• Marcel E. Dorken,

    1. Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, United Kingdom
    2. E-mail:
    3. Department of Biology, Trent University, Peterborough, ON, K9J 7B8, Canada
    Search for more papers by this author
  • Edward T. A. Mitchard

    1. Department of Plant Sciences, South Parks Road, University of Oxford, Oxford OX1 3RB, United Kingdom
    2. School of Geosciences, University of Edinburgh, Edinburgh EH8 9XP, United Kingdom
    Search for more papers by this author


Separate sexes can evolve under nuclear inheritance when unisexuals have more than twice the reproductive fitness of hermaphrodites through one sex function (e.g., when females have more than twice the seed fertility of hermaphrodites). Because separate sexes are thought to evolve most commonly via a gynodioecious intermediate (i.e., populations in which females and hermaphrodites cooccur), the conditions under which females can become established in populations of hermaphrodites are of considerable interest. It has been proposed that resource-poor conditions could promote the establishment of females if hermaphrodites are plastic in their sex allocation and allocate fewer resources to seed production under these conditions. If this occurs, the seed fertility of females could exceed the doubling required for the evolution of unisexuality under low-, but not high-resource conditions (the sex-differential plasticity hypothesis). We tested this hypothesis using replicate experimental arrays of the aquatic herb Sagittaria latifolia grown under two fertilizer treatments. The results supported the sex-differential plasticity hypothesis, with females having more than twice the seed fertility of hermaphrodites under low-, but not high-fertilizer conditions. Our findings are consistent with the idea that separate sexes are more likely to evolve under unfavorable conditions.

The evolutionary transition from hermaphroditism to dioecy (the co-occurrence of females and males) is thought to have occurred at least 100 times in the flowering plants (Charlesworth 2002; and see Renner and Ricklefs 1995; Weiblen et al. 2000). This transition has a long history of study by evolutionary biologists, including Darwin, who claimed to have “much difficulty in understanding why hermaphrodites should ever have been rendered dioecious” (Darwin 1877). Nevertheless, Darwin identified two of the major advantages of dioecy: the avoidance of inbreeding (Darwin 1875, p. 108) and the reallocation of resources from one sex function to another (resource compensation; Darwin 1877, p. 279). Indeed, these ideas form the foundation of our current understanding of the evolution of dioecy (Lewis 1941; Charlesworth and Charlesworth 1978; Lloyd 1982; Charlesworth 1999). Although inbreeding avoidance has been implicated in the evolution of dioecy for several groups (e.g., Kohn and Biardi 1995; Schultz and Ganders 1996; Sakai et al. 1997; Dorken et al. 2002; reviewed in Webb 1999), because dioecy can only evolve in populations capable of outcrossing, Darwin appeared to favor the notion that resource compensation promoted the evolution of dioecy. He reasoned that “if a species were subjected to unfavourable conditions… the production of the male and female elements … might prove too great a strain on its powers, and the separation of the sexes would then be highly beneficial” (Darwin 1877, p. 279).

The occurrence of dioecy has been associated with arid, resource-poor habitats (reviewed in Sakai and Weller 1999), and this would seem to provide support for Darwin's idea that “unfavourable conditions” can promote the evolution of dioecy. Why should this association occur?Delph (1990a, 2003) argued that changes in the proportion of resources allocated to female versus male function by hermaphrodites across environmental gradients promote the maintenance of unisexual females. Because dioecy commonly evolves via a gynodioecious intermediate (Charlesworth and Charlesworth 1978; Weiblen et al. 2000), Delph's (1990a, and see also Delph and Wolf 2005) argument provides a testable framework for examining the first step in the evolution of dioecy, the invasion of a population of hermaphrodites by females. Females can invade a population of hermaphrodites if they have more than double the seed fertility of hermaphrodites (under nuclear sex-determination; Lewis 1941). Lloyd (1976) showed that, under nuclear inheritance, the equilibrium frequency of females in an outcrossing population, P, is given by P= (1 − 2C)/(2 − 2C), where C is the average seed fertility of hermaphrodites divided by the average seed fertility of females. Thus, female frequencies are negatively associated with C and are nonzero only with C < 0.5. If hermaphrodites change their allocation to female versus male reproduction across environmental gradients, then, under certain conditions, females may exceed the doubling of seed fertility required for their establishment (the “sex-differential plasticity hypothesis”; Delph 2003). Moreover, if this doubling occurs in resource-poor, but not in more productive environments, this would provide empirical support for Darwin's arguments regarding the evolution of dioecy.

