Evolution and maintenance of pollen-colour dimorphisms in Nigella degenii: habitat-correlated variation and morph-by-environment interactions

Authors

  • Tove Hedegaard Jorgensen,

    Corresponding author
    1. Department of Ecology, Section of Plant Ecology & Systematics, Sölvegatan 37, Lund University, SE-22362 Lund, Sweden
    Search for more papers by this author
  • Stefan Andersson

    1. Department of Ecology, Section of Plant Ecology & Systematics, Sölvegatan 37, Lund University, SE-22362 Lund, Sweden
    Search for more papers by this author

Author for correspondence: Tove H. Jorgensen Tel: +46 46 2220912 Fax: +46 46 2224423 Email: Tove.Hedegaard@sysbot.lu.se

Summary

  • • Dimorphism in pollen colour is rare among flowering plants, but occurs in two geographically and morphologically distinct subspecies of Nigella degenii (Ranunculaceae). We evaluated the role of genotype-by-environment interactions in the maintenance of two pollen morphs within each of these subspecies.
  • • Morph frequencies in a number of populations were related to current habitat conditions, and an extensive common-garden experiment involving both optimal and stressful conditions (drought and nutrient deficiency) was carried out.
  • • The putatively derived (dark) pollen morph of N. degenii ssp. barbro has a higher frequency on slopes facing north or east than on slopes facing south or west. Plants of the dark morph also have a higher mortality under drought stress or nutrient deficiency. Data available for N. degenii ssp. jenny provide little evidence for habitat-correlated variation in morph frequency or morph-specific differences in fitness under optimal and stressful growth conditions.
  • • Our results suggest that morph-by-environment interactions in mortality could contribute to the maintenance of pollen-colour dimorphisms in N. degenii ssp. barbro.

Introduction

A major goal of evolutionary biologists is to quantify the factors responsible for maintaining the great diversity exhibited by natural populations of most organisms. In the case of fitness-related characters, within-population variation may be a manifestation of mutation–selection balance or some form of balancing selection. For self-incompatibility and other self-recognition systems, negative frequency-dependent selection leads to the maintenance of high allelic diversity (Richman & Kohn, 2000). A number of empirical and theoretical studies have tied the persistence of heritable variation to spatially or temporally varying selection and genotype-by-environment interactions (Levene, 1953; Gillespie, 1974, 1975; Ellner & Hairston, 1994; Abbott et al., 1998; Stratton & Bennington, 1998). For genotype-by-environment interactions to be an effective evolutionary force maintaining diversity, there must be large changes in the fitness ranking of genotypes across environments, such that a single genotype does not have the highest fitness in all environments (Levene, 1953).

Inferring agents of selection and the genes on which they act is difficult when different traits are influenced by the same genes (pleiotropy), or by different genes in linkage disequilibrium. Such among-trait correlations can reverse the direction of selection response from expectation, or promote evolutionary responses of characters not under direct selection, depending on the sign and magnitude of the correlations and the relationship between each trait and fitness (Lynch & Walsh, 1998). For example, genetic data from plants indicate the potential for floral characters to evolve by selection on genetically correlated characters expressed before or after flowering, such as vegetative size and germination behaviour (Andersson, 1997; Abbott et al., 1998; Andersson, 2001; Conner, 2002). Thus the evolution of floral morphology is likely to reflect a pluralistic process, involving not only plant–pollinator interactions but also flower herbivores, seed predators, resource costs and various aspects of the plant's abiotic environment (Brody, 1992; Abbott et al., 1998; Galen et al., 1999; Galen, 1999a, 1999b; Andersson, 2001).

Polymorphism in flower colour is a conspicuous feature of many plant populations and has been a topic of long-standing interest for both pollination ecologists and evolutionary biologists (Clegg & Durbin, 2000). Animal pollinators have been shown to act as a strong selective agent on flower colour, as evidenced by the detection of morph-specific differences in pollinator visitation, female fecundity, outcrossing rate or siring ability (Waser & Price, 1981, 1983; Brown & Clegg, 1984; Stanton, 1987; Stanton et al., 1989; Rausher & Fry, 1993; Jones, 1996; Gigord et al., 2001; Jones & Reithel, 2001; Irwin & Strauss, 2005). Some authors have found that pollinators impose negative frequency-dependent selection on flower colour (Brown & Clegg, 1984; Gigord et al., 2001), thus promoting the maintenance of flower colour polymorphisms. In other cases, pollinator-mediated selection is either absent or too weak to explain the co-occurrence of different colour morphs within the same population (Jones, 1996; Stone, 2000; Schemske & Bierzychudek, 2001).

