M. S. Olson, Institute of Arctic Biology, University of Alaska Fairbanks, PO Box 757000, Fairbanks, AK 99775-7000, USA. Tel.: 907-474-7640; fax: 907-474-6967 e-mail: firstname.lastname@example.org
Fine scale spatial structure (FSSS) of cytoplasmic genes in plants is thought to be generated via founder events and can be amplified when seeds germinate close to their mother. In gynodioecious species these processes are expected to generate FSSS in sex ratio because maternally inherited cytoplasmic male sterility genes partially influence sex expression. Here we document a striking example of FSSS in both mitochondrial genetic markers and sex in roadside populations of Silene vulgaris. We show that in one population FSSS of sexes influences relative fruit production of females compared to hermaphrodites. Furthermore, FSSS in sex ratio is expected to persist into future generations because offspring sex ratios from females are female-biased whereas offspring sex ratios from hermaphrodites are hermaphrodite-biased. Earlier studies indicated that pollen limitation is the most likely mechanism underlying negative frequency dependent fitness of females. Our results support the theoretical predictions that FSSS in sex ratio can reduce female fitness by decreasing the frequency at which females experience hermaphrodites. We argue that the influence of FSSS on female fitness is complementary to the influence of larger scale population structure on female fitness, and that population structure at both scales will act to decrease female frequencies in gynodioecious species. Better comprehension of the spatial structure of genders and genes controlling sex expression at a local scale is required for future progress toward understanding sex ratio evolution in gynodioecious plants.
Gynodioecious species (those with females and hermaphrodites in populations) are good candidates for exhibiting the influence of FSSS on fitness via variation in local sex ratio (McCauley & Taylor, 1997; Olson et al., 2005). Variation in local sex ratio is expected to alter the frequency at which females experience hermaphrodites; flowers on females are more likely to be pollinated when hermaphrodite neighbours are common than when they are rare (Lewis, 1941; Lloyd, 1974; McCauley & Taylor, 1997; McCauley & Brock, 1998; Taylor et al., 1999; McCauley et al., 2000a, b). Theoretical studies predict that with increased variation in local sex ratio, females have lower fitness through seed than under a scenario of random dispersion (McCauley & Taylor, 1997; Olson et al., 2005). The influence of FSSS on the relative fitness of females and hermaphrodites is of particular concern in gynodioecious species because this structuring may influence sex ratio evolution at larger scales. Although recent empirical studies have documented reduced female fitness resulting from low frequencies of hermaphrodites in the local neighbourhood (McCauley & Brock, 1998; Graff, 1999; McCauley et al., 2000a, b, but see Asikainen & Mutikainen, 2003, 2005), only one study was spatially explicit and conducted at a within-population scale (Graff, 1999). For insect pollinated plants, pollen is not usually spread evenly among potential donors; instead, pollinators often visit plants that are in close proximity (Marden & Waddington, 1981; Chittka et al., 1997). Thus, patterns of sex ratio variation at the scale of pollinator movement (meters) may better reflect the frequency at which females encounter pollen donors than previous studies conducted at larger scales (100’s of meters).
Population variation in sex ratio is thought to be generated both through random founder effects and differential selection on females and hermaphrodites that occurs after colonization of a new site (Frank, 1989; Gouyon et al., 1991; Couvet et al., 1998; Taylor et al., 1999; Laporte et al., 2001). Both processes also are likely to influence patchiness of sex ratio within populations. Selection on CMS and Rf genes results from female compensation (females produce more seeds than hermaphrodites Darwin, 1877; Gouyon & Couvet, 1987), pollen limitation (Graff, 1999; McCauley et al., 2000a, b) and sex ratio selection (Jacobs & Wade, 2003). One class of theoretical models predicts that high frequencies of females result from the epidemic spread of CMS types in the absence of their specific Rf genes (Frank, 1989; Couvet et al., 1998). If localized epidemic spread occurs commonly, we would expect to see populations or patches of individuals with high frequencies of females as was observed in Thymus vulgaris populations in a site that was recolonized after fire (Manicacci et al., 1996) and in some populations of S. vulgaris (McCauley et al., 2000a, b). Because this localized escape of CMS from Rf genes may generate female biased patches of individuals and female biased patches may suffer fitness reductions from the lack of pollen sources, CMS genes have been characterized as ‘selfish’ for acting in their own best interest (increased transmission through seed) to the detriment of the total fitness of the individual (Hatcher, 2000; but see Jacobs & Wade, 2003).
