Lack of strong selection pressures maintains wide variation in floral traits in a food-deceptive orchid

Abstract

Background and Aims

Angiosperms vary remarkably in traits such as colour, size and shape of flowers, yet such variation generally tends to be low within species. In deceptive orchids, however, large variation in floral traits has been described, not only between but also within populations. Nonetheless, the factors driving variation in floral traits in deceptive orchids remain largely unclear.

Methods

To identify determinants of variation in floral traits, we investigated patterns of fruit set and selection gradients in the food-deceptive orchid Orchis purpurea, which typically presents large within-population variation in the colour and size of the flowers. Using long-term data, fruit set was quantified in two populations over 16 consecutive years (2004–2019). Artificial hand pollination was performed to test the hypothesis that fruit set was pollinator-limited and that selfing led to decreased seed set and viability. Annual variation (2016–2019) in selection gradients was calculated for three colour traits (brightness, contrast and the number of spots on the labellum), flower size (spur length, labellum length and width) and plant size (number of flowers, plant height).

Key Results

Fruit set was, on average, low (~12 %) and severely pollinator-limited. Opportunities for selection varied strongly across years, but we found only weak evidence for selection on floral traits. In contrast, there was strong and consistent positive selection on floral display. Selfing led to reduced production of viable seeds and hence severe inbreeding depression (δ = 0.38).

Conclusion

Overall, these results demonstrate that the large variation in flower colour and size that is regularly observed in natural O. purpurea populations is maintained by the consistent lack of strong selection pressures on these traits through time.

Introduction

Angiosperm flowers are renowned for their enormous variation in size, colour, odour and shape (Sprengel, 1793; Barrett, 2010). While variation in floral traits is obvious at the species level, substantial variation in floral morphology can also be found among populations within a single species and even among individuals within a single population (Galen, 1999). One plant family that shows extreme variation in floral traits within species is the orchid family (Orchidaceae) (Dormont et al., 2019). This variation is particularly noticeable in deceptive orchid species, i.e. orchids that offer no reward to their pollinators, and can range from continuous variation (e.g. Sletvold et al., 2016) to more extreme cases of discrete variation, where two or more floral morphs can be discerned among (e.g. Delle-Vedove et al., 2011) or even within populations (Gigord et al., 2001; Kellenberger et al., 2019).

The forces driving variation in floral traits are complex and can involve multiple selection pressures that not only comprise pollinators (Sletvold, 2019), but also local abiotic conditions (Caruso et al., 2019), antagonistic agents such as herbivores (Frey, 2004; Ågren et al., 2013) or the presence of other (rewarding) plant species (Hopkins and Rausher, 2012). However, because the colour, odour, shape and size of flowers are the main signals that plants use to attract pollinators searching for rewards (Chittka and Raine, 2006; Schiestl, 2015) and because the sensory systems of insects are specifically adapted to detect and learn variation in floral traits (Chittka and Menzel, 1992), animal pollination has commonly been regarded as the most likely factor driving floral variation (Darwin, 1862; Fægri and van der Pijl, 1966; Harder and Johnson, 2009). In deceptive orchid species, pronounced variation in the colour, size or shape of flowers may be a successful strategy to lure pollinators and to guarantee fruit set as it will take some time before pollinators realize that the flowers offer no reward (Heinrich, 1975; Nilsson, 1980; Cropper and Calder, 1990; Ackerman and Galarza-Pérez, 1991; Ackerman et al., 2011). In the most extreme case, this variation can lead to the establishment of different colour morphs within populations (Gigord et al., 2001; Salzmann and Schiestl, 2007).

