Divergent selection on flowering phenology but not on floral morphology between two closely related orchids

Abstract Closely related species often differ in traits that influence reproductive success, suggesting that divergent selection on such traits contribute to the maintenance of species boundaries. Gymnadenia conopsea ss. and Gymnadenia densiflora are two closely related, perennial orchid species that differ in (a) floral traits important for pollination, including flowering phenology, floral display, and spur length, and (b) dominant pollinators. If plant–pollinator interactions contribute to the maintenance of trait differences between these two taxa, we expect current divergent selection on flowering phenology and floral morphology between the two species. We quantified phenotypic selection via female fitness in one year on flowering start, three floral display traits (plant height, number of flowers, and corolla size) and spur length, in six populations of G. conopsea s.s. and in four populations of G. densiflora. There was indication of divergent selection on flowering start in the expected direction, with selection for earlier flowering in two populations of the early‐flowering G. conopsea s.s. and for later flowering in one population of the late‐flowering G. densiflora. No divergent selection on floral morphology was detected, and there was no significant stabilizing selection on any trait in the two species. The results suggest ongoing adaptive differentiation of flowering phenology, strengthening this premating reproductive barrier between the two species. Synthesis: This study is among the first to test whether divergent selection on floral traits contribute to the maintenance of species differences between closely related plants. Phenological isolation confers a substantial potential for reproductive isolation, and divergent selection on flowering time can thus greatly influence reproductive isolation and adaptive differentiation.


