1. Unreliable pollinator service is thought to promote the evolution of self-compatible plant breeding systems, because selfing may provide reproductive assurance when outcrossing opportunity is limited. The recurrent evolution of self-compatible homostyly from obligately outcrossing heterostylous species has been regarded as a classic example of evolutionary response to lack of pollinators or mates, as homostyly frequently occurs in pollinator-limited or marginal environments. However, male and female sexual organs of homostylous species may display spatial separation (herkogamy), an arrangement presumed to promote outcrossing. It is largely unknown to what extent variation in herkogamy affects opportunities for autonomous selfing and reproductive assurance in self-compatible, homostylous species.
2. Using the homostylous Primula halleri, restricted to alpine environments, we investigated whether herkogamy occurs and varies during anthesis, among individuals, and populations, and compared the effects of herkogamy on seed set among three experimental treatments, to elucidate how herkogamy affects reproductive strategies in a homostylous species.
3. Herkogamy decreases during anthesis, but the ultimate expression of herkogamy in mature flowers differs among individuals and populations. Caging caging experiments indicate that herkogamy reduces a plant's potential for autonomous selfing, and emasculation and open-pollination treatments demonstrate that herkogamy markedly decreases total seed set and the potential for reproductive assurance.
4. Herkogamy early in anthesis may enhance outcrossing potential, while its decrease later could enable reproductive assurance via delayed autonomous selfing in some, but not all plants. Conversely, pronounced herkogamy in older flowers comes at the cost of reduced total reproductive output and imposes pollinator dependence for reproduction, but may promote the genetic diversity of populations.
5. Our study suggests that even small amounts of herkogamy can have large effects on the reproductive strategy of homostylous species, by enabling more outcrossing than generally thought to be typical of homostyly.
Scarcity of pollinator services has major consequences for plant reproduction and evolutionary processes, as recognized by early naturalists (e.g. Müller 1881; Schröter 1926). The pollinator fauna of alpine (i.e. above tree line) environments is generally depauperate, in terms of number of species and individuals, as compared to that of lower altitudes (e.g. Arroyo, Primack & Armesto 1982; Warren, Harper & Booth 1988). Moreover, the short flowering season and fluctuating weather conditions typical of alpine ecosystems further impair the reliability of pollinator services (Totland 1994; Bergman, Molau & Holmgren 1996; Körner 2003). Similar trends in pollination conditions occur with increasing latitude towards the poles (Hocking 1968; Kevan 1972).
When lack of pollinators limits outcrossing opportunity and reproductive output (García-Camacho & Totland 2009), autonomous selfing (i.e. autogamy unaided by pollinators) may boost total seed production (i.e. reproductive assurance; Eckert, Samis & Dart 2006). Similarly, autonomous selfing may be beneficial when mates are scarce, as during colonization processes (Baker 1955), and in geographically or ecologically marginal habitats (Lloyd 1980). Reproductive assurance has often been invoked as an explanation for the evolution of selfing from primarily outcrossing ancestors (Darwin 1876; Fausto, Eckhart & Geber 2001; Kalisz, Vogler & Hanley 2004; Eckert, Samis & Dart 2006; Moeller 2006), widely recognized as one of the most frequent evolutionary transitions in flowering plants (Stebbins 1950; Grant 1981). Despite the ecological and evolutionary importance of reproductive assurance, experimental demonstrations remain scarce (reviewed by Eckert, Samis & Dart 2006).
Selfing is often associated with negative fitness effects because of reduced survival and fertility of the offspring (inbreeding depression; Charlesworth & Charlesworth 1987; Charlesworth & Willis 2009), as well as with long-term negative consequences on the genetic variability and viability of populations, potentially representing an evolutionary dead end (reviewed by Takebayashi & Morrell 2001). Therefore, autogamy may decrease fitness when ovules and pollen that could otherwise be outcrossed are self-fertilized (i.e. gamete discounting; Herlihy & Eckert 2002; Eckert & Herlihy 2004). Discounting costs can be incurred when autonomous selfing takes place prior to or competing with outcrossing (Lloyd 1992), or when pollinators mediate selfing concurrently with outcrossing by foraging within flowers (facilitated selfing) or between flowers of the same plant (geitonogamy; Vaughton & Ramsey 2010). However, if autonomous selfing occurs at the end of floral life after opportunities for outcrossing have been exhausted (i.e. delayed autonomous selfing; Lloyd 1992), it affords the benefits of autogamy, while avoiding discounting costs, a ‘best-of-both-worlds’ scenario that seems ideally adaptive in alpine/arctic habitats (Kalisz & Vogler 2003; Moeller 2006; Duan et al. 2010; Vaughton & Ramsey 2010). The relative timing and mode of selfing and outcrossing events may thus be important for overall reproductive fitness and long-term evolutionary survival (Lloyd 1992; Eckert, Samis & Dart 2006; Vaughton & Ramsey 2010).
