Has adaptation occurred in males and females since separate sexes evolved in the plant Silene latifolia?

Abstract

The evolution of separate sexes may involve changed expression of many genes, as each sex adapts to its new state. Evidence is accumulating for sex differences in expression even in organisms that have recently evolved separate sexes from hermaphrodite or monoecious (cosexual) ancestors, such as some dioecious flowering plants. We describe evidence that a dioecious plant species with recently evolved dioecy, Silene latifolia, has undergone adaptive changes that improve functioning in females, in addition to changes that are probably pleiotropic effects of male sterility. The results suggest pervasive adaptations as soon as males and females evolve from their cosexual ancestor.

1. Background

When separate sexes evolve from hermaphroditism, each sex is freed from constraints that may have prevented it reaching its optimum in terms of physiology and gene expression. Such a change is therefore predicted to trigger adaptive evolution, and genes throughout the genomes of many species are indeed expressed differently in the two sexes [1,2]. The time scale of such changes cannot, however, be studied in sexually dimorphic animals, many of which have genetically determined separate sexes controlled by systems established long ago in their evolutionary history. Moreover, the current state in many animals, such as mammals and insects, evolved through many subsequent changes. In such animals, a single ‘master sex-determining gene’ controls gender, and downstream effects on many other genes control further aspects of development. Control of many genes' expression involves sex-specific factors, such as the sex hormones of some species, or splicing variants in some insects (reviewed by [3]). Even if the master gene is replaced by a new gene that takes over control of sex determination (e.g. [4–7]), the new gene controls only the developmental decision to become male or female, while the older established downstream machinery controlling gene expression may function unchanged. Sex biases of individual genes can also evolve, potentially resulting in new phenotypes becoming sex-biased or sex-limited in their expression, for example, when a new sexually dimorphic trait evolves, as has occurred in several clades of beetles [8]. Finally, most animal taxa are either functionally hermaphroditic (cosexual) or have separate sexes (often called gonochoristic), and evolution of gonochorism from cosexuality is known to have occurred in few taxa. It thus remains unclear whether the predicted changes occur over a long time scale, or in a burst soon after separate sexes first appear.

By contrast, flowering plants have evolved separate sexes many times, within a much more recent evolutionary time span than in most animal taxa (reviewed by [9]), offering systems in which to test what kinds of changes the evolution of separate sexes (dioecy) triggers. Here, we studied changes in gene expression in a species in section Melandrium of the genus Silene, in which dioecy evolved from hermaphroditism in the past approximately 5 million years [10–12]. This plant has distinct sexual dimorphism, with males producing more, but smaller, flowers than females, and differing in leaf properties [13].

The sex-biased expression has now been documented in Silene latifolia [14] and two distantly related dioecious plant species with highly dimorphic inflorescences, Mercurialis annua [15] and Salix purpurea [16], and differences between the sexes appear to be much commoner in flower buds than leaves. We previously showed that expression of many genes in S. latifolia buds differs from that in hermaphrodites of the related gynodioecious species S. vulgaris [14]. These changes could either be adaptations to being male or female, or pleiotropic effects of male or female sterility. To distinguish between these possibilities, we studied the expression of the same set of genes in S. vulgaris females. Here, we show that the dioecious species females exhibit larger changes than those in S. vulgaris females, suggesting that adaptations to the female state have occurred in S. latifolia.

2. Material and methods

Most S. vulgaris natural populations are gynodioecious, with cytoplasmic sterility factors (CMS) and nuclear restorer genes [17,18]. The androecium of females is reduced compared with hermaphrodites’ stamens and is highly reduced in the dioecious species S. latifolia (see, for example, [19]).

For this study, we generated new expression data from flower buds of four S. vulgaris females, three (V1_05, V1_08, V1_09) from the same cross as the hermaphrodites previously studied [14], plus one derived from the same natural population (population See_02), but probably carrying a different CMS factor. Direct effects of loss of bud parts complicate interpretations of the changes (see below), which could be avoided by studying sex biases in leaf tissue. However, many evolutionarily important effects, such as trade-offs between developing male and female parts, probably only occur in buds. At this stage, stamens of males and hermaphrodites are still developing, and contain no mature pollen.

