Variation in sex ratio, morph-specific reproductive ecology and an experimental test of frequency-dependence in the gynodioecious Kallstroemia grandiflora (Zygophyllaceae)

Authors

  • E. CUEVAS,

    1. Facultad de Biología, Universidad Michoacana de San Nicolás de Hidalgo, Morelia, Michoacan, México
    2. Departamento de Ecología de la Biodiversidad, Estación Regional del Noroeste, Instituto de Ecología-UNAM, Hermosillo, Sonora, México
    Search for more papers by this author
  • I. M. PARKER,

    1. Department of Ecology and Evolutionary Biology, University of California, Santa Cruz, CA, USA
    Search for more papers by this author
  • F. MOLINA-FREANER

    1. Departamento de Ecología de la Biodiversidad, Estación Regional del Noroeste, Instituto de Ecología-UNAM, Hermosillo, Sonora, México
    Search for more papers by this author

Eduardo Cuevas, Facultad de Biología, Universidad Michoacana de San Nicolás de Hidalgo, Morelia, Av. Francisco J. Mújica S/N, C.P. 58030, Michoacán, México.
Tel.: (443) 3 16 74 12; fax: (443) 3 16 74 12; e-mail: ecuevas@ecologia.unam.mx

Abstract

An enduring puzzle in gynodioecious species is the great variation in female frequency seen among populations. We quantified sex ratio in 44 populations of gynodioecious Kallstroemia grandiflora. Then, we measured pollinator visitation, pollen deposition, autonomous selfing rate and pollen limitation of females. Finally, using experimental populations, we tested whether female fitness responds to the frequency of female plants. We found broad variability in sex ratio among populations (0–44% female). Hermaphrodite flowers received more pollinator visits and pollen grains than females, and bagged hermaphrodite flowers produced fruits. However, we found no evidence of pollen limitation in females. In experimental populations, female plants showed no evidence of frequency-dependent pollinator visitation, fruit set, seed set or total seed mass. These results do not support frequency-dependent variation in fitness as a major mechanism affecting female frequencies in K. grandiflora. Within the context of this study, pollinators are abundant and pollinator movement appears to operate at a large enough scale to overcome the potential reproductive disadvantages of producing solely female flowers.

Introduction

Plants show a wide range of reproductive strategies. Gynodioecy, in which hermaphrodites and females (male-sterile plants) coexist, is a special case of polymorphism that has captivated the attention of scientists for more than a century (Darwin, 1877; Kaul, 1988). As with other cases of polymorphism, explaining how variation in reproductive strategies is maintained presents a central challenge to our understanding of both selective mechanisms and other factors that shape the patterns of diversity.

Because females depend upon hermaphrodite pollen to reproduce, pollination ecology should play a major role in determining the patterns of relative reproductive success in gynodioecious species. When hermaphrodites are self-compatible, they have a potential advantage over females in terms of reproductive assurance. As a result, females may be more susceptible to pollen limitation than hermaphrodites (Maurice & Fleming, 1995; Ronce & Olivieri, 2004). In addition, in animal-pollinated species, the behavioural response of pollinators to differences between flower morphs in the production of rewards may strongly influence the evolution and maintenance of sex ratios (Neiland & Willcock, 1998; Castillo et al., 2002). Pollinators have been shown to prefer hermaphrodite flowers in a number of species, responding to different cues such as larger size (e.g. Eckhart, 1992; Asikainen & Mutikainen, 2005), the presence of stamens (Ashman et al., 2000) or higher nectar production (Delph & Lively, 1992).

If females have a tendency to be pollen limited, either because of a lack of reproductive assurance through selfing or because of discrimination by pollinators, they may be vulnerable to negative frequency-dependent selection.