While the importance of pollinators as a driving force in shaping variation in floral traits is widely accepted, their role as selection agents probably depends on local environmental and climatic conditions, and therefore selection pressures on floral traits can be expected to vary among species (Trunschke et al., 2017), and even among populations within a single species (Weber and Kolb, 2013; Ågren et al., 2013; Chapurlat et al., 2015). Sletvold et al. (2013), for example, showed that in the food-deceptive orchid Dactylorhiza lapponica, pollinators mediated strong selection for taller plants in habitats with tall vegetation. In habitats with short vegetation, no significant selection on plant height was observed and pollinator-mediated selection on the number of flowers and spur length was substantially reduced. In general, the opportunity for selection is expected to be stronger under severe pollen limitation (Harder and Aizen, 2010) or under low plant densities (Caruso, 2002). Trunschke et al. (2017) investigated 12 orchid species and showed that species experiencing the highest pollinator limitation showed the strongest selection pressures on floral traits, whereas species showing high levels of fruit set were less affected by selection. Because pollinator visitation rates are likely to vary among years as a result of differences in climatic conditions and/or pollinator abundance, it can be assumed that within a single plant population, selection on floral traits varies strongly across years (Caruso et al., 2003; Siepielski et al., 2009; Kimball et al., 2012; Scopece et al., 2017). Therefore, to better understand the effects of selection on floral variation in natural populations, studies ideally need to be performed across multiple populations and over several consecutive years (Scopece et al., 2017).

Although floral traits that increase attractiveness to pollinators and hence increase fruit production may be under positive natural selection, the selection of these traits may also involve fitness costs, such as a reduced seed set, seed viability or seedling growth, as a result of increased selfing through geitonogamous pollination (Husband and Schemske, 1996). In this case, increased fruit production due to higher attractiveness towards pollinators may be counterbalanced by reduced seed production and progeny fitness, resulting in no difference in reproductive success between attractive and less attractive plants and no or little selection on floral traits. Previous research has already indicated that early-acting inbreeding depression can be strong in orchids and therefore contributes substantially to variation in fitness (e.g. Johnson et al., 2003; Peter and Johnson, 2009; Sletvold et al., 2012; Brys et al., 2016).

The overall aim of this study was to investigate the strength and direction of natural selection on floral traits in the food-deceptive orchid Orchis purpurea. This species is known to display considerable variation in flower colour and size (Fig. 1) (Kretzschmar et al., 2007). The basic colour of the labellum ranges from almost white to light purple, with all possible intermediate colours present. Dark purple paintbrush-like papillae are arranged in dots and spots on the labellum, and often provide a pronounced contrast (Fig. 1). Fruit set in this species is generally low (Jacquemyn et al., 2009; Jacquemyn and Brys, 2010), providing ample opportunities for selection (Trunschke et al., 2017). We assessed variation in flower traits (brightness and contrast, number of spots, labellum length and width, and spur length) and plant size (number of flowers and inflorescence height) in two populations of O. purpurea in Belgium and examined patterns of fruit set across a period of 16 years (2004–2019). Hand-pollinations were conducted to assess the costs of inbreeding and the extent of pollinator limitation. Finally, selection was quantified in both populations across multiple years to test whether the wide variation in colour and size can be explained by divergent selection pressures on these traits.

Fig. 1.

Example of variation in flower colour and size that can be observed in a single population of the food-deceptive orchid Orchis purpurea. All pictures were taken from flowers collected from plants growing in the grassland population investigated in this study.

MATERIALS AND METHODS

Study species

Orchis purpurea Huds. (lady orchid) is a tuberous, diploid (2n = 42), perennial orchid that has a wide distribution in the Mediterranean (Kretzschmar et al., 2007). The species can be found predominantly in calcareous grasslands, often in the immediate vicinity of trees and shrubs, or in calcareous forests, where it is often found in light gaps or along the forest edge. It occurs scattered through France and Central Europe and extends to Corsica and the mountains of central Italy, where it is one of the most common woodland orchid species (Rose, 1948). The species reaches the limit of its north-western distribution in the northern part of Belgium (Flanders), the Netherlands and Denmark (Kretzschmar et al., 2007). A limited number of populations also occur in the UK (Rose 1948). In Flanders, the lady orchid is rare and threatened (Jacquemyn et al., 2005).