Plants have evolved a wide range of floral traits that may influence the dynamics of sexual reproduction, providing the morphological and physiological basis of plant reproductive strategies. The study of the function and loss of complex floral polymorphisms has supplied key model systems for understanding the evolution of selfing (Darwin 1877; Barrett 2003, 2010). A prime example is the evolution of self-compatible homostylous species from obligately outcrossing heterostylous species (Ganders 1979; Barrett 1992; Barrett & Shore 2008; Cohen 2010). Heterostyly is thought to promote cross-pollination and reduce selfing and sexual interference, via the reciprocal placement of male and female organs in different floral morphs and an incompatibility system that prevents pollen germination within the same flower or floral morph (Darwin 1877; Charlesworth & Charlesworth 1979; Ganders 1979; Barrett 1992; Barrett, Jesson & Baker 2000; Barrett 2002; Barrett & Shore 2008; Cohen 2010). Therefore, heterostylous flowers depend on pollinators and mates for sexual reproduction. Homostylous species evolved multiple times independently from heterostylous ancestors in many of the c. 28 plant families with heterostyly (e.g. in Amsinckia, Boraginaceae, Schoen et al. 1997; Houstonia, Rubiaceae, Church 2003; Narcissus, Amaryllidaceae, Graham & Barrett 2004; Pontederiaceae, Kohn et al. 1996; Primula, Primulaceae, Mast, Kelso & Conti 2006; Turnera, Turneraceae, Truyens, Arbo & Shore 2005). Homostylous species have only one floral morph (i.e. monomorphic) and are self-compatible; hence self-fertilization is possible (reviewed by Ernst 1962; Ganders 1979; Barrett 1992; Barrett & Shore 2008; Cohen 2010). Although the term homostyly is sometimes applied to any plant species with stigmatic surface and pollen presentation at the same level, we refer here exclusively to monomorphic species that evolved within the context of heterostylous groups.
Because of the advantages of selfing, homostylous species have been hypothesized to be more successful than heterostylous relatives under ecological conditions that limit pollinator abundance, visitation activity or mate density (e.g. in Amsinckia, Ganders 1975; Plumbaginaceae, Baker 1966; Primula, Kelso 1992; Richards 2003; Guggisberg et al. 2006; Psychotria, Sakai & Wright 2008; Turnera, Barrett & Shore 1987). However, spatial separation between male and female sexual organs (i.e. herkogamy; Webb & Lloyd 1986) has been reported in several homostylous species (e.g. in Primula, Ernst 1962; Al Wadi & Richards 1993; Tremayne & Richards 1993; Amsinckia, Johnston & Schoen 1996; Turnera, Barrett & Shore 1987; Narcissus, Medrano, Herrera & Barrett 2005; Larrinaga et al. 2009). Herkogamy can negatively affect the relative selfing rate (e.g. shown in Aquilegia, Brunet & Eckert 1998; Herlihy & Eckert 2007; Clarkia, Holtsford & Ellstrand 1992; Datura, Motten & Stone 2000; Mimulus, Karron et al. 1997; Nicotiana, Breese 1959; Turnera, Belaoussoff & Shore 1995; but see Medrano, Herrera & Barrett 2005 on Narcissus), as it may decrease autonomous or facilitated selfing (Webb & Lloyd 1986; Barrett 2002). Importantly, herkogamy is usually heritable and may thus respond to selection (e.g. Shore & Barrett 1990; Lennartsson et al. 2000; Herlihy & Eckert 2007; Bodbyl Roels & Kelly 2011). However, it remains unclear exactly how variation in herkogamy may influence the potential for autonomous selfing under conditions of limited pollinator availability (Moeller 2006).
In primroses (Primula), the classic model for homostyly (Scott 1865; Darwin 1877), homostylous species have been predicted to be better adapted than their heterostylous relatives to the ecological settings typical of alpine and arctic environments (e.g. Kelso 1992; Richards 2003; Guggisberg et al. 2006; Carlson, Gisler & Kelso 2008; Guggisberg, Mansion & Conti 2009). However, the potential role of herkogamy on the reproductive behaviour of homostylous primroses has never been considered: the alpine Primula halleri J.F.Gmel. (Fig. 1) provides an ideal study system to investigate it. P. halleri represents a classic example of the loss of heterostyly in alpine environments (e.g. Darwin 1877; Schröter 1926; Richards 2003), and extensive crossing experiments conclusively demonstrated that it is self-compatible (Ernst 1951). Early studies reported the occurrence of herkogamy in the species (Schröter 1926) and remarked that it may vary during floral anthesis (Ernst 1925). It is thus conceivable that this developmental variation, if sufficiently large to affect floral function, might offer contrasting mating opportunities at different stages of anthesis, including the possibility of delayed autonomous selfing. Mating opportunities that shift with the age of a flower have been mentioned in several species, although experimental evidence is generally limited (reviewed by Marshall et al. 2010).