The biosynthetic pathway involved in the production of anthocyanins, the main pigments responsible for blue, red or purple flower colours (Shirley, 1996; Mol et al., 1998), also produces compounds that are important in, for example, plant–microbe interactions, herbivore defence, stress tolerance, UV protection and pollen viability (Mo et al., 1992; Koes et al., 1994; Shirley, 1996; Graham, 1998; Steyn et al., 2002). Furthermore, numerous studies have documented pleiotropic relationships between floral colour and the expression of anthocyanin-based pigments in leaves or stems (reviewed by Armbruster, 2002). Thus a wide variety of ecological factors could exert direct or indirect selection on flower colour polymorphisms. As yet, only a few studies have examined the potential for selective forces other than pollinators to influence the relative fitness of different floral colour morphs (Levin & Brack, 1995; Schemske & Bierzychudek, 2001; Armbruster, 2002; Coberley & Rausher, 2003).

In contrast to the large number of species reported to possess intraspecific variation in petal colour (Warren & Mackenzie, 2001), few studies have documented colour polymorphisms involving primary sexual organs. Exceptions include Rafinski (1979), who investigated a stigma colour dimorphism in Crocus scepusiensis; and Darwin (1877), Thomson (1986), Wolfe (2001) and Lau & Galloway (2004) who documented pollen-colour polymorphisms in species of Lythrum, Erythronium, Linum and Campanula, respectively. To our knowledge, only one study has been performed to investigate the role of selective forces such as pollinators in the maintenance of colour polymorphisms in primary sexual organs. Pollinators showed no preference for particular pollen colour morphs in Campanula americana, but differed in their response to the amount of pollen exposed by the different colour morphs (Lau & Galloway, 2004). Given the pleiotropic effects associated with anthocyanin production, and the potentially weak influence of pollen colour on a plant's visual display, it becomes meaningful to determine whether ecological factors other than pollinators contribute to the evolution of such dimorphisms (see also Lunau, 1991).

We have performed a series of field and glasshouse studies to elucidate the genetic and selective mechanisms underlying the origin and maintenance of pollen-colour dimorphism in the annual herb Nigella degenii (Ranunculaceae). This colour dimorphism contrasts with the uniformly light pollen grains of related species and occurs in slightly different forms in two geographically and morphologically distinct subspecies of N. degenii (Strid, 1970). Natural populations of both subspecies are more divergent in the frequency of the two colour morphs than in the frequency of different genotypes at 149 putatively neutral AFLP marker loci, suggesting a role for diversifying selection in maintaining pollen-colour dimorphisms in this species (unpublished data). Investigations of fertilization ability have revealed a general selective advantage of the dark pollen type in single-donor pollinations. This advantage appears to be disrupted under strong pollen competition or in certain genetic backgrounds (unpublished data). In a garden experiment with plants of N. degenii, pollinators sometimes visited one pollen morph more frequently than the other, although the favoured morph depended on site, date and type of pollinator (unpublished data). No attempt has yet been made to examine how different abiotic factors influence the relative fitness of the two colour morphs in this species.

In the study presented here, we related patterns of morph frequency variation in natural populations of both subspecies to local habitat conditions in order to identify abiotic factors of potential selective importance for the persistence of pollen-colour dimorphisms in N. degenii. We also obtained family-structured data from an extensive glasshouse experiment to test for morph-specific differences in vegetative and phenological performance variables including survival rate, flower number and flowering date. The glasshouse study, hereafter referred to as the stress-manipulation experiment, was performed under four different combinations of nutrient and water stress to explore the existence of morph-by-environment interactions. Water stress appears to be an important ecological factor driving the evolution of reproductive characters in Nigella (Strid, 1969), and both drought and nutrient levels have been found to affect the relationship between anthocyanin production and plant performance in other plant species (Bonguebartelsman & Phillips, 1995; Warren & Mackenzie, 2001).

Materials and Methods

Study species

Nigella degenii Vierh. (Ranunculaceae) is an annual, self-compatible, diploid herb with four local races (referred to as subspecies) occurring in a variety of lightly disturbed habitats (along roadsides and stone walls, in phrygana vegetation and on sea shores) on different islands in the Cyclades (Greece). The 15–25-mm-wide, bee-pollinated flowers (Strid, 1970) of N. degenii have a double perianth with five petal-like sepals, eight conspicuously coloured, nectariferous petals and a variable number of stamens, which shed their pollen as the filaments curve outwards during the male phase. The gynoecium consists of five to 10 partly united carpels (follicles) with styles c. 15 mm long. Protandry, coupled with spatial separation of anthers and stigmas, enhances the potential for outcrossing even though the maturing styles sometimes become twisted around the dehiscing anthers, which results in self-fertilization. Fertilized flowers develop into capsules, each consisting of 20–40 seeds (occasionally up to 100 seeds) with no special mechanisms for dispersal. Seed viability declines significantly after 2–3 yr (Strid, 1970).

