Can Sexual Selection Cause Divergence in Mating System–Related Floral Traits?

Premise of research. The wide diversity of floral traits seen among plants is shaped by neutral and selective evolutionary processes. In outcrossing species, sexual selection from competing pollen donors is expected to be important for shaping mating system–related traits, but empirical evidence is scarce. In a previous evaluation of experimental evolution lines crossed with either one or two pollen donors (monogamous, M, or polyandrous, P, lines) at early floral stages in mixed-mating Collinsia heterophylla (Plantaginaceae), P evolved enhanced pollen competitive ability and reduced maternal seed set compared with M, in accordance with the sexually antagonistic evolution of pollen. Here, we asked whether the presence of sexual selection during pollen competition affects mating system–related floral traits in the same lines. Methodology. We compared flowering start, timing of anther-stigma contact (as an indication of timing of self-pollination), timing of stigma receptivity, and first seed set between M and P and with a source line, S (starting plants). The first three traits are later in outcrossers than in selfers of Collinsia. The last trait was expected to be earlier in P than in M, resulting from sexual selection for early seed siring of pollen. Pivotal results. Artificial polyandry for four generations resulted in later flowering start and later anther-stigma contact in P compared with M, and the latter trait was intermediate in S. Thus, P appeared more “outcrossing” than M. Timing of stigma receptivity did not differ between lines. First seed set was earlier in P than in M, as expected from sexual selection. Conclusions. Our results from the C. heterophylla experimental evolution lines suggest that a component of sexual selection during outcross pollination could enhance the patterns of floral divergence commonly found between outcrossers and selfers.


Introduction
The wide floral diversity in angiosperms is largely considered to be caused by geographic variation in pollinator-mediated selection (Kay and Sargent 2009;Armbruster 2014;Van Der Niet et al. 2014). Because differences in mating systems (e.g., outcrossing vs. self-fertilization or their combination in mixed mating) are correlated with floral and developmental traits (Karron et al. 2012;Barrett 2013), selective forces influencing mating systems also contribute to angiosperm floral diversity. Traits such as small flower size and reduced separation of male-female functions in space and time (herkogamy and dichogamy) are thought to directly favor selfing as reproductive assurance when pollinator visits are unpredictable (Lloyd 1979;Lloyd and Schoen 1992;Opedal 2018). These traits can also be connected to rapid maturation or to reduced investment in cross-pollination in selfers (Snell and Aarssen 2005;Sicard and Lenhard 2011).
It has recently been argued that not only the maternal outcrossing rate but also mate diversity and individual variation in mating success could contribute to selection on the mating system (Barrett and Harder 2017). The influence of sexual selection on mating system evolution should therefore be important to consider, particularly because the degree of sexual selection is expected to be higher in outcrossing taxa than in selfing taxa (Mazer et al. 2010), which could further enhance differences in mating systems. For example, because of parental conflicts or stronger sexual selection in outcrossers than in selfers, previous studies have suggested that pollen competitive ability during pollen competition in the pistil (Mazer et al. 2018), pistil barriers to hybridization (Brandvain and Haig 2005), and male versus female antagonistic influence on seed provisioning (Willi 2013;Raunsgard et al. 2018) are increased in outcrossers compared with selfers. While sexual selection can influence floral traits (Delph and Ashman 2006;Moore and Pannell 2011;Dai and Galloway 2013), the relative contribution of this selective force on floral trait divergence between outcrossers and selfers is usually not considered.