When females are in low frequencies, pollen availability and female fecundity should be high; in fact, if females show reproductive compensation, their reproductive success will be higher than that of hermaphrodite plants (Shykoff et al., 2003). In contrast, as the relative frequency of females increases locally, we expect a decrease in the relative fitness of female plants, caused by increasing pollen limitation (McCauley & Taylor, 1997). Empirical studies of a number of gynodioecious species have found evidence for frequency-dependent variation in the reproductive success of females (McCauley & Brock, 1998; Graff, 1999; Williams et al., 2000). For example, with increasing relative female frequency, female fruit set decreased in Silene vulgaris (McCauley & Brock, 1998), and pollen deposition decreased in Geranium richardsoni (Williams et al., 2000). However, other studies have looked for evidence of frequency dependence and have not found it (Williams & Fenster, 1998; Alonso, 2005; Asikainen & Mutikainen, 2005). The importance of frequency-dependent selection for the maintenance of gynodioecy is still an unresolved question.

Kallstroemia grandiflora is an annual gynodioecious herb in which hermaphrodites are self-compatible, and both morphs attract pollinators with nectar (E. Cuevas, pers. obs.). Developmental abnormalities during pollen production suggest that male sterility in K. grandiflora shows a cytoplasmic component (Cuevas et al., 2005). A course-grain survey across a large proportion of the distributional range of K. grandiflora in Mexico found geographical variation in sex ratio, with male-sterile phenotypes reaching higher frequency in the northern part of the range (Cuevas et al., 2005).

In this study, we start with a detailed survey of sex ratios in natural populations of K. grandiflora within the Sonoran desert region of Mexico. Then, using a range of experimental and observational approaches, we investigate the potential role of pollinators in contributing to the relative fitness of hermaphrodite and male-sterile plants. We ask whether hermaphrodites show potential for full reproductive assurance through autonomous selfing, and whether flowers on female plants are pollen limited. Finally, we create experimental populations to evaluate whether frequency dependence is a major determinant of reproductive fitness components in females.

Methods

Study species

Kallstroemia grandiflora (Zygophyllaceae) ranges from the Sonoran and Chihuahuan deserts of the USA and Mexico, south to the semiarid west coast of Mexico along the limits of the states of Guerrero and Michoacán (Porter, 1969). This annual herb produces showy flowers during the rainy season from July to October and is a conspicuous feature of the desert landscape. It can be found in patches of a few hundreds plants, or in large populations of thousands of plants (E. Cuevas, pers. obs.). In the region that was the focus of the present study, most populations were small and widely spaced.

Each plant produces several flowers per day. Individual flowers of K. grandiflora last 1 day and are solitary and pentamerous, with each flower bearing 10 ovules and five nectariferous glands (Porter, 1969). Flowers from female plants have reduced anthers and smaller corollas (2.8 ± 0.51 cm; mean ± 1 SD) than the perfect flowers of hermaphrodites (4.4 ± 0.54 cm; Cuevas et al., 2005). In the hermaphrodite flowers, pollen release and stigma receptivity are simultaneous, and hermaphrodite flowers are self-compatible (Porter, 1969; Cuevas et al., 2005). Across the range of K. grandiflora, flowers are visited by more than 40 species of insects, mainly bees and wasps, but only about 10 of these species are considered effective pollinators (Cazier & Linsley, 1974, 1975; Osorio-Beristain et al., 1997).

Sex ratio variation

We estimated the frequency of male-sterile plants in 44 different populations within the state of Sonora during the flowering seasons of 1999 and 2000 (Fig. 1). In a subset of populations censused in multiple years, variation in sex ratio was low (r = 0.63, F1,4 = 11.07, P = 0.02); therefore data from the 2 years were combined. Sampled populations were located along roadsides, at least 3 km apart. In each population we sampled 200 plants, and the gender of each plant was determined using three or more open flowers. All statistical analyses were performed using jmp (SAS Institute, 1997). We performed regression analysis to test whether the association between female frequency and latitude observed by Cuevas et al. (2005) would also be evident within this single region.

Figure 1.

 Map showing the location and frequency of females for the 44 survey populations of Kallstroemia grandiflora in the Sonoran Desert of Mexico. The ‘x’ shows the location of the natural population where experimental populations were set up. At the right corner, the frequency distribution of females is shown.

