In many gynodioecious species, cytoplasmic male sterility genes (CMS) and nuclear male fertility restorers (Rf) jointly determine whether a plant is female or hermaphrodite. Equilibrium models of cytonuclear gynodioecy, which describe the effect of natural selection within populations on the sex ratio, predict that the frequency of females in a population will primarily depend on the cost of male fertility restoration, a negative pleiotropic effect of Rf alleles on hermaphrodite fitness. Specifically, when the cost of restoration is higher, the frequency of females at equilibrium is predicted to be higher. To test this prediction, we estimated variation in the cost of restoration across 26 populations of Lobelia siphilitica, a species in which Rf alleles can have negative pleiotropic effects on pollen viability. We found that L. siphilitica populations with many females were more likely to contain hermaphrodites with low pollen viability. This is consistent with the prediction that the cost of restoration is a key determinant of variation in female frequency. Our results suggest that equilibrium models can explain variation in sex ratio among natural populations of gynodioecious species.
A common dimorphic sexual system in plants is gynodioecy (Richards 1986), in which populations contain female and hermaphroditic individuals. Biologists since Darwin (1877) have asked how female plants can persist given that they reproduce only via seeds, whereas hermaphrodites reproduce via both seeds and pollen. The answer to this question depends on whether the sex-determining genes are located in the cytoplasm or the nucleus (reviewed in Bailey and Delph 2007a). When sex is determined strictly by nuclear genes, models (reviewed in Charlesworth 1999) suggest that females will persist if they produce more and/or higher quality seeds than hermaphrodites. This fitness advantage is known as reproductive compensation (Darwin 1877). However, in many gynodioecious species, sex is jointly determined by cytoplasmic and nuclear genes. Cytoplasmic male sterility (CMS) genes cause male sterility and matching nuclear genes (Rf) restore male fertility (Kaul 1988); multiple CMS types and Rf alleles segregate within natural populations (e.g.,Koelewijn and van Damme 1995; Charlesworth and Laporte 1998; Dudle et al. 2001; Garraud et al. 2011). When sex determination is cytonuclear, models (e.g., Gouyon et al. 1991; Bailey et al. 2003; Bailey and Delph 2007b) suggest that the frequency of females in a population at equilibrium depends not just on reproductive compensation, but also on the cost of male fertility restoration.
The cost of male fertility restoration is a negative pleiotropic effect of Rf alleles on hermaphrodite fitness (Charlesworth 1981). Although equilibrium models differ in their assumptions about the mode of this cost (reviewed in Delph et al. 2007), they generally predict that females will not be maintained unless hermaphrodites that carry excess (i.e., nonmatching) Rf alleles have lower fitness than hermaphrodites that do not carry excess Rf. Because costs limit the spread of Rf alleles within populations, the frequency of females at equilibrium is predicted to be higher when the cost of restoration is higher (reviewed in Bailey and Delph 2007a). Despite the predicted importance of a cost of restoration, it has been estimated empirically in only four gynodioecious species [Plantago lanceolata (de Haan et al. 1997), Lobelia siphilitica (Bailey 2002; Case and Caruso 2010), Beta vulgaris (Dufaÿ et al. 2008; Delph and Bailey 2010; de Cauwer et al. 2012), and Phacelia dubia (del Castillo and Trujillo 2009)]. The cost of restoration has been difficult to estimate because Rf genotypes have not been identified in natural populations of any gynodioecious species (reviewed in Delph et al. 2007), but instead must be inferred from controlled crosses.
