Abiotic factors may explain the geographical distribution of flower colour morphs and the maintenance of colour polymorphism in the scarlet pimpernel

Authors


Summary

  1. Flower colour polymorphism is traditionally attributed to pollinator selection although other factors, such as indirect selection on correlated traits, can play an important role.
  2. Lysimachia arvensis is a widespread annual species with two colour morphs differing in anthocyanin composition. We explored the hypothesis that colour polymorphism is maintained by selection related to environmental heterogeneity. Morph frequencies and environmental traits were recorded in 51 populations along a wide geographical range. To explore the existence of morph-by-environment interactions, we conducted an experimental study comparing the two morphs under treatments differing in water and light availability.
  3. A geographical pattern was found with a negative association between blue frequencies and latitude. The proportion of the blue morph increased with temperature and sunshine hours, but decreased with precipitation. Flowering onset and flower size differed between morphs and scarcely varied across treatments. In contrast, several fitness components such as germination, seedling survival, seedling mass and flower production showed important morph-by-environment interactions. The blue morph showed higher overall male and female fitness in all the treatment combinations excepting in sun-wet conditions where the red morph had higher fitness.
  4. Synthesis. Our results indicate that the mechanism of selection on flower colour seems to be related to differences in fitness of both morphs due to abiotic factors. These differences could explain the geographical distribution of flower colour morphs and the maintenance of the colour polymorphism. The marked difference in flowering time between morphs leaves open the potential for assortative mating and speciation in Lysimachia arvensis.

Introduction

Colour polymorphism is a widespread phenomenon in plant species affecting both vegetative and reproductive organs (e.g. Rafiński 1979; Smith 1986; Jorgensen & Andersson 2005; Whitney 2005; Rausher 2008). The co-occurrence of different colour morphs has ecological and evolutionary interest and can be due to a variety of evolutionary processes, from direct or indirect selection mediated by pollinators, frugivores, herbivores, pathogens or abiotic environmental effects, to random genetic drift (Epling & Dobzhansky 1942; Wright 1943; Rausher 2008).

Flower colour is a very conspicuous trait that is supposed to act as a selective target for pollinators. Directional selection by pollinators combined with random genetic drift should lead to the loss of floral colour polymorphism (Waser & Price 1981; Levin & Brack 1995; Campbell, Waser & Melendez-Ackerman 1997; Jones & Reithel 2001) but spatio-temporal fluctuations of pollinator spectrum could lead to divergent selection and maintenance of colour polymorphism (Subramanian & Rausher 2000; Schemske & Bierzychudek 2001). However, there are only relatively few direct demonstrations of pollinators exerting selection on flower colour variation, and recent studies have emphasized the need to critically reconsider the role attributed to pollinators in modelling floral characters (e.g. Clegg & Durbin 2000; Warren & Mackenzie 2001; Brown 2002; Strauss & Whittall 2006; Rausher 2008). It has been proposed that the evolution or predominance of a particular floral colour is mainly due to differences in the fitness associated with other correlated features (e.g. Levin & Brack 1995; Armbruster et al. 1997; Armbruster 2002; Frey 2004). Anthocyanins are omnipresent in angiosperms and probably evolved in early land plants long before the evolution of flowers. These pigments may have arisen in vegetative tissues in response to increased ultraviolet light, drought stress and herbivore pressures and were then subsequently co-opted by flowers to attract pollinators (Hanley, Lamont & Armbruster 2009; Whittal & Carlson 2009). In many cases, these pigments still maintain their original stress-related functions while also attracting pollinators (Winkel-Shirley 2002; Dick et al. 2011). The most common and studied flower colour polymorphism is the loss of anthocyanins by loss-of-function (LOF) mutations that cause pigmented flowers to become white. Unpigmented morphs often have lower tolerance to drought and heat than pigmented morphs (Schemske & Bierzychudek 2001, 2007; Strauss & Whittall 2006), and morphs with differential tolerance to abiotic conditions may be maintained by fluctuating environmental conditions or by geographical segregation (Dick et al. 2011). Although less frequent, the blue-purple/red-orange colour polymorphism is also common. This polymorphism can occur without any alteration of the responsible pigments involved, for example through vacuolar pH changes (Grotewold 2006). However, most frequently, the transition from blue-purple to red-orange is due to the production of less hydroxylated anthocyanins as a consequence of the inactivation of branches in the anthocyanin pathway (Rausher 2008). The biosynthetic pathway of anthocyanins comprises several branches differing in the B-ring hydroxylation level of their intermediate flavonoids and their final anthocyanins (Tanaka, Sasaki & Ohmiya 2008). As far as we are aware, environmentally related differences in fitness between morphs differing in such blue-red anthocyanin compositions have not been reported.