The mixed-mating herb Collinsia heterophylla belongs to a genus with extensive variation in mating systems, from selfpollinating to mixed-mating species (Armbruster et al. 2002;Kalisz et al. 2012). Most outcrossing species have larger flowers, delayed selfing brought about by the loss of herkogamy at late developmental stages, and delayed stigma receptivity, separating the timing of male and female reproductive functions ( fig. 1A). These floral traits also appear to be associated with slower plant developmental rates and later flowering start (Elle et al. 2010). In C. heterophylla, variation in the outcrossing rate is substantial, and populations with higher outcrossing rates have later timing of stigma receptivity . In line with the results at the genus level, delayed selfing measured as the timing of anther-stigma contact was shown to be genetically correlated with the timing of stigma receptivity and tended to be genetically correlated with flowering start (Lankinen et al. 2017a). Flower size, on the other hand, was uncorrelated to both selfing rate and timing of stigma receptivity in previous studies of C. heterophylla (Lankinen et al. 2017a;. The timing of stigma receptivity has been proposed to be affected by a sexual conflict in C. heterophylla (Lankinen and Kiboi 2007;fig. 1B). A sexual conflict involves opposing selection pressures in males and females due to divergent evolutionary interests of the sexes (Parker 1979). When such conflicts occur between different loci in the two sexes (interlocus conflict), selection is expected to move trait values of one sex closer to its fitness optimum, causing a direct fitness cost in the other sex (Parker 1979;Arnqvist and Rowe 2005). Sexual selection for a trait value leading to increased reproductive success can generate sexual conflict (Kokko and Jennions 2014). The sexual conflict over timing of stigma receptivity in C. heterophylla involves (i) the ability of pollen to sire seeds early to secure paternity when stigmas are partially receptive and (ii) a recipient cost of reduced early seed set (Lankinen and Kiboi 2007;Madjidian et al. 2012). We recently studied the evolutionary outcome of this conflict by producing experimental evolution lines by crossing recipients at early floral development with two pollen donors (polyandrous, P) or with one pollen donor (monogamous, M; fig. 2A) for four generations (Lankinen et al. 2017b). Recipients always contributed one offspring to the next generation, thus limiting selection on recipients rather than on pollen donors. We showed that plants in the P line produced pollen with a higher proportion of successful crosses at early floral stages and produced a faster tube-growth rate and reduced seed set compared with M plants (Lankinen et al. 2017b). These results are in accordance with enhanced sexual conflict and antagonistic evolution of P pollen in response to sexual selection (Arnqvist and Rowe 2005).
In the current study, we continued analyzing the experimental evolution lines in C. heterophylla (Lankinen et al. 2017b), asking whether the presence of sexual selection during pollen competition also affected mating system-related floral traits ( fig. 2B). Because of the importance of the timing of stigma receptivity in relation to both the mating system and sexual conflict in C. heterophylla, we hypothesized that sexual selection At later stages, stigma receptivity occurs, and the pistil grows through its own pollen, allowing delayed selfing. Stage 0 p stage of flower opening, stages 1-4 p stage with number of dehisced anthers, respectively. Gray arrows point to the location of the stigma. B, Model of sexual conflict over timing of stigma receptivity, involving opposing selection pressures in male and female function due to their divergent evolutionary interests. Moving trait values of the male function closer to its fitness optimum (blue arrow) in C. heterophylla in an experimental evolution experiment caused a direct fitness cost in the female function in terms of reduced seed set (Lankinen et al. 2017b). Photograph in A by Josefin A. Madjidian. LANKINEN & STRANDH-SEXUAL SELECTION AND FLOWERS 997 on early pollen competitive ability could affect this trait, as well as two other correlated floral traits: flowering start and timing of anther-stigma contact. We hypothesized that the P line would show later flowering start, later anther-stigma contact, and later stigma receptivity but earlier timing of first seed set-that is, stigma receptivity was influenced by both pollen and pistil (table 1). The prediction for later stigma receptivity is based on our previous finding of a negative correlation between male and female influence on first seed set (Hersh et al. 2015), suggesting later stigma receptivity when selecting for earlier pollen influence on stigma receptivity, given that this association is genetically determined. We also compared M and P with a source line, S, representing starting plants outcrossed at late floral stages for one generation (Lankinen et al. 2017b), that is, S was not selected for early performance of pollen and pistil traits. A comparison with S could give an indication of the direction of evolutionary change of M and P from the source.

