Pistillate flowers experience more pollen limitation and less geitonogamy than perfect flowers in a gynomonoecious herb

Authors


Summary

  • Gynomonoecy, a sexual system in which plants have both pistillate (female) flowers and perfect (hermaphroditic) flowers, occurs in at least 15 families, but the differential reproductive strategies of the two flower morphs within one individual remain unclear.
  • Racemes of Eremurus anisopterus (Xanthorrhoeaceae) have basal pistillate and distal perfect flowers. To compare sex allocation and reproductive success between the two flower morphs, we measured floral traits, pollinator preferences, and pollen movement in the field.
  • Pollen limitation was more severe in pistillate flowers; bee pollinators preferred to visit perfect flowers, which were also capable of partial self-fertilization. Pollen-staining experiments indicated that perfect flowers received a higher proportion of intra-plant pollen (geitonogamy) than pistillate flowers. Plants with greater numbers of pistillate flowers received more outcross pollen. The differential reproductive success conformed with differential floral sex allocation, in which pistillate flowers produce fewer but larger ovules, resulting in outcrossed seeds.
  • Our flower manipulations in these nectarless gynomonoecious plants demonstrated that perfect flowers promote seed quantity in that they are more attractive to pollinators, while pistillate flowers compensate for the loss of male function through better seed quality. These results are consistent with the outcrossing-benefit hypothesis for gynomonoecy.

Introduction

The gender of flowers can be variable within an individual flowering plant, with various combinations of pistillate (female), staminate (male), and/or perfect flowers producing diverse sexual systems. While 72% of plant species have only perfect (hermaphroditic) flowers, the others have at least one type of unisexual flower (Yampolsky & Yampolsky, 1922; Charnov, 1982). Studies of the evolution and maintenance of two types of flower within an individual began with Darwin (1877). Compared with andromonoecy (staminate and perfect flowers within one plant) and monoecy (separate staminate and pistillate flowers on the same plant), the adaptive advantage of gynomonoecy (pistillate and perfect flowers on the same plant) remains little studied. Gynomonoecy occurs in at least 15 families, accounting for 2.8–4.7% of flowering plant species (Yampolsky & Yampolsky, 1922; Lu & Huang, 2006).

Several hypotheses have been proposed to explain the evolution and maintenance of gynomonoecy (Bertin & Kerwin, 1998; Lu & Huang, 2006; Bertin et al., 2010). First, as in other systems with unisexual flowers, the presence of the two flower types may permit flexible allocation of resources to female and male reproductive functions in response to variation in environmental factors (Charnov & Bull, 1977; Lloyd, 1979; Willson, 1983). However, experimental tests of this sex allocation hypothesis in Asteraceae showed that the floral sex ratio of the capitula was stable, with little plasticity in various environments (Abbott & Schmitt, 1985; Bertin & Kerwin, 1998; Bertin et al., 2010; Zhang et al., 2012). Secondly, pistillate flowers may outcross more than perfect flowers. In the self-compatible species Senecio vulgaris, Marshall & Abbott (1984) demonstrated that the outcrossing rate of pistillate ray florets of the radiate morph was significantly greater than that of perfect disc florets of either the radiate or nonradiate morph. However, most or all Aster species appear to be self-incompatible, making the hypothesis of outcrossing unimportant (Bertin & Kerwin, 1998). Thirdly, in the absence of clear evidence in Asteraceae in support of the above two hypotheses concerning gynomonoecy, Bertin & Kerwin (1998) proposed the hypothesis that the advantage of the pistillate flowers in aster heads lies in their attractiveness to pollinators (Leppik, 1977). Fourthly, pistillate flowers would be favored if perfect flowers are susceptible to biased predation by florivores (Wise et al., 2008; Bertin et al., 2010; Zhang et al., 2012).