The present study addresses the current gap of knowledge on how variation in herkogamy may affect pollinator dependence and opportunity for reproductive assurance in homostyly. The traditional focus on the proximity of sexual organs and self-compatibility has led to an interpretation of homostylous taxa as being primarily selfing and adapted to unreliable pollinator services, while the possible effects of herkogamy have been largely overlooked. Using P. halleri as our study system, we test whether: (i) the phenotypic expression of herkogamy changes during anthesis and variation between individuals and populations occurs and (ii) herkogamy affects seed set in open-pollinated, caged and emasculated plants. More broadly, the present study examines the effects of intraspecific and developmental variation in sexual organ distance on different components of plant reproductive success and contributes to understanding plant reproductive strategies in alpine environments.
Materials and methods
Primula halleri J.F.Gmel. (synonym Primula longiflora All.; sect. Aleuritia Duby; Richards 2003) is a herbaceous perennial bearing 3–19 flowers in an umbel (Fig. 1), with one or two flowers opening per day (J. M. de Vos and S. T. Isham, pers. obs.). The anthers, attached to a short filament, are positioned c. 1 mm below the mouth and dehisce when the corolla opens. The stigma is placed either among, above (protruding 1–4 mm, occasionally up to 8 mm) or, rarely, below the anthers (see below; Ernst 1925). Flowers have simultaneous male and female phases (no dichogamy; Lloyd & Webb 1986; Lennartsson et al. 2000; Isham 2010; see Fig. S1, Supporting Information). Flowers wilt after 6–12 days of anthesis; manual pollination does not induce wilting. Because flowers develop sequentially, a single scape can bear open flowers for up to 3 weeks. The hummingbird-hawkmoth (Sphingidae: Macroglossum stellatarum Linnaeus, 1758) is the main pollinator (Schulz 1890), but appears to visit flowers very infrequently (observed briefly on three of 25 consecutive days of field monitoring; J. M. de Vos and S. T. Isham, pers. obs.). Primula halleri occurs in the Alps, Carpathian mountains and Balkan region, between (1000-) 1800 and 2400 (-2900) m (Lüdi 1927); its closest relative is the heterostylous, largely self-incompatible Primula farinosa (Guggisberg et al. 2006; Guggisberg, Mansion & Conti 2009).
After initial surveys in 2008, three populations (named A, B and C; see Table 1) of P. halleri in the Swiss Alps (Central and Upper Canton Valais) were selected as study sites. All populations occurred on steep (20°–45°), south facing, nutrient-poor grassy slopes on limestone at 2300–2350 m altitude, where the species is most abundant (Lüdi 1927). Populations A and C included several thousand individuals at a density of up to c. 10 plants m−2; population B included c. 1000 individuals at slightly lower density. Populations were sufficiently distant from busy trails and pastures to minimize the possibility of anthropogenic disturbance. The rarity and recent decline of the species (Wohlgemut, Boschi & Longatti 2006) and time-consuming ascents restricted the number of populations that could be studied. All fieldwork was performed in June–September 2009.
Table 1. Interfloral variation among mature flowers for anther position, stigma position and herkogamy within populations A, B, C: mean trait values (±standard deviation, SD), results of Kruskal–Wallis tests for differences among plants and partitioning of variance among populations and among plants nested within populations (see text and Fig. 4)
Mean ± SD (mm)
Significance of variation among plants
Mean ± SD (mm)
Significance of variation among plants
Mean ± SD (mm)
Significance of variation among plants
Among populations, %
Among plants, %
27·0 ± 2·8
χ2 = 70·1, d.f. = 24, P < 0·001
26·6 ± 2·2
χ2 = 32·3, d.f. = 12, P < 0·001
26·1 ± 2·1
χ2 = 37·7, d.f. = 14, P < 0·001
28·5 ± 2·7
χ2 = 65·9, d.f. = 24, P < 0·001
27·2 ± 2·1
χ2 = 32·3, d.f. = 12, P = 0·004
25·9 ± 2·0
χ2 = 35·7, d.f. = 14, P = 0·001
1·5 ± 1·4
χ2 = 60·0, d.f. = 24, P < 0·001
0·6 ± 1·3
χ2 = 32·3, d.f. = 12, P = 0·001
−0·2 ± 1·7
χ2 = 37·1, d.f. = 14, P < 0·001
Sample size (# flowers/# plants)
75 fl./25 pl.
39 fl./13 pl.
45 fl./15 pl.