We tested whether variation in the cost of restoring male fertility was correlated with variation in female frequency among populations of gynodioecious L. siphilitica. Lobelia siphilitica is an excellent system for this study for two reasons. First, there is empirical evidence for a cost of restoration. Bailey (2002) used a breeding design to manipulate genotypes at restorer loci and found that L. siphilitica hermaphrodites that carried excess Rf alleles had significantly lower pollen viability than hermaphrodites that carried fewer of these alleles. In addition, Case and Caruso (2010) used offspring sex ratios to infer the relative number of Rf alleles segregating in controlled crosses. We found that variation in pollen viability among L. siphilitica hermaphrodites reflected the relative number of Rf alleles carried, rather than inbreeding depression. Specifically, L. siphilitica hermaphrodites with low pollen viability always carried a relatively large number of Rf alleles; the combination of low pollen viability and relatively few Rf alleles was not observed. Taken together, these data suggest that the presence of L. siphilitica hermaphrodites with low pollen viability indicates a high cost of restoring male fertility. Second, there is evidence that the cost of restoring male fertility varies significantly among L. siphilitica populations (Case and Caruso 2010). We found that the cost of restoration was higher in a small population with many females relative to large populations with few females. However, these data were preliminary because we were unable to sample populations spanning a broad range of sex ratios (see Case and Caruso  for details).
We estimated pollen viability of greenhouse-grown hermaphrodites from 26 L. siphilitica populations containing 0–69% females. If variation in female frequencies among these populations reflects variation in the cost of restoring male fertility, then we expect to find some hermaphrodites with low pollen viability in L. siphilitica populations with high female frequencies. In contrast, hermaphrodites from L. siphilitica populations with low female frequencies should have consistently high pollen viability. This will result in higher variance in pollen viability among hermaphrodites from high female L. siphilitica populations than among hermaphrodites from low female populations.
Materials and Methods
Lobelia siphilitica is a short-lived, herbaceous perennial wildflower that grows in wet meadows and woods throughout eastern North America (Johnston [1991a] and references therein). Its 3-cm-long blue flowers are primarily pollinated by Bombus spp. (Beaudoin Yetter 1989). Plants flower from early August until early October and fruits dehisce from mid-September through November. Dry, dehiscent fruits contain numerous seeds that are passively dispersed (C. M. Caruso and A. L. Case, pers. obs.). Perfect flowers are protandrous and pollen is shed from a tube formed by the fused anthers and filaments (Johnston 1991a). Although L. siphilitica is self-compatible (Johnston 1992), the staminate and pistillate phases of development do not overlap. This reduces the opportunity for autonomous self-fertilization of perfect flowers (Johnston 1991b). Lobelia siphilitica can also reproduce clonally via rosettes that overwinter and produce a flowering stalk the following summer (Beaudoin Yetter 1989).
Female frequency varies widely among L. siphilitica populations (0–100%; Caruso and Case 2007) and gender is determined by both cytoplasmic and nuclear genes (Dudle et al. 2001). There are at least three different CMS types segregating within L. siphilitica (Dudle et al. 2001; Bailey 2002), but variation in the frequency of these different types has not been established. However, when L. siphilitica females are crossed with hermaphrodites from distant populations, they frequently produce hermaphrodite offspring (Dudle et al. 2001; Case and Caruso 2010; C. M. Caruso and A. L. Case, unpubl. data). These data suggest that different CMS types and their matching Rf alleles are widely distributed in L. siphilitica.
As part of a larger project to determine the causes of variation in population sex ratio, we sampled open-pollinated seeds from 26 L. siphilitica populations located in the midwestern United States and adjacent Canada (Table S1). Nineteen populations were sampled in 2009 and seven populations were sampled in 2010. We collected mature fruits from 7–24 (mean [1 SE]= 18.54 [0.87]) haphazardly selected plants per population, resulting in a series of open-pollinated maternal families. This sampling design allowed us to estimate typical costs of restoration across a range of possible restorer genotypes for a sample of CMS types within each population. For 20 of the populations, we estimated the population sex ratio during the same growing season that we sampled fruits. For the other six, we used our most recent estimate of the population sex ratio. To estimate sex ratio, we visited each population and sexed all flowering plants by determining whether they produced pistillate or perfect flowers. Female frequency was calculated as the percentage of plants that were female.