Current understanding of the adaptive advantages of gynomonoecy is largely limited to the Asteraceae, although a few studies have considered species in other families. Using electrophoretic analysis, Collin & Shykoff (2003) estimated the outcrossing rate in pistillate flowers on female and gynomonoecious plants and perfect flowers on hermaphroditic and gynomonoecious plants in Dianthus sylvestris (Caryophyllaceae). In contrast to expectation, pistillate flowers were observed to be more frequently selfed than perfect flowers in gynomonoecious plants. Perfect flowers were highly outcrossed, perhaps as a consequence of pollinators preferring to visit the larger perfect flowers (Collin & Shykoff, 2003). Perfect flowers in Silene noctiflora were observed to be capable of autonomous selfing, providing reproductive assurance under conditions of pollinator limitation, while pistillate flowers could enhance the likelihood of producing outcrossed seeds (Davis & Delph, 2005). Gynomonoecy has evolved several times in flowering plants via the invasion of plants able to produce both pistillate and perfect flowers into populations containing plants that produce only perfect flowers (Lu & Huang, 2006; Bertin et al., 2010).

To understand the significance of the presence of pistillate flowers in hermaphrodite plants, we investigated the sexual system of Eremurus anisopterus, a perennial desert herb. Here we report that in E. anisopterus large individuals are gynomonoecious, with both pistillate and perfect flowers, an unknown sexual system in the family Xanthorrhoeaceae, in which species are generally hermaphroditic. We used a combination of pollinator observation, flower manipulation and pollen-staining experiments to address the following questions. Do pistillate flowers reallocate resources to female function, compensating for the loss of pollen production? Do pistillate and perfect flowers differ in attractiveness to pollinators? Do pistillate flowers experience more severe pollen limitation than perfect flowers? Is the species capable of self-fertilization as in S. noctiflora (Davis & Delph, 2005), in which perfect flowers provide reproductive assurance under a scarcity of pollinators? Do pistillate flowers experience less geitonogamy than perfect flowers?

Materials and Methods

Eremurus anisopterus Regel. (Xanthorrhoeaceae, formerly Liliaceae) is a perennial, early-spring, ephemeral herb, mainly distributed in sand dunes in Kazakhstan and northwestern China (Chen et al., 2000). Each flowering individual produces an erect raceme with dozens of flowers, emerging from a rosette of basal leaves. It flowers from late April to early May and seeds mature in early June. The flowers are nectarless and bowl-shaped, with six pinkish to white petals (Fig. 1). A single flower lasts 1–2 d. Perfect flowers have six boldly yellow anthers that are offered as the only reward for pollinators. We observed that some of the larger plants (c. 15%) are gynomonoecious with pistillate (male-sterile) flowers at the bottom of racemes that possess six flat white anthers or just vestigial white filaments (Fig. 1). Our field work was conducted in a large population with thousands of individuals located in the Gurbantunggut Desert in Xinjiang Uyghur Autonomous Region in northwestern China (44°42′37.0″N, 86°02′02.8″E, 344 m above sea level).

Figure 1.

Habitat, plants, racemes and flowers of Eremurus anisopterus. (a) A flowering population of E. anisopterus in the typical habitat of fixed dune; (b) a halictid bee collecting pollen from perfect flowers; (c) pistillate flowers with aborted stamens. Flowers are nectarless with green bands at the base of petals.

Floral allocation

To compare flower size between pistillate and perfect flowers in gynomonoecious individuals, we measured the corolla-mouth diameter of one fully open pistillate and one fully open perfect flower from 70 plants using digital calipers to 0.01 mm at about noon during peak-flowering periods. To measure resource allocation to floral organs, dry weight, pollen production and ovule production were compared in the two flower morphs. Thirty newly opened flowers (with anthers undehisced) of each flower morph were randomly collected from 30 gynomonoecious individuals. The petals, stamens and pistils were dried separately in an oven at 50°C for 48 h before weighing of their dry mass using an electronic balance (0.1 mg). Additionally, one pistillate and one perfect flower each from 24 gynomonoecious plants were collected to count ovules and pollen grains per flower. Anthers were split and pollen grains were suspended in 10 ml of water. Ten drops of the pollen grain solution were then counted under a microscope and pollen production per flower was estimated. In these flowers, the length and width of four randomly selected ellipsoid ovules per flower were measured, given that ovule size looks different between pistillate and perfect flowers. To compare seed size, one pistillate and one perfect flower were randomly selected from each of 20 gynomonoecious plants. Five seeds were then randomly selected from each flower and the dry mass per seed was measured (to 0.1 mg). Independent t-tests were used to compare the above measurements between pistillate and perfect flowers.