This study involves populations of two morphologically distinct subspecies: N. degenii ssp. barbro Strid, endemic to the island of Mykonos and a few adjacent islands, and ssp. jenny Strid, endemic to the island of Syros. These subspecies differ from related taxa in being dimorphic for pollen colour. Plants of ssp. barbro have either yellow or violet pollen grains, while plants of ssp. jenny produce white or dark violet pollen grains (Strid, 1970). Differences in pollen colour are not associated with differences in external pollen morphology or the expression of anthocyanin pigments in leaves and stems (T.H.J., personal observation). It does, however, show a strong association with anther colour in N. degenii ssp. jenny (Strid, 1970). Differences in pollen colour appear to be controlled by a single major gene with dominance towards the dark morph (Andersson & Jorgensen, 2005).

Field study

Twenty populations of N. degenii ssp. barbro and 21 populations of N. degenii ssp. jenny were selected for the present study (Fig. 1). These populations covered the main distributional range of each subspecies. Data on morph frequencies were obtained during peak flowering (late May to early June) by recording the pollen colour for a maximum of 200 flowering plants along a 2-m-wide and up to 400-m-long transect within each population. A global positioning system was used to obtain the following data for each population: location (latitude, longitude); altitude (m a.s.l.); slope direction (angle between direction of steepest ascent and N–S line); and minimum distance from the sea (m). The majority of populations were located on slopes with low-to-moderate steepness, so this parameter was not measured. The minimum distance from the sea was determined with a map (scale 1 : 40 000) if the distance exceeded 500 m. Geographical location provided data on interpopulation distances, whereas the other measures were used as proxies for unmeasured ecological variables such as humidity, solar radiation and degree of salt exposure. A bulked sample of soil (c. 100 ml), representing four or five spatially separated points (sampling depth 0–3 cm), was obtained from each population to provide data on soil acidity (pH-KCL) and the concentration of phosphate, determined after extraction with 0.5 m NaHCO3 at pH 8.5 (Olsen et al., 1954). These extraction conditions release phosphates bound to calcium, the predominant form of phosphorus available to plants growing on alkaline or neutral soils with low organic content (Tyler, 2002). Soil pH has a profound effect on the availability of nutrients in soil solution, and phosphorus is known to be a limiting element in many limestone soils (Tyler, 1992). Hence the combination of soil pH and phosphate content was assumed to provide a general view of weathering state and nutrient status.

Figure 1.

Frequencies of dark and light pollen morphs in populations of Nigella degenii on the islands of Mykonos (ssp. barbro) and Syros (ssp. jenny). The two islands belong to the Cyclades (black area on inserted map) in the Aegean Sea.

Stress-manipulation experiment

The experimental populations were based on plant material from one population of N. degenii ssp. barbro (Mykonos, c. 2.5 km north-north-west of the town) and one population of N. degenii ssp. jenny (Syros, c. 300 m south of the village Kini), sampled in 1993 and maintained for three generations by random outcrossing within each population (involving a minimum of 150 plants per population and generation) in a glasshouse at the University of Lund, Sweden. When the present study was initiated the two study populations still segregated for pollen colour, despite a relatively low frequency of the dark pollen morph in the initial population samples (c. 14% in both cases, S.A., personal observation).

Before initiating the stress-manipulation experiment, we established a series of full-sib families from a large number of intermorph crosses within each of the two study populations, and scored the resulting progeny for pollen colour. Some full-sib families contained only plants with dark pollen (putative heterozygotes), while others showed clear segregation for pollen colour with morph frequencies approaching a 1 : 1 ratio. The latter progenies were assumed to contain a mixture of heterozygotes (plants with dark pollen) and recessive homozygotes (plants with light pollen). To ensure the occurrence of both morphs in the stress-manipulation experiment, we obtained 10 segregating progeny families (hereafter ‘families’) for each population by crossing plants with different pollen colour (putative heterozygotes × recessive homozygotes), each pair representing a distinct combination of segregating families from the initial crosses.

In early March, 135–185 seeds from each family were planted individually in 180 ml plastic pots containing a mix of nutrient-poor peat soil (80%) and sand (20%), and placed in the glasshouse chamber with regulated watering and 12 h day lighting (60% humidity = 24°C). When most of the seeds had germinated (c. 15 d after sowing), between 87 and 123 seedlings per family were distributed across six blocks per population, each divided into four subblocks (different treatments, see below). The blocks were distributed across six benches in the same glasshouse chamber, each bench having one block with Mykonos plants (N. degenii ssp. barbro) and one with Syros plants (N. degenii ssp. jenny). In each of the six blocks, the four subblocks consisted of 45–48 randomly positioned plants from the same population (10 families represented by four to five plants each). Between four and 51% of the seeds sown in each family remained ungerminated after 3 wk and had to be excluded from the experiment.