To generate plants for pollen sampling, we germinated seeds from 19 populations in November 2009 and eight populations in January 2011 (Table S1). One population (Yellowwood I) was included in both experiments. The methods used to break dormancy differed between the two experiments. For the 19 populations germinated in 2009, we exposed seeds from one haphazardly chosen fruit per family to a 16:1:1 mixture of distilled water, bleach, and 70% ethanol, with four drops of triton X-100 detergent added per 2 L of solution (Dudle et al. 2001). Seeds were then rinsed with distilled water and kept at 2°C for two days, until planting. For the eight populations germinated in 2011, seeds from one haphazardly chosen fruit per family were placed on moist filter paper in a petri dish, wrapped in parafilm, and stratified at 4°C for eight weeks prior to planting (Johnston 1992).
In November 2009 and January 2011, we sowed 100–200 seeds per maternal family into plug trays filled with Sunshine mix (Sun Gro Horticulture Canada Ltd., Vancouver, British Columbia, Canada). Trays were placed in standing water and their position on the greenhouse bench was randomized. In December 2009 and February 2011, we transplanted 20–30 seedlings from each of 7–20 (mean [1 SE]= 11.35 [0.71]) families per population into 656 mL3 Deepots (Stuewe and Sons Inc., Tangent, Oregon, USA) filled with Sunshine mix (N= 4461 seedlings in 2009 and N= 2101 in 2011). Each plant was randomly assigned a position on the greenhouse bench. Plants were watered as necessary, fertilized with eight pellets of Nutricote Total 13–13-13 (Plant Products, Brampton, Ontario, Canada), and exposed to supplemental light (16-h days).
At flowering, we sexed each plant by determining whether it produced pistillate or perfect flowers. We then collected pollen from 1–22 (mean [1 SE]= 9.95 [0.2]) hermaphrodites from each of 5–17 (mean [1 SE]= 10.31 [0.51]) families per population. The samples of both families within populations and hermaphrodites within families were haphazardly chosen. Pollen was collected from two flowers per hermaphrodite. Whenever possible, these two flowers were selected from the first 10 flowers produced. We used fine forceps to squeeze pollen from the anther cylinder into a microcentrifuge tube, and immediately vortexed the pollen in 25 μL of lactophenol-aniline blue stain (Kearns and Inouye 1993). Forceps were cleaned with 70% EtOH and dried between each sample. Samples were stored at room temperature until pollen was counted. We collected pollen from N= 2036 hermaphrodites in the 2009 experiment and N= 643 hermaphrodites in the 2011 experiment. A small number of plants (N= 12 across both experiments) produced both pistillate and perfect flowers (i.e., gynomonoecy), but they were excluded from our pollen sampling.
To estimate pollen viability, we viewed two 7 μL subsamples from each pollen sample at 40× magnification. We tallied the number of viable and inviable grains in each sample by starting at one corner of the cover slip, moving the microscope stage horizontally, and scoring pollen grains that intersected the center point of an ocular reticle grid. Once the edge of the cover slip was reached, the stage was moved forward one-quarter turn to make another horizontal pass until 200 pollen grains were scored for viability. Grains were considered viable if they stained dark blue and appeared filled. We considered grains inviable if either of these conditions were not met.
We used aniline blue to estimate pollen viability for three reasons. First, pollen samples in aniline blue can be stored for extended periods at room temperature. This differs from methods such as MTT and peroxidase, where samples are not shelf stable (Rodriguez-Riano and Dafni 2000; Bailey 2002). Consequently, by using aniline blue, we were able to estimate the pollen viability of a much larger sample of hermaphrodites. Second, aniline blue likely provides a conservative estimate of pollen viability. This is because inviable grains may contain some starch and thus will appear viable when stained with aniline blue. Third, estimates of L. siphilitica pollen viability using aniline blue were very similar to estimates using other methods. Specifically, when subsamples of pollen from the same L. siphilitica flowers were stained with aniline blue and MTT, the viability estimates were significantly positively correlated across methods (Spearman rank correlation on mean viability of two flowers per plant; rs= 0.66, N= 14 plants, P < 0.01).