Pollinator visitation and pollen deposition

In 2004, we carried out pollination studies in a natural population within the Plains of Sonora vegetational subdivision of the Sonoran Desert (Shreve, 1964) (28°59′N, 110°54′W). The frequency of female plants in this population was 20%. Observations were made over 3 days in 20-min periods between 9:00 and 12:00 (3 h total), when pollinators were most active. We selected pairs of adjacent hermaphrodite and female plants and observed one flower per plant. We defined a ‘visit’ as when the insect came in contact with either the anthers or stigma. We recorded the identity of each pollinator, number of visits and the gender of the visited flower. The time spent per flower was also recorded for a subset of observation periods. A paired t-test was used to analyse the number of visits, with observation period as the unit of replication, and a nested anova was used to test for differences between morphs for the time spent per flower, with visits nested within plants.

We then estimated the number of pollen grains deposited per insect visit on the stigmas of flowers, using five plants per gender and four flowers per plant. We bagged mature buds and emasculated half (n = 2) of the experimental buds of hermaphrodite plants. When each flower opened, we removed the bag and checked whether the stigma was free of pollen. After a single visit by the most frequent visitor (see Results), we removed the flower and counted the number of pollen grains on the stigma with a 10× magnifying glass. We calculated the proportion of autogamous self-pollen on hermaphrodite stigmas as [(control − emasculated)/control]. We also tested whether pollen deposition on control flowers was significantly greater than that on emasculated flowers (anova with plants as blocks). Then we compared pollen deposition for female vs. hermaphrodite flowers in the absence of autogamous pollen from within the same flower: we used nested anova (flowers nested within plants) to compare the number of pollen grains deposited on stigmas of emasculated hermaphrodite flowers with the number of pollen grains deposited on female flowers.

Autonomous self-pollination and pollen limitation

In the same year, we explored two aspects of K. grandiflora reproductive ecology that are important in determining the role of pollinators in the relative fitness of hermaphrodites and females. First, we assessed the ability of hermaphrodites to self in the absence of pollinators (autonomous self-pollination). On 10 hermaphrodite plants, four flowers were bagged starting the day before anthesis, with four unbagged flowers on the same plants as controls. Second, we estimated the degree of pollen limitation experienced by female flowers. On 10 female plants, four flowers were hand-pollinated with a fresh anther from a random hermaphrodite, and four control flowers were open-pollinated. Counts of developing fruits 1 week after pollination treatments were used to estimate fruit set (i.e. the proportion of flowers that result in a fruit), and mature fruits were collected 15 days after the pollination treatments to obtain the number of seeds per fruit and total seed mass per fruit. To improve normality, proportion fruit set was arcsine square-root transformed and number of seeds was log-transformed. Fruit set was compared between treatments within plants using a paired t-test, and seed number and total seed mass were compared using nested anova (flowers nested within treatments, nested within plants).

Frequency-dependent pollinator visitation and fitness

To test for frequency-dependent variation in fitness, we manipulated the sex ratio of a series of experimental field populations (Fig. 1). In August 2004, we set up six experimental populations, each composed of 20 plants in a 2-m diameter patch. We selected 120 plants in a natural population and transplanted 20 plants into each experimental population, selecting 10 female and 10 hermaphrodite plants (except in the single morph populations). In August, we fertilized two times with Osmocote (nitrogen, phosphorus and potasium) to promote flower production, and we watered plants every third day throughout the experiment. The experimental populations were approximately 1 km away from the nearest natural population and 100 m apart from each other, to minimize pollen movement among populations.

From 6–13 September 2004, we manipulated the sex ratio of flowers in each patch. The total number of flowers open each day was fixed at 20. The sex ratio of open flowers was manipulated to range from all female to all hermaphrodite in 20% intervals (i.e. 20/0, 16/4, 12/8, 8/12, 4/16, 0/20 hermaphrodite/female flowers). Manipulations (tagging flowers and eliminating extra flowers) were done daily to maintain each experimental population with the same sex ratio throughout the 8-day experiment.

Pollinator response to sex ratio was assessed by counting pollinator visits to hermaphrodite and female flowers, as described above. The entire patch was observed simultaneously, in five randomized, 20-min observation periods for each patch (10 h total). Visitation rate was calculated on a per-flower basis and regressed on sex ratio for both female and hermaphrodite flowers.