In early April, when most plants had formed rosettes with seven to 10 leaves, the following four treatments were assigned to the four subblocks within each block: (1) high-nutrient/high-water (no stress); (2) low-nutrient/high-water (nutrient stress); (3) high-nutrient/low-water (drought stress); (4) low-nutrient/low-water (nutrient and drought stress). Plants in high-water treatments were watered two or three times a week depending on weather conditions, while plants in low-water treatments were subjected to a drought lasting 5–7 d every second or third week. Plants in high-nutrient treatments were watered with a 0.5% nutrient solution (NPK 6-1-5, SuperbaS), whereas plants in low-nutrient treatments received pure tap water. Plants in the low-water/high-nutrient group were watered with a stronger nutrient solution (usually 1%) once or twice after a drought period to ensure that they received the same amount of nutrients as plants in the high-water/high-nutrient group. All plants were watered with a dosage device (Dispensette 1–100 ml) to ensure that equal amounts of water were added to all plants within each treatment. Plants within each treatment (subblock) were assigned to new random positions once a week. The treatments were continued until all plants had wilted or ceased to initiate new flowers.

To assess the relationship between pollen colour and measures of plant performance, we recorded whether or not a plant had died before flowering (survival status), and scored each flowering plant for pollen colour, date of first flowering (number of days from sowing), and number of flowers with fully developed anthers and pistils. Early flowering indicates high resource status before flowering, at least under glasshouse conditions (S.A., personal observation), whereas survival rate and flower number represent major components of fitness. One of the Mykonos families contained mainly plants with light pollen and was therefore excluded from the analyses. A total of 1033 Mykonos plants and 1095 Syros plants provided data for the statistical analyses.

Statistical analyses

Field data  A partial regression method based on distance matrices was employed to test for a relationship between each of the ecological variables and the spatial pattern of morph frequency variation. This technique allows for hypotheses (differences in explanatory variables represented as dissimilarity matrices) to be tested against observed patterns (differences in morph frequencies as dissimilarity matrices) (Legendre et al., 1994). Geographical distances between populations, as well as differences in altitude, distance from the sea, slope direction, soil acidity and phosphate content, were all included as explanatory variables.

Geographical distances were calculated from longitude and latitude using the r package (Casgrain, 2002). The difference in proximity to the sea of two populations was assigned a value of 0 if both populations were classified as either coastal (<100 m from the sea) or inland (>100 m from the sea); or a value of 1 if they represented different categories. Distance matrices expressing differences in slope direction were constructed by assigning two populations a distance of 0 if the direction of the slope differed by <45°; a distance of 1 if the difference was 45–90°; 2 if the difference was 90–135°; and 3 if the difference was 135–180°. Distance matrices based on altitude and soil variables were expressed as the absolute difference between estimates for each population pair.

Geographical and ecological distance matrices were considered against the morph frequency matrix (expressed as differences in frequency of the dark morph) using the backward elimination procedure suggested by Legendre et al. (1994) with a P-to-remove value of 0.10. Analyses were performed with the software program permute! ver. 3.2 (Casgrain, 1995). When the dependent variable is represented as a simple distance matrix as in the present study, this program performs significance tests of the regression parameters in the same way as the Mantel permutation test (Mantel, 1967), while keeping independent matrix variables fixed against one another. Significance levels were determined by 999 matrix permutations. A relationship was declared significant when P was <0.05 after Bonferroni correction. Data from the two subspecies were analysed separately.

Stress-manipulation experiment  The frequency of plants that survived to the flowering stage (data pooled over blocks) was analysed in hierarchical log-linear analyses based on a model with survival status, family, nutrient level and water level as categorical factors. In a log-linear analysis, the logarithm of the expected frequency of observations in a given cross-classification category is modelled as a linear function of parameters, each corresponding to an interaction between two factors or a higher-order interaction between more than two factors. Testing for the significance of an interaction term requires fitting two models, one with the term and one with it omitted, and computing a goodness-of-fit statistic (G value) for each model. The difference between the two G values (expressed as a likelihood ratio, χ2) is used to test the significance of the term being left out (Sokal & Rohlf, 1995). All two-, three- and four-way interactions were entered in the original model and tested for significance using a backward elimination procedure with a P-to-remove value of 0.05. A similar approach was used to analyse the effects of family, nutrient level and water level on the frequency of surviving plants with dark vs light pollen (based on a model with pollen colour, family, nutrient level and water level as categorical factors).

As a complementary approach, we carried out a replicated goodness-of-fit analysis using families as replicates to compare the observed morph frequencies in each of the three stress treatments with morph frequencies in the optimal environment. The observed morph frequency for a given family in a particular stress treatment was compared with an expected frequency calculated by multiplying the observed morph ratio in the optimal treatment with the actual number of surviving plants in the stress treatment. This approach was preferred over a comparison of observed frequencies with the general expectation of a 1 : 1 morph ratio in each family, because any difference in seedling survivorship or germination time between morphs, if present, could bias morph ratios before the start of the experiment.

