Relative fitness of females and hermaphrodites in a natural gynodioecious population of wild radish, Raphanus sativus L. (Brassicaceae): comparison based on molecular genotyping


Keiko Miyake, Planning Division, Support Office for Female Researchers, Niigata University, 8050 Ikarashi-2, Nishi-ku, Niigata 950-2181, Japan.
Tel.: +81 25 262 7570; fax: +81 25 262 7572; e-mail:


In many gynodioecious species, sex determination involves both cytoplasmic male-sterility (CMS) genes and nuclear genes that restore male function. Differences in fitness among genotypes affect the dynamics of those genes, and thus that of gynodioecy. We used a molecular marker to discriminate between hermaphrodites with and without a CMS gene in gynodioecious Raphanus sativus. We compared fitness through female function among the three genotypes: females, hermaphrodites with the CMS gene and those without it. Although there was no significant difference among the genotypes in seed size, hermaphrodites without the CMS gene produced significantly more seeds, and seeds with a higher germination rate than the other genotypes, suggesting no fitness advantage for females and no benefit to bearing the CMS gene. Despite the lack of fitness advantage for females in the parameter values we estimated, a theoretical model of gynodioecy shows it can be maintained if restorer genes impose a cost paid in pollen production. In addition, we found that females invest more resources into female reproduction than hermaphrodites when they become larger. If environmental conditions enable females to grow larger this would facilitate the dynamics of CMS genes.


Gynodioecy is a breeding system in which hermaphrodites and females (male steriles) co-exist within a population (Darwin, 1877). Evolutionary biologists have been interested in explaining how females that accrue no fitness through pollen can persist among hermaphrodites that gain fitness through both seeds and pollen.

In most gynodioecious species, the sex phenotype is determined by an interaction between a cytoplasmic male sterility (CMS) gene located on a mitochondorial genome and nuclear genes restoring the male function (hereafter called Rf genes, for restorer of fertility). In such nucleo-cytoplasmic systems, theoretical models have shown that the maintenance of gynodioecy depends on two key parameters: a seed fitness advantage for females (i.e. compensation; Darwin, 1877) and a cost of restorer genes (Charlesworth, 1981; Delannay et al., 1981; Frank, 1989; Gouyon et al., 1991).

Because CMS genes are transmitted only through seeds, females can be maintained if they have only slightly higher seed fitness than do hermaphrodites (Lewis, 1941; Lloyd, 1975; Charlesworth & Charlesworth, 1978; Frank, 1989; Couvet et al., 1998). Thus seed fitness has been compared between different sex phenotypes (females vs. hermaphrodites) to determine if there is a fitness advantage for females (reviewed in Delph et al., 1999; Shykoff et al., 2003). However, the relative seed fitness of hermaphrodites that bear the CMS gene, as well as that of females, has a crucial impact on the dynamics of CMS genes and hence on the maintenance of nucleo-cytoplasmic gynodioecy. This has been overlooked in most cases because the primary phenotypic effect of CMS genes (male sterility) is masked in the presence of Rf genes. A model for nucleo-cytoplasmic gynodioecy by Dufaÿet al. (2007) assumed a cytotype effect on fitness through female function, ws, which could be either positive or negative. If ws is positive, hermaphrodites with CMS genes have higher seed fitness than hermaphrodites without CMS genes, and CMS genes can spread even if the female advantage is very small. However, this assumption has never been empirically investigated because the cytotypes of hermaphrodites cannot be determined without molecular markers for CMS genes, which are not available in most gynodioecious plants.