Scarlet pimpernel (Lysimachia arvensis (L.) U. Manns and Anderb., formerly Anagallis arvensis L.; Manns & Anderberg 2009) is a widely distributed annual herb, that was originally described as a red-flowered plant from Central Europe, but which also presents blue-flowered forms (Fig. 1). The two morphs of L. arvensis differ in anthocyanin composition, with malvidine and pelargonidine being mainly responsible for the blue and red coloration, respectively (Wiering & de Vlaming in Harborne 1968; Ishikura 1981). A search of L. arvensis in various Floras (e.g. Ferguson 1972) for the Mediterranean and European areas revealed that the red morph is most frequent in North and Central Europe, whereas in the Mediterranean region the blue morph is the most common, although some populations with both floral morphs occur in both areas. The climate of the Mediterranean region clearly differs from the Oceanic climate of North and Central Europe, being typically sunnier, hotter and dryer. Moreover, in an Oceanic climate, plants are rarely subjected to water stress since rainfall is distributed regularly throughout the year, while in the Mediterranean, besides an extended summer drought, there are shorter water stress periods in the rainy season due to unpredictability of rains. In L. arvensis, the colour polymorphism could be maintained by environmental heterogeneity if each morph performed better in different environments. In fact, in mixed Mediterranean populations, we have also observed that red plants show a tendency to inhabit wetter or shadier microsites (river banks, irrigated orchards or under tree canopy) than blue plants.

Figure 1.

The blue and red morphs of Lysimachia arvensis and their frequencies throughout the western Mediterranean and south of Europe.

There are no studies on pollinator discrimination between morphs in the scarlet pimpernel, but their role as selective agents might be minimized because the flowers show nastic movements and self-pollinate automatically (Gibbs & Talavera 2001). In this study, we have surveyed floral-morph proportions and environmental traits in 51 populations of scarlet pimpernel along a wide geographical range to test the hypothesis that geographical and local variations in flower colour are related to environmental variation. Moreover, to explore the existence of morph-by-environment interactions, we compared fitness components of blue and red morphs by growing plants under treatments simulating differences in water and light availability. In this way, we explored the hypothesis that flower colour polymorphism may be maintained by selection related to environmental heterogeneity.

Materials and methods

Study Species

Lysimachia arvensis is an annual species that constitutes small populations in open habitats such as cultivated fields, degraded sites and marine sands. It is probably native to the Mediterranean region, but is dispersed world-wide as an introduced plant. It is a hermaphrodite and self-compatible species (Gibbs & Talavera 2001) that produces capsules with numerous seeds. The flowers do not produce nectar (Gibbs & Talavera 2001; Raine & Chittka 2007) and show nastic movements, with the petals opening in the morning and closing in the evening, such that flowers that have not received visitors self-pollinate automatically (Gibbs & Talavera 2001). Nevertheless, some cross-pollination also occurs (Marsden-Jones & Weiss 1960). According to F. E. Weiss (unpubl. data) crosses between plants with the distinct colours produce a F1 with all red flowers, whereas in the F2, plants with red or blue flowers appear in a proportion expected for a character controlled by a single gene. Populations in the south of Spain receive very few pollinator visitors, mainly solitary bees (Gibbs & Talavera 2001; P. L. Ortiz, R. Berjano, M. Talavera & M. Arista unpubl. data), but those in Germany receive visits from Bombus terrestris (Raine & Chittka 2007).