Study Species and Experimental Evolution Lines
Collinsia heterophylla Buist (Plantaginaceae) is a hermaphroditic, self-compatible, winter annual herb native to California (Newson 1929;Neese 1993). The species is pollinated by longtongued, nectar-feeding bees (Armbruster et al. 2002). Flowers are zygomorphic with five-lobed corollas forming an upper and a lower lip arranged in whorls on spikes. They contain four epipetalous stamens and one single-style pistil that develop into seed capsules containing up to 20 seeds (Armbruster et al. Fig. 2 Overview of experimental design and results of experimental evolution in Collinsia heterophylla. A, Experimental evolution lines M (monogamous, one pollen donor) and P (polyandrous, two pollen donors) were produced by controlled hand-pollinations in the greenhouse for four generations starting from the S (source) line. In both M and P, four flowers per recipient were repeatedly crossed at stages 1 and 2 (unreceptive or partially receptive pistils) involving two pollen donors (gray vs. white pollen grains). In M, each flower received pollen from the same donor, while in P, each flower received pollen from two different donors. B, Overview of estimated trait values related to different life stages and mating systems in S, M, and P lines. Pollen traits (proportion early siring and pollen tube growth rate) represent the male reproductive function, and pistil traits (anther-stigma contact and stigma receptivity) represent the female reproductive function. First seed set is the result of the combined influence of male-based and female-based influence on stigma receptivity and seed set. Results for traits in white boxes are reported in the current study, and results in gray boxes are reported in Lankinen et al. (2017b). Later anther-stigma contact in P because of prediction of later stigma receptivity in P (as these traits are genetically correlated) Later stigma receptivity in P because of earlier pollen-based influence in P (as pollen and pistil components of It should be noted that we studied the evolutionary changes in our experimental populations exposed to either of the two conditions of monogamy versus polyandry rather than performing artificial selection on certain phenotypes (experimental evolution; Kawecki et al. 2012). The evolved lines were unreplicated. We sampled a large number of genotypes within each treatment to capture a high degree of the natural genetic variation. This was expected to minimize the impact of genetic drift in experimental evolution (Fuller et al. 2005). To impose sexual selection on early siring success of pollen in partially receptive pistils, we conducted crosses twice per flower at early floral stages 1 and 2 (i.e., days 1 and 2 after flower opening; fig. 2A). We hand-pollinated four flowers per recipient plant involving two different pollen donors (see a more detailed description in Lankinen et al. 2017b). One of the seeds generated per plant gave rise to the next generation, thus reducing selection on the female reproductive function compared with selection on the male function. All experimental plants were raised from cold-stratified seeds and grown in an insect-free greenhouse. Thus, both S and evolved lines M and P were grown under similar conditions before assessment of traits in 2013.

Estimates of Floral Traits among Experimental Lines
We assessed four floral traits of the S line and of the two evolved lines M and P at the same time ( fig. 2B). We recorded flowering start as the number of days since the first plant, independent of line, started flowering (n families p 21 in S, 25 in M, 29 in P, n p 3-8 plants per family). Timing of antherstigma contact, as an indication of timing of self-pollination, was assessed by noting the floral developmental stage (pnumber of dehisced anthers) when the stigma was in contact with the open anthers (n families p 20-22 per line, n p 2 plants per family, n p 2 flowers per plant and floral stages 1-4). Timing of stigma receptivity was determined in a droplet of 3% hydrogen peroxide (Kearns and Inouye 1993) in emasculated flowers at stages 1-4 (pday after flower opening; n families p 12-17 per line, n p 1 plant per family, n p 1 or 2 flowers per plant and floral stage). Vigorous bubbling on the stigmatic surface (unharmed and pollen-free tissue) suggests activity of stigmatic peroxidase, which has been shown to correlate with the presence of pollen tubes in the pistil following hand-pollination in this species ).
Timing of first seed set was calculated from one-donor handpollinations performed in emasculated flowers at days 1-4 after flower opening (n p 9 recipients and 6 pollen donors per line, n p 16 crosses per recipient involving 2 donors and 2 flowers per stage and donor; Lankinen et al. 2017b). The stigma and upper part of the style were removed 4 h after the crosses to ensure that seed formation occurred only in flowers with stigmas receptive at the time of the cross.
