2 Present address: Department of Biology, Indiana University, Bloomington, Indiana 47405 U.S.A. E-mail: cherlihy@indiana.edu.


The mating system of flowering plant populations evolves through selection on genetically based phenotypic variation in floral traits. The physical separation of anthers and stigmas within flowers (herkogamy) is expected to be an important target of selection to limit self-fertilization. We investigated the pattern of phenotypic and genetic variation in herkogamy and its effect of self-fertilization in a broad sample of natural populations of Aquilegia canadensis, a species that is highly selfing despite strong inbreeding depression. Within natural populations, plants exhibit substantial phenotypic variation in herkogamy caused primarily by variation in pistil length rather than stamen length. Compared to other floral traits, herkogamy is much more variable and a greater proportion of variation is distributed among rather than within individuals. We tested for a genetic component of this marked phenotypic variation by growing naturally pollinated seed families from five populations in a common greenhouse environment. For three populations, we detected a significant variation in herkogamy among families, and a positive regression between parental herkogamy measured in the field and progeny herkogamy in the greenhouse, suggesting that there is often genetic variation in herkogamy within natural populations. We estimated levels of self-fertilization for groups of flowers that differed in herkogamy and show that, as expected, herkogamy was associated with reduced selfing in 13 of 19 populations. In six of these populations, we performed floral emasculations to show that this decrease in selfing is due to decreased autogamy (within-flower selfing), the mode of selfing that herkogamy should most directly influence. Taken together, these results suggest that increased herkogamy should be selected to reduce the production of low-quality selfed seed. The combination of high selfing and substantial genetic variation for herkogamy in A. canadensis is enigmatic, and reconciling this observation will require a more integrated analysis of how herkogamy influences not only self-fertilization, but also patterns of outcross pollen import and export.

Evolution of the tremendous diversity of mating systems exhibited by flowering plants involves natural selection acting on phenotypic variation in plant traits such as floral morphology, development, and display. Most plant species are hermaphroditic and many engage in a mixture of self-fertilization and outcrossing. However, the evolutionary stability of these “mixed” mating systems has been controversial (Goodwillie et al. 2005). The short-term stability of an observed mix of selfing and outcrossing can be tested by (1) manipulating flowers, inflorescences, or whole plants to determine if an alternative mating pattern yields higher fitness; and (2) quantifying the mating consequences and genetic basis of variation in floral traits that selection might act on to achieve this fitness increase (Schoen and Lloyd 1992; Eckert and Herlihy 2004).

Given that most plant species bear both male and female sex organs within individual flowers, traits that separate pollen presentation and pollen receipt may be important targets of selection on the mating system. Herkogamy, the presentation of dehiscing anthers and receptive stigmas at different positions within a flower, occurs in a large proportion of flowering plant species (Webb and Lloyd 1986). Evolutionary differentiation in the degree of selfing and outcrossing is often associated with marked differences in herkogamy between closely related species (e.g., Ennos 1981; Ritland and Ritland 1989) or among populations within species (e.g., Rick et al. 1977; Schoen 1982; Holtsford and Ellstrand 1992; Belaoussoff and Shore 1995). However, herkogamy may also reduce other forms of interference between male and female functions, and thus largely serve to enhance pollen export rather than reduce self-pollination (Barrett 2002).

In this study, we investigate phenotypic variation in herkogamy, its morphological and genetic bases, and whether it limits self-fertilization within natural populations of columbine, Aquilegia canadensis (Ranunculaceae). This species is one of the best-documented examples of mixed mating. It is a spring-flowering, short-lived perennial found on rocky outcrops throughout eastern North America (Whittemore 1997). Flowers are fully self-compatible and can achieve near full seed set by autonomous self-pollination in the absence of pollinators (Eckert and Schaefer 1998; Routley et al. 1999). Hence plants engage in a broad mixture of selfing and outcrossing (mean proportion of seeds selfed = 0.75, range = 0.15–1.00, n= 38 populations, 63 estimates; Eckert and Herlihy 2004).

Results from a detailed analysis of the costs and benefits of selfing in A. canadensis pose a challenge for our understanding of mating system evolution (Eckert et al. 2006). Selfing provides reproductive assurance in these small patchy populations by increasing total seed production, but it also reduces outcrossed seed production (seed discounting, Herlihy and Eckert 2002). This is costly because selfed seeds suffer inbreeding depression and appear to survive to reproductive maturity one-tenth as frequently as outcrossed seed (Routley et al. 1999; Herlihy and Eckert 2002, 2005). Therefore, self-fertilization appears maladaptive in A. canadensis (Eckert and Herlihy 2004). Paradoxically, our casual observations suggest substantial phenotypic variation in herkogamy within natural populations. Such a variation, if it influenced the mating system and had a genetic basis, could be acted on by selection to reduce selfing.

Few studies have quantified phenotypic variation in herkogamy within natural populations (Medrano et al. 2005) and how much of this variation is partitioned among individuals: a prerequisite for natural selection on herkogamy (Luijten et al. 1999; Lennartsson et al. 2000; Takebayashi et al. 2006). Studies on a several species have detected significant heritable variation in herkogamy among plants taken from natural populations and grown in a uniform environment (Breese 1959; Ennos 1981; Shore and Barrett 1990; Holtsford and Ellstrand 1992; Carr and Fenster 1994; Robertson et al. 1994; Chang and Rausher 1998; Lennartsson et al. 2000; Motten and Stone 2000; Lendvai and Levin 2003). However, herkogamy can vary strongly in response to environmental factors (e.g., Motten and Stone 2000; Elle and Hare 2002). Few studies have shown that a variation in herkogamy as expressed within natural populations is heritable (van Kleunen and Ritland 2004), or tested whether heritability is consistent across populations (Ritland and Ritland 1996).