For each tagged flower, we quantified fruit set, seeds per fruit and total seed mass. For each of these fitness measures, we regressed relative female fitness (mean female trait value/mean hermaphrodite trait value, per population) on the frequency of females in the population. We also used regression to analyse absolute and relative fitness values for both females and hermaphrodites as a function of female frequency.

Results

Sex ratio variation

In 44 populations of K. grandiflora across the state of Sonora, the frequency of female plants averaged 10.3%, ranging from 0 to 44%. Most populations had <10% females, but only three populations (7%) contained no female plants at all (Fig. 1). Unlike the pattern seen across a broad latitudinal distribution (Cuevas et al., 2005), we found no relationship between female frequency and latitude within this region (F1,42 = 0.19, r2 = 0.004, P = 0.65).

Pollinator visitation and pollen deposition

In 3 h of observation in a natural population, 1295 visits were observed. The great majority of visitors were Melissodes bees (97.7% of total visits), with diurnal butterflies (0.8%) and flies (1.5%) making up the rest. The dominance of Melissodes was also seen in the experimental populations (see below), and may be typical for K. grandiflora in this region (see also Cazier & Linsley, 1974). Visitation rates were significantly higher to hermaphrodite flowers (2.2 ± 0.2 mean ± 1 SE here and hereafter) than to female flowers (1.6 ± 0.15, t16 = 5.96, P < 0.0001). The average time spent per visit did not differ between hermaphrodite flowers (5.5 ± 0.16) and female flowers (5.2 ± 0.17; F1,725 = 1.73, P = 0.188).

Female flowers received significantly fewer pollen grains per visit than emasculated hermaphrodite flowers (Fig. 2; F1,9 = 19.59, P < 0.001). Both genders received on average more pollen grains in one visit than the number of ovules that they produce (Fig. 2). Pollen grain deposition in emasculated flowers was, on average, 0.75 ± 0.01 of the pollen deposition in control flowers, and the two treatments were not significantly different (F1,8 = 0.73, P = 0.4; Fig. 2). In other words, the majority of the pollen deposited by insect visitors to hermaphrodite flowers came from a different flower.

Figure 2.

 Number of pollen grains deposited on the stigma per insect visit to hermaphrodite and female flowers. Values are the averages ± 1 SE. EMASCUL, emasculated flowers. Dashed line shows ovule number per flower.

Autonomous self-pollination and pollen limitation

Hermaphrodite flowers were able to reproduce in the pollinator-exclusion treatment through autonomous self-pollination. Proportional fruit set was significantly lower in the bagged treatment (0.43 ± 0.07) than in the control (0.65 ± 0.07; t18 = 3.6, P < 0.01). However, the number of seeds per fruit was not significantly different between bagged flowers and control flowers (bagged = 8.4 ± 0.61, control = 9.1 ± 0.24; F1,9 = 1.19, P = 0.27). Pollen addition to female flowers produced no evidence of pollen limitation. Fruit set and seed set were high regardless of treatment. Fruit set of female flowers with pollen added (0.95 ± 0.05) did not differ significantly from controls (0.85 ± 0.06; t9 = 1.2, P = 0.26), although this test had low power (β = 0.24). In addition, seed number per fruit was not significantly different between pollen addition (7.63 ± 0.32) and control flowers (8.05 ± 0.34; F10,52 = 1.51, P = 0.16), and total seed mass per fruit was not significantly different between pollen addition (0.041 ± 0.002) and control (0.045 ± 0.002; F1,9 = 0.49, P = 0.88). Finally, using just control flowers to compare female plants to hermaphrodite plants, fruit set was marginally significantly higher in females (0.85 ± 0.06) than hermaphrodites (0.65 ± 0.07; t18 = 2.00, P = 0.052), while seed number per fruit was significantly lower for females (8.05 ± 0.34) than for hermaphrodites (9.17 ± 0.24; F10,37 = 2.12, P = 0.047). Total seed mass of females (0.045 ± 0.002) was not significantly different from that of hermaphrodites (0.047 ± 0.003; F10,37 = 1.93, P = 0.07). The overall measure of seeds produced per plant was not significantly different between females (27.4 ± 2.66) and hermaphrodites (21.1 ± 2.52; F1,18 = 2.67, P = 0.11), although our power to detect significance was limited by small sample size (β = 0.34 for the observed mean difference).