Experimental manipulations resulting in changes to the relative seed fertility of females and hermaphrodites under field conditions have been performed for Hebe subalpina (Delph 1990b) and Nemophila menziesii (Barr 2004). In both cases, when plants were subjected to less favorable conditions (defoliated for H. subalpina or no supplemental watering for N. menziesii) females had more than double the seed fertility of hermaphrodites, but these differences disappeared under less stressful conditions, consistent with the ideas outlined above. Moreover, observations of females and hermaphrodites in gynodioecious populations under field conditions have shown associations between site quality (or measures of plant vigor) and the relative seed fitness of females and hermaphrodites (Delph 1990a; Delph and Lloyd 1991; Barrett 1992; Wolfe and Shmida 1997; Ashman 1999; Delph and Carroll 2001; Asikainen and Mutikainen 2003; Case and Barrett 2004; Vaughton and Ramsey 2005). In most of these studies, there was also an association between site quality (or plant vigor) and female frequency (and see Alonso and Herrera 2001). Collectively these studies support the idea that environmental gradients affect the relative fertility of females and hermaphrodites, and thus the frequency of females in a population (Delph and Wolf 2005; but see Case and Ashman 2007). Studies in which environmental conditions are manipulated under controlled conditions would provide further support for the sex-differential plasticity hypothesis, particularly if it could be shown that the threshold value of C < 0.5 holds under some conditions but not others.

Here, we evaluate the sex-differential plasticity hypothesis by examining the value of C under two resource treatments in five replicate mating arrays of Sagittaria latifolia comprised of females and hermaphrodites. Specifically, we ask: is the value of C less than 0.5 under low-resource conditions but greater than 0.5 under more benign conditions?



Sagittaria latifolia (Alismataceae) is a clonal aquatic herbaceous plant with monoecious and dioecious populations common to a variety of wetland habitats in North America (Wooten 1971; Dorken et al. 2002; Dorken and Barrett 2004a). In northern regions of the species range (from which the source material for our experiment was obtained), individual shoots are annual and genets overwinter as underground corms. Experimental populations comprised of females and monoecious hermaphrodites (i.e., plants that produce a combination of unisexual female and male flowers; hereafter referred to as hermaphrodites) were obtained using crosses between plants from monoecious and dioecious populations. This was done to generate hermaphrodites and females that had the same genetic background, and thus avoid comparisons of plants collected directly from monoecious and dioecious populations, which have different patterns of allocation to growth and reproduction (Dorken and Barrett 2003).

Sex expression in S. latifolia is governed by the simple Mendelian segregation of sex-determining alleles (Dorken and Barrett 2004b) and families that segregate female and hermaphrodite offspring are first obtained in F2 progenies of crosses between plants from monoecious and dioecious populations. Specifically, at the putative male-sterility locus, plants in dioecious populations are either heterozygous (males) or homozygous for an allele conferring male sterility. Plants from monoecious populations are homozygous and male fertile (for details see Dorken and Barrett 2004b). To generate these F2 seed families, we used those crosses described in Dorken and Barrett (2004b) that segregated females and hermaphrodites in 1:3 ratios, respectively, plus additional families formed using the same classes of parents. In brief, these progeny were generated by crossing males from dioecious populations with hermaphrodites from monoecious populations. The resulting hermaphroditic progeny, all of which were heterozygous for the male-sterility allele, were crossed with other hermaphrodites with the same classes of parents (i.e., also heterozygous for the male-sterility allele) to form progeny arrays that segregated females and hermaphrodites in 1:3 ratios (e.g., crosses K, L, & M in Dorken and Barrett 2004b). A total of 20 genotypes were used to generate the 15 F2 families used in our experiment.