Individual data on flowering date and flower number were subjected to mixed-model factorial analysis of variance (anova) using type III sum of squares (PROC GLM, sas ver. 8) with block, family, colour morph, nutrient level and water level as main factors and initial plant size (leaf number) as a covariate. Particular attention was given to the overall difference between the two colour morphs (as determined by the main effect of colour morph) and significant morph-by-treatment interactions. Inclusion of family and block provided statistical control for genetic background effects and spatial variation within the glasshouse, respectively, whereas leaf number controlled for differences in plant size when the treatment differences were imposed. All factors except family were considered as fixed. The error term for each F test was synthesized with the Satterthwaite approximation using the TEST option in the RANDOM statement of PROC GLM (sas ver. 8), the only exceptions being the main treatment effects. The latter effects were tested against the block × water × nutrient interaction to account for the fact that different treatments were applied to groups of plants (subblocks) rather than individual plants randomized within blocks. Residual plots confirmed that data were approximately normally distributed.

Results

Field study

The populations visited in the field survey (Fig. 1) represented a broad range of altitudes, both on Mykonos (0–170 m) and Syros (20–250 m), although this was less than the full range of altitudes on these islands (maximum altitude 373 and 432 m for Mykonos and Syros, respectively), because of poor accessibility or late flowering of populations at the highest altitudes. Populations on Syros (N. degenii ssp. jenny) showed no bias with respect to slope direction, whereas the Mykonos populations (N. degenii ssp. barbro) occurred less often on slopes facing north. Populations on Syros and Mykonos did not differ significantly from one another in the concentration of soil phosphate (F1,43 = 2.51, P = 0.12; one-way anova), which ranged from 0.048 to 0.993 mol g−1 d. wt (mean = 0.257, SD = 0.244, data pooled across islands). Soil pH was high and differed significantly between Mykonos (mean pH = 6.84, SD = 0.43, range 6.14–7.68) and Syros (mean pH = 7.55, SD = 0.33, range 6.91–8.20; F1,43 = 34.80, P < 0.001, one-way anova).

The frequency of the dark pollen morph varied from 0 to 0.93 on Mykonos, and from 0.02 to 0.71 on Syros (Fig. 1). Judging from the partial regression analyses, the local morph frequency was not significantly affected by altitude, distance from the sea or soil chemistry (P > 0.05 in all cases) but varied significantly with slope direction on Mykonos (standardized partial regression coefficient = 0.16, P = 0.017). On average, the dark morph was more frequent on slopes facing north or east (mean frequency = 0.49, SD = 0.31, n = 7) than on slopes facing south or west (mean frequency = 0.24, SD = 0.25, n = 13) (Fig. 2). Data from Syros showed a weakly significant positive relationship between the difference in morph frequency and the geographical distance separating the populations (standardized partial regression coefficient = 0.12, P = 0.046). The proportion of variance in morph frequency explained by the regression models was low for both islands (R2 = 0.024, P = 0.039 for the Mykonos populations; R2 = 0.014, P = 0.099 for the Syros populations).

Figure 2.

Association between morph frequency and slope direction among populations of Nigella degenii ssp. barbro on the island of Mykonos.

Stress-manipulation experiment

Plant survival  Survival rates were usually higher in the optimal nutrient and water treatment than in the three stress environments (Fig. 3). Log-linear analyses with survival status, family, nutrient level and water level as categorical factors revealed significant three-way interactions between survival status, nutrient level and water availability (likelihood ratio χ2 = 13.0, df = 1, P < 0.001 for the Mykonos population; χ2 = 45.0, df = 1, P < 0.001 for the Syros population). These observations not only confirm that our low-water and low-nutrient treatments induced significant levels of stress, but also indicate that drought and nutrient deficiency had interactive effects on survival rate. Families responded differently to nutrient availability, as shown by a significant interaction between survival status, family and nutrient level (likelihood ratio χ2 = 58.5, df = 8, P < 0.001 for the Mykonos population; χ2 = 52.3, df = 9, P < 0.001 for the Syros population). The interaction between survival status, family and water availability failed to reach significance (P > 0.05).

Figure 3.

Bars showing numbers of Nigella degenii plants that died before flowering (hatched); light-pollen plants (light grey); dark-pollen plants (dark grey); and plants with male-sterile flowers (white), for each family in the different environments in the stress-manipulation experiment.