We analysed expression as previously described for males and females of S. latifolia and S. vulgaris hermaphrodites, with six libraries per sex type [14]. The new and old RNA-seq data are available in the European Nucleotide Archive (ENA accession number PRJEB14171). As previously, the S. latifolia reference transcriptome was used to map reads using BWA [20]. Read counts were imported into edgeR [21]. After normalizing all libraries together, we retained only contigs with more than 1 count per million reads in at least half of the libraries, and identified contigs with expression differences between S. latifolia males and females using simple contrasts (as described in the edgeR manual), with p < 0.05 after correction for multiple tests [22]. Only the 17% of contigs showing sex-biased expression in S. latifolia were further analysed here, including small fold changes. Our analysis cannot reliably identify individual genes with changed expression but can reveal patterns of differences between categories, and detect evolutionary forces [23]. We used the same approach to identify the subset of these contigs whose expression in S. latifolia males or females differed from that in S. vulgaris hermaphrodites.

As described previously [14], S. latifolia contigs were classified by analysing genotypes in a family using the SEX-DETector program [24], into autosomal versus sex-linked (excluding X-hemizygous contigs, which will have low expression in males). Of note, 1513 autosomal and 199 sex-linked non-hemizygous (or XY) genes with sex-biased expression in S. latifolia yielded expression estimates that could be compared between all four sex types, including females as well as hermaphrodites of S. vulgaris. Expression levels of contigs were highly correlated across the two sexes and species, indicating that estimated expression level differences represent biological differences rather than an experimental error.

(a). Selection on different categories of change

Under the null hypothesis of neutrality, phenotypic differences between species should correlate with the variability of each trait among individuals within species, whereas the involvement of selection should increase between-species differences, relative to variation within species. A test based on this principle compares the proportions of changes in categories of expression change that are predicted to be more (or less) likely to be adaptive [25]. We used the means for females or males of S. latifolia and S. vulgaris hermaphrodites, and the standard deviation values for S. latifolia females or males, to compute the quantity:

3.2

3. Results

Almost all changes producing significant sex-biased expression in S. latifolia flower buds involved changes exclusively (as far as could be determined) in one sex (denoted in figure 1 by F < H = M or F = H > M), or, in many cases, in the same direction in both sexes, but larger in one sex than the other (F << H > M or F < H >> M in figure 1). Expression estimates of contigs in the two sexes are highly correlated (electronic supplementary material, table S1), conforming to the widely accepted view that many genetic variants affecting phenotypic traits affect both sexes similarly, though often to different degrees [23,26–28]. Only about 6% of genes showed opposite directions of expression change in the two sexes, which would suggest sexually antagonistic (SA) changes. A few male-biased genes had increased expression in males and decreased in females (F < H < M in figure 1), and even fewer female-biased genes had increased expression in females and decreased expression in males (F > H > M). As previously reported [14], most changes in S. latifolia were in females (figure 1a versus b).

Figure 1.

Summary of numbers of contigs with expression in both sexes of S. latifolia and detectable sex biases in expression between males and females. The changes inferred, relative to the hermaphrodite, are classified into different types, showing (a) changes largely in females, and (b) those mainly in males. The category ‘other types’ shows the total numbers of all other types possible. Each bar distinguishes the numbers of contigs inferred by Zemp et al. [14] to be autosomal and sex-linked (though the numbers are shown here for the first time, as are the numbers of contigs with expression changes in the same direction in both sexes). When only one sex showed a significant change, the other sex is shown as having expression equal to that of the S. vulgaris hermaphrodite (= symbol); if both sexes showed significant changes, the sex with the smaller change is indicated by < or >, and the sex that changed more by << or >>. (Online version in colour.)