Raphanus sativus L. (Brassicaceae) provides an ideal opportunity for overcoming this difficulty. In this species, the sex phenotype of individuals is determined by the presence or absence of only one CMS gene and by one (or more) dominant Rf gene(s) that interacts with the CMS gene (Yamagishi & Terachi, 1994; Yasumoto et al., 2008). The molecular basis of CMS has been well characterized; a gene (called orf138) specific to the Ogura cytotype is responsible for the expression of male sterility (Ogura, 1968; Bonhomme et al., 1992; Grelon et al., 1994; Krishnasamy & Makaroff, 1994). By using a set of PCR primers that amplifies a fragment that includes orf138, we can determine the presence or absence of the CMS gene for each individual (Murayama et al., 2004). In this manner, we can directly compare relative seed fitness between phenotypic hermaphrodites with different cytotypes, in addition to a comparison between hermaphrodites and females.

In this study, we addressed the following questions in order to better understand the dynamics of the CMS gene.

  • 1Do females of R. sativus have a fitness advantage through seeds?
  • 2Do hermaphrodites that bear the CMS gene have more seed fitness than hermaphrodites without the CMS gene?

Materials and methods

Plant and study site

Raphanus sativus is an entomophilous, self-incompatible annual widely distributed in coastal habitats of Japan. Most seedlings appear in early September and grow to young plants that remain in the form of a rosette during winter, whereas a minority of seedlings appear in the end of winter or the following spring. Irrespective of germination time, plants bloom from mid March to early May. In natural populations near the study site, female frequencies are relatively low, ranging from 0% to 21%, while the CMS gene frequencies are highly variable among populations, ranging from 0% to 100% (Murayama et al., 2004). In 2004, we conducted this study in a population in a pine forest on Shikanoshima Island in northern Kyushu, Japan, which is one of the populations studied in Murayama et al. (2004). The population consisted of over 1000 individuals, and the female frequency in this population was 20%.

Determination of genotypes

In early April, 302 individuals were marked for genotyping. We collected flower buds and a young leaf from each marked individual to determine sexual phenotype and cytotype. To determine sexual phenotype, pollen grains were stained with 0.4% (w/v) Fuchsin Basic solution. Females were defined as individuals lacking stainable pollen grains in flower buds.

Three PCR primers, 6C (5′-GACATCTAGAAAGTTAAAAAT-3′), 6B (5′-CCACCCATGGTACAGAGTGT-3′), 6D (5′-TTGCGGAAGATGTCTTATCACG) were used to determine the presence of the CMS gene (Terachi et al., 2001). The primer 6C is located in orf138, and PCR using primers 6C and 6B amplifies a 1.4-kb fragment specific to Ogura cytotype. Plants with this fragment were considered to have the CMS gene (Ogura cytotype). For plants without this fragment, we conducted PCR using primers 6D and 6B, which amplify a 1.3-kb fragment specific to the Normal cytotype (i.e. no CMS gene). Plants with no PCR products from either primer pair were excluded from the experiments.

Using information on sexual phenotype (female or hermaphrodite) and cytotype (Ogura or Normal), individuals were classified into the following three types; (1) females with the CMS gene (FCMS), (2) hermaphrodites with the CMS gene (HCMS) and (3) hermaphrodites with normal cytotype (HN). The last category includes plants with and without the Rf gene because we could not discriminate them. We will refer to these three types as ‘genotypes’ in this study. For comparing reproductive parameters, a total of 161 flowering plants were chosen (57 FCMS, 56 HCMS and 48 HN).

Reproductive parameters

In late May, when plants had finished setting fruit, each marked plant was collected and brought to the laboratory. After air-drying, each individual was divided into fruits vs. vegetative tissue (the rest of the plant), and weighed. All fruits from each plant were counted and bulked for weighing. The dry weight of fruits plus vegetative tissue was considered to be the total biomass of the individual. The number of peduncles on each plant was counted and used as a measure of flower number. All seeds were removed from each fruit and maldeveloped seeds were removed from the analyses. All mature seeds were weighed for each individual. Seed size was calculated as the total mature seed weight divided by the total seed number for each individual.