Mating system

To examine the mating system in E. anisopterus, we randomly labeled 60 pistillate and 120 perfect flower buds (with distinctive yellow anthers) on 60 individuals and enclosed some of them in fine nylon mesh nets to exclude flower visitors. Each of these flowers was randomly assigned to one of the following four pollination treatments (30 buds for each group): (1) flowers emasculated and bagged to examine the possibility of apomixis; (2) flowers emasculated without bags to examine the possibility of insect pollination; (3) flowers bagged without emasculation to examine self-compatibility and the possibility of autogamy; (4) flowers open-pollinated as controls. Fruits were collected 4–5 wk later and seed set per flower was calculated. Given that pistillate flowers produce sterile anthers, they were not emasculated. We compared seed set between pistillate and perfect flowers under different treatments using one-way ANOVA. Significant differences in post-hoc multiple comparisons were identified using Tukey's method.

Pollinators and preference

To examine the effect of floral morphs on pollinator visitation, we compared pollinator visits to four different floral displays (Huang et al., 2006). Individuals of E. anisopterus generally have two to four flowers opening daily. We constructed four-flower displays by manipulating flower morphs in racemes from 12 plants at similar height: four perfect flowers, two perfect flowers in a distal position and two pistillate flowers in a basal position, two pistillate flowers in a distal position and two perfect flowers in a basal position, and four pistillate flowers in one plant. Opening flowers or anthers from hermaphrodite flowers were removed from the plants if necessary to construct these four-flower displays. These racemes were manipulated early in the morning before pollinators visited. Four observers recorded insect visitors to each group of displays for a total of at least 45 h. Only visits by floral visitors that collected pollen or contacted stigmas were recorded. New floral displays were constructed on the 12 manipulated plants over 5 d, giving a total of 60 racemes (5 d ×4 treatments ×3 racemes per day). Visitation rate was calculated separately for bees and flies and then pollinator preferences between floral morphs and treatments were compared by one-way ANOVA.

Tracking pollen movement

To examine whether differential pollination occurs in the two flower morphs, we compared pollen movements in pistillate and perfect flowers through a pollen-dyeing experiment. In the morning from 2 to 6 May 2012, we used clean forceps dipped into dye solution to stain pollen in newly dehisced anthers (Huang & Shi, 2013). Pollen grains within anthers were stained red, green, and violet with aqueous solutions of safranine (1%), fast green (1%), and crystal violet (1%), respectively. Every day we randomly selected three to six small patches of E. anisopterus that were at least 100 m apart to minimize pollen transfer from the same colorization. In each patch, pollen grains in newly opened flowers of three adjacent individuals were stained with three types of dye in the morning. Within each patch, one to four flowers on one raceme were manipulated to be pistillate by removing anthers from perfect flowers if necessary to examine the effect of pistillate flower production on geitonogamy. We harvested a total of 152 stigmas from 39 pistillate flowers and 113 perfect flowers in the 19 patches (57 plants) at sunset on 5 d. The gender and number of flowers on each plant were recorded, and pollen deposition on each stigma was observed immediately under a microscope. Pollen grains were counted, including stained pollen with different colors and natural pollen. Stained pollen of the same color deposited on stigmas within a plant is from geitonogamous pollination, while different colored pollen deposited on stigmas must have come from different individuals and demonstrates cross-pollination. To investigate the effect of flower morph on geitonogamy, we divided 57 plants with stained pollen into three groups: no pistillate flowers (perfect flowers only; 27 plants), a low ratio of pistillate flowers (< 0.5; 21 plants) and a high ratio of pistillate flowers (≥ 0.5; nine plants).