Log-linear analyses of the frequency of plants that survived to flowering, including colour morph, family, nutrient level and water level as categorical factors, revealed weakly significant family × colour morph interactions (likelihood ratio χ2 = 15.2, df = 8, P = 0.055 for the Mykonos population; χ2 = 18.5, df = 9, P = 0.030 for the Syros population), indicating slight differences between families in the proportions of plants with dark and light pollen (Fig. 3). No other interaction involving colour morph was significant (P > 0.05) in the log-linear analyses. Family-level frequencies of the dark morph in the optimal nutrient and water treatment varied between 0.37 and 0.70 for the Mykonos population, and between 0.29 and 0.67 for the Syros population (Fig. 3). These morph frequencies were significantly different from the 1 : 1 expectation in the Mykonos population (G = 17.9, df = 9, P < 0.05) but not in the Syros population (G = 15.0, df = 10, P > 0.05; replicated goodness-of-fit analysis). Given these observations, we used the morph ratios in the optimal treatment rather than the 1 : 1 ratio as a baseline against which to test the morph frequencies in the stress environments (see Statistical analyses).

Replicated goodness-of-fit analyses of the Mykonos data revealed a significant difference between the proportion of morphs in the optimal nutrient and water treatment and the morph frequencies in the nutrient-deficiency (G = 18.0, df = 8, P < 0.05) and drought-stress (G = 34.6, df = 8, P < 0.001) treatments. This contrasts with the nonsignificant change in morph frequencies for plants subjected to both nutrient and drought stress (G = 11.6, df = 8, P > 0.05). As shown in Fig. 4, most families had an excess of the light morph after nutrient or drought stress, indicating that plants with dark pollen had a relatively low survival rate before flowering. In some families, the dark morph did poorly under drought stress but well in the nutrient stress treatment compared with other families.

Figure 4.

Observed and predicted morph frequencies (absolute numbers) for Nigella degenii plants that survived to flowering in the three stress treatments. Each symbol represents the same full-sib family in all three treatments of one subspecies. Predicted frequency of dark morph was calculated from observed frequency of dark morph in the optimal environment. Open symbols represent families showing a significant change in morph frequency (P < 0.05) as determined by a separate goodness-of-fit test for each family (significance levels corrected for multiple testing using Šidák's multiplicative inequality). Fisher's exact test was applied when the expected frequency was <5.

In the Syros population, we found a significant difference in morph frequency between the optimal nutrient and water environment and the nutrient-poor environment (G = 23.1, df = 9, P < 0.01). Most of this difference can be attributed to a single family that showed a drastic reduction in the frequency of the dark morph after nutrient deficiency (Fig. 4). Morph frequencies in the other stress treatments corresponded with morph frequencies in the optimal treatment (G < 9, P > 0.05).

Flowering date and flower number  Pollen colour did not significantly affect the number of days to flowering or the total number of flowers produced, regardless of whether the plants experienced nutrient stress, drought stress, or a combination of both (no nutrient-by-morph, water-by-morph, or water-by-nutrient-by-morph interactions; Tables 1 and 2). According to the mean squares, the most important sources of variation were initial leaf number, block, the two treatment factors, the block-by-treatment interactions, and the water-by-nutrient interaction. Judging from the treatment means (data not shown), nutrient and/or water deficiency delayed the mean flowering date by 9–16 d compared with the optimal treatment, whereas the mean number of flowers per plant declined from 19 (Syros) or 26 (Mykonos) in the optimal treatment to one to three (Syros) or two to six (Mykonos) in the three stress treatments. The main effect of family was significant for date of first flowering in the Mykonos population (Table 2) and there was a weakly significant family-by-water-by-nutrient interaction (P < 0.05) for flower number in both populations (Table 1).