To evaluate seed germination rates, seeds of 30 plants (10 plants from each genotype) were stored at room temperature until October 2004. Three replicates of 30–40 seeds (90–120 seeds in total) were examined for each individual. For germination experiments, we adopted the DT regime of Washitani (1987) that simulates temperature change from summer to winter. Seeds were sown in plastic petri dishes, and incubated in the dark. They were exposed to 36, 32, 28, 24, 20, 16, 12, 8 and 4 °C successively for 1–8 days. Before the incubator temperature was reduced to the next level, seed germination was scored, and germinated seeds were removed. After the end of the DT regime at 4 °C, seeds were exposed to 25 °C for 5 d to simulate spring. We excluded data from seeds from four parents (2 FCMS, 1 HCMS and 1 HN) because seeds sustained severe fungal damage and germination was extremely low. Consequently, a total of 2820 seeds from 26 individuals were used for the analysis.

Statistical analyses

Statistical analyses were performed with R version 2.6.2 for Macintosh ( and JMP version 5.0.1J (SAS Institute Inc., Cary, NC, USA). To test for differences in seed fitness among genotypes (FCMS, HCMS and HN), reproductive parameters were compared. Plant biomass, total fruit weight, total seed weight and seed size were logarithmically transformed to meet normality assumptions, and then compared by anova. For total flower, fruit and seed numbers, a generalized linear model was employed assuming a Poisson distribution and setting genotype as an explanatory factor, because the data did not meet the statistical assumption of anova (homogeneity of variance and normality). Analysis of deviance comparing χ2 statistics was conducted to test for generalized linear model fits. Holm’s posteriori test was performed to evaluate differences between each genotype pair.

The effect of biomass on differences among genotypes in total fruit weight and total seed weight was evaluated by analysis of covariance (ancova) with genotype as an explanatory factor and biomass as a covariate. For the effect of biomass on differences among genotypes in flower, fruit and seed numbers, we performed generalized linear models with analysis of deviance assuming a Poisson distribution to test the effect of genotype alone, genotype with biomass and the full model including the interaction between genotype and biomass. Holm’s posteriori test was performed to evaluate difference between each genotype pair.

Seed germination rate (the number of germinated seeds divided by the total seed number used in the experiment), early seed germination rate (the number of germinated seeds during the DT regime divided by the total seed number used in the experiment) and the fraction of seeds germination during ‘spring’ (the number of seeds that germinated during the final incubation at 25 °C which simulated spring, divided by the total number of germinated seeds) were compared among genotypes by logistic regression analyses. Holm’s posteriori test was performed to evaluate difference between each genotype pair. Spearman’s rank correlation test was used to evaluate the relationship between seed weight and seed germination rate.


Of the 161 flowering target plants, 57 were FCMS, 56 were HCMS and 48 were HN. All females had the amplified fragment with orf138, which was consistent with the assumption that this is the only CMS gene present in this species.

Compared to FCMS and HCMS, HN had significantly higher plant biomass, total fruit weight, and total flower, fruit, seed numbers (Fig. 1a–e). Comparison between FCMS and HCMS showed that HCMS had more flowers but FCMS set more seeds (Fig. 1b, e). Seed size was not significantly different among the genotypes (Fig. 1g). Although there were no significant differences between any genotype pairs (Holm’s posteriori test) for total seed weight, the overall comparison showed a significant difference among the genotypes, with HN having the highest total seed weight (Fig. 1f).

Figure 1.

 Reproductive parameters for females (F/CMS), hermaphrodites with the CMS gene (H/CMS) and hermaphrodites without the CMS gene (H/N). (a) Biomass. (b) Total flower number. (c) Total fruit number. (d) Total fruit weight. (e) Total seed number. (f) Total seed weight. (g) Seed size. (h) Seed germination rate. The box itself contains the middle 50% of the data. The upper edge (hinge) of the box indicates the 75th percentile of the data set, and the lower hinge indicates the 25th percentile. The line in the box indicates the median value of the data. The error bars represent the 5th and 95th percentile values. Comparison was performed using either anova after logarithmic transformation (a, d, f, g), analysis of deviance assuming Poisson distribution was conducted with a generalized linear model (b, c, e) or a logistic regression analysis (h). Statistics and probabilities are shown above each graph. Significant differences among the genotypes are denoted by different letters (Holm’s posteriori test, < 0.05). No significant differences are found in any pairs for total seed weight (f), while the overall comparison is significant.