To investigate the effect of flower gender and pistillate flower ratio on pollen receipt, we used general linear models (GLMs). Pollen counts were transformed (loge(x + 1)) before GLM analysis. Models included transformed pollen count as the dependent variable, and fixed effects of flower gender and pistillate flower ratio and their interaction. We also added daily display size into the models as a covariate. We ran GLM analysis four times for total pollen receipt, geitonogamous pollen receipt (the color of pollen received was the same as the pollen color of the stained recipient plant), outcross pollen receipt (the color of pollen received was different from the pollen color of the stained recipient plant) and the proportion of the same color pollen on the stigma.

All data analyses were conducted in spss, version 19.0 (SPSS Inc., Armonk, NY, USA). All means are presented ±SE.

Results

Floral allocation

Mean corolla diameter and the dry weight of petals were not significantly different between pistillate and perfect flowers (Table 1). While pistils of pistillate flowers were not significantly heavier than those of perfect flowers, stamens of pistillate flowers were significantly lighter than those of hermaphrodite flowers, consistent with the male sterility in pistillate flowers (Table 1). No visible pollen grains were observed in pistillate flowers (Fig. 1). Perfect flowers produced 121 320 ± 37 680 pollen grains per flower. Ovule number was lower in pistillate flowers than in perfect flowers, but ovule length and width were significantly larger in pistillate flowers than in perfect flowers ( 1).

Table 1. Comparison of corolla size, dry weight of different floral organs, ovule number and size (mean ± SE) between two flower morphs in Eremurus anisopterus
Flower morphCorolla diameter (mm)Dry weight of floral organ (mg)Ovule numberOvule size (mm)
PetalsStamensPistilsLengthWidth
  1. Significant differences are shown in bold type.

Perfect20.10 ± 0.1514.20 ± 0.581.74 ± 0.076.34 ± 0.1618.3 ± 0.20.77 ± 0.020.38 ± 0.00
Pistillate20.34 ± 0.2114.59 ± 0.650.48 ± 0.036.83 ± 0.2216.8 ± 0.40.90 ± 0.010.52 ± 0.06
T 1.0680.57216.7341.7753.0443.1298.221
P 0.2880.571 < 0.001 0.085 0.006 < 0.001 0.002

Mating system

Both the bagged pistillate and the bagged perfect flowers under emasculation treatment set no seeds, indicating no apomixis in E. anisopterus. Seed set was significantly lower (Z = −32.42; = 0.004) in bagged perfect flowers without emasculation (autogamy) than in open-pollinated, emasculated perfect flowers (xenogamy; Fig. 2), indicating that this species was capable of partial autonomous self-fertilization but seed production largely depended on insect-mediated pollination. Seed set of open-pollinated pistillate flowers was significantly lower (Z = −59.02; < 0.001) than that of open-pollinated perfect flowers, indicating that pollen limitation was more severe in pistillate flowers than in perfect flowers. Seed set of open-pollinated pistillate flowers was also significantly lower (Z = −34.27; = 0.001) than that of open-pollinated, emasculated perfect flowers, suggesting that early-flowering pistillate flowers were less attractive to pollinators than emasculated perfect flowers in this species. However, the mass of individual seeds from pistillate flowers (14.9 ± 0.1 mg) was significantly greater than that from perfect flowers (12.7 ± 0.1 mg; = 12.865; < 0.001).

Figure 2.

A comparison of seed set between pistillate and perfect flowers (mean ± SE) under different pollination treatments in Eremurus anisopterus (F3,59 = 14.726; < 0.001). Different letters above means indicate significant differences.

Pollinator preference

Halictid bees (Lasioglossum niveocinctum and Halictus elegans) and honeybees (Apis mellifera) were the most frequent floral visitors, while syrphid flies (Syrphus pyrastri and Helophilus pendula) visited E. anisopterus at very low frequency. Bees significantly preferred perfect flowers over pistillate flowers (F1,268 = 5.914; = 0.016; mean visits per flower per hour 1.34 ± 0.11 versus 0.97 ± 0.10, respectively). Syrphid flies did not discriminate between perfect and pistillate flowers (F1,268 = 0.461; = 0.498; visits per flower per hour 0.14 ± 0.05 versus 0.11 ± 0.03, respectively).