| INTRODUC TI ON
In angiosperms, flowering time and flower morphology critically influence mating patterns because of their effects on pollen transfer.
Adaptive divergence occurs when selection drives the evolution of traits toward different optima in different populations or species.
Depending on the current trait distributions in relation to these respective optima, divergent selection can be linear in different directions (e.g., Hall & Willis, 2006) or stabilizing with different optima (e.g., Benkman, 2003). Divergent selection on flowering phenology has been documented between lowland and montane populations of Mimulus guttatus (Hall & Willis, 2006), between lowland and alpine populations of Arabidopsis lyrata (Sandring, RiihimäKi, Savolainen, & Ågren, 2007), and between diploid and tetraploid Heuchera grossulariifolia (Nuismer & Cunningham, 2005). Divergent selection on floral morphology has been detected in several studies, including traits that influence the efficiency of pollen transfer such as tube or spur length (Gómez, Perfectti, Bosch, & Camacho, 2009;Rymer, Johnson, & Savolainen, 2010), and traits that influence the attraction of pollinators such as corolla size (Campbell, 2003;Gómez et al., 2009) and number of inflorescences (Sandring et al., 2007). Most of these studies provide examples of divergent selection within species, and only a few studies have tested whether floral differentiation between species is maintained by divergent selection. While there was divergent selection on corolla size between two Ipomopsis species visited by hummingbirds and hawkmoths (Campbell, 2003), this was not the case between two Lobelia species specialized on hummingbirds and bumblebees, respectively (Johnston, 1991). To elucidate which traits contribute to the maintenance of species boundaries, it is necessary to study selection on floral traits that are differentiated between closely related taxa.
In this study, we quantify phenotypic selection on flowering phenology, three floral display traits and spur length in the closely related orchids Gymnadenia conopsea s.s. and Gymnadenia densiflora on the island of Öland, southern Sweden. These two species constitute an excellent system to study divergent selection on floral traits. First, the two species differ in flowering phenology and flower morphology, but also exhibit partly overlapping quantitative variation in these traits in the wild (Jersáková et al., 2010;Stark, Michalski, Babik, Winterfeld, & Durka, 2011). Second, both orchids depend on pollinators for successful fruit set, and significant pollinator-mediated selection on flowering phenology, floral display, and spur length has been documented in G. conopsea s.s. (Chapurlat, Ågren, & Sletvold, 2015;Sletvold, Trunschke, Wimmergren, & Ågren, 2012). Third, the pollinator communities partly differ between the two species, and on Öland, G. conopsea s.s. is mainly visited by nocturnal pollinators, while G. densiflora is mainly visited by diurnal pollinators with shorter proboscis than the nocturnal ones (Chapurlat, Anderson, Ågren, Friberg, & Sletvold, 2018;this study). Fourth, genetic studies suggest interspecific gene flow and introgression between the species, where introgression is associated with reduced fitness (Gustafsson & Lönn, 2003;Lönn, Alexandersson, & Gustafsson, 2006). Our objective is to test for divergent selection on flowering phenology and floral morphology between the two Gymnadenia species. On Öland, G. conopsea s.s. flowers earlier than G. densiflora, produces shorter inflorescences with fewer flowers and longer spurs, and is pollinated by species with longer proboscis (see below). If trait differences are adaptive, we expect optimal flowering to be earlier, optimal flower production and plant height to be lower and optimal spur length to be longer in G. conopsea s.s. than in G. densiflora.
Given sufficient trait variation, this should be evident as directional selection of opposite sign, or stabilizing selection with different optima in the two species.
The Gymnadenia conopsea (L.) s.l. complex is highly variable with regard to morphology, scent production, flowering phenology, and habitat (Gustafsson & Lönn, 2003;Jersáková et al., 2010;Soliva & Widmer, 1999;Stark et al., 2011). The most recent classification based on genetic data recognizes two taxa within the G. conopsea (L.) s.l. complex: G. conopsea (L.) R.Br. s.s. and G. densiflora A. Dietr (Bateman et al., 2003;Stark et al., 2011). These two taxa were previously considered subspecies based on morphological similarity, but they do not even have a sister-species relationship as phylogenetic analyses of the genus have shown that G. odoratissima is the sister species of G. conopsea s.s. (Bateman et al., 2003;Brandrud, Paun, Lorenz, Baar, & Hedrén, 2019;Sun et al., 2015). Gymnadenia odoratissima differs from the other taxa in color, floral scent, and morphology and was thus not previously included in the G. conopsea (L.) s.l. complex. Furthermore, variation in ploidy levels ranging from diploids to hexaploids has been reported in G. conopsea s.s., with diploids and tetraploids being the major cytotypes (Trávníček et al., 2012). No tetraploid G. conopsea s.s. has been found in Sweden, where diploids dominate, even though some triploid individuals have been identified (Stark et al., 2011;Travnicek et al., 2012).
Both species produce a single inflorescence of ca 10-100 fragrant pink flowers ( Figure 1) that open sequentially from the bottom to the top of the inflorescence. Individual flowers remain open for up to a week while individual plants may flower for a month. A narrow spur contains nectar that is produced throughout anthesis (Stpiczynska & Matusiewicz, 2001). Each flower contains two pollinaria which are situated above the spur entrance. Both species are self-compatible, but depend on pollinators for successful fruit set (Sletvold, Grindeland, Zu, & Ågren, 2012). The available literature indicates that diploid G. conopsea s.s. flowers earlier than G. densiflora (Jersáková et al., 2010) and produces shorter inflorescences with fewer flowers (Stark et al., 2011). The two species also differ in floral scent (Jersáková et al., 2010). In contrast, there is no consistent difference in spur length, as G. conopsea s.s. had shorter spurs than G. densiflora in a study conducted in the Czech Republic (Jersáková et al., 2010), while the opposite has been reported in Germany (Stark et al., 2011).

| Study sites and pollinator communities
The ten study populations are located on the calcareous island in close proximity (20-100 m) but with slight habitat separation, as well as in truly mixed populations. Flow cytometry conducted on leaves (see below) revealed that the two species grow in sympatry (populations ≤100 m apart) at five of the sites (Gråborg, Igelmossen, Ismantorp, Kalkstad, Melösa) but, except at Gråborg, selection was quantified in only one of the species at each site.
On Öland, the two species share several nocturnal pollinators,  Table S2). While flowers of both species are visited both diurnally and nocturnally, nocturnal visitors are more frequent than diurnal ones in populations of G. conopsea (mean visits per hour, 6.8 vs. 0), whereas the opposite trend is observed in populations of G. densiflora (0.6 vs. 1.9), based on 123 hr video recordings at night, and 68 hr at day, in two populations of each species. Nocturnal pollinators also contribute more than diurnal pollinators to reproductive success of G. conopsea s.s. (Chapurlat et al., 2015(Chapurlat et al., , 2018).  ; Table S1). There was no