Addressing the FSSS for sexes is straightforward, but for CMS genes it is difficult because neighbouring females (or hermaphrodites) may carry different CMS genes (de Haan et al., 1997; M. S. Olson, Unpublished data). Nonetheless, initial studies of within population patch structure can proceed by assessing molecular markers that cosegregate with CMS genes. Because the entire mitochondrial genome is inherited as a single linkage unit (but see McCauley et al., 2005), CMS genes and random mitochondrial (mtDNA) polymorphisms should be in linkage disequilibrium, creating nonrandom associations between mtDNA haplotypes and sex expression (Belhassen et al., 1993; Cuguen et al., 1994; Olson & McCauley, 2002).
Understanding the interplay between neighbourhood sex ratio and female fitness within large populations is required to identify the relevant spatial scale at which to study the ecology of sex ratio selection. Silene vulgaris is one of the best studied gynodioecious plant species in terms of our knowledge of sex ratio and mitochondrial population structure. In Giles and Craig Counties, Virginia populations exhibit high among-population variation in mitochondrial genetic markers and sex ratio (McCauley et al., 2000a, b; Olson & McCauley, 2002). Moreover, among population female frequency is negatively associated with average population female seed fitness (McCauley & Brock, 1998; McCauley et al., 2000a, b). Many populations outside of western Virginia are very large, spreading across kilometres (Storchova & Olson, 2004). Within-population spatial clumping of females and hermaphrodites has been documented in one small population from western Virginia (Olson & McCauley, 2002; McCauley & Olson, 2003), but the extent of this pattern is heretofore unknown.
Here we study FSSS of sex ratio in S. vulgaris to two large roadside populations. Our aim is to explore the potential for direct effects of local sex ratio on fitness through seed at localized neighbourhood scales to better understand localized feedbacks on sex ratio evolution in gynodioecious plants. Within a section of each population we have mapped every individual, assessed their mtDNA, and measured several components of seed fitness. Our study addresses the following hypotheses. Are sexes and/or mtDNA spatially structured within populations? Is sex expression related to variation in mtDNA? Is the seed fitness of individuals related to the sex of their neighbours? The results suggest that spatial structuring of sexes and mtDNA within populations may be ubiquitous and that this structuring indeed affects seed fitness in some instances, but not others.
Study organism and study site
Silene vulgaris is a gynodioecious short-lived perennial herb that is native to Europe and introduced to North America in the 19th century (McCauley et al., 2003). In Virginia S. vulgaris flowers in early June and has set seed by late July. Hermaphrodites are protandrous but are capable of self-pollination either by geitonogamous pollination or a breakdown in protandry. Outcross pollination requires an insect vector; small bees have often been observed visiting the flowers and no specialist pollinator of S. vulgaris is known in Virginia.
The two populations in this study, Mountain View (MV) and Old Mountain View (OMV) were chosen because these populations were larger than any other in Giles County. These populations were located at different ends of a two mile section of state road 622 near Eggleston, Virginia. Although each population was delineated by the absence of additional plants for at least 200 m on either side of the population, additional patches of S. vulgaris plants were present along the roadside between MV and OMV and beyond the ends of each population. We had previous knowledge of high mitotype diversity along SR622 near these populations (MV is population 11 cited in Olson & McCauley, 2002), but no prior knowledge of mitotypes or sex ratios in these specific locations.
Population mapping and fitness components
At each population, all plants were tagged and mapped to the nearest 1 cm by measuring the distance from the plant to two fixed points. Measurements between plants and fixed points were taken using Sonin distance measuring devices (Sonin Inc., Charlotte, NC, USA). The relative locations of each plant were calculated using triangulation and trigonometric calculations. Maps were displayed on a Cartesian plane using Sigmaplot 8.0 (SPSS Inc., Chicago, IL, USA).