The expected negative relation between selfing and herkogamy has been found for some (Barrett and Shore 1987; Brunet and Eckert 1998; Motten and Stone 2000; van Kleunen and Ritland 2004; Takebayashi et al. 2006) but not all species (Eckert and Barrett 1994; Baker et al. 2000; Medrano et al. 2005). A likely explanation for these mixed results is that herkogamy is only expected to limit autogamous (within-flower) self-pollination. Hence it will have little effect of the mating system if most selfing occurs via self-pollination between flowers in large floral displays. For example, in highly clonal, mass-flowering Decodon verticillatus,Eckert and Barrett (1994) did not find differences in self-fertilization among style morphs differing in herkogamy, largely because self-fertilization occurs almost entirely through geitonogamy in this species (Eckert 2000). Very few studies have determined how selfing occurs in natural populations (reviewed in Herlihy and Eckert 2004), and an effect of herkogamy on autogamous selfing, specifically, has not been demonstrated.

Finally, herkogamy is a composite trait, depending on the relative position of dehiscing anthers and receptive stigmas, which depends on the relative lengths of stamens and styles, respectively. The morphological basis of variation in herkogamy may determine how it influences other components of the mating system by determining how pollinators contact anthers and stigmas. For instance, variation in herkogamy due to anther position might affect pollen export (e.g., Murcia 1990; Harder and Barrett 1993; Kudo 2003) and outcross siring success. In contrast, variation in herkogamy due to stigma position may affect outcross pollen deposition (e.g., Eckert and Barrett 1995; Barrett 2003) and, consequently, the number of seeds outcrossed. As a result, phenotypic variation in herkogamy may be subject to selection via its influence on several components of reproductive success. Yet, the morphological basis of variation in herkogamy remains unexplored.

The larger goal of this study is to better understand the maintenance of high levels of self-fertilization and high inbreeding depression in A. canadensis, when herkogamy could potentially limit this costly self-fertilization. To this end, we study a broad sample of natural populations to quantify patterns of phenotypic variation in herkogamy in comparison to other floral traits; and examine the covariation between herkogamy, pistil length, and stamen length to determine the morphological basis for herkogamy. We test whether phenotypic variation in herkogamy has a heritable genetic basis by estimating the correlation between herkogamy of plants expressed under natural conditions and that of their progeny grown in a common greenhouse environment. Finally, we test whether herkogamy influences self-fertilization in natural populations by comparing total selfing and the autogamous component of selfing between groups of plants differing in herkogamy.

Materials and Methods

We studied 10 populations of A. canadensis located in eastern Ontario, Canada in the northern portion of the species' geographical range, and eight populations in the southern Appalachian region at the center of the species' range in Virginia and North Carolina, U.S. (Appendix 1). These populations spanned the range of population size, plant density, and habitats typical of this species, from large, dense patches of up to 2000 individuals on exposed rocky outcrops to small, scattered clusters of a few dozen individuals under a forest canopy.


Aquilegia canadensis produces large, red, nodding flowers consisting of four to six petals modified into ∼30 mm nectar spurs that produce copious, dilute nectar, the same number of ∼20 mm long sepals that flare out to form a skirt-like structure around the opening of the spurs, and four to six septate carpels (each with about 20 ovules) surrounded by a column of 30–40 stamens that elongate and dehisce in small groups over two to four days. Stigmas are exserted from the corolla ∼3 days before anthers dehisce; but do not become receptive until anther dehiscence (Griffin et al. 2000). Flowers senesce a few days after the last anther dehisces. Although the flowers appear adapted for pollination by hummingbirds, flower visitors consist of several species of bumblebees (Bombus spp.) along with the ruby-throated hummingbird (Archilochus colubris), and both types of visitors are very infrequent in populations we studied.

To examine the phenotypic variation in floral morphology, a single observer measured (to 0.1 mm) herkogamy as the minimum distance between stigmas and anthers midway through anthesis, maximum pistil length from the receptacle to the stigma, and the length of one nectar spur from the nectar gland to the tip of the lamella on one flower per plant. Stamens could not be measured without destroying flowers (see below). On a subset of 71 plants, we measured each flower twice and assessed measurement error as the intraclass correlation coefficient (rI). All floral measurements were highly repeatable for this (herkogamy rI= 0.989; pistil length rI= 0.978; spur length rI= 0.991; all P < 0.0001) and all other analyses reported below. We used spur length as a general index of floral size, because it correlates positively with the size of other floral organs (e.g., sepals and pistils, Herlihy and Eckert 2005). We sampled ∼65 and ∼80 plants per northern population in 1999 and 2000, respectively, and ∼50 plants per central population in 2000. We collected fruits produced by all measured flowers for genetic analysis (see below).

To contrast the amount of phenotypic variation between different floral traits, we used a modification of Levene's median ratio test (Schultz 1985) in which individual values (i) for a floral measurement are expressed as absolute relative deviations from the population median


We compared the population mean deviations (MdRi) of herkogamy to those for pistil length and spur length using paired t-tests. We compared individual MdRi values between floral traits within each population using nonparametric Wilcoxon tests (because sample variances in individual MdRi values were not homogeneous).