Female and male flowers of S. latifolia open for one day, and hermaphroditic ramets are synchronously protogynous, with female flowers typically opening at least one day before male flowers. Hermaphroditic plants are self-compatible and in natural populations outcrossing rates have been observed to vary between 0.37 and 0.96 (Dorken et al. 2002). Sagittaria latifolia is clonal and selfing is most likely to occur via interramet geitonogamy. Previous estimates from monoecious and dioecious populations indicate that there is likely to be substantial inbreeding depression in this species (ranging between δ= 0.52 and δ= 1.0; Dorken et al. 2002). In our experiment, we prevented selfing through interramet geitonogamy by growing each plant as a single flowering ramet (i.e., one ramet per genet was grown). In the native range, flowers are visited by a variety of insect pollinators, including bees, flies, and wasps (Muenchow and Delesalle 1994), and these classes of pollinators visited the arrays described below.


Plants were grown under glasshouse conditions and transplanted into outdoor arrays of 5-L buckets (one ramet per bucket) before they began to flower. Plants were randomly placed into one of five mating arrays using a split-plot experimental design. All of the plants used in an array were transplanted simultaneously, but the deployment of arrays was staggered over three weeks, starting in mid-July 2006. For each array (whole plots), plants were subjected to one of two levels of a fertilizer treatment (subplots), with each subplot receiving one level of the fertilizer treatment. Each subplot consisted of 81 plants arranged in a 9 × 9 grid (total number of plants = 810, 162 plants in each array), and the subplots that made up an array were arranged in pairs on a large, unshaded patio at the University Field Laboratory at Wytham, University of Oxford. Plants in high-resource subplots were grown in 5″ pots and fertilized weekly with Phostrogen© (Bayer CropScience Limited, Cambridge, U.K.) at the recommended dosage for general use. Plants in low-resource subplots were grown in 4″ pots and fertilized once, at the beginning of the experiment.

To ensure that the treatments had been given enough time to take effect, and to standardize data collection across plants at similar developmental stages, our measures of hermaphrodite and female seed fertility were taken from the second inflorescence produced by each plant. For each plant, we counted the number of female and male flowers produced on this inflorescence and measured the mid-vein length of the leaf subtending the inflorescence, which correlates strongly with other measures of plant size (Sarkissian et al. 2001). Note that because the arrays were set out at different times, arrays were subject to somewhat different growth conditions, resulting in variation in plant size (measured as mid-vein length) across arrays for each resource treatment. Overall, the range of mid-vein lengths observed in this study was 1.9–21.4 cm (average = 10.8 cm ± 0.16 SE, N= 645). This range closely corresponds with that found in natural populations, where mid-vein lengths have been observed to vary between 0.63 and 23.4 cm (average = 9.3 cm ± 0.12 SE, N= 985; M. E. Dorken, unpubl. data from 36 populations in southern Ontario—the source region for the material used in this study).

Each of the five arrays was observed over a four-week period and plants that failed to flower over this period were not included in the analyses presented below, resulting in the exclusion of 159 of 810 plants. The sex ratio of the plants that flowered was 0.72:0.28, consistent with expectations for a 3:1 hermaphrodite:female ratio (Dorken and Barrett 2004b), and there was no heterogeneity in the sex ratio of the different arrays (Gpooled= 2.4, df = 1, P > 0.10; Ghet= 8.4, df = 6, P > 0.20). There was also no significant heterogeneity in the sex allocation of plants from different families (measured as the proportion of flowers that were female; data not shown).