(a). Decreases only, or largely, in S. latifolia females

The decreased expression might be predicted to be the commonest change for plants carrying male-sterility mutations, in which androecium loss should dominate expression changes. Most of the 932 autosomal and 95 sex-linked genes with male-biased expression in the dioecious species indeed have the expected lower expression in females than in the S. vulgaris hermaphrodites (some also decreased in both sexes, with smaller decreases in males than females; figure 1a). Importantly, our analyses excluded all 839 S. latifolia contigs whose expression appeared male-limited, and 64 expressed only in females, because these are likely to be expressed specifically in either stamens or pistils. Differences from the expression in the hermaphrodite for such genes are therefore likely merely to reflect the primary loss of the androecium or gynoecium [14] (including increasing normalized expression levels of all other genes in females); restricting our analyses to genes expressed in both sexes of S. latifolia minimizes such effects. As expected, contigs that were male-limited in S. latifolia were indeed expressed at significantly lower levels in the S. vulgaris females than hermaphrodites.

Adaptive changes causing gene expression divergence from S. vulgaris for reasons unrelated to the evolution of dioecy should rarely cause significantly sex-biased expression, which we required for inclusion in our comparisons. The simplest possibility for contigs whose expression decreased specifically in females is therefore that they are expressed strongly in the hermaphrodites' androecium, but not the gynoecium; however, given that we included only genes expressed in both sexes, bud tissues other than the androecium, such as developing petals, must also express the genes analysed. Expression changes due to loss of the androecium or gynoecium, or changes in expression in those tissues, might then go undetected if the expression is low compared to that in other bud tissues. The many genes showing changes in the dioecious species (figure 1) must, therefore, mostly have high expression in male- or female-specific tissues. Many contigs with female-specific decreases in the dioecious species’ flower buds could therefore reflect primary sex differences additional to those that produced fully sex-limited expression.

However, some of the changes producing male biases in the dioecious species could be female-specific adaptive mutations that reduced expression. This category may also include non-sex-specific reductions in expression that were favoured in females when they first arose (e.g. by decreasing androecium development); some of these may have had SA effects that become resolved by changes in males, restoring their expression to values closer to that in hermaphrodites, so that the changes now appear female-specific. This illustrates the difficulty of determining the changes that actually occurred.

To help distinguish between effects of losing female flower parts and adaptive changes in expression, we quantified expression (figure 2). Changes in gynodioecious species females should predominantly reflect the effects (both direct and pleiotropic) of loss of stamens. Figure 2a shows results for genes that have a male-biased expression in the dioecious species due to decreased expression in females (F < H = M in figure 1a). As expected, these genes' expression in males of the dioecious species is similar to that in S. vulgaris hermaphrodites (figure 2a). Also as expected if reduced expression in females of both species is due to a shared cause—loss of male structures—the expression changes in the two species’ females are correlated; for contigs that decreased only in the dioecious species females (F < H = M) R² = 0.40, and R² = 0.36 for contigs that also decreased in males (F << H > M). Importantly, however, changes in the dioecious species females are consistently larger than those in female S. vulgaris. Purely pleiotropic effects of loss of male structures cannot explain this difference, suggesting that adaptive evolution has occurred in the dioecious species females.

Figure 2.

Summary of the estimated expression intensities in contigs with detectable sex biases in expression in S. latifolia males and females due to changes specific to females or males. The figure indicates the bias caused by each type of change in each part (a–d; changes that are mainly in females or males are shown in the corresponding parts of electronic supplementary material, figure S1), and the changes in each sex, relative to the expression in the hermaphrodite, are indicated using the same notation as in figure 1 (for example, ‘F < H = M’ indicates a decreased value in females, and no significant difference between the males and hermaphrodites). The changes are classified into the same major types as in figure 1, but exclude the few contigs with other types of changes. Differences significant with p < 0.05 using Wilcoxon tests, after adjusting for multiple testing [22], are indicated by different letters in the figure. (Online version in colour.)

The patterns of changes are similar for the many male-biased genes with significantly decreased expression in S. latifolia females and also in males (F << H > M; electronic supplementary material, figure S1A). Similar possibilities to those just outlined for contigs without changes in males can explain this category and include adaptations to unisexuality that became fixed in S. latifolia.