In five northern populations studied in 2000, a single observer measured all flowers on each plant to partition variation in herkogamy into its within-plant and among-plant components. In each population, we measured 2–16 flowers (mean = 3.6) on each of 16–71 plants (mean = 51), and used restricted maximum likelihood (REML) to calculate within- and among-plant variance components.

In 2001, we investigated the morphological basis of herkogamy by examining the relative lengths of pistils and stamens in relation to herkogamy within the same five northern populations. The base of each stamen is deep within flowers, so that their length cannot be accurately measured on intact flowers. Consequently, we randomly selected a single flower on each of 30 plants per population (25 in QLL3), and performed replicate measurements of spur length, pistil length, and herkogamy in the field, as above. Flowers were then dissected and digitally photographed (640 × 480 pixels, at about 5× magnification). Using ImageJ software (http://rsb.info.nih.gov/ij), we measured (to 0.001 mm) herkogamy in intact flowers as well as the lengths of one spur, one pistil, and one stamen, which were dissected out and pressed between microscope slides. A single image was taken of each organ. This image was then measured twice. Measurements of different floral traits on the same flower, and replicate measurements of the same image were not made consecutively. Floral measurements made on the same set of digital images were highly repeatable (herkogamy rI= 0.997; spur length rI= 0.993; pistil length rI= 0.996; stamen length rI= 0.978; all P < 0.0001), and measurements of the same floral trait in the field and from digital images correlated strongly (mean r=+0.86). Measurements used in the analysis below are means of the two replicate measurements. We analyzed correlations between herkogamy and floral organs using ANCOVA, with herkogamy as the response variable, population as a categorical effect, and another floral trait (pistil length, stamen length, or spur length) as the covariate. This enabled us to estimate the overall correlation between floral traits and test for heterogeneity in correlations among populations by testing population × covariate interactions. Data for all floral traits were transformed as Y′= log10(Y+ 1) to eliminate associations between residuals and predicted values from the ANCOVA models.


We examined the genetic basis for herkogamy in three northern (QFP1, QLL3, and QOR1) and two central (VACBW1 and VANPT1) populations, using randomly sampled maternal seed families. Seeds were stored at 4°C until planting in fall 2001. We planted 20 seeds per maternal family from ∼50 families per population sown in randomized locations in 30 mm plug trays containing Sunshine Mix #2 and kept at 20–25°C with 16 h light/day. Seedlings were transplanted into randomly positioned 55 mm pots seven weeks after germination. Eight weeks after transplanting, when foliage began to senesce, the aboveground growth was cut back and the plants were moved to a dark, 4°C cold room, because A. canadensis must experience an overwintering period before flowering. Plants were returned to the greenhouse six months later, and began flowering almost immediately. Two observers measured herkogamy, pistil length, and nectar spur length, as above, on one flower per plant. There was no tendency for plants within the same family to be measured by the same observer hence any interobserver differences would simply cause random error in the analyses below. Of the 4840 seeds sown, 2515 germinated and survived to the second year, of which 1760 flowered.

We investigated whether phenotypic variation in herkogamy and other floral traits had a heritable genetic basis by testing for (1) floral variation among maternal families grown in this common greenhouse environment; and (2) associations between floral traits of progeny measured in the greenhouse and maternal plants measured in natural populations. Accurate estimates of heritability using maternal families derived from natural pollination are complicated by three factors. First, each family probably consists of a mixture of selfed and outcrossed progeny, and the outcrossed progeny could be full- or half-siblings. The proportion of seeds self-fertilized in these populations was previously estimated as 0.76 in QFP1, 0.63 in QLL3, 0.51 in QOR1, 0.80 in VACBW1, and 0.71 in VANPT1 (Herlihy and Eckert 2005). In the analyses below, we treat the progeny within families as full-siblings as this yields the most conservative estimate of heritability. Second, genetic and nongenetic maternal effects may increase the resemblance among relatives, causing heritability to be overestimated (Roach and Wulff 1987). Third, heritability will be overestimated if maternal parents vary in the proportion of progeny produced through self-fertilization and inbreeding depression affects the expression of the trait(s) under study (Kelly 1999). However, supplementary analyses (not shown) failed to detect any association between parental herkogamy and progeny performance in the relatively benign greenhouse environment.

We partitioned floral variation expressed in the fully randomized greenhouse experiment into its among-family (Vf) and within-family (Vw) components using REML (Lynch and Walsh 1998), and calculated broad-sense heritability (following Falconer and Mackay 1996) as: H2= 2Vf/(Vf+Vw) and the coefficient of genetic variation as: CVG= 100(sqrt[2Vf]/M), where M is the population mean of the trait in question. CVG is particularly useful for comparing the genetic potential for the evolution between traits of different size. We calculated genetic correlations (rG) as the Pearson correlations between traits using family means estimated with REML. Statistical significance of rG and differences in rG between traits and populations was evaluated using Fisher's Z-test (Sokal and Rolhf 1995).