We collected fruits as they matured, allowed them to dry to constant mass, and weighed the mass of each fruit. Because each fruit contained thousands of seeds, we estimated total seed production by dividing fruit mass by the average seed mass, measured by weighing the mass of 20 seeds for a subset of the fruits (i.e., 200 fruits; one fruit from the basal whorl of flowers from 10 randomly chosen plants of each sex per subplot). We used the average seed mass per subplot, calculated separately for females and hermaphrodites, to generate each of these estimates. For each subplot, the magnitude of C was estimated by calculating the average seed fertilities of hermaphrodites and females, yielding a total of five estimates of C for each level of the fertilizer treatment. These values were compared across fertilizer treatments using analysis of variance with array and fertilizer treatment included as the main effects. Two other measures of female fertility were examined using analysis of variance for split-plot experimental designs: the number of female flowers produced per plant and the total mass of fruits produced per plant.


The resource treatment had a substantial effect on the relative seed fertility of females and hermaphrodites. Consistent with our predictions, the average value of C was 84% greater among subplots in the high-fertilizer treatment compared to the low-fertilizer treatment (F1,4= 19.9, P < 0.05; Fig. 1). The average value of C in the high-fertilizer treatment observed in this experiment (average C= 0.54 ± 0.12 SD, N= 5) closely matched the threshold value of 0.5 predicted to limit the spread of females. In contrast, values of C from the low-fertilizer treatment were consistently lower than 0.5 (average value of C= 0.29 ± 0.08 SD, N= 5; test of the null hypothesis that the average value of C in low-fertilizer subplots is equal to 0.5: t=−5.55, df = 4, P < 0.01). Similar results were obtained for other measures of female fertility, such as female flower production and total fruit mass (Table 1; Fig. 2A, B). For these measures, we found significant interactions between the fertilizer treatment and sex that resulted from contrasting responses by females and hermaphrodites to the fertilizer treatments. Female flower production by hermaphrodites increased approximately threefold from low- to high-fertilizer conditions (from an average of 2.1 ± 0.1 SE in low-fertilizer subplots to 6.2 ± 0.1 SE flowers per plant in high-fertilizer subplots), whereas for females, the increase was approximately twofold (from 6.1 ± 0.4 SE to 12.9 ± 0.3 SE). Similarly, the total mass of fruits produced by hermaphrodites increased almost sixfold from low- to high-fertilizer conditions (from 0.33 g± 0.03 SE in low-fertilizer subplots to 1.9 g± 0.1 SE in high-fertilizer subplots), whereas for females the increase was less than fourfold (from 0.85 g± 0.06 SE to 3.1 g± 0.2 SE). These interactions resulted from lower values for the relative fertility of hermaphrodites versus females in low-fertilizer compared to high-fertilizer subplots (Table 2; Fig. 2A, B), as was found for C.

Figure 1.

The average value of C (±SE) calculated across each of the five arrays increases in value from low- to high-fertilizer conditions.

Table 1. Split-plot analyses of variance of the number of female flowers produced per plant, the combined mass of all fruits produced per plant, plant size measured as the mid-vein length, and the average seed mass. Array, which was nested within treatment, was treated as a random effect. F-test degrees of freedom for the numerator and denominator are shown as subscripts for each value of F. Synthetic denominator mean squares and degrees of freedom were constructed using Satterthwaite's approximation for unequal sample sizes (Sokal and Rohlf 1995).
EffectFlower numberFruit massMid-vein lengthSeed mass
  1. *P<0.05; **P<0.01; ***P<0.001.

Treatment F 1,8.1=39.5*** F 1,8.0=19.4** F 1,8.0=23.2** F 1,8.1=47.3***
Array[treatment] F 8,8=8.7** F 8,8=23.3*** F 8,8=25.1*** F 8,8=1.4
Sex F 1,8.5=275.9*** F 1,8.7=55.8*** F 1,8.7=2.8 F 1,8.1=5.0
Sex × treatment F 1,8.5=17.9** F 1,8.7=6.3* F 1,8.7=1.3 F 1,8.1=0.0
Figure 2.