The plants were revisited twice during the growing season, once in early June to collect leaf tissue for mitochondrial DNA analysis and count open flower numbers, and a second time in mid-July to count the numbers of fruit produced and collect three capsules from each plant. Fruit set was estimated as the numbers of mature fruit counted in July divided by the numbers of open flowers in June. Seed production was estimated by counting all mature seeds within each of three fruits from each plant using a dissecting microscope in the laboratory. Seeds were counted as ‘mature’ if they exhibited enlargement and a characteristic bumpiness on the seed coat. Germination rates were estimating by randomly drawing 20 seeds from a pooled sample from the three fruits from each plant. Seeds were planted in conical tubes (conetainers, Stewe and Sons, Corvallis, OR, USA) containing a soil-less mix of equal parts coconut coir, perlite and vermiculite. Humidity was kept high by placing racks of 200 container tubes inside large transparent plastic bags. Germination rate was calculated as the numbers of seeds that germinated divided by the number of seeds planted. Two days after germination, plants were removed from the plastic bags and grown under constant light in the greenhouse until flowering. At the time of flowering, up to five flowers on each plant were examined for sex expression. If all examined flowers lacked anthers, the plant was labelled a female; otherwise, the plant was considered a hermaphrodite.
DNA was isolated from fresh or frozen leaf tissue and the mtDNA of each individual was determined using RFLP-Southern blot methods (Storchova & Olson, 2004). Briefly, genomic DNA was digested with HindIII for >6 h at 37°C, the digested products were electrophoresed on a 0.7% agarose gel and transferred onto a positively charged membrane using capillary blotting. This membrane was probed with a labelled 1.5 kb fragment of the cytochrome oxidase I gene amplified from S. vulgaris by PCR as detailed in Storchova & Olson (2004). Mitochondrial haplotypes (mitotypes) were scored by determining the size(s) of the DNA fragment(s) with which the labelled probe hybridized most strongly after exposure on film. In most cases, one strong band was visible; exceptions were mitotype n in which four equally strongly hybridizing bands were visible, and mitotype p in which two equally strong hybridizing bands were visible. Mitotypes were labelled as in Olson & McCauley (2002) with the exception of mitotypes q and r which were previously unidentified. Mitotype q had a restriction fragment size of approximately 2.25 kb and mitotype r had a restriction fragment size of approximately 6.1 kb.
Spatial patterns of sex expression and mitotypes
Variation in sex ratio across the populations was determined using a sliding window analysis implemented by a program created for this study in R 1.7.1 (Ihaka & Gentleman, 1996). A ‘window’ was defined as either a 10 m (MV) or 20 m (OMV) region along the x-axis of the Cartesian plane of each two-dimensional map of the plants. Window sizes differed because plant density differed at MV and OMV. The frequency of females and 95% confidence intervals (CI) around the mean were calculated using all plants within the window (Zar, 1999). The window was shifted 1 m along the x-axis and the sex ratio and CI were recalculated, and so on, until all windows along the x-axis were completed.
We tested for spatial autocorrelation of mitotypes and sexes using Moran's I which detects departures from spatial randomness (Sokal & Oden, 1978). Calculations of Moran's I and 95% confidence intervals were performed using Spatial Genetic Software (SGS) Version 1c (Degen et al., 2001).
Nonparameteric Kruskal–Wallis tests were used to investigate the relationship between mitotype and sex expression within each of the natural populations using the individual plants as the units of replication (Zar, 1999).