We estimated the slope (m) of family means for floral traits of progeny (measured in the greenhouse) regressed over the trait values of the maternal plants (measured in natural populations). Heritability is typically calculated as m when progeny means are regressed over the mean of both parents and 2m when the regression involves only one parent (Falconer and Mackay 1996). Because a large proportion of these progenies were produced by self-fertilization (i.e., the mother is also the father), we used m as an estimate of heritability. We tested for heterogeneity in parent–offspring regressions among populations by evaluating the interaction term in an analysis of covariance (ANCOVA) with population as a fixed effect and parental trait values as a covariate. We calculated standardized coefficients of regression (β) between parents and progeny to facilitate comparison among traits (as above). The genetic correlation (rG) between traits (xi and xj) was estimated following Falconer and Mackay (1996):


where COV is covariance and subscripts O and P denote offspring and parents, respectively. Significance of rG was assessed using the Z-test. All analyses were performed using JMP (ver. 6, SAS Institute Inc., Cary, North Carolina). All means are presented ±1 SE unless otherwise indicated.


We assessed the effect of herkogamy on self-fertilization within populations by dividing randomly sampled plants into two groups with respect to herkogamy (high and low herkogamy, Appendix 2), and jointly estimating the proportion of seeds self-fertilized (s) for each group. For six Ontario populations studied in 1999, we also examined the effect of herkogamy on autogamous self-fertilization specifically by partitioning self-fertilization into components due to autogamy versus geitonogamy. The proportion of seed produced via autogamy (a) was determined by comparing self-fertilization (s) between intact (I) flowers capable of autogamy and flowers that were emasculated (E) before anther dehiscence to remove the capacity for autogamy (Schoen and Lloyd 1992). We have previously shown that emasculation does not damage flowers, (Eckert and Schaefer 1998; Griffin et al. 2000) reduce floral longevity (Griffin et al. 2000), or appear to cause any substantial change in pollinator behavior (C. R. Herlihy and C. G. Eckert, unpubl. data). Emasculation eliminates autogamy, while allowing outcrossing (t) and nonautogamous inbreeding (g*), which includes both geitonogamous self-fertilization and biparental inbreeding. Estimates of self-fertilization in emasculated flowers do not include a, thus sE= g*E. However, g*E > g*I, because all types of nonautogamous pollen will experience higher absolute siring success in the absence of competition from autogamous pollen. However, the ratio of g* to t pollen deposited on stigmas will likely be the same in both intact and emasculated flowers (Schoen and Lloyd 1992), so that


Thus, a in I flowers can be estimated as: aI=sIg*I. Consequently, to estimate autogamy in the high and low herkogamy groups, self-fertilization in intact flowers from each group (sI-high and sI-low) was compared to self-fertilization in all emasculated flowers (sE). This is justified as herkogamy does not influence selfing in emasculated flowers, and the plants selected for emasculation were a random sample with respect to herkogamy.

Self-fertilization (s) was estimated by assaying ∼9 seeds per fruit (hereafter, maternal family) for two polymorphic allozyme loci, isocitrate dehydrogenase (IDH, EC, and peroxidase (PER, EC, following the electrophoretic procedures in Routley et al. (1999). In each population, we assayed ∼17 maternal families from intact flowers in each of the high and low herkogamy groups per population (ranges: high: 11–32, low: 9–38) and ∼40 (range: 29–57) families from emasculated flowers (Appendix 2). Progeny allele frequencies (p) were within the range useful for estimating mating system parameters (i.e., 0.1 < p < 0.9). Multilocus estimates of s were obtained using the maximum-likelihood program MLTR (Ritland 1990) with Newton–Raphson iteration. For all analyses, iterations always converged on only one maximum-likelihood estimate. Standard errors for each estimate were calculated as the standard deviation of 1000 replicate bootstrap estimates, with the progeny array as the unit of resampling.

Differences in estimated self-fertilization (s) or autogamy (a) between high and low herkogamy groups were evaluated statistically by calculating the proportional overlap between the bootstrap distributions of high and low herkogamy groups, which is roughly equivalent to a P-value. Two estimates were considered significantly different if this overlap (P) was less than 5% for a one-tailed test of the directional hypothesis that s and a will be lower in high than low herkogamy groups. The overall effect of herkogamy across populations was assessed in two ways. First, we averaged randomly paired bootstrap estimates across populations, and compared the overlap of averaged bootstrap distributions for high and low herkogamy groups, as above. Second, we evaluated consistency in the direction of the effect of herkogamy on self-fertilization among populations using a one-tailed Wilcoxon signed-rank test. For two populations in which we estimated s in two years, differences in the effect of herkogamy between years in each population were evaluated as the proportional overlap of bootstrap distributions of the difference in s (sLowsHigh) from each year, and were considered statistically significant if the overlap of bootstrap distributions was less than 2.5% (two-tailed test).



Total phenotypic variation was much greater for herkogamy than other floral traits in all populations. Population mean median ratios (Fig. 1) were much greater for herkogamy (mean of population mean MdRi± 1 SE = 0.518 ± 0.014) than pistil (0.090 ± 0.002; paired t-test, t= 12.1, df = 18, P < 0.0001) or spur length (0.068 ± 0.002, t= 12.9, df = 18, P < 0.0001). Within all populations, median ratios for herkogamy greatly exceeded those for either pistil or spur length (Wilcoxon signed-rank test, all P < 0.0001). Pistil length exhibited more relative variability than spur length across populations (Fig. 1; paired t= 6.8, df = 18, P < 0.0001). However, within populations, the differences in median ratios between pistil and spur length were not significant.

Figure 1.