Association between plant size (measured as mid-vein length) and the relative fertility of hermaphrodite and female plants per subplot, measured as: (A) the number of female flowers per plant on hermaphrodites/females; (B) the mass of fruits produced by hermaphrodites/females; and (C) the value of C. Filled circles are for high-fertilizer subplots, unfilled circles for low-fertilizer subplots. Error bars indicate ± 1 SE. Note that C was calculated separately for each plot. Accordingly no error bars are presented for this measure of the relative hermaphrodite:female fertility.

Table 2. Estimates of female fertility for each combination of plant sex, fertilizer treatment, and array. These values were used to calculate the relative fertility of hermaphrodites versus females presented in Figures 1 and 2.
ArrayTreatmentSexNumber of plantsFlower numberTotal fruit mass (g)Average number of seeds per plant
1 High Hermaphrodite 60  6.6 1.74 3041
1LowFemale15 7.91.053472
1 Low Hermaphrodite 56  2.5 0.50 1225
2 High Hermaphrodite 34  8.3 3.566 5660
2LowFemale20 5.80.682040
2 Low Hermaphrodite 52  1.8 0.26  710
3 High Hermaphrodite 58  6.0 2.06 3769
3LowFemale22 7.51.214257
3 Low Hermaphrodite 45  3.2 0.55 1505
4 High Hermaphrodite 52  5.3 1.59 2615
4LowFemale14 3.60.521696
4 Low Hermaphrodite 32  1.0 0.10  295
5 High Hermaphrodite 53  5.4 1.27 1936
5LowFemale 9 4.10.37 987
5 Low Hermaphrodite 28  1.0 0.09  231

There was no difference in our measure of plant size (mid-vein length) between the sexes (average mid-vein length of females: 11.2 cm ± 0.3 SE; hermaphrodites: 10.6 cm ± 0.2 SE), nor was there an interaction between sex and fertilizer treatments for plant size (Table 1). However, the fertilizer treatment itself had a substantial effect on plant size (average mid-vein length under low-fertilizer conditions: 7.7 cm ± 0.2 SE; high fertilizer: 13.3 cm ± 0.2 SE; Table 1) and plant size was positively correlated with all measures of the relative female fertility of hermaphrodites and females (Flower production: r= 0.96, N= 10, P < 0.0001; Fruit mass: r= 0.93, N= 10, P < 0.0001; C: r= 0.89, N= 10, P < 0.001; Fig. 2). The fertilizer treatment also affected seed mass, with larger seeds under high-fertilizer conditions (average seed mass = 0.47 mg ± 0.01 SE) compared to low-fertilizer conditions (average = 0.32 mg ± 0.01 SE; Table 1).


Environmental conditions clearly affected the relative seed fertility of female versus hermaphrodite plants of S. latifolia. In all five arrays, our estimates of C were lower under low- compared to high-fertilizer conditions, providing experimental support for Delph's (2003) sex-differential plasticity hypothesis. Similar results have been obtained from experiments conducted under field conditions in which plants may have experienced variation in their resource conditions (Delph 1990b; Barr 2004). Our experiment builds on these studies by demonstrating that females have higher seed fertility relative to hermaphrodites in replicate arrays of plants grown under controlled conditions. Because of the simple relationship between C and the equilibrium frequency of females under nuclear inheritance described above (Lloyd 1976), we can infer that, all else being equal, the average frequency of females should evolve toward 29% under low-resource conditions, but only 4% under high-resource conditions. Our results are therefore consistent with the idea, originally proposed by Darwin, that separate sexes are more likely to evolve under “unfavourable conditions.”