Table 1.  Mixed-model factorial anovas on number of fertile flowers produced by plants from Mykonos (Nigella degenii ssp. barbro) and Syros (N. degenii ssp. jenny) in the stress-manipulation experiment
Source of variationMykonosSyros
dfMSFdfMSF
Leaves  1  257.70 14.37***  1  176.55  16.46***
Block  5   79.12  4.41***  5    2.25   0.21
Family  8  119.57  1.23  9   47.98   1.43
Water  112639.76 93.41***  1 9148.392891.40***
Nutrients  124150.05178.47***  113625.144306.29***
Morph  1    1.49  0.10  1   19.35   3.77
Block × water  5  161.29  8.99***  5    4.65   0.43
Block × nutrients  5   93.77  5.23***  5    5.95   0.55
Block × water × nutrients  5  135.32  7.55***  5    3.16   0.29
Family × water  8   97.58  1.35  9   19.15   0.91
Family × nutrient  8   75.19  0.94  9   35.74   2.00
Family × morph  8   12.57  2.10  9    4.88   0.56
Water × nutrient  112969.94204.91***  1 9750.73 582.37***
Water × morph  1   26.94  2.78  1    2.67   0.33
Nutrient × morph  1    4.37  0.30  1    4.34   0.84
Family × water × nutrient  8   78.67  6.46**  9   17.08   4.23*
Family × water × morph  8    4.20  0.36  9    7.93   1.93
Family × nutrient × morph  8   12.97  1.05  9    4.88   1.19
Water × nutrient × morph  1   45.62  3.85  1    0.39   0.09
Family × water × nutrient × morph  7   11.74  0.65  9    4.12   0.38
Error759   17.93 670   10.73 
Table 2.  Mixed-model factorial anovas on date of first flowering for plants from Mykonos (Nigella degenii ssp. barbro) and Syros (N. degenii ssp. jenny) in the stress-manipulation experiment
Source of variationMykonosSyros
dfMSFdfMSF
Leaves  1 1883.07 34.19***  1 2633.24 33.66***
Block  5  179.84  3.27**  5  332.73  4.25***
Family  8  797.48  3.05*  9  727.74  7.81
Water  1  271.46 14.29*  1 3729.06 15.05*
Nutrients  111156.37587.12***  114098.44 56.89***
Morph  1  237.65  3.78  1    2.99  0.03
Block × water  5  180.01  3.27**  5  122.36  1.56
Block × nutrients  5   96.23  1.75  5  157.53  2.01
Block × water × nutrients  5   19.00  0.35  5  247.81  3.17**
Family × water  8  143.20  2.52  9   81.53  0.79
Family × nutrient  8  164.99  2.67  9   81.45  2.16
Family × morph  8   67.73  1.15  9  104.86  0.74
Water × nutrient  1 1370.01 28.16***  1 4878.68143.84***
Water × morph  1   68.01  1.56  1  136.97  1.02
Nutrient × morph  1   64.83  1.28  1  107.39  1.49
Family × water × nutrient  8   47.42  1.35  9   31.38  0.48
Family × water × morph  8   42.86  1.29  9  136.14  2.08
Family × nutrient × morph  8   49.60  1.38  9   71.85  1.10
Water × nutrient × morph  1    0.19  0.01  1  107.05  1.62
Family × water × nutrient × morph  7   33.73  0.61  9   65.40  0.84
Error756   55.07 637   78.23 

Discussion

All close relatives of N. degenii have pollen grains that are either white or yellow (Strid, 1970). Thus the dark pollen types in this species probably evolved in situ rather than being inherited from some unknown ancestor, and there is no reason to consider the light morph as a novel phenotype, as in many other floral colour polymorphisms (Levin & Brack, 1995; Warren & Mackenzie, 2001). Some ecological factors are therefore hypothesized to favor the dark pollen morphs in N. degenii, either as a result of direct selection on pollen colour or as a correlated genetic response to selection on other characters. Data on fertilization ability indicate a selective advantage of the dark morph (unpublished data), and several populations of both subspecies contain a high frequency of plants with dark pollen; however, none of the populations has reached fixation for the dark pollen type, and there are several populations in which the light morph predominates. These patterns indicate either that different populations are at different stages on the way to the fixation of the dark morph, or that the relationship between pollen colour and fitness depends on environment in a way that favours the retention of the ancestral (light) phenotype. The latter hypothesis would be supported if the frequency of the two colour morphs varies in a habitat-correlated manner, and if the relative morph fitnesses varies across environments. Ecological and experimental data from the present study indicate a potential for morph-by-environment interactions to stabilize pollen-colour dimorphisms in N. degenii ssp. barbro. No such morph-by-environment interactions were detected in the other subspecies.

The spatial analyses of morph frequencies in N. degenii ssp. barbro demonstrated a weak, though significant, correlation between slope direction and the frequency of the dark morph, pointing to the possible influence of environmental factors such as solar radiation and humidity on the relative fitness of the two colour morphs. Plants with dark pollen grains were most frequent on slopes facing north or east, in populations with presumably lower sun influx and higher humidity than other populations. Thus one would expect the light pollen morph to outperform the dark morph under low-water conditions, in terms of either survival or siring ability. Results from our stress-manipulation experiment not only indicated higher mortality of the dark morph in the drought treatment, but also showed reduced survivorship of the dark morph under nutrient-poor conditions. Our data also suggest a genetic influence on the relative morph fitnesses in two of the stress treatments: in some families the dark pollen morph performed poorly under drought stress but well under nutrient deficiency compared with other families (and vice versa). Such interactions could contribute to the maintenance of pollen-colour dimorphisms under variable and stressful field conditions, and have also been invoked to explain the persistence of high genetic diversity in other plant systems (Abbott et al., 1998; Stratton & Bennington, 1998; Galen, 1999a, 1999b).

The lack of detectable relationship between morph frequency and nutrient status of the soil (as determined by pH and phosphate content) in the spatial analysis is not in agreement with the morph-by-nutrient interactions seen in the stress-manipulation experiment. Therefore if nutrient status plays a major role in determining morph frequencies in natural populations of N. degenii ssp. barbro, it must either be related to ecological factors other than pH and phosphate content, or be operating on a finer scale than that measured in our field study.