Comparison of relative phenotypic variation for herkogamy versus pistil length and spur length in natural populations of Aquilegia canadensis. Bars are population mean median ratios (±1 SE).

The proportion of phenotypic variation partitioned among individual plants (Table 1) was much greater for herkogamy (mean = 72%) than pistil length (45%; paired t-test, t= 6.3, df = 4, P= 0.003) or spur length (41%, t= 7.7, df = 4, P= 0.002). Among-plant variation did not differ between pistil and spur length (t= 0.5, P= 0.66).

Table 1.  Among-plant variation in herkogamy and other floral traits within five populations of Aquilegia canadensis studied in 2000. The mean (based on plant means), the coefficient of variation (CV) calculated among all flowers measured, and the% of total variation partitioned among individuals calculated using restricted maximum likelihood (REML) are presented for each floral trait. The number of plants sampled per population (with the average number of flowers measured per plant in brackets) is also presented.
PopulationPlants (flw/plant)HerkogamyPistil lengthSpur length
Mean (mm)Total CV (%)% variation among plantsMean (mm)Total CV (%)% variation among plantsMean (mm)Total CV (%)% variation among plants
QCA240 (3.0)2.37780.222.0 954.529.1 844.1
QFP169 (4.0)3.05169.421.91029.528.31043.8
QLL371 (3.2)4.86763.923.91042.029.01045.4
QMT116 (2.2)2.33977.521.01262.028.1 836.5
QOR158 (4.5)2.85767.922.11035.428.7 937.6
Mean51 (3.4)3.05871.822.21044.728.6 941.5


Herkogamy of individual flowers covaried strongly and positively with pistil length but not with stamen length (Fig. 2; Z-test for difference between two correlations: P < 0.0001). Both stamen and pistil length correlated positively with spur length (pistil: r=+0.44, P < 0.0001; stamen: r=+0.47, P < 0.0001) and with each other (r=+0.47, P < 0.0001), indicating that larger flowers possessed longer pistils and longer stamens. Although spur length correlated positively with herkogamy (r=+0.22, P= 0.008), this correlation was weaker than correlations between spur length and either pistil or stamen length (Z-test: both P < 0.038). When positive covariation with pistil length was controlled, herkogamy correlated negatively with stamen length (Fig. 2). Although there was modest heterogeneity among populations for all floral traits (among population r2= 0.10–0.17, all P < 0.005), none of the correlations between floral traits varied among populations (population × covariate interaction: all P > 0.10).

Figure 2.

Correlation of herkogamy with pistil and stamen lengths, and the association between herkogamy and stamen length with covariation between pistil length and herkogamy controlled. N= 145 flowers from five populations of Aquilegia canadensis. Points are residuals from ANOVA with the differences in all traits among populations removed.


Maximum-likelihood estimates of broad-sense heritability (H2) based on the variance in floral traits among families grown in a common greenhouse environment (Table 2) were moderate-to-high for herkogamy in all populations (mean = 0.42) and statistically significant for three of five populations. Across populations, estimates of H2 for herkogamy were usually higher, although not significant, than H2 for pistil length (mean = 0.21, paired t-test P= 0.085) or spur length (mean = 0.22, P= 0.23). Moreover, the differences in H2 among traits were not significant within any population (Table 2). Coefficients of genetic variation (CVG, Table 2) were consistently higher for herkogamy (mean = 31.8%) than either pistil length (6.5%, paired t-test P= 0.0092) or spur length (7.5%, P= 0.046).

Table 2.  Patterns of phenotypic variation and broad-sense heritabilities for herkogamy and other floral traits in five populations of Aquilegia canadensis. Coefficients of variation were calculated among maternal plants in the field (CVM), and among family means in the greenhouse (CVF). nfam is the number of maternal plants and families sampled. Broad-sense heritabilities (H2) and coefficients of genetic variation (CVG) were calculated from restricted maximum-likelihood (REML) estimates of the variance in floral traits among maternal families grown in a common greenhouse environment. 95% confidence intervals (CI) for each estimate of H2 are provided in brackets. The CVG for spur length could not be calculated (“na”) for population VACBW1 because the REML estimate of the among-family variance component was negative
PopulationnfamHerkogamyPistil lengthSpur length
QFP14866430.78 (0.36–1.19)5210 80.31 (0.06–0.57)8 7 7 0.32 (0.09–0.56) 8
QLL36478340.38 (0.15–0.61)3410 80.26 (0.06–0.45)711 8 0.25 (0.06–0.43) 7
QOR15276330.21 (−0.01–0.42)2610 60.04 (–0.10–0.18)3 8 8 0.16 (–0.04–0.36) 5
VACBW13550280.54 (0.05–1.03)2911110.20 (−0.16–0.56)7 8 9−0.04 (−0.32–0.25)na
VANPT13540240.20 (−0.06–0.46)1811 80.26 (−0.01–0.52)7 610 0.41 (0.06–0.77)10

For the two populations in which H2 was statistically significant for both herkogamy and pistil length, the genetic correlation (rG) between these traits was positive and significant (QFP1 rG=+0.55, P < 0.0001; QLL3 rG=+0.72, P < 0.0001). The genetic correlation between pistil length and spur length was also positive and significant in both these populations (rG=+0.67 and +0.61, respectively, both P < 0.0001) and neither of these correlations was different from rG between herkogamy and pistil length (Z-test: P= 0.33 and 0.27, respectively). However, the genetic correlation between herkogamy and spur length was much weaker (+0.11, P= 0.43 and +0.34, P= 0.006, respectively) and significantly lower than rG for herkogamy and pistil length (Z-test: P= 0.016 and 0.0022, respectively).