Orchis purpurea tubers hibernate during the winter and the leaves appear above ground in February. Plants have one to four (sometimes up to seven) basal elliptic–ovate to lanceolate leaves, 2–5 cm wide and 6–20 cm long (Jacquemyn et al., 2010b). Flowering takes place at the end of May. An individual rosette produces one single inflorescence, but on rare occasions plants can consists of multiple rosettes and produce more than one inflorescence. Under light conditions, plants may flower for two or more consecutive years, whereas under a closed canopy flowering is mostly followed by one or several years in a vegetative state (Jacquemyn et al., 2010a; Miller et al., 2012). Dormancy seldom occurs in this species. The species is long-lived (Jacquemyn et al., 2010b), and vegetative reproduction may occur, although this is limited to a few individuals.

Plants produce inflorescences that carry between 10 and 70 bright white to purple–brown flowers (Fig. 1). The height of the flowering stalk usually varies between 25 and 60 cm, sometimes reaching 80 cm (Jacquemyn et al., 2010a). The flowers are long-lived (21–28 d if unpollinated) and nectarless, but they produce a sweet odour. Like most other orchid species, O. purpurea is self-compatible, and in our study area flowers are pollinated by a wide range of generalist pollinators (most often bees and flies, and to a lesser extent butterflies and beetles). Spontaneous self-pollination does not occur in this species (Jacquemyn and Brys, 2010). There is also no specific nectar plant that acts as a model for mimetic resemblance. Seed capsules ripen by the end of June, followed by dehiscence and seed dispersal in August. From mid-August onwards, all above-ground resources are reallocated to the tuber and no living green parts are found above ground.

Study sites

Data were collected at two sites in Voeren (Belgium). The first site (hereafter referred to as the ‘forest population’) (50°44′21″N, 5°50′54 ″E) consists of a hornbeam (Carpinus betulus) forest with occasional occurrence of wild cherry (Prunus avium) and common ash (Fraxinus excelsior). Here, plants are found in a large canopy gap, which was created in 2000 in an attempt to restore the population to its original state. The second site (hereafter referred to as the ‘grassland population’) (50°43′57.3″N, 5°51′23.6″E) consists of a calcareous grassland bordering a beech forest. Here, most plants grow in the immediate shade of the forest, although some plants can be found in full sunlight as well. The two sites are situated about 1 km away from each other, and they have the same elevation (210 m a.s.l.).

Floral traits

To assess variation in floral traits, a single flower of each flowering individual was collected in both populations between 2016 and 2019 and brought to the laboratory for further inspection. Because flowers become gradually smaller from proximal to distal parts of the inflorescence, we collected a flower from the most central part of the inflorescence. Sampling one flower per inflorescence was deemed sufficient to assess floral traits as preliminary analyses investigating inter- and intraplant variation in flower traits have shown that variation in flower traits within plants was much smaller than variation among plants. Each flower was dissected and digital scans were made from each dissected flower using the same procedures as in Jacquemyn et al. (2012). Using the image analysis software ImageJ 1.33, three different flower characters were measured for each flower: spur length, width and length of the labellum. For each flower, we also determined the number of marked spots on the labellum. Achromatic brightness of the central part of the labellum and the spots on the labellum (Fig. 1) was assessed by calculating the mean RGB value of both flower parts as I = 1/3 [R + B + G] (Lambert and Carron, 1999; Sletvold et al., 2016). Brightness values vary between 0 (no light, black) and 1 (full light, white). The contrast between the spots (dark) and the background colour (light) was determined by calculating the difference between the brightness scores of the labellum and those of the spots. For each sampled plant, we also determined the height of the inflorescence and the number of flowers.