Geographical Pattern of Flower Colour Distribution

To plot the geographical distribution of flower colour in L. arvensis, we recorded morph frequencies in the field in 51 populations from West Europe and Macaronesia (Fig. 1). Most populations of this species are small (< 100 individuals), and in these, all the plants were censused, while in bigger populations the colour of at least 150 individuals was recorded along a 4 × 200 m transect. Location (latitude and longitude) and altitude were recorded in each population. Additionally, mean annual temperature, mean annual precipitation and mean annual sunshine hours were noted using historic climatic data (Worldwide Bioclimatic Classification System http://www.ucm.es/info/cif/data/indexc.htm and Climate data http://es.climate-data.org) (see Appendix S1 in Supporting Information).

Experimental Design

We collected seeds from seven red and seven blue plants from a wild population of Larvensis in the Doñana Natural Park (Hinojos, SW Spain). This population was selected due to the presence of both colour morphs and because it shows a clear spatial segregation of the two morphs, with red plants occurring in shadiest sites. The seeds were germinated in the laboratory and plants grown in a glasshouse. To minimize random maternal effects on plant traits, with each plant, we performed self-pollinations to obtain F1 self-seeds that were germinated and hand self-pollinated again to obtain seeds from two selfed generations. A total of 1232 second-generation seeds from red and 1232 from blue plants were used in the experimental design. These seeds were sown individually in plastic pots (6 × 6 × 8 cm) using about 97% peat and 3% vermiculite as substrate.

Sunshine and precipitation were the most important factors explaining blue morph proportion in our geographical survey (see results); moreover, red morphs in Mediterranean mixed populations mainly occur in the wettest or shadiest places. Thus, we implemented a fully factorial design with two levels of irradiance and two levels of water irrigation. In 100% open fields, maximum radiation is near 2000 μmol m−2 s−1, while radiation under tree canopy is frequently about 6% full sunlight (Gómez, Valladares & Puerta-Piñero 2004; Valladares 2004). We defined two levels of irradiance: high sunlight radiation (hereafter sun) provided by artificial light supplementation (from 08:00 to 20:00 h) to natural glasshouse light to reach full sunlight and reduced sunlight radiation with horticultural shade netting (hereafter shade) to simulate radiation under tree canopy. In the sun, photosynthetic photon flux density was 2033.6 ± 3.2 μmol m−2 s−1, while under the shade it was 96.2 ± 8.1 (mean ± SE, both = 20). We irrigated the seeds for the first 3 days with tap water for root system establishment before commencing the different water treatments. During the rest of the experiment, we irrigated the plants using two moisture levels: low water irrigation (hereafter dry) and high water irrigation (hereafter wet). In the wet treatment, plants were irrigated as much as necessary to maintain the soil constantly moist, avoiding water stress. In contrast, plants in the dry treatment were irrigated only once a week; in this treatment, the soil surface was completely dry before irrigations simulating short water stress periods. The combination of sun and dry treatments would induce the strongest drought effect. Maximum glasshouse air temperature ranged from 19 °C in January to 36 °C in May, with minimum night temperature ranging from 4 °C to 10 °C, respectively. Colour morphs and families were randomly placed within each treatment combination. Given that plants were in individual pots, root competition for water was avoided, and they were spaced sufficiently to avoid shadowing effects between neighbouring plants. We assume, therefore, that each potted plant can be considered as an independent experimental unit. The experiment began in October 2009, and the plants were harvested at the end of May 2010. During experiment, the following fitness components were measured in experimental plants:

  1. Germination.
  2. Seedling vigour: 30 seedlings from each combination of colour morph, irrigation and light availability were harvested when they were 21 days old. These seedlings were obtained from another similar set of seeds sown in the same conditions as described above to prevent influencing survival data, given the low germination rate of L. arvensis seeds (see results). Each seedling was dried at 45 °C in a stove until constant weight, and then weighed using an electronic balance (± 0.1 mg).
  3. Survival: number of plants still alive to reach the reproductive age.
  4. Onset of flowering: number of days to begin flowering from germination.
  5. Flower size measures. Corolla diameter, sepal length, pedicel length and stamen length were measured in the first flower of 35–40 plants from each colour–water–light combination using a digital caliper. Sepals alternate with petals and clearly contributed to the shape of flower. Stamen filaments exhibit long purple hairs that contrast with anthers and corolla, and so stamen length could improve floral attractiveness. Lastly, a longer pedicel makes the flower more conspicuous within the foliage. Thus, all these traits are related to floral attractiveness.
  6. Ovule and pollen production. We collected and fixed in ethanol 70% the first floral bud from 15 to 20 plants from each colour–water–light combination. In these buds, ovule number was counted under a dissecting microscope and pollen production was counted using a particle counter (Coulter Multisizer 3). For the latter, pollen of all anthers was dispersed in 50 mL Isoton II isotonic liquid (Beckman Coulter, Fullerton, CA, USA); then, the exact number of pollen grains was counted in three subsamples of 500 μL, and the mean was calculated. From this mean, the total number of pollen grains of each flower was determined.
  7. Number of flowers produced.