We detected significant positive regressions between herkogamy measured on maternal parents in the field and the mean herkogamy of their progeny grown in a common greenhouse environment for the three northern but not the two central populations (Fig. 3, Table 3). There was significant heterogeneity in the parent–offspring regressions among populations (ANCOVA: population × parental herkogamy interaction: F4,223= 4.71, P= 0.001). Much of this involved a marked difference between northern (mean m= 0.29) and central populations (mean m= 0.04). However, there was also heterogeneity in regressions among northern (F2,158= 4.15, P= 0.02) populations. Regressions slopes did not differ between central populations (F165= 0.05, P= 0.83).

Figure 3.

Parent–offspring regression for herkogamy in five populations of Aquilegia canadensis. Herkogamy of maternal parents was measured in the field, whereas offspring values are the mean herkogamy of open-pollinated progeny grown in a common greenhouse environment (mean number of progeny per maternal family = 7, range = 1–18). The slope of the regression (m) and the standardized regression coefficient (β) are presented above the panel for each population. The significance of each regression is indicated by superscripts on β: nsP > 0.1, *P < 0.05. **P < 0.01, ***P < 0.001.

Table 3.  Resemblance of floral traits between maternal parents naturally occurring in the field and their open-pollination progeny grown in a common greenhouse environment. Slopes of parent–offspring regressions plus their 95% confidence intervals along with standardized regression coefficients (β) are presented for each trait and population. Statistical significance indicated as *P < 0.05, **P < 0.01, ***P < 0.001, otherwise P > 0.10. Regressions for herkogamy are in Figure 3.
PopulationHerkogamyPistil lengthSpur length
  Slope (m)+0.47+0.30+0.16
  95% CI of m 0.31–0.64 0.07–0.49−0.11–0.43
  Standardized β+0.64***+0.37**+0.17
  Slope (m)+0.16+0.34+0.05
  95% CI of m 0.004–0.32 0.16–0.52−0.10–0.20
  Standardized β+0.25*+0.44***+0.08
  Slope (m)+0.23+0.08+0.14
  95% CI of m 0.08–0.39−0.07–0.24−0.08–0.36
  Standardized β+0.40**+0.14+0.17
  Slope (m)+0.03+0.14+0.15
  95% CI of m−0.14–0.20−0.15–0.43−0.19–0.50
  Standardized β+0.07+0.16+0.15
  Slope (m)+0.06+0.25+0.08
  95% CI of m−0.12–0.25 0.02–0.48−0.36–0.52
  Standardized β+0.11+0.36*+0.06

There were significant parent–offspring regressions of pistil length for two northern and one central population (Table 3), but no among-population heterogeneity in the parent–offspring regression (F4,224= 1.11, P= 0.35; overall β=+0.37, P < 0.0001). Regressions of spur length were not significant for any population (Table 3). The overall regression was not quite significant (overall β=+0.12, P= 0.08) and did not differ among populations (F4,223= 0.20, P= 0.94). For the two populations with significant parent–offspring regressions for both herkogamy and pistil length, genetic correlations between these traits were relatively strong and significant (QFP1 rG=+0.68, P < 0.0001; QLL3 rG=+1.02, P < 0.0001).

There was moderate concordance between measures of genetic variation based on variance among progenies (H2 and CVG) and the parent–offspring regression (m) for herkogamy (correlation between measures among five population: mean r=+0.70) and pistil length (+0.81) but not for spur length (−0.37).


Estimates of self-fertilization were lower for high herkogamy than low herkogamy flowers in 13 of 19 populations (Fig. 4; one-tailed Wilcoxon signed-rank test P= 0.013). Averaged across all populations, the proportion of seeds self-fertilized (±1 SE) was 16% lower for flowers with high (0.66 ± 0.03) than low (0.79 ± 0.02) herkogamy (P < 0.0001), and the difference between groups was significant within four individual populations. The difference in self-fertilization between low and high herkogamy flowers varied substantially among populations (Fig. 4) but the magnitude of the difference in self-fertilization did not correlate with the magnitude of the difference in herkogamy between groups (r=+0.09, n= 19, P= 0.72; data in Appendix 2).

Figure 4.

Comparison of the proportion of seeds self-fertilized between groups of Aquilegia canadensis flowers with high versus low herkogamy in seven Ontario populations studied in 1999 (n= 177 families/1893 progeny assayed), four Ontario populations studied in 2000 (n= 216/2026), and eight central populations studied in 2000 (n= 268/1680). Points are group maximum likelihood estimates ±1 SE derived from 1000 bootstraps. Statistical differences between groups are indicated as †P < 0.1, *P < 0.05, **P < 0.01.

In two populations (QOR1 and QFP1), we examined the effect of herkogamy on self-fertilization (s) in two consecutive years (1999 and 2000; Appendix 2, Fig. 4). The difference in herkogamy between high and low groups did not differ between years in either QOR1 (3.6 mm in 1999, 3.2 mm in 2000, 2-way ANOVA year × group interaction F1,72= 0.77, P= 0.38) or QFP1 (4.3 mm in 1999, 3.3 mm in 2000; F1,67= 2.6, P= 0.11). In QFP1, high herkogamy flowers selfed 38% less than low herkogamy flowers in 1999 and 30% less in 2000 (absolute difference in [slowshigh] between years ±1 SE = 0.048 ± 0.208; P= 0.56). However, in QOR1 high herkogamy flowers selfed 29% less in 2000, but 60% more in 1999 (absolute difference = 0.482 ± 0.213, P= 0.009).