46°22·94′N/8°13·72′E (2300 m)
46°20·96′N/8°09·16′E (2300 m)
46°02·62′N/7°49·86′E (2350 m)
Phenotypic variation in sexual organ position and herkogamy
Developmental variation during anthesis
We tested whether herkogamy changes during floral development using a regression approach in which the positions of sexual organs were modelled as a function of the number of days flowers had been open. Because precise measurements were only possible in fixed flowers, we harvested inflorescences in 70% EtOH when displaying a wide range of floral ages. We could determine relative floral age (i.e. opening order) from the position of flowers within each inflorescence, because they develop from the base to the top of the inflorescence and are positioned approximately in Fibonacci spirals (Fig. 1). Hence, the flower lowest in the inflorescence is the oldest flower, the one positioned at an angle of c. 110° from it is the second oldest, etc. We determined the absolute age of each flower (number of days since the corolla first opened), by combining the relative age with detailed field observations on the number of open and closed flowers in each inflorescence on alternate days. To determine the positions of sexual organs, the corolla tube and calyx of each flower were slit longitudinally, opened, and a high-resolution digital image was taken with a Canon PowerShot A610 camera. The distance from the base of the ovary to (i) the base of the stigma (‘stigma position’) and (ii) the top of the anther (‘anther position’) were then measured to an accuracy of 0·01 mm using ImageJ 1.43 (http://rsbweb.nih.gov/ij/). The degree of herkogamy of each flower was calculated as the difference between (i) and (ii). Because the mean anther length of P. halleri was 2·4 mm (data not shown), the herkogamy values of flowers with stigmas between the anthers varied between 0 and c. −2·4 mm. We analysed only inflorescences in which ages of flowers could be determined unambiguously: 199 flowers from 25 individuals (4–18 flowers per inflorescence) of population A in total.
The statistical software r (v.2.14.1; R Development Core Team 2011) was used for all analyses in this study. To test whether herkogamy changed during anthesis, we applied a linear mixed effects model (LMM) fitted with restricted maximum likelihood (REML), implemented in nlme (Pinheiro et al. 2012). We treated floral measurement (i.e. both anther position and stigma position) as response variable. As fixed effects, we employed floral age, organ type (anther or stigma) and their interaction to test whether the positions of anthers and stigma changed with floral age and at different rates, thus causing changes in herkogamy during floral development. We used plant identity and flower identity nested within plant identity as random effects to account for variation between individuals and for hierarchical data structure. Finally, we calculated Spearman rank correlations between herkogamy and floral age to assess the rate and significance of decrease in herkogamy within each inflorescence, which allowed us to establish whether developmental changes in herkogamy differed between individuals.
Variation within and between populations
We investigated whether variation in anther position, stigma position, and herkogamy was explained by non-developmental, interfloral components in 75, 39 and 45 flowers, respectively, from randomly selected inflorescences of populations A, B, C (see Table 1). To correct for developmental variation during anthesis, we analysed flowers from the same developmental stage. We included only the three oldest, non-wilting flowers of each inflorescence (hereafter called ‘mature flowers’), representing the ultimate expression of herkogamy. Positions of floral organs and herkogamy were determined as described above.
First, we used Kruskal–Wallis tests to assess whether the anther and stigma positions and the herkogamy of mature flowers differed significantly between individuals within populations. Secondly, we tested whether the examined floral traits differed significantly between populations by building a LMM for each population pair and floral trait. LMMs included one of the three floral traits as response variable, population as fixed effect and plant identity as random effect. Thirdly, we assessed how variances for anther position, stigma position and herkogamy are partitioned within and between populations by extracting the variance components for each floral trait from linear models fitted with REML, implemented in lme4 (Bates, Maechler & Bolker 2011). The models included population and plant identity nested within population as two random effects. Finally, we used two linear regression models to analyse correlations between herkogamy and floral organ positions, with herkogamy as response variable, population as a categorical effect and either anther or stigma position as quantitative predictor. Similarly, we used a third linear regression model to analyse the correlation between anther and stigma position among populations. We tested for heterogeneity in correlations among populations by including population × sexual organ interactions.
Effects of herkogamy on seed set
We evaluated whether interindividual variation in herkogamy affects reproductive assurance and pollinator dependence in P. halleri, by analysing total seed set of plants that differed in the herkogamy of mature flowers under four experimental treatments: open pollination (control), caging (i.e. pollinator exclusion), emasculation and caging + emasculation. Caged plants can only set seed via autonomous selfing, whereas emasculated plants can only set seed through outcrossing. Open-pollinated plants can potentially do both and can also produce seed via pollinator-mediated selfing. Plants that are both caged and emasculated can potentially produce seed asexually (i.e. apomixis). The occurrence of reproductive assurance can thus be determined by testing whether seed set is higher in open-pollinated than emasculated plants for a particular herkogamy class. If homostyly is associated with predominant selfing, as often presumed, we expect to find no significant difference in seed set between caged and control treatments, and higher seed set in control than emasculation treatments. If, on the other hand, herkogamy affects the reproductive strategy of homostylous species by diminishing selfing, we expect to find, as herkogamy increases, lower seed set in caged plants and a smaller difference in seed set between emasculated and control plants.