The sex-differential plasticity posits that female invasion is favored by environmental conditions that cause hermaphrodites to be male-biased in their sex allocation. A theoretical model by Seger and Eckhart (1996) provided a broadly similar conclusion, with females able to invade populations of hermaphrodites with male-biased sex allocation. However, in their model, hermaphrodites were male biased because they experienced accelerating fitness gains through their male function. In contrast, the sex-differential plasticity hypothesis involves saturating fitness gains for male investment (Ashman 2006). Under saturating fitness gains through male function, one would predict higher male allocation under low-resource availability and lower male allocation under high-resource availability (Klinkhamer et al. 1997). This expectation matches the results of our experiment, but is not consistent with the accelerating male fitness gains modeled by Seger and Eckhart (1996), under which we would expect male-biased sex allocation of hermaphrodites to be most pronounced under high-resource conditions. The little information on male gain curves available suggests that male gain curves are saturating (Campbell 2000), although evidence from animal systems indicates that the shapes of these curves are not fixed and can vary with the mating environment (Yund 1998).

An alternative to female invasion in populations of hermaphrodites with male-biased sex allocation is the evolutionary readjustment of hermaphrodite allocations to increase their seed production. This scenario would require additive genetic variation for hermaphrodite sex allocation under the environmental conditions that cause hermaphrodites to be male-biased in their sex allocation. Additive variation for hermaphrodite sex allocation is known to exist in a variety of plant taxa (reviewed in Campbell 2000; and see Ashman 2003), and there is evidence for substantial broad-sense heritability of hermaphrodite sex allocation in S. latifolia (Dorken and Barrett 2004c; although the extent to which this broad-sense heritability reflects additive variation is not known). The occurrence of such heritable variation could lead to the reequilibration of hermaphrodite sex allocation, such that hermaphrodites are no longer male-biased under low-resource conditions. Thus, because the initial spread of male-sterility alleles is expected to take many generations, especially under recessive male sterility (Charlesworth and Charlesworth 1978), the occurrence of additive variation for hermaphrodite sex allocation could prevent the spread of females.

Sex allocation in monoecious populations of S. latifolia is size-dependent (Sarkissian et al. 2001; Dorken and Barrett 2004c). In our study, this resulted in a positive, linear association between plant size and the relative seed fertility of hermaphrodites versus females (Fig. 2). This finding is consistent with studies conducted in other gynodioecious populations showing that the seed fertility of hermaphrodites is related to plant size (Delph 1990a; Ashman 1999; Barrett et al. 1999; but see Barr 2004; Case and Ashman 2007). In these populations, therefore, gradients in environmental conditions are thought to regulate female frequencies through their effects on hermaphrodite seed production, with females occurring at higher frequencies under conditions in which hermaphrodites produce less seed (reviewed in Delph and Wolf 2005; and see Vaughton and Ramsey 2005). Our study provides further evidence to support this idea by demonstrating that the ability of females to invade populations of hermaphrodites might decrease linearly as plant size increases (Fig. 2C; and see Barrett et al. 1999).

The prediction that females are more likely to invade populations subject to low-resource conditions in which hermaphrodites have low male allocation depends on two assumptions: (1) the average sex allocation of hermaphrodites remains constant in any given environment; (2) the relative fertility of females and hermaphrodites through their female function is determined solely by seed quantity. The first of these assumptions has been evaluated in a previous experiment on S. latifolia (Dorken and Barrett 2004c) where it was found that clonally replicated genotypes have consistent patterns of sex expression in a given set of environmental conditions. However, the second assumption could be violated if there are changes in seed quality between females and hermaphrodites across environmental gradients. For example, in natural populations, clonality is expected to lead to selfing via geitonogamous pollen transfer (Dorken et al. 2002), and it has been shown that rates of clonal expansion are higher under high-resource conditions (Dorken and Barrett 2004c). Indeed, the combination of geitonogamous selfing and substantial inbreeding depression (Dorken et al. 2002) might provide the greatest advantage to females under the opposite conditions (i.e., high resource availability) to those that provide the greatest advantage to females via resource compensation (i.e., low-resource availability). Thus, although resource compensation and inbreeding avoidance are not mutually exclusive and may both promote the evolution of females (Charlesworth 1999), it seems unlikely that both would simultaneously have strong effects in S. latifolia.