Because plant size often affects reproductive output, we used biomass as a covariate in our analyses. Along with the genotype, biomass had a significant effect on total seed and fruit weight (Table 1). The interaction between the genotype and biomass was also significant for total weight of both fruits and seeds: all genotypes showed an increase in fruit and seed production with increasing biomass, but slopes differed among genotypes. As for the total numbers of flowers, fruits and seeds, the models including biomass showed significantly better fit than those including only genotype as an explanatory variable (χ21 = 7638, < 0.001 for flowers, χ21 = 3496, < 0.001 for fruits, χ21 = 7746, < 0.001 for seeds), suggesting that not only the genotype but also biomass had a significant effect on these reproductive parameters. Likewise, the model including genotype, biomass and their interaction as explanatory variables fit better than that including no interaction (χ22 = 241, < 0.001 for flowers, χ22 = 226, < 0.001 for fruits, χ22 = 630, < 0.001 for seeds), indicating a differential effect of biomass on flower, fruit and seed production among the genotypes. The regression curves, estimated by the generalized linear model, between biomass and either flower number, fruit number or seed number for each genotype suggest that females have higher reproductive output than hermaphrodites when plants are large (Fig. 2).

Table 1.   Results of ancovas investigating the effects of biomass and genotypes on total seed weight and total fruit weight.
Source of variationdfSSMSFP
Total seed weight
 Biomass171.7271.7294.76< 0.001
 Genotype ×   biomass27.343.674.850.009
Total fruit weight
 Genotype27.643.829.18< 0.001
 Biomass176.1276.12182.98< 0.001
 Genotype ×   biomass25.942.977.140.001
Figure 2.

 Scatter plots of the relationship between biomass and (a) the flower number, (b) the fruit number and (c) the seed number for females (filled circles), hermaphrodites with the CMS gene (open circles) and hermaphrodites without the CMS gene (open triangles). The regression curves for females (solid lines), hermaphrodites with the CMS gene (dashed lines) and hermaphrodites without the CMS gene (dotted lines) were obtained with a generalized linear model assuming Poisson distribution.

The germination rate of seeds produced by HN was higher than for seeds produced by the other two genotypes (Fig. 1h). In the seed germination test, we simulated the temperature change from summer to winter, and then spring. More than 50% of germination occurred during the period with decreasing temperature simulating autumn, with most germinating by 16 °C. There was an additional peak in germination during the final 25 °C simulating spring. The timing of germination will affect the biomass at reproduction. The fraction of seeds germinating during ‘spring’ differed among the three genotypes (χ22 = 17.64, < 0.001), with HN having the greatest proportion of ‘spring’-germinating seeds (35.0 ± 16.0% [mean ± 1 SD] vs. 24.8 ± 13.1% for HCMS and 27.7 ± 20.6% for FCMS). However, early (= autumn) germination rate did not vary among genotypes (χ22 = 1.62, = 0.45). Seed germination rates were not correlated with mean seed size for any genotypes (Spearman’s rank correlation test, ρ = −0.02, = 9, = 0.97 for HN, ρ = 0.13, = 9, = 0.73 for HCMS, ρ = −0.48, = 8, = 0.23 for FCMS).


By using a molecular marker for a CMS gene, we were able to distinguish hermaphrodites with the CMS gene (HCMS) from those without it (HN) in gynodioecious Raphanus sativus. Consequently, we were able to categorize individuals into three genotypes: females (FCMS), hermaphrodites with the CMS gene (HCMS) and those without it (HN). If FCMS individuals have higher fitness through female function than the other two genotypes, it means that females show compensation. In addition to this conventional comparison of females (FCMS) and hermaphrodites (HCMS + HN), we were able to compare HCMS with HN in order to evaluate the ‘cytotype effect’: the effect of bearing the CMS gene. Although the CMS gene cannot show its primary phenotypic effect (i.e. male sterility) in HCMS because of the presence of the male fertility restorer (the Rf gene), the relative fitness of HCMS can have a large effect on the evolutionary dynamics of the CMS gene, and thus the maintenance of nucleo-cytoplasmic gynodioecy (Dufaÿet al., 2007).