Frequency-dependent pollinator visitation and fitness

Pollinators in the experimental populations comprised the same set of species, with similar taxonomic composition, as in the natural populations (91.1%Melissodes bees, 7.1% butterflies and 1.7% flies). The frequency of female flowers had no significant effect on either visitation rate to females (F1,3 = 0.005, r2 = 0.001, P = 0.94), or relative visitation rate (calculated as number of visits per female/number of visits per hermaphrodite) (F1,2 = 2.71, r2 = 0.57, P = 0.24).

Fruit set in experimental populations was relatively high. From 868 tagged flowers (444 female and 424 hermaphrodite) 682 set fruit (78%). Fruit set did not differ significantly between female flowers (0.81) and hermaphrodite flowers (0.75; F4,60 = 0.53, P = 0.71). Seed number was higher for hermaphrodite flowers (8.79 ± 0.1) than female flowers (7.28 ± 0.1; F4,659 = 11.57, P < 0.0001), as was total seed mass (0.045 ± 0.0001 vs. 0.044 ± 0.0001; F4,659 = 3.82, P < 0.01).

We found no evidence for frequency dependence in the relative fitness of female flowers for any of the three estimated fitness components. Linear regressions of relative female fitness on sex ratio were not significant for fruit set (F1,2 = 0.005, r2 = 0.002, P = 0.95), seed number (F1,2 = 0.42, r2 = 0.17, P = 0.58) or total seed mass (F1,2 = 0.09, r2 = 0.043, P = 0.79; Fig. 3). In addition, no relationship was found between the sex ratio and the absolute values of these fitness components for either female flowers (P = 0.37) or hermaphrodite flowers (P = 0.35; data not shown).

Figure 3.

 Relative female fitness values (female/hermaphrodite) for fruit set, seed set and seed mass in experimental populations as a function of female frequency.

Discussion

We combined surveys, manipulations in natural populations, and the construction of experimental populations to ascertain variation in sex ratio and the contribution of pollination ecology to frequency dependence and the maintenance of gynodioecy in K. grandiflora. In the Sonora region, sex ratio among populations varied from 0% to 44%, and only a few populations were completely lacking in females. Hermaphrodites showed the potential for reproductive assurance expected from autonomous self-pollination. In addition, we found strong evidence that pollinators discriminated against female flowers, and that pollen deposition was nearly three times lower in females. Yet our experiments showed that females were not pollen limited. Pollinator visitation to females did not decline, as the frequency of females increased in small experimental populations. In contrast to theoretical predictions, we found no evidence of frequency-dependent variation in fitness in experimental populations.

Relative advantages of female and hermaphrodite flowers

Consistent with expectation, we found that females showed several disadvantages relative to hermaphrodite flowers. First, hermaphrodites were capable of nearly full seed set even when pollinators were excluded, providing a means for reproductive assurance. Second, hermaphrodites attracted more pollinator visits than females. Pollinator preference for hermaphrodite flowers has also been reported in the gynodioecious species Sidalcea oregana and Hebe strictissima (Ashman & Stanton, 1991; Delph & Lively, 1992). In both species, hermaphrodite flowers are larger and produce more nectar. In K. grandiflora, hermaphrodite flowers are also larger (Cuevas et al., 2005). Both morphs produce nectar (E. Cuevas, pers. obs.), but we do not know whether hermaphrodite plants produce more nectar than females. Finally, even though pollinator visits were not longer at hermaphrodite flowers than female flowers, pollen deposition per visit was much higher on hermaphrodite stigmas than on female stigmas. This advantage of hermaphrodites may reflect some difference in pollinator behaviour or flower morphology that we could not directly observe. Pollinator identity did not vary in this study and therefore was not involved. Because the hermaphrodite flowers were emasculated, we know that the difference cannot be explained by deposition of self-pollen from within the flower. However, it is common for K. grandiflora plants to have several flowers open simultaneously (Cuevas et al., 2005), and pollen loads probably decline as pollinators move among the flowers on female plants. The disadvantage to females in terms of pollen receipt may represent an advantage in terms of offspring quality, which we consider below.