Overall male and female fitness was calculated as total pollen grain number and total ovule number per plant, by multiplying the number of flowers per plant by the number of pollen grains and ovules per flower, respectively. In these estimates, fitness of seedlings that did not survive until reproduction, or that did not flower, was computed as zero. Germination was not included in the calculation of overall fitness because a large number of seeds failed to germinate, and assigning a fitness of zero to seeds that did not germinate required the assumption that ungerminated seeds were dead, whereas they might be dormant. Under natural conditions, female fitness is usually estimated as the number of seeds produced by the plant and male fitness as the number of seeds sired by pollen (Strauss, Conner & Lehtilä 2001). However, in the glasshouse both pollen dispersal and receipt are constrained by the lack of pollinators. Thus, we used ovule production and pollen production per plant, respectively, as estimates of lifetime female and male fitness (Busch 2005). Although these are imperfect fitness estimates, plants producing fewer ovules or pollen grains would have fewer opportunities to ripen or to sire seeds, respectively (Strauss, Conner & Rush 1996; Friedman & Barrett 2011).

Data Analyses

A partial regression method based on distance matrices was employed to test for a relationship between geographical locality and the blue morph frequency variation. Geographical distances were calculated from longitude and latitude using the ade4 R package (Casgrain 2002). Distance matrices based on latitude or on longitude were expressed as the absolute difference between each population pair. Geographical distances between populations, as well as differences in latitude or in longitude were all included as explanatory variables and considered against the morph frequency matrix (expressed as differences in frequency of the blue morph) using Mantel permutation tests (Mantel 1967). Moreover, to test the relationship between the blue morph frequency and ecogeographical variables (longitude, latitude, altitude, mean temperature, mean precipitation and mean sunshine hours, Appendix S1), a stepwise multiple regression analysis was carried out.

In the experimental study, germination, seedling survival, seedling mass, flowering onset, pollen and ovule production, and overall male and female fitness were analysed by GLMs with colour, light and water as fixed factors and family nested within colour. In order to test the hypotheses that there are differences in the responses of blue and red morphs to water and light availability, we considered only the interactions colour-by-water, colour-by-light and colour-by-water-by-light. Data were analysed using generalized linear models with different link functions and error distributions depending on the type of response variable modelled (Crawley 2005). Binomial distribution of errors and logit link function were used to analyse germination and seedling survival. Gaussian distributions were used to analyse seedling mass, reproductive age, flower production, pollen and ovule production, and overall male fitness and female fitness. All these analyses were conducted using GLzM module of spss (IBM SPSS Statistic 20, 2011, USA) with Type III test, and significance levels were adjusted using Bonferroni correction (assuming P value < 0.05/7). When the GLMs showed significant differences, the means of treatments were compared using t-tests based on the standard errors calculated from the specific model. To know the possible correlations between corolla diameter, sepal length, stamen length and pedicel length, we performed Pearson correlations between each pair of variables. Then, the four variables were analysed by means of a manova analysis with colour, water and light as fixed factors and considering the interactions colour-by-water, colour-by-light and colour-by-water-by-light. These three interactions were not significant, and so, they were removed from the analysis.