Different floral displays had a significant effect on visitation rate for bees but not for flies (Fig. 3; F3,266 = 14.718; < 0.001 for bees; F3,266 = 0.936; = 0.424 for flies). With an increase in the proportion of pistillate flowers in a display, bee visit frequency decreased (Fig. 3). Interestingly, we observed that bees' visitation was influenced by the relative position of pistillate and perfect flowers on the racemes. Gynomonoecious displays with two perfect flowers in the upper position received significantly more visits than displays with two pistillate flowers in the upper position. Racemes with perfect flowers both basally and distally received significantly more visits than racemes with basal pistillate flowers.

Figure 3.

A comparison of visitation rate by bees (closed bars) and syrphid flies (open bars) to four-flower displays of Eremurus anisopterus with four different combinations of the two flower morphs. Flower position on the racemes and flower morph are shown in diagrams above bars (P, perfect flower; F, pistillate flower). Different letters indicate significant differences (< 0.05).

Geitonogamy

Display size did not significantly affect either pollen receipt or the proportion of same-colored pollen deposition. We then ran all the models without display size as a covariate. Perfect flowers received significantly more pollen grains (including geitonogamous pollen) than pistillate flowers (Fig. 4a,b, Table 2), but outcross pollen receipt (the number of different-colored pollen grains) was not significantly different between pistillate and perfect flowers (Fig. 4c, Table 2). Similarly, flower morph affected the proportion of geitonogamous pollen receipt. The proportion of same-colored pollen on stigmas was also significantly lower in pistillate than in perfect flowers (Table 2), although both total pollen receipt and same-colored pollen receipt were lower in pistillate flowers. There were no significant effects of the ratio of pistillate flowers on total pollen receipt, number of same-colored pollen grains received, and the proportion of same-colored pollen receipt per flower (Fig. 5a,b, Table 2). However, flowers from displays with a high ratio of pistillate flowers received significantly more different-colored pollen than the other types of floral displays (Fig. 5c, Table 2).

Table 2. General linear model: effect of flower morph (pistillate versus perfect), pistillate ratio (no pistillate flowers, a low proportion of pistillate flowers or a high proportion of pistillate flowers) and their interaction on pollen receipt (total pollen, same-colored pollen and different-colored pollen receipt) and the ratio of same-colored pollen on stigmas in Eremurus anisopterus
Source of variationdfMS F P
  1. Data for pollen receipt were transformed (loge(x + 1)) before GLM analysis. Effects with < 0.05 are shown in bold type.

Total pollen receipt
Display size10.0740.0730.787
Sex17.9327.904 0.006
Pistillate ratio20.6750.6730.512
Sex × pistillate ratio10.1450.1450.704
Error1461.004  
Same-colored pollen receipt
Display size10.3430.1330.716
Sex166.20825.611 < 0.0001
Pistillate ratio22.3170.8960.410
Sex × pistillate ratio11.2620.4880.486
Error1462.585  
Different-colored pollen receipt
Display size11.9924.2050.042
Sex10.0000.0000.997
Pistillate ratio22.5125.300 0.006
Sex × pistillate ratio10.3200.6740.413
Error1460.474  
Ratio of same-colored pollen on stigmas
Display size10.0370.2630.609
Sex11.98914.125 < 0.0001
Pistillate ratio20.2341.6620.193
Sex × pistillate ratio10.2291.6270.204
Error1460.141  
Figure 4.

The effects of floral morph (pistillate versus perfect flowers) on pollen receipt (mean number of pollen grains ±SE) detected by staining pollen in anthers of Eremurus anisopterus. (a) Total pollen receipt, (b) same-colored pollen receipt (i.e. geitonogamy), and (c) different-colored pollen receipt (outcrossing). All data for pollen receipt were transformed (loge(x + 1)). Different letters indicate significant differences (< 0.05).

Figure 5.