Despite their apparent disadvantages in pollinator attraction and reproductive assurance, female flowers experienced no pollen limitation of reproductive output. Visitation was sufficient to ensure maximal fruit set and seed set. This result should be interpreted with some caution, as pollen limitation may vary substantially from year to year (e.g. Baker et al., 2000; Vanhoenacker, 2006), and our pollen addition experiment was only carried out in 1 year and one population. A previous study did find significant pollen limitation of fruit set (but not seeds per fruit) for this species in a different population in 2001; interestingly, females were pollen limited whereas hermaphrodites were not. Natural fruit set, however, was not significantly different between the morphs (Cuevas et al., 2005). What is striking in our study is that reproduction was not limited by pollinators, even under conditions of strong discrimination against females by pollinators.

Theoretical models for the maintenance of females in populations predict some female advantage in terms of seed production or offspring quality through inbreeding avoidance (Lewis, 1941; Gouyon & Couvet, 1987). We did not detect consistent differences in reproductive success between females and hermaphrodites in the control pollination treatments or in the experimental populations. There was a suggestion of higher fruit set but lower seed set per fruit in female plants. Small sample sizes suggest caution in interpreting these results; however, a more detailed study of two populations also showed a lack of reproductive compensation in total number of flowers or fruits produced (Cuevas et al., 2005). Compensation could occur at or after germination; we have been unable to germinate seeds of K. grandiflora to compare the germination success of seeds from females and hermaphrodites, however, seed mass was lower in females, not higher, in both of the experiments.

One potential advantage of females over hermaphrodites is the production of offspring of higher genetic quality. It has been suggested in other gynodioecious species that inbreeding depression may maintain the presence of females in populations (Medrano et al., 2005; Weller & Sakai, 2005). Because females are obligately outcrossing, they will tend to produce fewer inbred offspring than hermaphrodites, which may experience geitonogamous pollen transfer among flowers. In addition, although from single insect visits we estimated that the proportion of pollen coming from anthers within the same flower is fairly low (about a quarter), we also found that hermaphrodite flowers are able to self-pollinate autonomously to a high degree when pollinators are excluded. Therefore, there is potential for hermaphrodites to experience high rates of self-fertilization, although the degree of realized selfing in natural populations requires further investigation. In a previous study, hand-pollinations of hermaphrodites with self and outcross pollen resulted in similar fruit and seed set, but inbreeding depression was not quantified for other aspects of the life cycle (Cuevas et al., 2005).

Frequency-dependent variation in fitness

Even though frequency-dependent variation in fitness has been proposed as a major mechanism involved in the evolution of sex ratios in gynodioecious species (McCauley & Taylor, 1997; McCauley & Brock, 1998), we were not able to find evidence for it in K. grandiflora. Neither pollination visitation rates, nor fruit or seed set responded linearly to the sex ratio of flowers in experimental populations. Because of the difficulty of maintaining population-wide sex ratios and collecting pollination data simultaneously in multiple populations, our sample size was small; we would not be able to detect a decrease in female fitness with increasing local female frequency if it produced only a weak signal. Other empirical studies have also failed to find evidence for frequency dependence in gynodioecious species (Williams & Fenster, 1998; Alonso, 2005; Asikainen & Mutikainen, 2005). For example, no frequency-dependent variation in fitness was detected in Chamaecrista fasciculata, a plant that provides no nectar reward; consequently bees discriminated against females regardless of density (Williams & Fenster, 1998). In K. grandiflora, various factors might contribute to explain the lack of frequency-dependent variation in fitness. First, both morphs offer nectar, increasing the attractiveness of females to pollinators. Despite discrimination against females, pollinators still delivered sufficient pollen to effect full fruit set and seed set (Graff, 1999). Second, in contrast to other species in which frequency-dependent variation in fitness has been shown (e.g. McCauley & Brock, 1998), each flower of K. grandiflora contains a small number of ovules, only 10 per flower for both hermaphrodites and females (Porter, 1969; Cuevas et al., 2005). Therefore, even restricted pollen receipt may be enough to achieve full seed set. Female flowers received 2.1 visits per hour even in the female-only population; if we estimate, conservatively, that bees visit over a 4-h period per day, then a female flower should receive about 80 pollen grains before senescence. This should be easily sufficient for full fertilization. We did not measure whether mean pollen deposition declines with the frequency of females, as one might expect it would. In very large populations with high frequencies of females, pollen deposition might eventually reach sufficiently low levels to cause pollen limitation of females, but we saw no evidence of this in our experimental populations.