Results

Ecogeographical Pattern of Flower Colour Distribution

Most sampled populations (74.5%; Appendix S1) were monomorphic in colour although mixed populations appeared throughout the study area (Fig. 1). A significant negative relationship was found between similarity in blue morph proportions and the geographical distance between pairs of populations (standardized partial regression coefficient = 0.1334, = 0.002). Significant correlations were observed between blue morph frequency and all the ecogeographical variables investigated in this study excepting longitude (= 0.144). This reflects the fact that most of these variables covary to greater or lesser extent; sunshine hours, latitude and mean temperature were most strongly correlated between them (< 0.001 in all the cases), and they were also most significantly correlated with blue morph frequency (< 0.001 in all the cases). The multiple regression of the blue proportion and the ecogeographical variables was significant with high predictive power (R2 = 0.707; < 0.001). The best predictor of blue morph proportion was sunshine hours (< 0.001) that alone explained 63.4% of the total variance, followed by mean rainfall (= 0.010) that explained 4.3% and by elevation (= 0.043) that explained only 3% (Fig. 2). Regression function for the blue proportion was: 0.72 × sunshine hours – 0.23 × mean rainfall + 0.18 × elevation (F3,45 = 33.836, < 0.001). The remaining variables were not significant predictors (= 0.259 for latitude and = 0.144 for mean temperature).

Figure 2.

Effects of annual sunshine (hours), annual precipitation (mm) and elevation (m.a.s.l.) on blue morph proportion (multiple R2 = 0.716, < 0.001). The above partial regression plots plotted all variables as residuals of the labelled variable, given the two factors not shown in the plot (Neter et al. 1996).

Effects of Light and Water on the Two Floral Colour Phenotypes

The germination of L. arvensis seeds was only around 14% (= 342), and no differences between morphs were found in final germination (Table 1). Light affected germination since it was markedly higher in the shade than in the sun (17% and 11%, respectively; Table 1). Families within each colour also showed differences in germination, which ranged from 2% to 49% in the red morph and from 4% to 31% in the blue morph. Significant differences in germination were found among colour morphs when light was considered; in sun conditions, germination was similar in both morphs, but in shade, germination of red seeds was significantly higher (Fig. 3). Seedling survival was in general high (= 301) and similar between treatments (Table 1). Differences between morphs were only significant when water and light were considered jointly (L × W × C interaction). In shade, survival of the two morphs was similar irrespective of water availability; however, in sun, survival of red plants was markedly lower than that of blue plants in dry conditions (Fig. 4).

Table 1. Summary of GLM results of different effects light (low/high), flower colour (blue/red), water (wet/dry) and their interactions on fitness components of Lysimachia arvensis
Effectsd.f.GerminationSurvivalSeedling massReproductive ageNo. of flowersOvules per flowerPollen grains per flower
Wald chi-square P Wald chi-square P Wald chi-square P Wald chi-square P Wald chi-square P Wald chi-square P Wald chi-square P
  1. ns, not significant.

  2. *< 0.007; **< 0.001.

Colour11.19ns0.0ns9.17 * 42.59 ** 11.04 * 48.78 ** 3.64ns
Light136.69 ** 6.41ns506.2 ** 71.32 ** 9.92 * 0.70ns35.26 **
Water11.71ns3.91ns57.25 ** 27.89 ** 4.64ns52.01 ** 0.003ns
Family (colour)12297.7 ** 14.9ns21.93ns19.25ns56.03 ** 28.01 * 22.27ns
L × C16.79 * 0.16ns5.19ns0.86ns9.69 * 12.00 * 3.89ns
W × C15.54ns2.9ns0.0ns0.95ns17.05 ** 4.04ns0.49ns
L × W × C24.31ns11.61 * 37.93 ** 8.35ns25.32 ** 5.19ns4.19ns
Figure 3.

Two-way interaction graphs for the effects of light (sun/shade) and water availability (wet/dry) on different fitness components of blue and red Lysimachia arvensis plants (only traits with significant interactions are presented). Female fitness is measured as ovules per plant and male fitness as pollen grains per plant. Means ± 1 SE are shown. In each graph, P values correspond to tests of significance of the four simple main effects involved in each interaction (see Table 1 for significance tests of main effects and interactions).

Figure 4.

Three-way interaction graphs for the effects of light (sun/shade) and water availability (wet/dry) on different fitness traits of blue and red Lysimachia arvensis plants (only traits with significant interactions are presented). Female fitness is measured as ovules per plant and male fitness as pollen grains per plant. Significant differences between blue and red morphs in a particular combination of water and light availability treatment are denoted by asterisks (see Tables 1 and 2 for significance tests of main effects and interactions).