Patterns of fitness variation
In the present study three factors were identified that may influence plant seed fitness: (1) environment (population membership), (2) sex expression and (3) the sex expression of neighbours. The residuals from parametric analyses of fruit set and seed set conformed to the assumption of normality, but the residuals from germination percentage and composite seed fitness (fruit set * seed set * germination rate) were not normally distributed (Wilks–Shapiro P < 0.0001 in both cases). For these reasons, parametric two-factor anova was used to analyse the influence of population and sex expression on fruit set and seed set, whereas nonparametric Kruskal–Wallis sign rank tests were used to assess the influences of population and sex expression on total fruit production, germination percentage and composite seed fitness. Treatment effects for the two factor anovas were Population (Pop), Sex Expression (Sex), and the Pop by Sex interaction; population was treated as a random variable for hypothesis testing. Tukey's HSD procedure, which uses a family wise error rate, was used to assess differences among means in cases with >2 treatments (Zar, 1999). The two-factor Kruskal–Wallis analysis with interactions is considered unreliable (Zar, 1999); therefore, we constructed nested hypotheses to first determine the influence of population on seed germination and composite fitness, and then to determine the influence of sex expression on these dependent variables within each population independently. The critical values for hypothesis testing for the two planned tests of the influence of sex expression within each population were adjusted to α = 0.05/2 = 0.025 to avoid inflation of type I error. All analyses were conducted using jmp software (SAS Institute Inc., Cary, NC, USA)
To assess whether the fitness of individuals was related to the sex expression of their neighbors a program was written in R (Ihaka & Gentleman, 1996) to calculate the correlation between an individual's fitness and the neighbourhood sex ratio in a given radius around each plant in the population. To adjust for potential local environmental influences on seed fitness, each individual's fitness was compared to that of its neighbours by dividing the focal plant's fitness component by that of the average value for all hermaphrodites within a given radius around the focal plant.
Progeny sex ratios
Sex expression was determined for 2181 progeny from 211 maternal families. Influences of maternal population origin (Pop), mtDNA, and maternal sex expression (DamSex) on progeny sex ratios were analysed using a generalized linear mixed model assuming a binomial distribution (proc glimmix, Schabenberger, 2005). In addition, progeny sex ratios from plants with mitotype b or d within two patches (PATCH) in the MV population were analysed in two separate analyses with PATCH, DAMSEX and PATCH*DAMSEX as treatment effects. Maternal families were modelled as random effects in all progeny analyses.
Spatial patterns of sex expression and mtDNA
One hundred and seventy-eight individuals (119 hermaphrodites and 59 females; 33.1% female) were mapped along 50 m of roadside at the MV site (Fig. 1a) and 101 individuals (65 hermaphrodites and 36 females; 35.6% female) were mapped along 130 m of roadside at the OMV site (Fig. 1c). Sex ratio varied across each population. At the MV site, sex ratios of plants within 10 m windows ranged from 5 to 70% females. Examination of Fig. 2a reveals that neighbourhoods of individuals at the left side of MV (1st–11th windows) were significantly female biased compared to the overall mean, whereas neighbourhoods of individuals in the 14th–25th windows and past the 39th window were significantly hermaphrodite biased compared to the mean. At MV, neighbours within 9 m of each plant were more likely to be of the same sex than individuals drawn at random from the entire population (Fig. 3a). At OMV individuals were less dense than at MV, so a 20 m sliding window was used to assess neighbourhood variation in sex ratio. At OMV sex ratios within the 20 m sliding window varied from 5 to 56% females (Fig. 2b). Comparisons of the 95% confidence intervals around the mean sex ratio for each 20 m window indicated that this population was female biased in the 9th–11th windows (left side of Fig. 2b) and hermaphroditic biased after the 100th window (right side of Fig. 2b), but throughout the majority of the population local sex ratios did not vary significantly from the mean. Moran's I calculations also discerned spatial autocorrelation of sexes at OMV out to 3 m (Fig. 3c).
Eight mitotypes were identified in this study using HindIII digests probed with coxI; five mitotypes were present at MV and six were at OMV (Fig. 1b, d). Three mitotypes (a, d and n) were present in both populations, whereas two were found only at MV (c and q) and three were found only at OMV (l, r, and p; Fig. 1b, d). Inspection of Fig. 1b and d reveals that mitotypes were strongly structured in space. For instance, at MV all individuals with mitotype n were within 2 m of one another and all individuals with mitotype q were within 9 m of one another. The correlogram for mtDNA at MV indicated that on average mitotypes were spatially autocorrelated at less than 4 m (Fig. 3b). Clustering of mitotypes at OMV was not as visually obvious, yet the average scale of mitotype spatial autocorrelation at OMV was about 4 m, similar to that at MV (Fig. 3d).
