted designs can result in less predictable reward distribution within an inflorescence, which in turn may cause subsequent visitors to leave inflorescences sooner. Thus, tightly twisted inflorescences may also benefit from reduced geitonogamy by experiencing relatively inconsistent movements of pollinators.

Effects of inflorescence architecture on pollinia removal and fruit maturation

Loosely twisted arrangements exhibited a higher probability of pollinia removal at both sites in 2007 and 2008 (Table 2). These results were consistent with observed bee behaviours, considering that bees more frequently visited loosely twisted inflorescences and, while there, successively probed more flowers (Table 1); that is, individual flowers on loosely twisted inflorescences must receive more visits than those on highly twisted inflorescences. Thus, the probability of pollinia removal per flower probably reflects the pollinator-visitation rate per flower, indicating that inflorescence architecture affects male reproductive success in S. sinensis. However, a higher probability of pollinia removal per flower is not necessarily related to advantages of loosely twisted inflorescences, as higher pollinia removal is attributable in part to increased numbers of successive probes, which can induce pollen and seed discounting and inbreeding depression in selfed offspring.

The effect of helical angle on fruit maturation was ambiguous at site A, although a marginally negative effect was observed at site B. Fruit set at site A was higher than that at site B, suggesting that the effect of helical angle on fruit set became more obvious under pollen-limited conditions.

Whereas the effects of helical angle on pollinia removal and fruit maturation differed depending on pollination conditions, the effects of display size and flower position on pollinia removal and fruit maturation were consistent. More specifically, flowers on larger inflorescences and flowers at lower positions exhibited higher probabilities of pollinia removal and fruit maturation. One possible reason for this observation is that large display sizes attract more pollinators per flower, and flowers at lower positions receive more visits than those at upper positions; such a visitation gradient may yield display size- and position-dependent pollinia removal and fruit maturation. Alternatively, size- and position-dependent patterns of fruit maturation may be regulated by resource limitation under pollinator-abundant conditions. Many studies have demonstrated size- and position-dependent fruit maturation in relation to resource availability for individual flowers (e.g. Lloyd & Bawa, 1984; Diggle, 1995).

Fruit set was higher than the pollinia removal rate in all cases, probably because the granular pollinia of S. sinensis from a single flower can be distributed to many flowers, unlike the hard pollinia that are most common among orchids (Dafni et al., 2000). This trait of the pollinia may mitigate the effect of pollinator limitation on fruit set in Ssinensis.

Adaptive significance of inflorescence architecture

The study population of S. sinensis exhibited various inflorescence architectures (Figs 1, S2). The most probable explanation for this variation is that the benefits and costs of mating success for each inflorescence type are balanced. Loosely twisted floral arrangements would have an advantage of increased attractiveness, which could result in enhanced pollen import and export, and mate diversity. However, this advantage of loosely twisted inflorescences would be counterbalanced by increased successive probes arising from a decreased inter-flower distance, which could cause geitonogamy (Harder & Barrett, 1996). The reverse is true for tightly twisted inflorescences. Thus, these conflicting factors may maintain the diverse inflorescence architectures in S. sinensis populations. Another possibility is that the selective advantage of a certain inflorescence type is ecologically context-dependent. For example, twisting decreases the opportunity for geitonogamous pollination under pollinator-abundant conditions, whereas decreased twisting should be advantageous when pollinators are rare, as long as the benefit of increased assurance of some reproductive output mitigates the negative effects of geitonogamy. If pollinator abundance varies temporally and spatially, diverse inflorescence architectures would be maintained.

As observed for S. sinensis, flowering plants might face a dilemma between the regulations of pollinator attraction and successive probes. We found that display size also posed a similar dilemma, in which larger displays increased pollinator attraction, and consequently pollinia removal and fruit set, but also enhanced successive probes within inflorescences in S. sinensis (cf. Ishii & Harder, 2006). A balanced point of the conflict would depend on the environments and characteristics of individual plants, such as pollinator abundance, self-compatibility and sexual segregation. In this situation, the fitness of individuals with different display architectures could vary with plant condition and consequently allow several optimal strategies in plants. This context-dependent suitability of floral arrangement may have driven the evolution of interspecific diversity in inflorescence architecture in angiosperms.

Acknowledgements

We thank an anonymous reviewer for kindly giving us many constructive comments and suggestions during the reviewing process. We very much appreciate the help of C. Kamei, Y. Iwata, J. Iwata, M. Iguchi, S. Kenji, Y. Uematsu and M. Tsuji with data collection. Identifications of observed bees were kindly provided by M. Kato. This research was partly supported by a grant-in-aid for young scientists (No. 17770023) from the Japan Society for the Promotion of Science.

Ancillary

Advertisement