Light was the most important factor affecting seedling mass, since it was markedly heavier in sun than at shade (8.2 mg and 1.05 mg, respectively; Table 1). Seedlings were also significantly heavier in dry conditions (5.7 mg vs. 3 mg), and those from blue plants were heavier than those from red ones (5.06 mg vs. 4.2 mg). However, the interaction L × W × C was also significant because blue seedlings were heavier than red ones only when they were growing in sun-wet conditions (Fig. 4).

Throughout the experimental period, 254 plants reached flowering, and their reproductive age was markedly affected by water, light and flower colour, but not by family within each colour (Table 1). Plants in light or dry conditions flowered about 23 days earlier than plants in shade or wet conditions, respectively. Surprisingly, blue plants always flowered earlier than red plants (121 days after germination for blue plants vs. 141 for red plants), and this pattern was consistent irrespective of growing conditions (112 vs. 130 in sun, 130 vs. 153 in shade, 126 vs. 149 in wet and 116 vs. 134 in dry). Flower yield was the trait most affected by the different treatments and also varied among families within each colour morph (Table 1). Blue plants showed a mean of 9 flowers more than red plants but differences were significant only when growing in shade or in dry conditions (Fig. 3). Moreover, the three-way interaction was also markedly significant since differences in flower production between morphs changed when the two treatments were combined (Fig. 4). Ovule production per flower ranged from 16 to 44 (= 142) and was affected by water availability and colour morph (Table 1). Flowers of red plants significantly produced more ovules than those of blue plants (30.5 vs. 27.8), and this trend was observed in every treatment combination although the light by colour interaction was significant (Fig. 3). The production of pollen per flower ranged from 2073.7 to 26 285.4 grains (= 143) and showed significant differences between light conditions (Table 1) being 20% higher in plants growing in sun than in those in shade (13 100 vs. 10 150).

All the morphological floral traits measured showed positive significant correlations (< 0.01 in all cases, = 306). Flower attractiveness measured by corolla diameter, and stamen, sepal and pedicel lengths showed significant differences among colour morphs and in relation to light and water availability, but interactions were not significant. Plants in wet conditions produced bigger flowers than those in dry (F4,299 = 6.10, < 0.001), and plants in sun also produced bigger flowers (F4,299 = 51.14, < 0.001). The two colour morphs differed significantly in flower size, with the red plants producing bigger flowers than the blue plants (F4,299 = 14.98, < 0.001).

Overall male and female fitness showed important differences between morphs and among families within each morph (Table 2). In general, the blue morph showed higher male and female fitness than the red morph although all interactions were significant. Significant differences were found among colour morphs when water was considered; in wet conditions, both morphs showed similar male and female fitness, but they were markedly higher for the blue morph in dry conditions. In the interaction colour-by-light, in shade, both male and female fitness were markedly higher for the blue morph, while in sun, these differences were significant only for male fitness (Fig. 3). However, the interactions L × W × C were also significant (Table 2) as the blue morph showed higher male and female fitness in all the treatment combinations excepting in sun-wet conditions where the red morph had higher fitness (Fig. 4). In both morphs, the lower fitness was found in the shade–wet combination.

Table 2. Summary of GLM results of different effects light (low/high), flower colour (blue/red), water (wet/dry), seed family and their interactions on overall male (pollen grains per plant) and female fitness (ovules per plant) of Lysimachia arvensis. After Bonferroni correction, an effect is significant when < 0.007
Effectsd.f.Male fitnessFemale fitness
Wald chi-square P Wald chi-square P
Colour113.780.0009.060.003
Light18.990.0031.420.232
Water10.090.7672.7540.097
Family (colour)1251.790.00033.590.001
Light × colour17.970.00510.810.001
Water × colour125.060.00031.950.000
L × W × C244.470.00041.320.000