Annual variation in fruit set

Data on fruit production were collected over 16 consecutive years (2004–2019) at both sites. All plants in both populations were mapped to the nearest centimetre and individually labelled (for more details see Jacquemyn et al., 2010a, b). Each year, all plants were carefully inspected to see whether they were flowering and for each flowering plant the number of flowers was counted at peak flowering (mid-May). At the same time, plant height was measured as the distance from the ground to the top of the flowering stalk. In early July, when fruits had matured, all plants were visited again and the number of fruits was counted. For each plant, fruit set was calculated as the percentage of flowers that produced a fruit.

The impact of selfing on seed set and seed viability

For both populations, ten flowering individuals were randomly selected, ensuring that they were located at least 5 m from each other. During flowering, two pollination treatments were randomly assigned to 20 flowers of each plant with a target of ten replicates per treatment and plant. Pollination treatments were supplemental cross- and self-pollination during anthesis after flowers were emasculated immediately after opening. Treated flowers were marked with a colour pencil at the base of the flower to indicate the different pollination treatments (red and black for selfing and outcrossing, respectively). Cross-pollen was obtained from individuals growing in the same population, while anthers that were removed in emasculation were used for self-pollination. Two pollinia from two different individuals were used for the cross-pollination treatment and from two different flowers for the self-pollination treatment to saturate all available ovules. When the fruits were ripe, fruit set following the different pollination treatments was determined per plant, and all fruits were harvested to manually count all seeds per fruit in the laboratory using a dissecting microscope. Average seed production was then determined for each plant and pollination treatment. A tetrazolium treatment (Van Waes and Debergh, 1986) was used to determine the viability of selfed and crossed seeds. Three randomly collected batches of seeds were inspected under a dissecting microscope per fruit by counting up to 50 seeds per batch. Seeds were classified into viable seeds (seeds with a red coloured embryo) and non-viable seeds (seeds lacking an embryo or having an uncoloured and/or abnormal embryo). For each maternal plant and pollination treatment, we calculated the mean proportion of viable seeds and mean total female fitness at fruit level as: (average number of seeds) × (average proportion of viable seeds). Finally, we estimated inbreeding depression for each maternal plant as:

$${\displaystyle \begin{array}{l}{\displaystyle \delta =\frac{{\omega }_{\mathrm{O}}-{\omega }_{\mathrm{S}}}{\mathrm{m}\mathrm{a}\mathrm{x}({\omega }_{\mathrm{O}},{\omega }_{\mathrm{S}})}}\end{array}}$$

where ω_O and ω_S represent average total female fitness following supplemental outcrossing and selfing, respectively (Ågren and Schemske, 1993).

Data analysis

We used ternary plots and histograms to visualize colour variation in the collected flowers of O. purpurea. A t-test was used to determine whether brightness and contrast differed between the forest and grassland population. A principal component analysis (PCA) was used to visualize the morphometric data in a limited set of dimensions and to assess the amount of variation in morphology among the two populations. Because the first two axes were found to be representative of the higher order axes, only PCA scores of the first two axes were plotted. A non-parametric multivariate analysis of variance (perMANOVA) was used to test whether overall plant morphology differed between plants from the grassland and forest population. For each measured trait, we also calculated the coefficient of variation (CV) as the ratio of the standard deviation to the mean, CV = σ/µ, to determine whether some traits were more variable than others and whether the degree of interplant variation was consistent across the two populations.

For each year and population, we assessed the average fruit set by first calculating for each flowering plant the percentage of flowers producing a fruit, and then calculating the average fruit set across all plants. To test the hypothesis that fruit set in both populations varied in similar ways across years, we calculated Kendall’s W and $\phi =\frac{{\sigma }_{xT}^2}{{\left(\sum ^{}{\sigma }_{x_i}\right)}^2}$, a measure of synchrony proposed by Loreau and de Mazancourt (2008), using the R package synchrony (Gouhier, 2019). The significance of both parameters was tested by performing 9999 Monte Carlo simulations. We also calculated the opportunity for selection (I) as the ratio of the variance in fitness (number of fruits) to the square of mean fitness (Downhower et al., 1987). Low values indicate little opportunity for selection, whereas high values indicate greater opportunities for selection. For both populations, values of I were plotted against year to determine how the opportunity for selection varied across years and between the two populations.