Relative seed fitness of females to hermaphrodites

In many gynodioecious species, females produce more fruits and/or seeds than hermaphrodites (reviewed in Gouyon & Couvet, 1987; Delph et al., 1999; Shykoff et al., 2003). In R. sativus, however, we did not detect any female advantage in fruit or seed production. Similar results have been reported in Beta vulgaris (Boutin et al., 1988), Phacelia dubia (del Castillo, 1993), Plantago maritima (Dinnetz & Jerling, 1997) and Daphne laureola (Alonso & Herrera, 2001). In R. sativus, one reason for the lack of a female fitness advantage may be the difference in plant biomass (Table 1; Fig. 2). Since hermaphrodites, especially HN, are larger than females, hermaphrodites are able to invest more resources into female reproduction than females, even though hermaphrodites allocate a portion of resource toward male reproduction. Indeed, total fruit weight, total fruit and seed numbers and total flower number were affected by biomass. When biomass was taken into account, the different genotypes still had fitness differences (Table 1; Fig. 2). Further, females and hermaphrodites have different investment strategies: when plant size is large, females invest more resources into seed and fruit production than hermaphrodites.

Can females grow larger than hermaphrodites? R. sativus grows in a variety of habitats, such as open, sandy beaches or roadsides, and half-shaded forest margins, and females may grow to be larger than hermaphrodites under some environmental conditions. In Hebe strictissima, seed production by hermaphrodites was highly plastic and was significantly reduced in poor sites. The greater plasticity of hermaphrodites compared to females altered the relative seed fitness of the two morphs, creating a negative correlation between site quality and female frequency (Delph, 1990). Similar sexual dimorphism in plasticity (Dawson & Geber, 1999;Delph, 2003) is possible in R. sativus. Comparative studies of R. sativus in other habitats will help clarify whether the results observed in this population hold true under different environmental conditions.

It is still not clear why HN grew larger than other genotypes in this population. One possible reason for the difference in biomass among genotypes may be the timing of germination. As expected based on field observations, a portion of the seeds in our experiment germinated in the simulated spring-like conditions, while many seeds germinated in ‘autumn’. If hermaphrodites were to germinate in autumn, and females germinate in the spring, hermaphrodites will have a longer growth period and can therefore become larger. However, the fraction of spring germination was actually larger in hermaphrodites (HN) than females, so this is not the explanation. As a next step, it may be worth investigating differences among genotypes in winter growth or mortality; in some species, sex morphs differ in their response to cold temperatures (Dawson & Bliss, 1989; Li et al., 2005).

Relative seed fitness of hermaphrodites with (HCMS) and without (HN) the CMS gene

In a previous study of multiple R. sativus populations, we found that female frequency was relatively low, with 0.21 being the maximum, while the CMS gene frequency was highly variable among populations, ranging from 0 to 1 (Murayama et al., 2004). These results suggest that most CMS genes exist in hermaphrodites (i.e. HCMS); 0.88 ± 0.08 (mean ± 1 SD, Npopulation = 12, when excluding populations with a CMS gene frequency of zero) of all CMS-bearing individuals are hermaphrodites (HCMS). Therefore, the relative fitness of HCMS should be important in the dynamics of CMS, and thus the maintenance of nucleo-cytoplasmic gynodioecy. If HCMS has higher fitness through seeds than HN, the CMS gene can spread to fixation in a population, even if females do not have any fitness advantage over hermaphrodites. In contrast, if HCMS has lower fitness through seed production than HN, the CMS gene frequency will decrease in the absence of a female fitness advantage. Our results suggest the latter; no reproductive parameters in HCMS surpassed those of HN, and for some parameters, especially total seed number, HCMS was inferior to HN (Fig. 1). These results, combined with the lack of any fitness advantage for females (FCMS), suggest that the CMS gene should not be maintained by selection unless relative fitness varies from year to year or among populations. It is thus puzzling that the CMS gene frequency varies and the CMS gene is sometimes fixed in populations of this species (Murayama et al., 2004). The low effective population size of cytoplasmic genes relative to nuclear genes, genetic drift, and the extinction and recolonization of populations in various habitats might account for it (Murayama et al., 2004).