Discussion

Colour polymorphism in plants has been traditionally considered to be a result of divergent direct selection driven by biotic agents. Polymorphisms of fruits and seeds are thought to be maintained by dispersers or by predators, and those of flowers by pollinators, although direct evidence is scarce (Whitney 2005; Rausher 2008; Porter 2013). Recently, the possibility of indirect selection by biotic or abiotic factors has become increasingly relevant as diverse cases are being reported. Polymorphisms of flowers, fruits and seeds can be caused by indirect selection due to biotic (Irwin et al. 2003; Whitney & Stanton 2004; Strauss & Whittall 2006) or abiotic agents (Willson & O'Dowd 1989; Traveset & Willson 1998; Schemske & Bierzychudek 2001; Whitney & Lister 2004). As we will show below, our results are consistent with the possibility that the flower colour polymorphism in L. arvensis is maintained by indirect selection driven by abiotic factors.

We found significant negative associations between blue morph frequency and latitude of populations, and between similarity in blue morph frequency and geographical distance of population pairs. This means that a geographical pattern of flower colour exists in L. arvensis, and it seems to be related to climatic features, which suggests that flower colour is not a neutral trait (Mayr 1965). The correlations found between blue morph frequency and the environmental variables studied indicate that blue plants are more frequent in dryer, hotter Mediterranean localities while red plants predominate in more temperate Oceanic areas. This could reflect a differential adequacy of morphs to environmental conditions that is also supported by the fact that red plants in southern mixed populations frequently occupied the wettest or shadiest places (M. Arista & P. L. Ortiz). The existence of a relationship between environmental gradient and geographical pattern does not prove a selective response to climatic traits and alternative explanations should be considered (Pope et al. 2013). The spectrum of pollinators of L. arvensis appears to differ in different localities; populations in the south of Spain receive very few visitors, mainly solitary bees (Gibbs & Talavera 2001; P. L. Ortiz et al. unpub. data), but those in Germany receive visits from B. terrestris (Raine & Chittka 2007). We have also found significant differences in flower size between colour morphs with red flowers bigger than blue ones without any morph-by-environment interactions. Flower colour preferences are variable for some pollinators (Chittka, Ings & Raine 2004), but they usually prefer large flowers to small ones (Galen & Newport 1987; Galen 1989; Vaughton & Ramsey 1998; Arista & Ortiz 2007). Pollinators could be the selection agents responsible for the geographical pattern observed if they covaried with environmental conditions. However, Larvensis could be relatively independent from pollinator activity because its flowers self-pollinate when petals close after the first day of anthesis. Even so, pollinator attendance could be important if xenogamous seeds had higher fitness than autogamous seeds. Given that we have not studied either pollinator attendance or differences in fitness between autogamous and xenogamous seeds, we cannot disregard the fact that pollinators play a role in maintaining flower colour polymorphism.

If flower colour itself is not the target of selection, pleiotropy could be the primary mechanism favouring one morph over the other in different environments. Differences in colour have been associated with differences in seedling survival, flower and seed production, plant biomass, and herbivore and pathogen resistance (Koes, Quattrocchio & Mol 1994; Koes, Verweij & Quattrocchio 2005; Jonson, Berhow & Dowd 2008). In our experimental study, germination, seedling mass, seedling survival, and flower and ovule production all showed different morph-by-environment interactions. The blue morph showed lower germination in the shade and higher seedling mass in the sun treatment, while the red morph showed lower survival in the dry–sun combination, more flowers in the sun–wet combination and more ovules at sun or wet treatments. Since some treatment effects on the components of plant performance may counteract each other, they are poor predictors of the overall effect when analysed separately. Only by considering overall fitness, instead of each trait separately, enables us to assess how each colour morph is affected by the treatments (García & Ehrlén 2002). Overall male and female fitness of blue morph was markedly higher in dry conditions, and this suggests a better tolerance to more xeric environments. However, our experimental study failed to find a clear pattern of adequacy of red morph to more mesic environments as only in wet-sun but not in wet-shade conditions was overall female fitness higher (male fitness was also higher but not significantly). In fact, the wet–shade combination seems to be the less favourable for L. arvensis as both morphs showed their lowest fitness. Thus, the Mediterranean environment seems to be more suitable for the blue morph, while the red morph seems to perform better in wet and sunny places, such as those where it usually occurs in central Europe. But, it is possible that other environmental factors not considered here could also be responsible for the geographic pattern found in our survey.