To avoid damage to or accidental hand pollination of the flowers during manual measurements of herkogamy, individual plants were assigned to one of four herkogamy classes defined upon visual inspection of mature flowers in the inflorescence (see also Medrano, Herrera & Barrett 2005). Class assignment was congruent between two independent observers and on different days. We used the following herkogamy classes: 0, 1, 2 and 3–4 mm. Because anthers are invariably attached c. 1 mm below the corolla mouth, we could use the position of the stigma relative to the corolla mouth to assign emasculated flowers to the corresponding herkogamy classes.
Cages consisted of a chicken-wire frame covered by a fabric used to protect crop plants from pest insects and were employed in populations A, B, C. Emasculations were performed in unopened flowers near anthesis, by making a small longitudinal slit in the corolla and removing the undehisced anthers with tweezers. Treatments involving emasculation were performed in population A. To capture seeds, we bagged wilted flowers before capsules opened. Ripe capsules were collected at the end of the season. The number of seeds and undeveloped ovules were counted from a digital image or directly under a dissecting microscope. Seed set was expressed as the proportion of ovules that developed into seeds, thus allowing to correct for the large variation in ovule number, ranging between 197 and 493 ovules per flower. Accounting for loss of replicates because of the damage to the plants or abnormal fruit development, seed set could be determined in a total of 361 flowers from 76 plants; over 60 000 seeds and ovules were counted (see Table 2 for details on sample sizes).
Table 2. Effects of herkogamy in mature flowers and open-pollination (control), caged (pollinator exclusion) and emasculation treatments on seed set; results of generalized linear mixed effect models. Top: mean seed set (± standard deviation) and treatment effect size (± standard error) within individual herkogamy classes and all herkogamy classes. Bottom: herkogamy effects within treatments, indicating significance. Seed set is expressed as the proportion of ovules in a flower that developed into seed. See also Fig. 5. Sample sizes: open pollination, 125 flowers/30 plants (10 plants per population); caging, 142 flowers/30 plants (10 plants per population); emasculation, 40 flowers/7 plants (1 population); emasculation plus caging, 40 flowers/9 plants (1 population)
Open pollination (control)
Caged (pollinator exclusion)
Mean seed set (median)
Mean seed set (median)
Treatment effect: Control vs. pollinator exclusion
Mean seed set (median)
Treatment effect: Control vs. emasculation
0·82 ± 0·31 (0·94)
0·59 ± 0·43 (0·885)
−0·97 ± 0·61
t23 = −1·59, P = 0·12
0·35 ± 0·30 (0·26)
−2·77 ± 0·96
t23 = −2·9, P = 0·008
0·74 ± 0·29 (0·815)
0·25 ± 0·36 (0·025)
−2·58 ± 0·54
t16 = −4·80, P < 0·001
0·22 ± 0·37 (0·06)
−2·60 ± 0·87
t16 = −2·99, P = 0·009
0·66 ± 0·24 (0·68)
0·07 ± 0·20 (0·00)
−3·87 ± 0·93
t10 = −4·16, P = 0·002
0·61 ± 0·27 (0·67)
0·00 ± 0·00 (0·00)
−4·81 ± 2·94
t32 = −2·10, P = 0·044
0·37 ± 0·39 (0·26)
−0·73 ± 0·50
t32 = −1·47, P = 0·15
0·74 ± 0·29 (0·89)
0·33 ± 0·41 (0·025)
−1·05 ± 0·50
t60 = −2·10, P = 0·040
0·32 ± 0·34 (0·23)
−0·46 ± 0·76
t60 = −3·26, P = 0·002
−0·41 ± 0·16 (t94 = −2·55, P = 0·012)
−1·53 ± 0·41 (t60 = −3·72, P < 0·001)
0·36 ± 0·36 (t60 = 1·00, P = 0·31)
We employed generalized linear mixed effects models (GLMMs), using a binomial error distribution with logit link function, because the response variable ‘seed set’ is a proportion. Population and plant identity nested in population were used as random effects. GLMMs were fitted using Penalized Quasi Likelihood, implemented in the r package MASS (Venables & Ripley 2002), because more than five observations were available for each group (Bolker et al. 2009). Significance was established using the Wald t-test, because it is robust to overdispersion (Bolker et al. 2009).
We generated a first GLMM for open-pollinated plants to test whether herkogamy had a significant effect on total seed set. We then built a full GLMM that allowed us to simultaneously assess the relative effects of herkogamy, caging and emasculation on seed set, with open pollination as the control. The model included herkogamy class, treatment and treatment × herkogamy interactions as fixed effects. We tested whether herkogamy reduced seed set in caged plants, by determining the significance of the caging × herkogamy interaction term. The full model also allowed us to test whether emasculated plants set less seed than open-pollinated plants (emasculation effect) and whether the effect of emasculation differed among herkogamy classes (emasculation × herkogamy interaction). Additionally, we used a series of four GLMMs to test whether seed set differed significantly between control and caged and between control and emasculated plants for each of the four herkogamy classes. Finally, to check whether seeds can be formed apomictically in P. halleri, we used summary statistics to establish whether seed set occurred in plants that were emasculated and caged.