Mitotype and sex expression
At both MV and OMV, the ratios of females and hermaphrodites varied with mitotype association (MV –Fig. 4a; Likelihood ratio = 51.2, P < 0.0001; OMV –Fig. 4b; Likelihood ratio = 10.8, P < 0.005). At MV mitotypes b and d were associated with similar numbers of females and hermaphrodites, whereas mitotypes n and q were associated only with hermaphrodites (Fig. 4a). Mitotype c was excluded from statistical analyses because it was found in only one individual. At OMV mitotype d was associated with approximately equal numbers of females and hermaphrodites, whereas b and n were each associated with fewer females than hermaphrodites (Fig. 4b). Statistical analyses were not conducted for mitotypes l, r and p because they each were found in six or fewer individuals. Mitotypes b, d and n were present in both populations; no significant differences were found in the sex ratios of individuals associated with the same mitotype across both populations (two tailed Fisher's exact tests P > 0.05 in all cases).
At MV the strong spatial segregation of groups of plants allowed within-population comparisons of individuals with the same mitotype. Plants carrying mitotype b occurred in two discrete patches in MV (one on the right and one on the left of Fig. 1b). Thirty-eight of 71 (55%) of plants with mitotype b in the large patch on the left of Fig. 1b were females, but in the patch to the right end of Fig. 1b, only two of 19 (11%) of individuals carrying mitotype b were females (two tailed Fisher's Exact test P < 0.0005). Thus, associations between gender and mitotype were not static even over spatial scales as small as 50 m. Plants carrying mitotype d were also patchily distributed at MV; however, the sex ratios of individuals carrying mitotype d in these two patches did not differ (two tailed Fisher's exact test P > 0.10).
Patterns of fitness variation
Generally, seed fitness component differences between sexes were greater at MV than OMV (Fig. 5). Differences in fruit set were expressed as an interaction between population and sex (Fig. 5a; Pop: F1,1 = 1.47, P = 0.44; Sex: F1,1 = 6.00, P = 0.19; Sex*Pop F1,266 = 13.28, P < 0.0005). Tukey's HSD comparisons indicated that females at MV had higher fruit set than either females at OMV or hermaphrodites at MV and OMV (Fig. 5a); females at OMV had higher fruit set than hermaphrodites at both MV and OMV. Differences in the numbers of seeds per fruit were expressed as an interaction between population and sex (Fig. 5b; Pop: F1,1 = 2.98, P = 0.33; Sex: F1,1 = 2.12, P = 0.38; Sex*Pop: F1,257 = 9.46, P < 0.005). Tukey's HSD comparisons indicated that females at MV had higher numbers of seeds per fruit than hermaphrodites and MV and OMV and females at OMV (Fig. 5b). At OMV, females did not produce a statistically significant higher number of seeds per fruit than hermaphrodites (Fig. 5b). Seed germination rate did not differ significantly across populations (Kruskal–Wallis = 0.76, P = 0.38), but the seeds produced by females had higher germination rates than those produced by hermaphrodites in both populations (Fig. 5c; MV: Kruskal–Wallis = 11.1, P < 0.001, OMV: Kruskal–Wallis = 6.6, P < 0.01). Finally, plants at MV had higher composite seed fitness (fruit set * seeds/fruit *% germination) than plants at OMV (Kruskal–Wallis = 12.0, P < 0.0005), and females had higher composite seed fitness than hermaphrodites in both populations (Fig. 5d; MV: Kruskal–Wallis = 56.9, P < 0.0001, OMV: Kruskal–Wallis = 16.2, P < 0.001).