The effects of pistillate flower ratio (no pistillate flowers, a low proportion of pistillate flowers or a high proportion of pistillate flowers) on pollen receipt (mean number of pollen grains ±SE) detected by staining pollen in anthers of Eremurus anisopterus. (a) Total pollen receipt, (b) same-colored pollen receipt (i.e. geitonogamy), and (c) different-colored pollen receipt (outcrossing). All data for pollen receipt were transformed (loge(x + 1)). Different letters indicate significant differences (< 0.05).

Discussion

Our study on E. anisopterus demonstrated that pistillate flowers experienced greater pollen limitation and enhanced female function through producing fewer but larger ovules (seeds), while perfect flowers received more pollinator visits but showed a higher proportion of geitonogamy. Individuals with more pistillate flowers received a higher proportion of outcross pollen and less geitonogamous pollen than those with more perfect flowers. Hence, the presence of pistillate flowers could help reduce selfing rates, in support of the outcrossing hypothesis of gynomonoecy.

Female flowers are generally smaller than perfect flowers in gynomonoecious and gynodioecious species (Darwin, 1877; Baker, 1948; Delph, 1996; Méndez & Munzinger, 2010). In two wind-pollinated species, Lactoris fernandeziana (Bernardello et al., 1999) and Rhoiptelea chiliantha (Sun et al., 2006), pistillate flowers were significantly smaller than perfect flowers, but the two lamelliform stigmas of pistillate flowers in the latter species were wider. Flower size dimorphism was also observed in the newly recognized gynomonoecious species Planchonella endlicheri, in which the dry mass of corollas and gynoecium of pistillate flowers was lower than that of perfect flowers. In the gynomonoecious annual Silene noctiflora, pistillate flowers are approximately the same size as perfect flowers, but in pistillate flowers the styles extend outward from the floral tubes, while the styles of most perfect flowers do not extend beyond the opening (Folke & Delph, 1997). We observed that corolla diameter and petal weight were not significantly different between pistillate and perfect flowers in E. anisopterus. Notably, ovules were fewer but larger in pistillate flowers, which corresponds to more severe pollen limitation and a higher proportion of outcross-pollen deposition in pistillate than perfect flowers. This pattern of floral allocation of two flower morphs is consistent with the prediction of the mating environment hypothesis (Brunet & Charlesworth, 1995; Huang et al., 2004), which proposes that sequentially blooming flowers within plants could adjust sex allocation in response to pollination opportunity to maximize reproductive success.

It has been postulated that pistillate flowers should reallocate resources to increase female function to compensate for loss of male function, while perfect flowers must allocate resources to pollen (Darwin, 1877; Lewis, 1941; Lloyd, 1976; Charlesworth & Charlesworth, 1978; Williams et al., 2000). Some species described as gynodioecious are truly gynomonoecious-gynodioecious, having three phenotypes: females, perfect-flowered hermaphrodites and gynomonoecious individuals. In one such species, Silene nutans, Dufay et al. (2010) observed that females had smaller and lighter flowers than hermaphrodites, with pistillate and perfect flowers being intermediate and not significantly different within the gynomonoecious plants. In another gynomonoecious-gynodioecious species, Dianthus sylvestris, Collin & Shykoff (2003) observed that the presence of pistillate flowers increased the outcrossing rate of perfect flowers because pollinators preferred to visit the larger perfect flowers before visiting the smaller pistillate flowers in gynomonoecious plants. If pollinators visit perfect flowers first in gynomonoecious plants, pistillate flowers may experience higher geitonogamy than they would if pollinators visited pistillate flowers first. In Asteraceae, the family with the most known gynomonecious species, insects generally visit peripheral ray florets in a capitulum before moving to the central hermaphrodite disc florets (Willson, 1983; Cheptou et al., 2001; Gibson & Tomlinson, 2002), which may help to reduce geitonogamy and pollen–pistil interference (Bertin & Kerwin, 1998).