(b). Female-biased genes with increased expression in S. latifolia females, or in both sexes

Fewer S. latifolia genes showed female-biased than male-biased expression. Most of these contigs showed increased expression in females (F > H = M, or F >> H < M, in the third and fourth bars from the top of figure 1a, rather than decreases in males as in the top two bars of figure 1b). It might be thought that increases in females cannot be direct effects of losing male structures, and must reflect increased expression of genes expressed mainly in tissues other than the androecium of the hermaphrodite, which experience no primary loss effect. However, one cannot exclude the possibility that androecium development in hermaphrodites may suppress the development of the pistil (and/or other bud tissues). Such competition between floral organs (Darwin's concept of ‘compensation’) [29] may include androecium-expressed genes suppressing the expression of some gene(s) in pistils. Increased expression of some genes in females might therefore reflect the release of such suppression due to primary loss of the androecium. However, this will not increase overall expression in females unless the suppressive effect of the androecium exceeds the expression level in the androecium (whose loss directly decreases expression). Many genes with increased expression in females therefore probably reflect adaptive secondary sexual changes. This is supported by the observation that expression of this set of contigs also increased more in S. latifolia than S. vulgaris females (figure 2b and f).

(c). Changes mainly, or only, affecting S. latifolia males

Changes in males of the dioecious species (figure 1b) again did not predominantly produce expression lower than in the hermaphrodites (M < H or M << H), as predicted for loss of the gynoecium. Indeed, more contigs increased than decreased in expression, and again many genes changed expression in both sexes (F < H >> M, creating female bias, or F > H << M, creating male bias). Unlike the situation for androecium loss in females, we cannot distinguish between direct and pleiotropic effects of gynoecium loss. However, as expected if contigs with decreased expression exclusively or mainly in males mostly reflect loss of the gynoecium, these contigs' expression levels are unchanged in females of either species (the small overall increases in expression in S. vulgaris females, in figure 2c and electronic supplementary material, figure S1C, are not statistically significant, as indicated by the letters in the figures). As before, contigs that increased exclusively in males (figure 2d) may reflect either release from pleiotropic effects in which presence of a gynoecium suppresses the expression of genes in developing stamens, or adaptive strong upregulation in males. Contigs that increased most in males, but also to a lesser extent in females (electronic supplementary material, figure 1C and D, respectively), could therefore include cases reflecting the adaptive evolution in males or partially resolved SA conflicts.

(d). Comparing selection on different categories of expression changes

The hypothesis that adaptation has occurred in the dioecious species leads to the expectation that selection promoted higher proportions of the expression changes in certain sets of contigs than others. As outlined above, specific predictions are that (i) selection will have affected a higher proportion of contigs whose expression increased in the dioecious species compared with those that decreased, and (ii) if SA selection and resolution of conflicts has been important, higher proportions will have changed in both sexes, compared with those with sex-specific changes. The ΔX test described in the Material and methods supports the second prediction (table 1). It also supports the previous conclusion for changes affecting only a single sex that more selected changes occurred in females than males [14]. However, significantly larger ΔX values were found for the categories with decreased expression in the dioecious species, rather than in the expected increase categories (table 1).

Table 1.

Comparisons of ΔX values between different categories of expression change for autosomal genes (each pair of rows shows a pair of categories to be compared). ΔX values were calculated for changes in the dioecious species sex that had changed specifically, or changed most. The other columns show the numbers with values exceeding a specified value, and tests of the significance of the differences in these numbers. The conclusions are unchanged if ΔX values exceeding 1.5 times the interquartile range are used (electronic supplementary material, table S2); the numbers of contigs are then smaller, and there were no outliers in the fourth comparison, so this could be tested only using the 75th percentile.