| Statistical analyses
All analyses were conducted with R 3.1.3 (R Core Team, 2015). Data from four of the study populations (G. densiflora at Gråborg, G. conopsea s.s. at Kvinneby, Långlöt, and Melösa) were also included in a previous study (Chapurlat et al., 2015). for the full models, indicating that the level of collinearity was not problematic (Quinn & Keough, 2002).
Phenotypic selection studies cannot distinguish the causal effects of focal traits from potential environmentally induced covariances between traits and fitness unless trait expression is manipulated (Mauricio & Mojonnier, 1997;Rausher, 1992). This is likely to be a problem mainly for size-related traits, and the best approach to deal with this if you cannot use genotypic selection is to include measures of overall plant size in the model. We included both plant height and number of flowers in our phenotypic selection models.
To test for divergent linear selection, we conducted for each floral trait a one-sided Welch t test on the linear selection gradients, with the alternative hypothesis being that selection gradients are greater in the species with the largest mean trait value. We examined whether there was stabilizing selection (presence of an intermediate optimum) graphically by the use of added-variable plots.

| Differences in floral traits and reproductive performance between the two species
Flowering start and floral display differed between the two species (Table 1). On average, Gymnadenia conopsea s.s. individuals flowered earlier (Figure 3), were shorter, produced fewer and smaller flowers but had longer spurs than G. densiflora individuals, although the difference in spur length was only marginally significant (Table 1; Figure S1). The observed phenotypic distributions overlapped between species, ranging from a small overlap for flowering start ( Figure 3) to a large overlap for the morphological traits ( Figure S1).
Floral traits were moderately positively correlated within each population, except flowering date, which tended to be negatively correlated with the other traits (Table S3). Number of fruits and fruit mass differed significantly between the two species (Table 1). Gymnadenia conopsea s.s. individuals produced fewer but heavier fruits than did G. densiflora individuals, which led to marginally significant higher average female fitness for G. densiflora.

Gymnadenia species
In both species, there was significant directional selection on all floral traits included in the analysis, but only flowering start tended to experience divergent selection between the two species ( Figure 4; Table S4). There was selection for earlier flowering in two G. conopsea s.s. populations and for later flowering in one G. densiflora population ( Figure 4; Table S4). In addition, selection for longer spurs tended to be stronger in G. conopsea s.s. than in G. densiflora, but selection on display traits did not differ between species (Figure 4; Table S4).
There was no indication of divergent stabilizing selection. Only two quadratic gradients were statistically significant; one positive for number of flowers in Kalkstad, and one negative for spur length in Ismantorp (Table S5). However, added-variable plots revealed that the negative quadratic selection gradient for spur length reflected curvature but no intermediate optimum. .079

| D ISCUSS I ON
In this study, we tested the hypothesis that floral divergence be- introgression into G. densiflora was associated with reduced fitness (Gustafsson & Lönn, 2003;Lönn et al., 2006). Interspecific pollen deposition during the overlapping flowering period may thus be costly and could potentially cause divergent selection, as has been hypothesized for diploid and tetraploid Heuchera (Nuismer & Cunningham, 2005 Sletvold, Moritz, & Ågren, 2015). In four of the included study populations, spatial variation in net selection on flowering start is partly explained by variation in pollinator-mediated selection (Chapurlat et al., 2015). The divergent selection observed between G. conopsea s.s. and G. densiflora could thus be caused by temporal variation in pollinator communities throughout the flowering season.
However, some of the net selection on flowering start is nonpollinator mediated in the Kvinneby population, suggesting that abiotic factors could also contribute to the selection gradients (Chapurlat et al., 2015). Phenological isolation between two plant taxa is the earliest premating barrier possible and has the greatest potential for reproductive isolation (Widmer, Lexer, & Cozzolino, 2009), and our results suggest that divergent natural selection should reinforce this barrier between the two Gymnadenia species.
The strength and direction of linear selection on spur length, a trait influencing the efficiency of pollination (Boberg & Ågren, 2009;Ellis & Johnson, 2010;Nilsson, 1988;Sletvold & Ågren, 2011;Trunschke et al., 2019), varied among populations, but there was little evidence of divergent selection between the two species.
Overall, selection on spur length tended to be stronger in the longer-spurred species, G. conopsea s.s., with significant selection for longer spurs in two of the six populations. In the shorter-spurred G.
densiflora, there was selection for longer spurs in one population.