Phenotypic Variation in Sexual Organ Position and Herkogamy
Developmental variation during anthesis
The positions of both anthers and stigmas increased significantly during floral development (floral age: t173 = 14·463, P < 0·001), but at different rates (organ type × floral age interaction: t197 = −8·981, P < 0·001), causing a general and significant decrease of herkogamy throughout anthesis (Fig. 2). While anthers started in a lower position than stigmas (intercepts ± SE of 22·68 ± 0·45 and 25·89 ± 0·49 mm, respectively), the former raised their position faster than the latter (slopes of 0·65 ± 0·04 and 0·41 ± 0·05 mm day−1, respectively; Fig. 2).
Herkogamy decreased in 22 of 25 inflorescences, as indicated by negative Spearman rank correlations, and the decrease was significant in eight cases at α = 0·05 (Fig. 3). In plants 13–25, no flowers reached a degree of herkogamy below 1 mm, while in plants 2–12, the oldest flowers had the stigma positioned among or just above the anthers. Only plant 1 presented one flower with the stigma below the anthers (Fig. 3).
Variation within and between populations
Anther position, stigma position and herkogamy of mature flowers differed significantly between individuals in populations A, B, C (all P < 0·005; Table 1). Between populations, however, anther position did not differ significantly, while stigma position and herkogamy differed significantly between some, but not all population pairs (Fig. 4; Table S1, Supporting Information). The variance of the three floral traits was partitioned mostly among plants within populations (87·2–54·8%) rather than among populations (24·3–<1%; Table 1). Stigma and anther positions were overall positively and significantly correlated with each other (Fig. S2; Table S2, Supporting Information). Overall, herkogamy was negatively correlated with anther position and positively, but not significantly, with stigma position. The correlations did not (anther position and herkogamy) or did (stigma position) significantly differ among populations.
Effects of Herkogamy on Seed Set
Seed set was absent in 49 of 54 flowers from nine plants that were emasculated and caged. The proportion of ovules that developed into seed (seed set ± SD) from the remaining five flowers of four different plants was 0·254 ± 0·317 (range 0·03–0·77). This low seed set is likely explained by emasculation error rather than apomixis, for seed set varied among flowers between inflorescences, while in the case of apomixis, it is expected to be similar across flowers within an inflorescence.
In open-pollinated plants, herkogamy correlated negatively with seed set (P = 0·012; Table 2; Fig. 5). The full GLMM indicated that seed set in caged (selfing only) and emasculated plants (outcrossing only) was significantly lower (P = 0·04 and P = 0·002, respectively) than in open-pollinated plants (selfing plus outcrossing). The interaction terms indicated that seed set decreased significantly with increasing herkogamy in caged plants (P < 0·001), but not in emasculated plants (P = 0·31). The four GLMMs within herkogamy classes indicated that seed set did not significantly differ between control and caged treatments when the stigma was placed between the anthers (herkogamy class 0 mm, P = 0·12; Table 2, Fig. 5). However, seed set was significantly higher in the control than in the caged treatment when herkogamy was >0 mm and the difference increased with herkogamy (effect size from −2·58 to −4·81), suggesting that the proportion of autonomous selfing that might contribute to total seed set in open-pollinated plants decreases with increasing herkogamy. Seed set was significantly higher in open-pollinated than in emasculated flowers of the two lowest herkogamy classes (P = 0·008 and P = 0·009, respectively), but it did not significantly differ between the two treatments in the highest herkogamy class (P = 0·15), implying that, at lower levels of herkogamy, seed in open-pollinated plants is not exclusively produced via outcrossing (Table 2, Fig. 5).
We here document a general significant trend of decreasing herkogamy during anthesis in the homostylous Primula halleri (Figs 1-3), a classic example of the loss of heterostyly in alpine habitats (Scott 1865; Darwin 1877; Schröter 1926; Mast, Kelso & Conti 2006). We also establish that herkogamy varies between individuals and between populations in this species (Figs 3 and 4). Variation in herkogamy among individuals (e.g. Karron et al. 1997; Brunet & Eckert 1998), among populations (e.g. Holtsford & Ellstrand 1992) or, like in P. halleri, at both levels (e.g. Medrano, Herrera & Barrett 2005; Herlihy & Eckert 2007) is also known to occur in other species, including homostylous Turnera ulmifolia (Barrett & Shore 1987) and Amsinckia spectabilis (Johnston & Schoen 1996). A few previous studies reported variation in herkogamy in homostylous primroses (e.g. Primula bellidifolia, Tremayne & Richards 1993; P. verticillata, Al Wadi & Richards 1993; see also Ernst 1962), but reproductive implications were not considered. Here, we discuss the consequences of variation in herkogamy for reproduction of the alpine, homostylous P. halleri.