Prior to data analysis, we calculated mean percent viable pollen across the two flowers scored per hermaphrodite, and then calculated mean percent viable pollen across all hermaphrodites scored within each family. We used family means for three reasons. First, hermaphrodite siblings are not statistically independent because they share CMS and Rf from their maternal parent. Second, if there is multiple paternity in L. siphilitica fruits, then siblings likely vary in the Rf alleles inherited from their paternal parent. Consequently, the mean pollen viability for an open-pollinated maternal family represents the cost of restoration incurred when the hermaphrodites that carry a particular CMS type sample Rf alleles from the population pool. Third, family means also allow us to include costs of Rf alleles inherited from female dams (i.e., silent costs; see Case and Caruso (2010) for details). This is important because silent costs of restoration are thought to be common (reviewed in Delph et al. 2007).
To increase the power of our analysis, we pooled pollen viability estimates from the 2009 and 2011 experiments. Although environmental conditions can affect pollen fitness (e.g., Peet et al. 1998), two lines of evidence suggest that it was reasonable to pool pollen viability data for L. siphilitica. First, previous work indicates that significant variation in pollen viability of L. siphilitica can be explained by variation in the number and/or identity of Rf alleles, rather than by inbreeding (Case and Caruso 2010). Second, we included offspring from six Yellowwood I families in both experiments and found that estimates of pollen viability were highly repeatable (Pearson correlation coefficient; r= 0.88, df = 4, P < 0.05).
We used two analyses to test whether pollen viability was more variable among open-pollinated maternal families collected from high-female populations, as expected if variation in female frequency reflects variation in the cost of restoration. First, we used Levene's test to evaluate the hypothesis that variance in pollen viability among families was not equal across populations. Second, we used a Spearman rank correlation (rs) to test whether among-family variation in pollen viability was higher in high female populations, as expected if the cost of restoration is an important determinant of variation in female frequency. We estimated variation in pollen viability among families within populations using the coefficient of variation (CV). We used the CV because mean pollen viability differed significantly among L. siphilitica populations (Welch's analysis of variance (ANOVA) assuming unequal variances; F25, 78.82= 2.23, P= 0.004); the magnitude of the CV, unlike other measures of dispersion, is not dependent on the mean (Zar 1999). However, our results were qualitatively similar if we used the variance to estimate among-family variation in pollen viability (data not shown).
We found that the amount of variation in pollen viability differed significantly among L. siphilitica populations (Levene's test, F25, 242= 2.49, P= 0.0002). This is because all populations contained families with high pollen viability, but some populations also contained families with low pollen viability (Fig. 1). The amount of among-family variation in pollen viability, as estimated by the CV, was significantly positively correlated with the population sex ratio (rs= 0.52, df = 24, P= 0.007). Specifically, the CV for pollen viability was higher in populations with higher female frequencies (Fig. 2); the mean (1 SE) CV for populations that contained ≥ 50% females (10.27 [3.12], N= 5) was ∼2.7 times larger than the mean for populations that contained ≤ 5% females (3.78 [0.35], N= 10).
Our data support the prediction that variation in female frequency among populations of gynodioecious species reflects variation in the cost of restoring male fertility. Although this result is consistent with Case and Caruso (2010), our previous study included data on only eight populations that spanned a much smaller range of variation in female frequencies (0–38%), and only included one high female population. Consequently, the data in the current paper represent the best test to date of the prediction that a cost of restoring male fertility is an important determinant of variation in female frequency. These data also suggest that equilibrium models can explain significant variation in sex ratio among natural populations of gynodioecious species (e.g., Gouyon et al. 1991; Bailey et al. 2003; Bailey and Delph 2007b). We are currently testing the additional model prediction that, relative to differences in reproductive compensation, a cost of restoration explains more of the variation in female frequency among L. siphilitica populations.