Selection was estimated following Lande and Arnold (1983), using multiple regression analyses with relative fitness (individual number of fruits divided by the mean number of fruits) as the response variable and the standardized trait values (with a mean of 0 and a variance of 1) as explanatory variables. We used the number of flowers, plant height, brightness, contrast, spur length, labellum width and the number of dark spots on the labellum as explanatory variables. To see how selection pressures differed among years, selection gradients were estimated in both study populations for four consecutive years between 2016 and 2019.

RESULTS

Variation in floral traits

Plants differed markedly in the colour of their flowers (Figs 2 and 3). In both populations, the colour of the central part of the labellum varied between white and light-purple. Brightness values varied between 0.63 and 0.98 (mean ±s.d.: 0.83 ± 0.08) in the grassland population, and between 0.60 and 0.98 (mean: 0.86 ± 0.08) in the forest population, and did not differ significantly between the forest and grassland populations (Supplementary Data, Fig. S1). The dots on the labellum were markedly darker (brightness values varying between 0.13 and 0.85 in the grassland population and 0.12 and 0.71 in the forest population). The contrast between the interior part of the labellum and the spots varied between 0.04 and 0.69 (mean: 0.37 ± 0.11) in the grassland population and between 0.08 and 0.70 (mean: 0.43 ± 12) in forest population (Fig. S1).

Fig. 2.

Colour variation in flowers of Orchis purpurea presented in a ternary plot and histograms. A summary of the variation in the background colour of the flower (right) and the colour of the spots on the lip (under) is presented in the histograms.

Fig. 3.

Variation in plant morphology in two populations of Orchis purpurea in Belgium. (A) Principal component analysis (PCA) of the morphological variation based on seven plant traits measured in a grassland and forest population of O. purpurea. (B) Coefficients of variation calculated for the same traits.

The first and second axis of the PCA explained 34.8 and 22.3 % of the total variation, respectively (Fig. 3A). The first axis mainly coincided with the size of the plant (inflorescence height and the number of flowers) and the flower (labellum width and length), whereas the second axis mainly corresponded to colour traits (brightness and contrast of the flower). The overall morphology of plants of the forest population was significantly different from that of the grassland population (perMANOVA: pseudo-F = 4.89, P < 0.007). Inflorescences of plants of the forest population tended to be larger than those of the grassland population and produced more flowers, but no differences in colour traits were observed between plants from forest and grassland populations. Coefficients of variation of the different traits measured varied between 0.13 and 0.38 (average: 0.24 and 0.23 for the grassland and forest population, respectively), with the number of dots per flower and the number of flowers per plant showing the largest CV and brightness and spur length showing the lowest CV (Fig. 3B).

Fruit set

Although average fruit set (i.e. the percentage of flowers developing into a fruit) varied substantially among years (Fig. 4A), the overall average across years was markedly similar in both study populations (11.63 and 12.04 % in the forest and grassland population, respectively). Moreover, variation in average fruit set among years was consistent across the two study populations (Kendall’s W = 0.74, P = 0.028, φ = 0.82, p = 0.012). The lowest fruit set was observed in 2007, when only 5.69 and 5.04 % of all flowers developed into a fruit. In 2017, on average 23.09 and 26.8 % of all flowers developed into a fruit (Fig. 4A). Opportunities for selection (I) varied between 0.50 and 2.25 (mean: 1.09) in the grassland population and between 0.41 and 1.62 (mean: 0.87) in the forest population (Fig. 4B). Years in which opportunities for selection were highest were 2007 and 2016, when fruit set was very low.

Fig. 4.

Annual variation in (A) average fruit set and (B) the opportunity for selection I based on fruit set data collected between 2004 and 2019 in two populations of Orchis purpurea in Belgium.