Fruit set of females was related to the frequency of hermaphrodites in the local neighbourhood at MV, but not at OMV. Figure 6 shows the correlation between the 3 m neighbourhood sex ratio and relative fruit set at MV; females had higher relative fruit set in neighbourhoods with more hermaphrodites (Kendall's τ = 0.489, P < 0.0001; Fig. 6a). This correlation also was significant at MV when the sex ratio and relative fruit set were calculated for 1, 2, 4, 5 and 10 m neighbourhoods (data not shown, P < 0.001 in all cases). At OMV the correlation between the 3 m neighbourhood sex ratio and relative fruit set of females was nearly significant (Kendall's τ = 0.238, P = 0.059; Fig. 6c), but the correlation was not close to significant at any other neighbourhood scale (1, 2, 4, 5 and 10 m all P > 0.10). Neither female seed set nor the germination rates of seeds produced by females were correlated with the frequencies of hermaphrodites in the local neighbourhood at MV or OMV at any neighbourhood scale (data not shown; in all cases P > 0.05.). Relative fruit set for hermaphrodites was not related to neighbourhood sex ratio at MV or OMV (Fig. 6b and d; Kendall's tau P > 0.05 in both cases). Likewise, seed set and germination rates were not related to neighbourhood sex ratio at MV or OMV at any scale (data not shown, Kendall's tau P > 0.05. in all cases).
Progeny sex ratios
Analyses of progeny sex ratios were restricted to individuals with the three mitotypes (b, d and n) that were present at both MV and OMV. Because there were no females with mitotype n at MV, we tested for influences among population, maternal sex, and mtDNA on progeny sex ratio using only mitotypes b and d (Table 1). Maternal sex expression strongly influenced progeny sex ratio (Table 1); maternal dams produced female biased progeny sex ratios [least square (LS) mean 0.42] whereas hermaphrodite dams produced hermaphrodite biased progeny sex ratios (LSMean 0.83). No other treatment or interaction among treatments had significant influence on progeny sex ratio (Table 1). In order to evaluate whether progeny sex ratios of maternal parents carrying mitotype n differed from those of individuals carrying mitotypes b or d, progeny from only hermaphrodite maternal parents were compared (Table 2). In this analysis, both mitotype and population origin influenced progeny sex ratio (Table 2). An a posteriori, contrast comparing progeny sex ratios from mothers with haplotype n to those from b and d (combined) indicated that progeny sex ratios from hermaphrodites mothers with mitotype n (LSMean 0.93) produced significantly more (F1,122 = 5.5, P < 0.03) hermaphrodites than individuals with haplotypes b and d (mitotype b LSMean = 0.81, mitotype d LSMean = 0.87). Also, progeny from MV mothers were more hermaphrodite biased (LSMean 0.91) than those from OMV mothers in this analysis (Table 2; LSMean 0.82).
Table 1. Type III tests of fixed effects on progeny sex ratios from maternal plants with haplotypes (b) and (d) using proc glimmix (SAS).
Pr > F
The maternal parent nested within the interaction Pop*Damsex*mtDNA was treated as a random effect and used as the denominator for calculating F. Damsex refers to the maternal parents sex expression.
Table 2. Type III tests of fixed effects on progeny sex ratios from hermaphrodite maternal plants with haplotypes b, d and n using procglimmix (SAS).
Pr > F
The maternal parent nested within the interaction Pop*mtDNA was treated as a random effect and used as the denominator for calculating F.
MtDNA, mitochondrial types.
Sex ratios of progeny from mothers carrying either mitotype b or d in the different patches at MV were not significantly different (mitotype b: F1,79 = 0.34, P = 0.56; mitotype d: F1,30 = 1.94, P = 0.17).