Our data also provide insight into the relative contributions of natural selection versus metapopulation forces to variation in sex ratio among populations of gynodioecious species. If sex ratio is primarily determined by extinction/recolonization dynamics, gene flow, and/or genetic drift, as suggested by metapopulation models (e.g., Pannell 1997; Couvet et al. 1998; Dufaÿ and Pannell 2010), then populations should rarely be at an equilibrium sex ratio determined by natural selection. However, we detected a significant correlation between the cost of restoration and female frequency in L. siphilitica (Fig. 2), as expected if the sex ratio was at or near equilibrium. Consequently, our data support the hypothesis that while metapopulation forces may be necessary to explain variation in female frequency of some gynodioecious species, they are not a sufficient explanation for this variation (Bailey and Delph 2007a).
One limitation of our study is that we cannot reject the hypothesis that variation in pollen viability in L. siphilitica in part reflects incomplete restoration (Ehlers et al. 2005; Dufaÿ et al. 2008). This is because the frequency of incomplete restoration, like the cost of restoration, should be positively correlated with the frequency of females in a population. However, our data do suggest that incomplete restoration is unlikely to explain variation in pollen viability in L. siphilitica. Specifically, the distribution of pollen viability in L. siphilitica is strongly skewed, with most individuals having high viabilities (compare Fig. 1 with Figs. 2 and 3 in Dufaÿ et al. 2008). In contrast, models of incomplete restoration predict that pollen viability will vary continuously between fully sterile and fully restored genotypes (Ehlers et al. 2005).
Another limitation of our study is that we could not determine whether variation in the cost of restoration among L. siphilitica populations reflects variation in (1) the distribution of Rf alleles or (2) the fitness cost of carrying these alleles. However, progeny sex ratios of within- and between-population crosses do not significantly differ (C. M. Caruso and A. L. Case, unpubl. data), suggesting that the distribution of Rf alleles in L. siphilitica is not geographically structured. Consequently, variation in the cost of restoration among L. siphilitica populations is likely caused in part by variation in the fitness cost of carrying excess Rf alleles.
Although we used greenhouse-grown plants to estimate variation in the cost of restoration among natural populations, the ecological context is likely to modify this cost by influencing the relationship between an individual's genotype and its phenotype (reviewed in Caruso et al. 2012). Such genotype by environment interactions could affect the expression of a cost of restoring male fertility by causing the relationship between the Rf alleles carried by an individual and its pollen fitness to differ between environments. For example, the fitness ranking of families that have different complements of Rf alleles could be consistent across environments, but the variance in fitness among these families could differ (variance G × E; reviewed in Conner and Hartl 2004). Such variance G × E would cause the cost of restoration, estimated as the difference in pollen fitness among individuals with different complements of Rf alleles, to differ between environments. Specifically, the cost of restoring male fertility would be higher in environments where variance in fitness among families was higher. Such an effect of G × E on the expression of cost could cause the frequency of Rf alleles, and thus female frequencies, to differ among populations growing in different ecological conditions, and could help explain why sex ratio varies along ecological gradients in many gynodioecious species (reviewed in Ashman 2006). Testing this hypothesis will require measuring the cost of restoration in contrasting ecological conditions, something that has yet to be done in any gynodioecious species.
Associate Editor: K. Bomblies
We thank A. Benscoter, J. Dennis, P. Euclide, N. Gale, A. Johnstone, E. Knox, H. Madson, M. Mucci, E. Seifert, T. Slimmon, S. Vernon, and A. Vitale for assistance in the field, greenhouse and/or lab. J. Allen, K. Barrett, S. Faulkenham, M. Jones, K. Kinnick, Q. Malott, K. Menard, R. Mitchell, L. Mottl, L. Petit, R. Phillips, D. Taylor, and K. van Zante provided permission to collect seeds. M. Bailey, H. Maherali, and two anonymous reviewers provided helpful comments on an earlier version of this manuscript. This work was funded by grants from the US National Science Foundation (DEB-0842280, awarded to ALC and CMC) and the Natural Science and Engineering Research Council of Canada (awarded to CMC). During the writing of this manuscript, CMC was supported by a sabbatical fellowship from the National Evolutionary Synthesis Center (NESCent, NSF #EF-0905606).