pairs of categories compared	numbers of contigs (proportions)		p-value of Fisher's exact test
	total exceeding 75th percentile value	total in the category
sex-specific versus changes in both sexes
F < H = M	320 (0.87)	368	0.265 ns
F << H > M	251 (0.99)	252
F > H = M	133 (0.54)	248	0.0125
F >> H < M	154 (0.79)	196
F = H > M	21 (0.32)	66	0.0015
F < H >> M	30 (1.0)	30
F = H < M	8 (0.058)	139	0.0002
F > H << M	16 (0.32)	50
decreases in expression versus increases
F < H = M	320 (0.87)	368	0.0002
F > H = M	133 (0.54)	248
F = H > M	21 (0.32)	66	0.0001
F = H < M	8 (0.058)	139
changes in females versus males
F < H = M	320 (0.87)	368	0.0001
F = H > M	21 (0.32)	66
F > H = M	133 (0.54)	248	0.0001
F = H < M	8 (0.058)	139

The use of ΔX values to compare the likelihoods of adaptive changes in these different categories requires that they are defined (as we did) by criteria other than different magnitudes of expression. The categories we compared might nevertheless differ in this manner, simply because highly expressed genes might also tend to undergo large changes that are particularly likely to be detectable in both sexes. If so, the test result might not reflect adaptation. However, changes should not be larger in one direction or the other, or in one of the two sexes than the other. The results therefore support the inference above, based on larger changes in females of the dioecious than the gynodioecious species, that adaptive changes occur among decreases in expression, in addition to direct and indirect effects of androecium loss.

4. Discussion

(a). Sex-biased expression in a dioecious plant and its gynodioecious relative

The dioecious species S. latifolia has surprisingly many contigs whose expression levels differ from those in S. vulgaris hermaphrodites (table 1 and figure 1), and many of these are also changed detectably in S. vulgaris females compared with hermaphrodites. Many contigs that decreased in expression in females did so in the same direction in females of both species, suggesting that some could be direct or indirect effects stemming from stamen loss. However, the consistently larger changes observed in S. latifolia females than S. vulgaris females (figure 2) suggest that some of the genes represented by these categories of contigs underwent further advantageous changes in females of the dioecious species.

It is more difficult to detect adaptive changes in males, although an important process in the evolution of dioecy probably often involves mutations increasing hermaphrodites’ male function, accounting for the presence in many dioecious plant populations of hermaphrodites known as ‘inconstant males’ showing a range of female function, ranging down to complete maleness (reviewed in [30]). However, we found considerably smaller numbers of contigs with changes in males than females. A possible explanation for this is that expression changes in males might be missed by our analyses, because some differences between S. vulgaris females and hermaphrodites could reflect adaptations of hermaphrodites to the frequent presence of females in S. vulgaris populations [31,32]. Silene vulgaris hermaphrodites have poor female functions, including fruit set (reviewed in [33,34]), possibly reflecting trade-offs with enhanced male function. It remains unclear whether male sterility in S. vulgaris has persisted for long enough for such adaptation to be likely [35], and it is difficult to test whether these hermaphrodites have evolved more male-like expression than the ancestor of S. latifolia.

(b). Correlated changes in both sexes and the possibility of sexually antagonistic selection in plants

Female-suppressing mutations may be advantageous in hermaphrodites and inconstant males in gynodioecious and sub-dioecious populations, but will decrease females' fitness unless they are sex-specifically expressed. Many contigs changed expression in both sexes in S. latifolia, relative to the S. vulgaris hermaphrodites, and expression was strongly positively correlated in the two sexes [14]. This suggests that many genes are expressed in both male and female flower bud parts, and could be important for the development of both.

A major reason for studying expression changes in evolving dioecious plants is that, if the evolutionary change from hermaphrodite to male and female states triggers many adaptive changes in genes with correlated expression patterns, SA effects are likely, as they arise readily when there are sex differences in the direction of selection on phenotypes that both sexes express [36]. In the hermaphrodite ancestor of a plant evolving dioecy, genes expressed in both the gynoecium and the androecium may often exhibit trade-offs, and maintain expression levels consistent with both sex functions. In a derived dioecious species, however, SA mutations can become fixed if the fitness benefits in one sex outweigh the fitness loss in the other, favouring further (sex-specific) adaptive changes, to resolve the conflicts [2].