Comparison of current morph frequencies in N. degenii ssp. barbro with the corresponding data from a previous study (Strid, 1970) revealed a significant increase in the frequency of the dark pollen morph in four out of seven populations over a 35-yr period, the remainder showing a nonsignificant change (two populations) or a significant difference in the opposite direction (one population) (unpublished data). Although these observations agree with a scenario involving directional selection for plants with dark pollen, they also provide support for the existence of counteracting selection forces in some localities.

The significant differences in reproductive and vegetative performance between the dark and light pollen morphs of N. degenii ssp. barbro contrast sharply with the lack of morph-specific fitness differences in N. degenii ssp. jenny. The Syros population showed some evidence for a morph-by-environment interaction in survival rate, arising from the different response of a single family in one of the stress environments, but there was little change in the ranking of morph fitness. According to the spatial analyses, the frequencies of the dark and light morphs of N. degenii ssp. jenny varied independently of local habitat conditions, indicating that nonselective factors (genetic drift, restricted gene flow) play a more important role in determining morph frequencies in this subspecies.

The natural populations of both N. degenii ssp. barbro and N. degenii ssp. jenny show greater divergence in the frequency of the two pollen-colour morphs than in putatively neutral marker genes (unpublished data). In N. degenii ssp. jenny, there is also evidence for a weak, but significant, positive association between morph frequency and geographical distance between populations, contrasting with the lack of distance effects seen in marker gene analyses (unpublished data). Consequently, we cannot dismiss the possibility that some unmeasured selective factor is contributing to the morph frequency variation in N. degenii ssp. jenny. In this context, we also note that experimental data on pollination and fertilization success from both subspecies generally support a scenario involving the co-occurrence of both pollen-colour morphs, particularly under conditions of high pollen competition, a variable genetic background, and/or spatial or temporal variation in the pollinator fauna (unpublished data).

Genetic associations between floral and nonfloral characters have also been observed in other plant species (Andersson, 1997; Abbott et al., 1998; Armbruster et al., 1999). When considering polymorphisms in flower colour, there is growing evidence for links between floral pigmentation and measures of plant performance (Jones, 1996; Warren & Mackenzie, 2001; see also Irwin & Strauss, 2005). For instance, individuals of Ipomoea purpurea with unpigmented (white) petals were less vulnerable to herbivory than conspecific plants with dark petals (Simms & Bucher, 1996), and had a reduced male and female fertilization success at high temperatures but not at low temperatures (Coberley & Rausher, 2003). White-flowered plants of Phlox drummondii had lower survivorship and flower production than the pigmented flowers (Levin & Brack, 1995), and a well documented flower colour dimorphism in Linanthus parryae (Epling & Dobzhansky, 1942) may persist because of an interaction between flower colour and water availability (Schemske & Bierzychudek, 2001). According to a phylogenetically informed analysis of species in Dalechampia (Armbruster, 2002), the expression of pink or purple pigments in the floral bracts is consistent with a correlated response to selection on stem and leaf pigmentation, and not to pollinator-mediated selection on bract colour.

The presence of genetically based associations between floral and vegetative characters indicates that floral and nonfloral structures sometimes share resources, developmental history or common systems of growth regulation, or that the different characters are controlled by different genes in linkage disequilibrium. We have no direct evidence as to the mechanism underlying the relationship between pollen colour and survival rate in nutrient- and water-stressed plants of N. degenii ssp. barbro. However, the synthesis of anthocyanin-based pigments is known to produce compounds that are important in stress tolerance (Koes et al., 1994; Shirley, 1996) and to changes in response to environmental conditions such as nutrient availability (Bonguebartelsman & Phillips, 1995). Moreover, the majority of the nonflowering individuals in the stress treatments died just before flowering (T.H.J., personal observation), at a stage when the synthesis of floral pigments could affect plant vigour because of a pleiotropic relationship. In this context, we also note that pollen colour had a nonsignificant influence on the remaining performance variables (flowering date, flower number), and that the morph effect varied in magnitude between families. The latter finding indicates genotype-specific differences in the amount of linkage disquilibrium or in the strength of pleiotropic relationship between pollen colour and survivorship.

Acknowledgements

The present work was supported by a fellowship from the Danish Natural Science Research Council (T.H.J.), and grants from the Swedish Natural Science Research Council (S.A.) and the Julie von Müllens foundation, the Svend G. Fiedlers foundation and the Frimodt-Heineke foundation (T.H.J.). Thanks to Olympia Tassopoulou and colleagues at the Department of Environment, The Prefecture of Cyclades, for encouraging the project and to David S. Richardson for comments on the manuscript.

Ancillary