Although only a handful of studies on cytoplasmic FSSS have been conducted, every published study of which we are aware on chloroplast or mitochondrial FSSS has detected it (McCauley et al., 1996; Tarayre & Thompson, 1997; Levy & Neal, 1999; this study, Laporte et al., 2001; Olson & McCauley, 2002; Klaas & Olson, in press). Several studies also have shown that cytoplasmic genes are more highly structured than nuclear genes at metapopulation and within-population scales (McCauley, 1994, 1998; McCauley et al., 1996; Levy & Neal, 1999; Laporte et al., 2001). Fine scale patchiness of cytoplasmic genotypes may be the rule rather than the exception in plants, but generalizations are premature because most of the above studies were conducted on gynodioecious or dioecious species. If the mitochondria is inherited as a single linkage unit, mitochondrial genes other than the one studied here (coxI), such as CMS genes, also would be highly structured at localized spatial scales. However, extending the results from S. vulgaris to predicting patterns in CMS diversity must be tempered by the fact that the exact relationship between mitotypes and CMS types is not yet known.
Founder effects are considered the most plausible mechanism generating FSSS in plants (McCauley et al., 1995; Laporte et al., 2001), but for gynodioecious species local epidemic spread of CMS types also may generate these patterns. In the simplest theoretical explication of localised epidemic spread, progeny from a female will be 100% female assuming that neighbouring hermaphrodites do not carry restorers for the CMS she carries. It also is common for females of gynodioecious species to produce more seeds than hermaphrodites, a characteristic known as female compensation (Darwin, 1877; Jolls & Chenier, 1989, Ashman, 1994). Put together, the CMS and mitotypes that these females carry may spread locally, as will the female phenotype, until countered by a force such as pollen limitation or immigration of genes that restore male fertility. Although founder effects are undoubtedly important for generating cytoplasmic FSSS in gynodioecious species (Tarayre et al., 1997; Laporte et al., 2001), local epidemic spread may further elevate patchiness and FSSS of cytotypes.
The heritability of sex expression observed at MV and OMV is consistent with the persistence of patchiness of sex ratio into the next generation; progeny of females were primarily females, whereas most progeny of hermaphrodites were hermaphroditic. It is unclear, however, whether the persistence of patchiness results from selection (epidemic spread) or from slow breakdown of cytonuclear linkage disequilibrium of genes controlling sex expression that is generated in the initial founding event. If epidemic spread is indeed a strong force, it is not unreasonable that it may be detected in comparative data sets. An example of this approach is exemplified by a recent meta-analysis which showed that selfing species tend to have higher levels of spatial genetic structure for nuclear genes than mixed mating and outcrossing species (Vekemans & Hardy, 2004).
Whatever its basis, patchiness in sex ratio influenced the relative seed fitnesses of females and hermaphrodites in S. vulgaris. In particular, neighbourhood sex ratio (% hermaphrodites) was positively correlated with increased female fruit production in one population (Mountain View – MV). Theoretically, increased variation in sex ratio experienced by females decreases the frequency at which females experience hermaphrodite pollen donors; thus, with pollen limitation, female contribution to the global seed pool decreases (McCauley & Taylor, 1997). Assuming a linear relationship between sex ratio and relative female: hermaphrodite fruit set (Fig. 6a best fit linear regression: y = 0.44 + 2.55*x), we can predict that if all females had experienced a sex ratio equal to the average across MV (66.9%), average fruit set of females relative to hermaphrodites would be 2.08, compared to the empirically measured ratio of 1.69. Thus, at MV FSSS alone is predicted to have reduced female fruit production relative to hermaphrodites by nearly 19%. Although the influences of FSSS on fitness may be spatially or temporally variable (e.g. compare patterns at MV & OMV and see Graff, 1999), the overall influence is predicted to decrease the ability of females to spread in gynodioecious populations and decrease the global frequency of females compared to hermaphrodites.
CMS genes may differ in their molecular action and metabolic influences resulting in different overall effects on plant fitness. Some models suggest that these differences contribute to predicting selection patterns that allow the joint maintenance of polymorphism of CMS and male fertility restorers (Gregorius & Ross, 1984; Ross & Gregorius, 1985; Bailey et al., 2003). Assuming a 1 : 1 relationship between mitotype and CMS type, females carrying the same mitotype may express a different average fitness than females carrying another mitotype. However, distilling the influence of mitotype on fitness in this unmanipulated natural population was complicated by the spatial clustering of mitotypes and unequal sample sizes. Every seed fitness component was influenced by the population origin, indicating that the environment has strong influence on seed production and viability. This influence combined with the strong spatial clustering of mitotypes in both MV and OMV indicate that patterns of mitotype-specific fitness differences will be very difficult to disentangle from micro-environmental influences on fitness. When spatial position and mitotype are completely confounded, as in the current study, it is unclear whether genetics has more influence on fitness than environment, or vice versa, and experimental manipulations are warranted.