We analyzed differences in the measured traits (i) among S, M, and P or (ii) between M and P using ANOVAs (type III sums of squares) in SPSS (SPSS 2016). Because our main focus was to investigate M and P divergence, the latter analysis was performed when no significant differences were detected among the three lines to evaluate differences between M and P potentially masked by inclusion of S. When more than one plant was estimated per family, we used a nested model including line (fixed) and family (random) nested within line. In other cases, we included only line. Differences among lines were determined by Tukey tests. Additionally, we tested for differences in trait variance between treatments using a one-tailed F-test. Timing of first seed set was square root transformed. Models were evaluated for normality and homogeneity of the residuals.

Results
Flowering start and timing of anther-stigma contact showed divergence between M and P following four generations of experimental evolution (table 2; M-P: flowering start, P p 0:034; anther-stigma contact, P p 0:001). In accordance with our hypotheses, both traits were later in P than in M ( fig. 3A, 3B). Our source, S, had earlier flowering start than the evolved lines (S-M and S-P: P < 0:001; fig. 3A), but no difference was found for timing of anther-stigma contact (P > 0:13; fig. 3B).
Contrary to our hypothesis, timing of stigma receptivity did not differ among all three lines (table 2; fig. 3C) or between M and P (F 1, 28 p 0:026, P p 0:87). Variability was higher in P than in M (F-test; P p 0:037, n p 30) and in S (P p 0:002, n p 26), which was an unexpected result.
Day of first seed set following one-donor pollinations showed, as predicted, earlier formation of seeds in P than in M (Line: F 1, 16 p 4:90, P p 0:042; Recipient nested within line: F 16, 54 p 0:691, P p 0:79; fig. 3D). There was no significant difference among all three lines (table 2).

Discussion
While sexual selection has been studied in plants for several decades, this mode of selection is still not a well-integrated concept 1000 in plant evolution theory, including the theory on mating system evolution (Lankinen and Karlsson Green 2015). In the current study, we investigated a potential link between sexual selection and floral traits related to the mating system in Collinsia heterophylla, a mixed-mating species belonging to a genus with extensive variation in mating systems and associated floral traits (Armbruster et al. 2002;Kalisz et al. 2012). To study how the presence of sexual selection impacts evolution of floral traits, we compared monogamous (M) and polyandrous (P) experimental evolution lines, both outcrossed at early floral stages (Lankinen et al. 2017b). A previous analysis of M and P showed that the presence of sexual selection in P led to higher levels of sexual conflict, with increased pollen competitive ability and reduced seed set. With the help of other previous studies on genetic and phenotypic correlations between floral traits in C. heterophylla (Hersh et al. 2015;Lankinen et al. 2017a), we made specific predictions regarding floral traits of particular importance for mating systems in this species, including flowering start, timing of anther-stigma receptivity, stigma receptivity (all three later in P vs. M; table 1), and first seed set (earlier in P vs. M; table 1).  In C. heterophylla, flowering start and timing of anther-stigma contact, the latter as an indication of timing of self-pollination, showed divergence between M and P. Both traits were later in P than in M, which was in line with expectations from genetic correlations between these two traits and between timing of anther-stigma contact and stigma receptivity (Lankinen et al. 2017a; table 1). Thus, P appeared more "outcrossing" than M (Elle et al. 2010;Kalisz et al. 2012;. The source line (S) had earlier flowering start than both M and P. This result may indicate evolution of later flowering start in both M and P, which were produced at early floral stages compared with S. We cannot, however, exclude that the earlier flowering start in S could be related to the previously found reduction in the seed germination rate and the number of flowers in S (Lankinen et al. 2017b), which may have resulted from longer storage of S seeds than seeds from M and P. Anther-stigma contact in S was intermediate between M and P. For this trait, S was similar to that found in other greenhouse studies using plants from the same population (Madjidian and Lankinen 2009). This suggests that divergence in anther-stigma contact was caused by P becoming later and M becoming earlier than S. We can hypothesize that not only the presence of sexual selection but also the absence of sexual selection during outcrossing can influence divergence of pollen and floral traits. Interestingly, in Clarkia xantiana, strong pollen limitation resulted in disruptive selection through female and male fitness, as reduced herkogamy and protandry increased female fitness, while both large and small petal area increased male fitness (Briscoe Runquist et al. 2017).