To our knowledge, the only other study in which pistillate flowers have been shown to experience more severe pollen limitation than perfect flowers in gynomonoecious species was that on Silene noctiflora, in which perfect flowers were capable of autonomous selfing, conferring reproductive assurance under a scarcity of pollinators (Davis & Delph, 2005). Our pollination treatments indicated that pollen receipt as well as seed set of pistillate flowers was significantly lower than that of perfect flowers, as a consequence of pollinators significantly preferring perfect over pistillate flowers in E. anisopterus. Because perfect flowers provide a pollen reward but female flowers do not in this nectarless species, insect visits to pistillate flowers can be regarded as deceit pollination (Baker, 1976), given that the corolla sizes of the two flower morphs are similar. The seed set of open-pollinated pistillate flowers was also significantly lower than that of open-pollinated, emasculated perfect flowers, perhaps because pistillate flowers opened earlier in the flowering season when pollinators were very scarce.

Our pollen-tracking experiments using three colors of pollen dyes showed that perfect flowers received a higher proportion of same-colored pollen from the same plants (i.e. geitonogamous pollination) than pistillate flowers. If there were more pistillate flowers on one plant, the proportion of outcross pollen received would be higher than it would be if there were few pistillate flowers. These results indicate that the occurrence of pistillate flowers largely reduced geitonogamy and enhanced outcrossing in these gynomonoecious plants. Perfect flowers with a pollen reward received significantly more visits by bee pollinators than pistillate flowers. Such an unfavorable pollination environment for pistillate flowers may limit the extent to which E. anisopterus can increase female function through an increase of seed number. We observed that pistillate flowers produced fewer but larger ovules and received a higher proportion of outcross pollen than perfect flowers, suggesting that the gynomonoecious plants could increase female function via seed quality. Indeed, the seed mass of pistillate flowers was significantly greater than that of perfect flowers.

Pistillate flowers are at the base of racemes while perfect flowers are distal in E. anisopterus, suggesting that production of pistillate flowers is not a result of resource competition. Otherwise, the position of the two flower morphs should be reversed, with perfect flowers being basal. We found that gynomonoecious racemes with basal pistillate flowers received more bee visits than gynomonoecious racemes with basal perfect flowers, suggesting an adaptation of raceme architecture in this species. We noted that bees visiting flowers usually worked upward on racemes. Again, the racemes with basal pistillate flowers were more likely to produce outcrossed seeds.

The evolution and maintenance of gynomonoecy have rarely been studied in species outside of the Asteraceae (Davis & Delph, 2005; Méndez & Munzinger, 2010). The outcrossing hypothesis of gynomonoecy, that is, that pistillate flowers have a higher level of outcrossing than perfect flowers because of a lower proportion of nonself pollen deposition, has been questioned given that many Asteraceae species are self-incompatible (Bertin & Kerwin, 1998). We compared the level of geitonogamy between flower morphs using colored pollen to track pollen movement within and between individuals in the experimental population of E. anisopterus. Our pollinator observations and pollen-staining experiment indicated that pistillate flowers experienced higher pollen limitation but a lower proportion of geitonogamy. Further tests of the outcrossing-benefit hypothesis in the gynomonoecious system may involve direct genetic determination of outcrossing rates of perfect and pistillate flowers, and comparisons of germination and growth of seeds from the two flower morphs. We observed that pistillate flowers produced fewer but larger ovules and seeds. One would expect that pistillate flowers would benefit even more if they could produce a greater number of larger ovules and seeds. However, an increase of pistillate flowers or ovule production may increase the degree of pollen limitation in them, given that pistillate flowers were less attractive to pollinators and could not be autogamous as in perfect flowers. An investigation of floral sex ratio and reproductive success in wild populations of this desert herb may provide more insights into the maintenance of gynomonoecy.

Acknowledgements

We thank M. Xie, X. Li, and S-J. Liu for help in the field, and S. Corbet, L. Delph and A. Vey for valuable comments on the manuscript. This study was supported by National Science Foundation of China (U1130301 and 31030016), Xinjiang Agricultural University (XJCYB-2011-06), and the Key Program for International S & T Cooperation Projects of China (2011DFA31070).

Ancillary