We suggested above that some increases in expression in the dioecious species may reflect release from pleiotropic effects in which presence of a gynoecium suppresses the expression of genes in developing stamens, or vice versa, because increases seem unlikely for genes expressed in reproductive structures that are lost. However, although such situations are likely to involve SA selection, we cannot firmly conclude that it is involved. It would be very helpful to estimate expression in specific developing tissues, and the information is starting to become available using male-sterile strains with defects in tissues such as tapetum [37]. Microarray analyses of transcripts in synchronized buds of Arabidopsis thaliana reveal complex gene expression changes in early flower development [38], with decreased expression of many genes at the onset of flower formation, followed by increases during the differentiation of floral organs, including genes involved in specifying floral organ identities and patterning, and finally increases of genes involved in carpel or stamen primordia development. Overall, at least a quarter of the genes in this plant's genome change expression during flower formation [39]. We suggest above that increases in expression in the dioecious species reflect upregulation of genes in non-reproductive structures, which could be tested by detailed analysis of these tissues.

SA selection may often arise from sexual selection, and it is sometimes questioned whether it occurs in plants. However, competition for pollinator visits has led to the evolution of exaggerated floral displays that are similar, in terms of their evolutionary effects, to sexual selection in animals [40]. Moreover, in both animals and plants, SA effects may also arise from different physiological requirements of the two sexes, which are important in angiosperms, including maternal care during seed development and fruit ripening. Physiological differences between males and females in dioecious species have been detected [13,41,42], and differences in life-history traits, including flowering time (e.g. [43,44]), and differences during bud development also seem likely. Sex differences in expression in S. latifolia are much commoner in buds than in leaves and are also particularly common on the sex chromosomes [14]. However, unlike results from birds [45], sex-linked genes did not have a higher proportion of changes with large ΔX values than autosomal genes, which would have suggested the involvement of SA mutations.

Information is also needed about the likelihood of sex-specific mutations; if most mutations are sex-specific, SA is less likely, even if many adaptive changes have occurred. Current evidence suggests that sex-specific mutations are uncommon in animals. In Drosophila melanogaster, about 3% of mutations in 2433 genes studied had sex-specific effects, and sex-biased fitness effects were found for about 16% [46]; similarly, in mice about 10% of 2186 knockout mutations affected only one sex, and around 3% affected the two sexes differently [47]. In both studies, the estimates include sterility mutations, which often affect only one sex [46], so these may be over-estimates for these species genomes in general. Empirical evidence is lacking about sex-specificity of mutations in plants in general, including in plants evolving dioecy from hermaphroditism.

The situation with respect to sex-specificity is probably often very different in plants and animals. Specifically, sex-determining genes have repeatedly evolved de novo in plants, so that the sexes generally differ only by a few mutations; for example, females probably often differ from hermaphrodites by a single loss-of-function mutation affecting the androecium [30,48]. In a plant newly evolving separate sexes, flower development cannot yet involve sex-specific proteins that control the expression of downstream targets in a sex-specific manner. Expression of mutations might therefore be predicted to be similar in both sexes until the two sexes evolve gender-specific developmental differences. However, gene expression must often differ between male or female flower structures in hermaphrodite species (in monoecious plants, it may differ in plant regions that will develop female or male flowers). Sex-specific mutations, and secondary sex differences may therefore also be possible in plants. Estimates of the frequency of sex-specific mutations should be obtained for plants, and evidence about how sex-specificity arises.

Data accessibility

Data are available in the European Nucleotide Archive (ENA accession no. PRJEB14171).

Authors' contributions

A.W. and N.Z. conceived the experiment with input from D.C.; N.Z. performed the experiments and statistical analyses with input from D.C. D.C. wrote the manuscript in collaboration with A.W. and N.Z.

Competing interests

We have no competing interests.

Funding

This work was supported through SNF projects 141260 and 160123 to A.W.