Phenotypic selection

In both study populations, the number of flowers was subject to selection in all years (Fig. 5). The strongest selection on flower production (β _c = 0.66 and 0.51 in the grassland and forest population, respectively) was found in 2016, when fruit production was lowest. In the grassland population, none of the other flower traits (brightness, contrast, spur length and labellum width) were subject to selection (Fig. 5A). In the forest population, there was weak selection on spur length and flower brightness (Fig. 5B).

Fig. 5.

Selection gradients (β _c) calculated in (A) a grassland and (B) forest population of Orchis purpurea over four consecutive years (2016–2019).

Inbreeding depression

All hand-pollinated flowers successfully developed into a fruit, indicating that the studied populations were strongly pollinator-limited. Seed production was significantly higher after outcrossing than after selfing (t = 2.41, P = 0.02) (Supplementary Data, Fig. S2a). Similarly, the proportion of viable seeds was higher in outcrossed fruits (87 %) than in selfed fruits (74 %) (t = 2.34, P = 0.03) (Fig. S2b). As a result, the total number of viable seeds contained within a single fruit was significantly higher in outcrossed fruits (mean: 1279) than in selfed fruits (mean: 769) (Fig. S2c). The corresponding inbreeding coefficient (δ) averaged across all plants was 0.38.

Discussion

Variation in floral traits

It has been estimated that about one-third of all orchid species are deceptive and do not present a nutritional reward to pollinators (Tremblay et al., 2005) and wide variation in flower colour, odour and size has been described in deceptive orchid species (reviewed by Dormont et al., 2019). Our results are clearly in line with these observations as individuals of Orchis purpurea sampled from two populations displayed substantial variation in colour and size, although this variation was clearly continuous rather than discrete. The overall appearance of the labellum depends on the background colour, which varies between almost white and light purple, and on the brightness and the number of the coloured spots on the labellum, which provide the contrast with the background colour of the labellum and which vary between light purple and very dark purple. The observed colour variation did not differ significantly between the two investigated populations. Similar results have been shown in the related O. italica, where a similar pattern of colour variation was observed (Dormont et al., 2019).

Besides the observed differences in colour, pronounced differences in the size of the flowers were observed as well. Coefficients of variation for the different traits measured varied between 0.13 and 0.38 and were highest for the number of spots on the labellum. The observed values are somewhat higher than values reported by Scopece et al. (2017), who measured variation in labellum length, width and spur length in Orchis mascula and O. pauciflora. Coefficients of variation for these three traits varied between 0.11 and 0.19 in O. mascula, and between 0.10 and 0.17 in O. pauciflora. They are also higher than the average value of 15.2 % reported by Ackerman et al. (2011). Overall, these results show that the investigated plants in both O. purpurea populations produce flowers that display considerable variation in size and colour.

Patterns of fruit set

Overall fruit set was low: averaged over 16 years, only 12 % of all flowers developed into a fruit, which corresponds to an average of about four flowers out of 40 that develop into a fruit. These results corroborate previous analyses that have shown that the proportion of fruit set in food-deceptive orchid species tends be low (Tremblay et al., 2005). Nonetheless, we found considerable temporal variation in fruit set between years that ranged from 5 to >25 %. The reasons for this variation are not entirely clear, but it is reasonable to assume that variations in climatic conditions play an important role (Jacquemyn et al., 2009). Given that O. purpurea flowers in early May, unfavourable climatic conditions such as low temperatures or prolonged periods of rain may limit the number of insects flying around and restrict pollination activities. Interannual patterns of fruit set were highly consistent between the two study populations, which supports the idea that regional factors rather than local differences in insect communities play an important role in determining pollination in this food-deceptive orchid species. The observed variation in fruit set has also important implications for selection, as the opportunities for selection are largest when the variance in fruit set relative to the mean is large (Downhower et al., 1987). This situation typically occurs when the overall fruit set is low (~5 %).