Diverse studies have shown that flower colour correlates with anthocyanin content in vegetative tissues (Strauss & Whittall 2006) and that pigmented individuals often tolerate stressful conditions like drought and heat better than anthocyanin-less morphs (Warren & Mackenzie 2001; Strauss & Whittall 2006; Rausher 2008). However, the flower colour polymorphism of Larvensis clearly differs from such a situation as both morphs are pigmented. It has been reported that high-hydroxylated flavonoids (cyanidin and malvidin branches) are more effective than low-hydroxylated ones (pelargonidin branch) in protecting against stress due to excess light and drought (Daniel et al. 1999; Winkel-Shirley 2002; Tattini et al. 2004; Tanaka, Sasaki & Ohmiya 2008; Agati & Tattini 2010). It is tempting to hypothesize that flavonoids of the blue morph confer a better fitness in more xeric environments than those of the red morph but this hypothesis needs to be appropriately tested.

Although most of the traits studied were affected by the experimental treatments, onset of flowering was markedly earlier in the blue morph in relation to the red morph without any morph-by-environment interactions. This difference between morphs, regardless of growing conditions, is one of our most notable results and suggests that this trait is linked to flower colour and is genetically determined. This pattern of flowering can be also observed in natural mixed populations (pers. obs.). Flowering phenology is a trait that is usually influenced by biotic and abiotic interactions, and so it is difficult to evaluate selection because different agents affect different fitness components (Ehrlén & Münzbergová 2009). Notwithstanding, in Mediterranean habitats, the flowering of annual species takes place early in spring when soil water is available. In these conditions, an earlier flowering could confer reproductive advantage ensuring fruit ripening when water still remains in the soil. In contrast, in northern regions, low temperatures instead of water availability limit the flowering season (Rathcke & Lacey 1985), and there a later flowering could be advantageous, avoiding the risks of reduced fitness from frost (Inouye 2008). The divergent flowering patterns of the colour morphs could be selectively advantageous in different environments, and could contribute to maintain colour polymorphism via indirect selection, which would be consistent with the geographical and local distribution patterns of colour morphs.

Taking into account all our results, we found in L. arvensis many monomorphic populations that were spatially isolated, and some mixed populations with observational and experimental evidence of divergence in flowering times between morphs. It is obvious that long-term spatial segregation can generate reproductive isolation and trigger speciation (Mayr 1965; Coyne 1992; Doebeli & Dieckmann 2003). However, even in absence of spatial barriers, differences in flowering time between morphs could cause assortative mating, leading to a decrease in gene flow between them and eventually to allochronic speciation (Fox 2003; Weis et al. 2005; Savolainen et al. 2006; Gavrilets & Vose 2007). Moreover, autonomous autogamy may also contribute to reduce gene flow between morphs. Theoretical models suggest that incipient allochronic speciation could be a common phenomenon in populations of limited size (Devaux & Lande 2008) and so, this may be occurring in the scarlet pimpernel.

In conclusion, a geographical pattern of flower colour distribution is found in L. arvensis, and although we lack definitive evidence, the mechanism of selection on flower colour seems to be related to differences in the fitness of both morphs in alternate environments. These fitness differences could explain the geographical pattern of flower colour and the maintenance of colour polymorphism in L. arvensis. However, it is difficult to extrapolate results from glasshouse to natural populations, and a reciprocal field transplant study could be very useful to test local adaptation. Thus, further research is needed to identify the precise mechanisms of selection and to explore alternative selective factors given that morphs differ in traits that affect pollinator attendance. The marked difference in flowering time between colour morphs leaves open the potential for assortative mating and speciation in L. arvensis.

Acknowledgements

This work was supported with FEDER funds and grants from the Spanish MINCYT (CGL2009-08257) and MINECO (CGL2012-33270). The authors thank to S. Talavera, J.A. Mejías, M.A. Ortiz, R. Casimiro-Soriguer, M.L. Buide, R. Albadalejo and J.L. García-Castaño for helping in material collection, to Peter E. Gibbs for comments on the manuscript and English revision, to Doñana Natural Park to allow material collection and to the Servicio de Invernaderos of the Seville University.

Ancillary