High herkogamy flowers had reduced autogamous selfing (a) in all but one of the six populations in which we estimated a using floral emasculation (Fig. 5) Across populations, mean a (± 1 SE) was significantly lower for high (0.41 ± 0.12) than low (0.61 ± 0.07) herkogamy flowers (P= 0.037), and the difference between groups was significant within two populations. Again, the difference in a did not correlate with the difference in herkogamy between groups across populations (r=+0.28, n= 6, P= 0.58). Differences between groups were more pronounced for a than total s (mean 32% reduction in a vs. 17% reduction in s in high vs. low herkogamy flowers).

Figure 5.

Comparison of the proportion of seeds produced via autogamous self-fertilization between groups of flowers with high versus low herkogamy in six Ontario populations of Aquilegia canadensis studied in 1999 (n= 391 families and 4010 progeny). Points are group maximum likelihood estimates ±1 SE derived from 1000 bootstraps. Statistical differences between groups are indicated as †P < 0.1, *P < 0.05.



Natural populations of A. canadensis exhibited striking phenotypic variation in herkogamy, and most of this within-population variation occurred among individuals, rather than among flowers on individual plants. This is consistent with the few other studies that have quantified phenotypic variation in herkogamy among individuals within populations (e.g., Ennos 1981; Luijten et al. 1999; Lennartsson et al. 2000). For example, Luijten et al. (1999) found substantial variation in herkogamy both among populations, and among plants within populations, but no significant variation among flowers within plants of Gentianella germanica. We further show that, within the A. canadensis populations studied here, herkogamy was much more variable than all major floral organs, including stamens, pistils, nectar spurs, and sepals (data not shown) and that the among-plant component of variation was greater for herkogamy than other floral traits. The among-plant partitioning of variation is significant, as evolution of the mating system via selection depends on consistent differences in floral phenotype among individuals.

Estimates of broad-sense heritability based on the variation among families derived from natural pollination grown in a common greenhouse environment as well as the regression between parents measured in the field and progeny measured in the greenhouse strongly suggest a heritable genetic basis to the wide phenotypic variation in herkogamy within some natural populations of A. canadensis. A significant genetic component of phenotypic variation in herkogamy has consistently been found for the relatively few species studied to date, although most estimates are based on plants grown in uniform artificial environments (but see Ritland and Ritland 1996; van Kleunen and Ritland 2004). For example, parent–offspring analysis of herkogamy revealed H2 of 0.50 for Ipomoea purpurea (Ennos 1981), 0.65 for Mimulus guttatus (Carr and Fenster 1994), and 0.85 for Gentianella campestris (Lennartsson et al. 2000). Controlled breeding experiments yielded narrow-sense heritabilities (h2) of 0.60 for Ipomopsis aggregata (Campbell 1996), 0.50 for M. guttatus (Carr and Fenster 1994), and 0.30 for Datura stramonium (Motten and Stone 2000). Again, our analyses further suggest more genetic variation for herkogamy than other floral traits, but this interpretation depends on which measure of genetic variation is considered. Estimates of CVG were consistently higher for herkogamy than either pistil or spur length. However parent–offspring regressions, although stronger for herkogamy than spur length, did not differ consistently between herkogamy and pistil length. Estimates of broad-sense H2 based on variance among maternal families were usually higher for herkogamy than other traits, although individual comparisons were not significant.

Our results also suggest that the amount of genetic variation for herkogamy varies among populations of A. canadensis. In particular, parent–offspring regressions were moderate and significant for three northern populations but very weak for two central populations. Weaker regressions for central populations appear to be due to lower covariance between parents and offspring trait values (mean covariance: central = 0.00237, north = 0.01046) rather than lower trait variation among either parents (mean variance: central = 0.0577, north = 0.0346) or their progeny (central = 0.0141, north = 0.0153). It is notable that the mean herkogamy of plants from central populations (mean = 53 mm) is much higher than that of northern populations (25 mm). Herlihy and Eckert (2005) found the same trend in a broader comparison of 10 central versus 10 northern populations. Although this might suggest that more constant selection for increased herkogamy has reduced genetic variation for this trait, there is no difference in self-fertilization or inbreeding depression between central and northern populations (Herlihy and Eckert 2005), and herkogamy appears to reduce selfing to the same extent in both regions. This suggests that the strength of selection on herkogamy does not differ between regions (see below). Currently, we have no likely explanation for why the genetic component of phenotypic variation in herkogamy differs between regions.