Selection

In food-deceptive species, pronounced variation in the colour, size and odour of flowers may be a successful way to lure pollinators and to guarantee fruit set as it will take some time before pollinators realize that the flowers offer no reward (Heinrich, 1975; Nilsson, 1980; Cropper and Calder, 1990; Ackerman and Galarza-Pérez, 1991; Ackerman et al., 2011), and consequently selection on these traits can be expected to be weak (Dormont et al., 2019). Consistent with this hypothesis, we found no, or only weak, evidence for selection on floral traits such as floral colour or size. In contrast, there was strong and consistently positive selection on the number of flowers, particularly in years when fruit set was low and opportunities for selection were largest. This is not unusual and most studies investigating selection gradients in orchids report significant selection on floral display size, and the selection gradients obtained here are largely congruent with values reported elsewhere. For example, in the food-deceptive orchid Anacamptis morio, the selection gradient for the number of flowers was 0.59 (Sletvold et al., 2016), which is similar to the values reported here. However, in contrast to their study, no significant selection on labellum contrast was found here. One explanation for this difference may be pronounced differences in the insect communities pollinating both orchid species. While A. morio appears to be mainly pollinated by bumble bee queens (Bombus lapidarius), until now it has not been entirely clear which insects pollinate flowers of O. purpurea (Claessens and Kleynen, 2011). Our own observations suggest that there is no single species that specializes on the flowers of O. purpurea and occasional visits by bees, bumble bees, butterflies, beetles and flies have been recorded. The pronounced differences in insect morphology, size and foraging behaviour may therefore contribute to the lack of significant selection pressures on flower traits. Many other food-deceptive orchid species are visited by a large number of insect species (Claessens and Kleynen, 2011). In addition, there do not seem to be rewarding plant species that co-flower with O. purpurea and that serve as ‘magnet’ species that attract pollinators and hence increase reproductive success of the orchid.

We also found little interannual variation in selection gradients, suggesting that fluctuating selection does also not contribute much to the observed variation in colour variation. This finding is in contrast to the results of Scopece et al. (2017), who showed that selection gradients fluctuated substantially among years and contributed to the maintenance of the observed phenotypic variation in two related Orchis species in Italy. Although we found some evidence of fluctuating selection in the grassland population, the effects were very weak and, in most cases, not significant. In the forest population, no such variation was observed and no selection on floral traits was observed.

Another mechanism that may explain the observed variation in floral morphology is inbreeding depression. Inbreeding depression can act against selection if fruits from flowers that were favoured by pollinators result from geitonogamous pollination. In this case, it is expected that pollinators spend a longer time on plants with attractive flowers before flying to another plant. If this occurs, a substantial number of fruits will be the result of geitonogamous selfing and if selfing results in reduced reproductive success due to reduced seed set or lower seed viability, the higher fruit set in attractive plants will be counterbalanced by a lower number of viable seeds per fruit. Indeed, our results clearly showed that selfed fruits produced fewer seeds and that seeds were less viable, leading to a 36 % decline in total reproductive output of selfed fruits compared to outcrossed fruits. Unfortunately, despite many hours of field observations in both populations, very few pollinators have been observed and empirical evidence for this hypothesis is therefore scant.

CONCLUSION

Our long-term data indicated that fruit set varied substantially across years, which in turn affected opportunities for selection. Selection pressures on flower traits such as spur length, labellum size and flower colour were negligible or even completely absent. In contrast, consistent and strong positive selection on floral display was found. Overall, we conclude that the large variation in flower colour and form that is regularly observed in natural O. purpurea populations is maintained by the consistent lack of strong selection pressures on these traits through time.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Fig. S1. Variation in achromatic brightness of the central part of the labellum and the contrast between the spots and the background colour of flowers of Orchis purpurea growing in grassland and forest. Fig. S2. Effect of pollination treatment on the number of seeds per fruit, seed viability and the number of viable seeds in Orchis purpurea.