Decrease in Herkogamy During Anthesis
Although developmental variation in the separation of male and female organs is thought to be relatively common in angiosperms, experimental evidence is limited (reviewed by Fenster & Martén-Rodríguez 2007; Marshall et al. 2010). In P. halleri, the overall significant decrease in herkogamy was supported by the negative (but not always significant) regression slopes between sexual organ distance and floral age in 88% of the investigated inflorescences (Fig. 3). We were able to demonstrate that the faster increase in anther than stigma position during anthesis explains the lower herkogamy of older flowers (Figs 1 and 2). While some studies did not detect any changes of herkogamy during anthesis (e.g. Luijten et al. 1999; Medrano, Herrera & Barrett 2005; Larrinaga et al. 2009), other analyses, specifically on heterostylous and homostylous species, did, especially prior to anthesis (e.g. Stirling 1932; Li & Johnston 2010). Similarly to P. halleri, herkogamy significantly diminished during anthesis in Gentianopsis paludosa from the Qinghai–Tibetan Plateau (c. 3200 m; Duan et al. 2010), a pattern interpreted by the authors as preventing selfing and favouring outcrossing in younger flowers, while enabling self-fertilization in older ones (i.e. delayed selfing; Lloyd 1992). A comparable potential for delayed selfing was found among the large-flowered species of 20 investigated Collinsia and Tonella species (Armbruster et al. 2002).
In P. halleri, smaller separation between sexual organs at later stages of anthesis, coupled with lack of dichogamy (Isham 2010; Fig S1, Supporting Information), might also enable delayed autonomous selfing in older flowers. While herkogamy ranged between 2 and 4 mm in the first 2 days of anthesis (Fig. 2), enough to prevent seed set through autonomous selfing (caged treatment, Fig. 5; Table 2), about half of the analysed plants reached a level of herkogamy ≤1 mm at the end of anthesis (Fig. 3), the threshold at which autonomous selfing becomes possible (caged treatment; Fig. 5; Table 2). These plants thus might experience delayed selfing, proposed to be highly adaptive in alpine environments (Duan et al. 2010) where pollinator services are subject to high stochasticity (e.g. Arroyo, Armesto & Primack 1985; Bergman, Molau & Holmgren 1996; Arroyo et al. 2006).
Conversely, herkogamy was higher than 1 mm at all stages of anthesis in the other half of the plants analysed for developmental variation of this trait (Fig. 3), implying that such individuals rely on pollinators for reproduction throughout the entire period of anthesis. This conclusion is contrary to common expectations, because homostyly is usually interpreted as alleviating dependence on pollinators and mates (e.g. Baker 1966; Kelso 1992; Richards 2003). Interindividual variation of herkogamy also occurred in populations B and C (Table 1, Fig. 4) although their mean herkogamy was lower than in population A (Fig. 4), with possible consequences for the extent of delayed selfing and outcrossing in different populations. Observations on living plants further suggest that other homostylous primroses may also exhibit developmental variation in herkogamy (e.g. Primula eximia, P. incana, P. japonica, P. laurentiana, P. scotica and P. verticillata; see also Chen 2009 on P. cicutariifolia). As these species belong to five sections in three subgenera of Primula (Richards 2003), wider applicability of our results is suggested, but experimental confirmation is needed.
Effects of Herkogamy on Reproductive Assurance
Overall, P. halleri appears to benefit from the reproductive assurance enabled by the ability to self, as the higher seed set of open-pollinated vs. emasculated plants indicates (Table 2, Fig. 5). Similarly, in Collinsia verna, open-pollinated flowers set more seed than emasculated flowers, because autonomous selfing increased total reproductive output, thus providing reproductive assurance (Kalisz, Vogler & Hanley 2004). Conversely, significant effects of emasculation on reproductive output were not always found in other species, suggesting that, in these cases, the contribution of selfing to total reproduction was not considerable (e.g. Eckert & Schaefer 1998; see also Herlihy & Eckert 2002; Eckert, Samis & Dart 2006).