Most studies of the environmental conditions associated with the evolution of separate sexes have considered the conditions that favor the initial invasion of a population of hermaphrodites by females (reviewed in Delph 2003; Delph and Wolf 2005), and this is the framework followed in our study. However, as pointed out by several authors, the environmental conditions that promote the establishment of females might impede the evolution of males (Delph and Lloyd 1991; Barrett et al. 1999; Delph and Wolf 2005). Specifically, female-sterility mutations might not increase in frequency when hermaphrodites are already strongly male-biased in their sex allocation (e.g., because of sex-allocation plasticity). For this reason, it has been argued that sex-allocation plasticity of hermaphrodites might promote the maintenance of gynodioecy or subdioecy (reviewed in Ashman 2006). Indeed, if hermaphrodites are rendered phenotypically male under certain environmental conditions, it would seem unlikely that female-sterility mutations could evolve in a gynodioecious population. In our experiment we found that across the low-fertilizer subplots between 11% and 57% of hermaphrodites were phenotypically male. Such plants are unlikely to have been able to increase their male allocation via reductions in their female allocation. However, even if female-sterility mutations do not affect the pollen output of a subset of the plants in a population, we might still expect them to increase in frequency, as explained below.

The spread of mutations that result in male (or male-biased) phenotypes can occur if modified plants have higher male fertility than unmodified hermaphrodites (Charlesworth and Charlesworth 1978). If increased male allocation occurs through the reallocation of resources from female function, then the mutations leading to male-biased sex allocation must, therefore, also cause substantial declines in the female fertility of hermaphrodites if they are to increase in frequency (Charlesworth and Charlesworth 1978; Charlesworth 1999). However, if females have substantially greater seed production than hermaphrodites, modifiers that cause even small increases in the pollen production of hermaphrodites (and thus only small decreases in the female fertility of hermaphrodites) might still be able to spread (Charlesworth and Charlesworth 1978). Specifically, Charlesworth and Charlesworth (1978) have shown that genetic modifiers of pollen production can invade gynodioecious populations if the pollen production of modified hermaphrodites exceeds K*, where K*=[(1 +s− 2sδ)/(k+ 2sδ−s)]×k*, s is the selfing rate, δ is the magnitude of inbreeding depression, k is increase in ovule production of females compared to hermaphrodites, and the ovule production of modified hermaphrodites is 1 −k* (see equation [10] in Charlesworth and Charlesworth 1978). Using this equation, the value of the relative seed fertility of hermaphrodites and females obtained here and average estimates of the selfing rate (s= 0.41) and magnitude of inbreeding depression (δ= 0.83) in monoecious populations of S. latifolia obtained by Dorken et al. (2002), and assuming that k*= 1 (i.e., modifiers cause complete male-sterility), the evolution of modified hermaphrodites might require increases in pollen production as small as 15%–35% (depending on the estimate of C for each low-fertilizer half of an array). Such increases would seem to be possible for S. latifolia: on average, 20% of the flowers produced by hermaphrodites subject to low-fertilizer conditions were female and became fruits. Thus, even if male flowers are as costly to produce as female flowers (and fruits), males might be favored over hermaphrodites in gynodioecious populations subject to the same environmental conditions that would also appear to promote the evolution of females (i.e., low resources).

Associate Editor: J. Shykoff


We thank P. Smith for logistical support, C. Barr, S. Barrett, D. Charlesworth, L. Delph, J. Pannell, J. Shykoff, and one anonymous reviewer for helpful comments on previous versions of the manuscript. We also acknowledge the Gatsby Charitable Foundation for a Sainsbury Undergraduate Studentship to ETAM, and the University of Oxford and St. Hugh's College for a Career Development Fellowship to MED for supporting this work.