G. densiflora populations
considerably larger differences in spur lengths between taxa (e.g., Anderson, Alexandersson, & Johnson, 2010;Fulton & Hodges, 1999;Nilsson, 1983Nilsson, , 1988Sun et al., 2015). Furthermore, reports on mean spur length in the two species indicate that the direction of difference varies throughout their range (this study , Jersáková et al., 2010;Stark et al., 2011). Studies that characterize differences in pollinator communities and test for floral isolation between the two Gymnadenia species in several parts of their range could help elucidate whether local differences in spur length are adaptive.
There was no evidence of divergent selection on floral display traits, that is, plant height, number of flowers and corolla size, in spite of significant differences in these traits between the two Gymnadenia species, suggesting this differentiation is nonadaptive.
Rather, the differences in display traits may in part represent plastic responses to habitat differences between species. Gymnadenia densiflora, which on average produces larger floral displays than G. conopsea s.s., grows in more moist conditions (Gustafsson & Lönn, 2003), which could favor growth. Differentiation in these traits may also be caused by pleiotropic effects if they are genetically correlated with other floral trait(s) that have been subject to divergent selection. Because G. densiflora begins to flower later, it has more time to gather resources before flowering and may therefore be able to produce larger floral displays (cf. Elzinga et al., 2007;Mitchell-Olds, 1996). Although difficult to conduct in orchids, common-garden experiments with half-sib crossings would be the ideal way to test for genetic differences and genetic correlations among traits in the two species.
Both studied species are long-lived perennials, and potential trade-offs across the life cycle may cause selection estimated via a single fitness component to deviate from estimates via lifetime fitness (e.g., Gómez, 2008). Field experiments in G. conopsea populations in Norway demonstrate that maximizing fruit production via supplemental hand-pollination is associated with significant shortterm costs in terms of reduced survival, flowering probability, and fruit production the next year, compared to individuals with natural pollination and fruit production (Sletvold & Ågren, 2011b. However, using a combination of experimental and long-term demographic data, Tye, Dahlgren and Sletvold (2020) showed that such costs do not carry over to later years and are too weak to counteract the advantage of high seed production in the first year. This suggests a minor role of conflicting selection via other fitness components, and a substantial correlation between seed production in a single season and lifetime female fitness. Ideally, effects on male fitness should also be considered, but because pollen removal is often a poor predictor of pollen export (Johnson, Neal, & Harder, 2005) or siring success (Snow & Lewis, 1993), paternity analyses would be required to reliably quantify selection through male function.
While many studies have examined whether spatial variation in selection on floral traits can explain differentiation of these traits within species (Chapurlat et al., 2015;Gómez et al., 2008Gómez et al., , 2009Gross, Sun, & Schiestl, 2016;Hall & Willis, 2006;Sandring et al., 2007;Schueller, 2007), our study is among the first to test whether variation in selection on floral traits can explain the maintenance of floral trait divergence between closely related species (but see Campbell, 2003;Joffard, 2017;Johnston, 1991). Our results indicate that divergent selection contributes to the marked phenological differentiation between Gymnadenia conopsea s.s. and Gymnadenia densiflora, but also show that current selection patterns do not mirror morphological floral divergence between the two species. This suggests that nonadaptive processes such as genetic drift or pleiotropic constraints may play a role in the floral trait differentiation between the two species, or that selection has driven this differentiation historically but is not strong any longer (Harder & Johnson, 2009). Further investigations are needed to fully understand whether floral differentiation between G. conopsea s.s. and G.
densiflora is adaptive, and the extent to which phenological and floral isolation act as reproductive barriers between the two species.
Phenological isolation between two plant taxa has a substantial potential for reproductive isolation (Widmer et al., 2009), and divergent selection on flowering time reported here and in other studies (Hall & Willis, 2006;Nuismer & Cunningham, 2005;Sandring et al., 2007) can thus greatly influence reproductive isolation and differentiation.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.