Despite the overall importance of reproductive assurance in P. halleri, interindividual variation of herkogamy in mature flowers strongly influenced the magnitude of this phenomenon (Table 2, Fig. 5). With increasing herkogamy, the difference in seed set between emasculated and open-pollinated plants became smaller, and it was only significant for plants with herkogamy ≤2 mm (Table 2; Fig. 5), suggesting that plants in the highest herkogamy class (3–4 mm) may not experience reproductive assurance. The negative influence of high herkogamy on the amount of reproductive assurance relates to the diminishing effect of herkogamy on autonomous selfing, as median seed set decreased from 0·885 to almost zero in caged plants with herkogamy c. 1 mm (Table 2; Fig. 5). The strong effect of herkogamy on autonomous selfing further suggests that even low amounts of sexual organ separation early in anthesis may promote outcrossing, by keeping ovules available for cross-fertilization, although this benefit of herkogamy may be obscured if facilitated selfing is frequent (Vaughton & Ramsey 2010). To summarize, comparisons of seed set among different treatments in four herkogamy classes indicate that, in the homostylous P. halleri, plants with greater distance between sexual organs in mature flowers are likely to be mainly outcrossed. This interpretation is compatible with the results of genetic studies, which typically found positive correlations between outcrossing rates and herkogamy both within (e.g. Barrett & Shore 1987; Karron et al. 1997; Brunet & Eckert 1998; Herlihy & Eckert 2007) and between populations (e.g. Shore & Barrett 1990; Holtsford & Ellstrand 1992; Belaoussoff & Shore 1995), although the correlation was not always significant (e.g. Narcissus; Medrano, Herrera & Barrett 2005).
The higher total reproductive output associated with the lower herkogamy of mature flowers (Fig. 5; Table 2) may suggest that selection should favour plants with low herkogamy (Figs 3 and 4; Table 1). How then can we explain the great variation of herkogamy detected in P. halleri? Several explanations may be proposed. First, reduced fitness of selfed progeny may offset the advantages of higher total seed set associated with increased selfing in plants with lower herkogamy (Herlihy & Eckert 2002). Therefore, even small amounts of herkogamy may alleviate the potentially negative effects of inbreeding and promote genetic diversity of populations, considered beneficial for the long-term adaptive potential of species (reviewed by Takebayashi & Morrell 2001). Nevertheless, common garden experiments indicated no evidence of high inbreeding depression in P. halleri at the seed-germination stage (Ernst 1951), although exhaustive studies over the entire life cycle are lacking.
Secondly, the effects of variation in herkogamy on the actual mating system may fluctuate between years (Eckert et al. 2009), and the selective pressures on herkogamy may differ over time as a consequence of changes in pollinator conditions (Kulbaba & Worley 2008). Thus, plants that performed poorly in a particular year may outperform others in subsequent years. Moreover, the timing of snow melt in different patches within a populations may create phenological variation in flowering on small spatial scales (pers. obs.), with possible effects on pollinator availability and optimal floral morphology of different plants (Forrest et al. 2011). Hence, selective pressures may not unidirectionally drive towards a decrease in herkogamy.
Thirdly, herkogamy is a compound trait dependent on the positions of both male and female organs, which themselves may convey independent fitness effects, irrespective of their contribution to herkogamy. For instance, the position of anthers alone might affect the dynamics of pollen deposition on a pollinator and thus the patterns of pollen export (Harder & Barrett 1993). The total reproductive fitness associated with a particular herkogamy class thus depends on the reproductive effects of male organ position, female organ position and herkogamy at each stage of anthesis. Following the arguments explained by Johnston et al. (2009), analysis of individual fertility components (the capacities to sire seed through selfing, through pollen export, and through pollen import) may reveal that total reproductive fitness among herkogamy classes may deviate from patterns inferred from seed set alone, because the optimal floral design may differ for each fertility component. Here, it is interesting to note that developmental variation of herkogamy in P. halleri depends on differential rates of change in anther and stigma position (Figs 1 and 2), whereas only stigma position varies significantly between populations (Fig. 4), suggesting that complex selective pressures may act on different components of herkogamy at different hierarchical levels (Herlihy & Eckert 2007).
While Primula has served as the paradigmatic system for the study of homostylous species within a mainly heterostylous genus since Scott's (1865) pioneering work (see also Piper, Charlesworth & Charlesworth 1984; Barrett & Shore 2008; Cohen 2010), our study emphasizes for the first time the key role of variation in herkogamy in the reproductive ecology of a homostylous species. Our results suggest that a small distance between male and female organs early in anthesis can increase outcrossing opportunity, whereas ‘excessive’ herkogamy in older flowers comes at the cost of reducing total reproductive output and imposing pollinator dependence. This study provides new evidence that the reproductive strategies of homostylous species, which are self-compatible and derived from obligate outcrossing, heterostylous relatives, may be more complex than previously anticipated.
We thank the Dienstelle für Wald und Landschaft of Canton Valais, Switzerland, for permits to perform this study (decision of 13-06-2008), the G. & A. Claraz-Schenkung Foundation for funding fieldwork, Bert & Mieke de Vos, George Maentz, Shuqing Xu, and Daniel Gervasi for help in the field, Bert de Vos for the photograph of Fig. 1, and Gerard Oostermeijer, Hein Krammer and Rafael Wüest for valuable discussions. We thank the reviewers and editor for their valuable comments.