An earlier study of S. vulgaris populations in this same region of Virginia as was studied here showed that population sex ratio was heritable and it influenced the relative seed fitness of females compared to hermaphrodites (McCauley et al., 2000a, b). In this earlier study, some populations were as large as MV and OMV whereas others were smaller (∼20 m in diameter). Here we showed that patch structure (=FSSS) in sex ratio at MV may influence relative female: hermaphrodite fitness on small within population scales of ∼3 m. Although both studies are complementary, they raise the issue of whether sex ratio dynamics in the longer term are likely to be driven more by metapopulation dynamics (extinction-colonization) or by small-scale structure within a population. The primary difference between how we define newly colonized populations that form within a metapopulation and newly colonized patches within existing populations is the proximity to established populations. Proximity influences probability of establishment, genetic variation, and pollen limitation. Because seeds generally fall close to their mothers (Levin et al., 2003), we might expect that colonization close to existing populations (new patches) is more common than colonization far from existing populations (new populations) and that the degree of genetic variance among newly colonized patches (or populations) may increase with distance from established populations (Giles & Goudet, 1997). Moreover, erosion in variation among patches due to gene flow should proceed more slowly in more isolated patches (McCauley et al., 1996; Giles & Goudet, 1997; Giles et al., 1998; Richards et al., 1999). Since pollen dispersal also is expected to be greater among patches than among populations, pollen limitation at the population scale has the potential to achieve greater severity (or at least greater variability) than at the patch scale. Patch- and population-scale structure in sex ratio both hamper the ability of females to persist and spread globally. Ultimately, the relative influences of patch and metapopulation scale factors on sex ratio dynamics may reduce to the relative frequency of patch and population colonization in nature and whether the effects of proximity on patch (or population) properties that influence plant fitness are linear or nonlinear.
This and other studies of local populations of S. vulgaris in western Virginia show that nuclear and mitochondrial genetic variation and the influences of sex ratio on fitness vary from patch-to-patch and population-to-population (McCauley, 1998; Taylor et al., 2001; Olson & McCauley, 2002). The emerging spatial model of the evolution of CMS, restorers, and sex ratio in gynodioecious S. vulgaris might be best described as a complex mosaic (Gomulkiewicz et al., 2000) whereby different cyto-nuclear co-evolutionary patterns result from the different relative frequencies of CMS and restorer genes. For instance, where local neighbourhoods have monotypic sex expression and mtDNA, the local neighbourhood is probably fixed for both restorers and CMS type and there is no opportunity for cyto-nuclear co-evolution. Likewise, fixation of either restorers or CMS types, but not both, would provide a genetic environment in which one or the other component could evolve, but not both. Finally, true coevolutionary responses of both CMS and restorers should be found only in populations or patches with polymorphism in both components. Selection on sex ratio in gynodioecious plants is the outcome of selective factors acting at the individual and patch or population levels and these factors may be further modified by the particular CMS and restorer genes found in particular populations, the size and density of populations, and their proximity to adjacent populations. The maintenance of joint cytonuclear polymorphism in gynodioecious plants will remain a mystery until an empirical understanding of these complex systems develops to an extent that observations in nature can guide theoretical studies instead of vice versa, as is now largely the case.
We thank D. McCauley, N. Takebayashi, and D. Wolf for thoughtful discussions during the course of this research and manuscript preparation. A. Blair aided with seed collections in the field. The manuscript was aided by suggestions from G. Houliston, D. McCauley, D. Wolf, and four anonymous reviewers. Logistical support was provided through Mountain Lake Biological Station. This research was supported by NSF DEB-0361637 to MSO, the Institute of Arctic Biology and Alaska EPSCoR.