Although we focused on the dynamics of the CMS gene in this study, theory shows that an expression cost of Rf genes also has a crucial role in the maintenance of gynodioecy. Discrimination between the two types of HN (hermaphrodites having both CMS and Rf genes and those without CMS but with Rf genes) as well as fitness comparisons are needed to quantify the parameters described in a theoretical model (Dufaÿet al., 2007). A molecular marker was recently developed for the Rf in R. sativus cultivar Kosena (Koizuka et al., 2003). Unfortunately, we were unable to use this marker to genotyping the Rf locus in this population (Murayama, Yahara & Terachi, unpublished work). Based on the preceding studies implying that restoration has a cost only on pollen production but not on seed production in cultivated R. sativus (Yamagishi & Terachi, 1994; Delph et al., 2007), we estimated the parameters for the case where a cost of restoration acts on pollen and is dominant. Relative seed fitness of FCMS, HCMS and HN will be FA*ws, ws and 1, respectively, according to the model by Dufaÿet al. (2007), where ws is a proportional effect of a CMS cytotype for fitness through female function, and FA is the female advantage (but note that relative seed fitness of females will be FA*ws). We thus can estimate ws by comparing the fitness of HN and HCMS, and we can estimate FA by comparing the fitness of FCMS and HCMS. A rough estimate of ws from our current data is 0.65 (total fruit weight) to 1 (no significant differences in seed size) and that of FA is 1 (no significant differences in most parameters) to 1.08 (total seed number). By plugging these values into the model by Dufaÿet al. (2007), we find that the maintenance of gynodioecy is possible, depending on the cost of restoration (less than 0.5 when FA*ws is close but more than 1) and the value of FA*ws (more than 1). Considering seed number and seed size as estimators for FA*ws, it could be between 1 and 1.08, which will marginally fulfill the condition. However, as these parameters, especially FA, may vary among populations with different environmental conditions, further empirical studies with different environmental conditions are needed.


In summary, we found no female fitness advantage in a comparison between females and hermaphrodites without a CMS gene in a natural gynodioecious population of R. sativus. Moreover, comparison between hermaphrodites with and without the CMS gene suggests a substantial cost of the CMS gene for fitness through female function. These results may imply that there is no opportunity for the CMS gene to spread by positive selection in this species. However, the CMS gene is found in many populations and the frequency varies among populations (Murayama et al., 2004). By putting these results together with our previous study, we suggest that the CMS gene in this species may be maintained by two factors. First, as discussed in Murayama et al. (2004), lower effective population size for cytoplasmic genes relative to nuclear genes and genetic drift may play a role in the occasional fixation of the CMS gene in some populations. Second, females may have a seed fitness advantage in some environmental conditions where they grow large, because females invest more resources into fruit and seed production than hermaphrodites when biomass is large. Further empirical studies within different environmental conditions are needed to confirm this possibility.


The authors thank The National Park Resort Village in Shikanoshima Island for permission to conduct the experiment, Dr. S. Kobayashi, Dr. M. Kinoshita and Dr. K. Ohashi for valuable comments on the experimental designs, Dr. E. Kasuya for his valuable suggestions and advice in statistical analyses and Dr. D. E. Wolf and Dr. M. S. Olson for valuable comments on the manuscript.