Timing of stigma receptivity did not differ significantly between M and P. This result was contrary to expectation (table 1). First seed set, a proxy for stigma receptivity influenced by both pistil and pollen, was earlier in P than in M, as predicted from the previous analyses of M and P (Lankinen et al. 2017b), suggesting that P pollen was more successful at siring seeds in partially receptive pistils. This result, in combination with a detected negative relationship between male and female influence on first seed set within individual plants (Hersh et al. 2015), indicated that we could expect later stigma receptivity in P. While the mean time of stigma receptivity did not differ between M and P, variability of this trait was higher in P compared with both M and S. This result was surprising. One possible explanation is that there was disruptive selection acting on this trait in P, favoring either early or late stigma receptivity. In Drosophila melanogaster, experimental evolution, involving disruptive selection by alternating up and down selection, increased phenotypic variation in wing shape, while fluctuating and stabilizing selection instead decreased the variation (Pélabon et al. 2010). We do not have direct support for disruptive selection on stigma receptivity. We could, however, hypothesize contrasting effects of direct influence of pollen (early pollen-based influence will lead to earlier receptivity; Lankinen et al. 2017b) and indirect genetic covariance (negative correlation between pollen-and pistil-based influence on stigma receptivity; Hersh et al. 2015). Interestingly, several previous studies in C. heterophylla found that stigma receptivity is more variable than anther-stigma contact Madjidian and Lankinen 2009;Hersh et al. 2015).
Despite the timing of stigma receptivity not being significantly later in P, we surmise that the later response in both flowering start and timing of anther-stigma contact in P was a consequence of genetic correlations among traits (Lankinen et al. 2017a). While flowering start was not significantly genetically correlated with the other two traits (Lankinen et al. 2017a), it is possible that these three traits are genetically linked with plant developmental rate. Other studies indicate that selection for rapid development can be correlated with rapid flower maturation (Mazer et al. 2004;Snell and Aarssen 2005;Elle et al. 2010). However, a recent study suggested low levels of genetic covariances between herkogamy and other floral traits across 17 species representing 10 taxonomical families (Opedal et al. 2017). Evolvability of herkogamy was estimated to be 9.07%, which was an order of magnitude greater than evolvabilities of the male and female organs that are components of herkogamy and of flower size. Because these results suggest that herkogamy has a high potential to respond to natural selection, it is possible that this trait is only weakly influenced by a component of sexual selection in other study systems.
A weakness of our study is the lack of replication of the experimental evolution lines. This means that we cannot fully exclude genetic drift as a cause of the results, despite following the recommendation to include a large number of individuals within each line to reduce the influence of drift (Fuller et al. 2005). However, several additional experiments suggest the presence of a sexual conflict over the timing of stigma receptivity in C. heterophylla (Lankinen and Kiboi 2007;Madjidian and Lankinen 2009;Madjidian et al. 2012;Hersh et al. 2015), indicating that the reported differences between M and P in pollen competitive ability and seed set are caused by sexual selection rather than genetic drift (Lankinen et al. 2017b). Moreover, the link between floral traits is well documented both in C. heterophylla (Lankinen et al. 2017a) and among species in Collinsia (Armbruster et al. 2002;Kalisz et al. 2012), further supporting a scenario of selection rather than drift.
In conclusion, the influence of sexual selection in C. heterophylla experimental evolution lines resulted in divergence of mating system-related floral traits, at least for the timing of antherstigma contact, which is related to the timing of self-pollination and to herkogamy. P appeared more "outcrossing" than M. Thus, adding a component of sexual selection during outcross pollination could enhance the patterns of floral divergence regularly seen between selfers and outcrossers (Karron et al. 2012;Barrett 2013). It would be of great interest to learn whether sexual selection could also affect divergence in herkogamy, or other mating system-related floral traits, in other study systems. Such knowledge could lead to a better understanding of how multiple factors influence evolution of plant mating systems and floral divergence. In line with Barrett and Harder (2017), we suggest that future studies should consider sexual selection and mate diversity in relation to selection on, and divergence of, plant mating systems and the possibility that some of the floral trait divergence we see today between outcrossers and selfers is in fact a result of sexual selection.