We expect the degree of frequency dependence to be sensitive to the relative scales of pollinator movement and plant distribution. Our experimental populations were small and isolated by 100 m, and pollinators may not have experienced them as independent units. Nevertheless, these distances and patch sizes are relevant to what we have observed in natural populations, where it is common to find small patches of plants in a landscape mosaic, but not isolated by more than about 30 m (E. Cuevas, pers. obs.). Over several years of observation in the field, we have never seen female plants so isolated from hermaphrodites as to experience total reproductive failure (E. Cuevas, pers. obs.). In this study, we found no reduction in the fitness of female flowers even at sex ratios much higher than those observed in our survey (80%, 100% females vs. 44%). Because our experimental populations reflect a relevant scale of plant distribution, we believe that frequency-dependent variation in fitness is not likely a major mechanism regulating local sex ratio variation in K. grandiflora. Although gene flow is thought to be highly local for most herbaceous plants, pollinator movement and the scale of gene flow varies greatly from species to species (Schulke & Waser, 2001; Barthelmess et al., 2006). Our results suggest that pollinators of K. grandiflora move easily among isolated patches, efficiently fertilizing female flowers even when surrounded primarily by females. This facility of pollinator movement may relate to the open nature of the desert ecosystem, which may allow easy perception and location of a plant with relatively large, conspicuous flowers. The end result is that sex ratio is effectively averaged over large areas, and it may take very large-scale changes in female frequency to result in any negative frequency-dependent effects on female fitness.

Other factors in the maintenance of gynodioecy

We found variation in sex ratio among populations in the central region of Sonora. Although we did not find populations with very high frequencies of females, we also found few populations where females were entirely lacking. Given the reproductive disadvantages female plants potentially face, combined with their apparent lack of reproductive compensation, we are left with the question of why female plants are a common feature of K. grandiflora in this region.

Over a wider latitudinal range (19°–31°), K. grandiflora shows a strong positive relationship between female frequency and latitude, with females increasing in the cooler, drier environments of the north (Cuevas et al., 2006). In contrast, we did not find any relationship between sex ratio and latitude at the regional scale (28°00′–29°30′). However, local patterns of temperature and rainfall could be uncorrelated with latitude at this scale, and may influence the relative fitness of females and therefore the variation in sex ratio. In addition, female frequency could be related to population size (Nilsson & Ågren, 2006; Caruso & Case, 2007), but we were unable to estimate this parameter in our sample populations. Beside environmental factors, female frequency could be the result of stochastic processes generated by nucleo-cytoplasmic dynamics (Frank, 1989), which could explain why female frequency varies greatly among populations relatively close to each other. Under this scenario, female frequency is increased by the presence of male-sterile cytotypes when populations are lacking the corresponding restorer of male function. Even though we do not have a formal genetic study of the inheritance of male sterility, K. grandiflora shows some developmental similarities with species that show cytoplasmic male sterility (Cuevas et al., 2005). Our data suggest that such male-sterile lineages may not be at a reproductive disadvantage in K. grandiflora, creating opportunities for stochastic nucleo-cytoplasmic factors to govern local patterns of sex ratio.

Acknowledgments

The authors thank to José Martínez for field assistance and Anabel Martínez, Ma. De Los Ángeles Quintana and Ricardo Gutiérrez for pollinator observations. Mario Vallejo, Katrina Dlugosch, Pete Holloran, Alden Griffin and Sarah Swope made valuables comments to previous versions. Financial support was provided by CONACYT (34889-V) and UC Mexus.

Ancillary