There are two significant caveats associated with our analysis of genetic variation in herkogamy. First, we used of open-pollinated families made up of a mixture of selfed progeny and outcrossed progeny that probably varied in relatedness. This creates uncertainty around our estimates of heritability, although we have tried to account for this in our choice of estimators. Second, variation among progeny derived from open-pollinated flowers in natural populations may be influenced by nongenetic maternal effects, which might have contributed to the among-family variance components as well as the resemblance between parents and offspring (Roach and Wulff 1987). However, maternal effects are thought to result primarily from differences among parents in seed provisioning and should most strongly influence traits expressed early in development, rather than size-independent aspects of floral morphology like herkogamy. For instance, experimental manipulation of maternal resource status, which had a significant effect on seed mass, did not affect herkogamy (Kliber and Eckert 2004). Moreover, floral traits, including herkogamy, do not covary with habitat variables like canopy cover or low vegetation cover that consistently affect plant size (Herlihy and Eckert 2004). Finally, formal genetic analysis failed to reveal significant maternal effects for seedling traits in closely related Aquilegia caerulea (Montalvo and Shaw 1994). Our results, taken together, suggest that populations of A. canadensis, particularly in the northern portion of the species' geographic range, contain genetic variation for herkogamy, and thus scope for the evolution of floral morphology and the mating system.


Our results show that flowers with more pronounced herkogamy usually experience less self-fertilization in natural populations of A. canadensis. Results of similar analyses conducted in natural populations of other species are mixed (compare Barrett and Shore 1987; Brunet and Eckert 1998; Motten and Stone 2000; vs. Eckert and Barrett 1994; Baker et al. 2000; Medrano et al. 2005). It is possible that this discord among previous studies arises because the functioning of herkogamy, like most floral traits, depends on the pollination context (Motten and Stone 2000; Medrano et al. 2005; Takebayashi et al. 2006). We found that the effect of herkogamy on selfing varied among the 19 populations we studied and sometimes between years within populations. Perhaps some of this variation is due to spatial and temporal differences in outcross pollination (Herlihy and Eckert 2004). For instance, herkogamy may have relatively little effect of the proportion of seed self-fertilized when the rate at which outcross pollen is deposited on stigmas is particularly high (self pollen is always swamped by outcross pollen) or particularly low (self pollen is all there is). Given that all of the 19 populations we studied exhibited an average proportion of seeds selfed > 50%, the first condition is unlikely to occur in A. canadensis. There is, however, weak support for a declining effect of herkogamy with increasing average levels of self-pollination among the population we studied. Both the absolute difference in selfing between high and low herkogamy plants and the difference expressed as a proportion of the maximum possible effect (i.e., the level of selfing in low herkogamy flowers) correlated negatively with average level of selfing in the population (both r=−0.27), although neither correlation was significant (both 1-tailed P= 0.12). In addition, the average level of selfing was higher in populations in which herkogamy did not reduce selfing (mean ± 1 SE = 0.78 ± 0.05) compared to those in which herkogamy did reduce selfing (0.69 ± 0.03), although again the difference is not quite significant (t-test: 1-tailed P= 0.051). The context-dependent functioning of herkogamy could be tested more directly by experimentally manipulating outcross pollination within natural populations.

Whether herkogamy reduces self-fertilization depends strongly on how and when self-pollination occurs (Webb and Lloyd 1986), which has rarely been quantified in natural plant populations. In general, herkogamy may be effective at reducing self-fertilization through autogamous self-pollination, especially if this occurs during the same phase of floral development as outcrossing. Herkogamy will have little effect of self-fertilization occurring via geitonogamy or autogamous selfing in the bud before flowers open or during floral senescence after opportunities for outcrossing have passed (e.g., Dole 1992). Yet, no previous study has examined the effect of herkogamy on autogamy in particular. Our experimental emasculation of high versus low herkogamy flowers revealed that herkogamy effectively reduces this component of selfing, and that the effect of herkogamy on autogamy was greater than the effect on total selfing. Moreover, autogamous self-pollen strongly competes for fertilizations with outcross pollen in A. canadensis because both pollen types appear to arrive on stigmas at the same time (Griffin et al. 2000; Herlihy and Eckert 2002). Our results also mirror a previous analysis showing herkogamy correlates negatively among populations with autogamous selfing but not total selfing (Herlihy and Eckert 2004). This emphasizes the importance of investigating associations between floral traits and the specific components of the mating system they are predicted to effect (Schoen et al. 1996).


The larger goal of this study is to better understand how high self-fertilization is maintained in A. canadensis when inbreeding depression appears to be very strong, and selfing reduces opportunities for outcross seed production (Eckert and Herlihy 2004). Our results add to the enigma of self-fertilization in A. canadensis. Why has selection not acted on what appears to be genetic variation in herkogamy to reduce selfing, and in so doing create populations that exhibit more pronounced and uniform herkogamy?

Selection for enhanced herkogamy might be complicated by the development of genetic associations (identity disequilibrium) between the alleles influencing the level of selfing and the alleles that directly influence fitness, such that lineages produced by recurrent selfing exhibit lower δ than lineages with a recent history of outcrossing (Uyenoyama and Waller 1991). Although some studies have detected a covariation between selfing and δ among lineages within populations (Vogler et al. 1999), including cases in which δ correlated positively with herkogamy (Takebayashi and Delph 2000; Stone and Motten 2002), others have not (Carr et al. 1997; Mutikainen and Delph 1998; Rao et al. 2002) and the generality of these associations on theoretical grounds, is uncertain (Schultz and Willis 1995). The combination of predominant selfing, extensive genetic variation in herkogamy, and a relatively consistent negative association between herkogamy and selfing in populations may facilitate the development of such associations in populations of A. canadensis. However, this requires selection among selfed progeny so that the lineage least encumbered with deleterious mutations pass on their low genetic load to their offspring. Yet, the estimates of F close to zero for mature plants in the populations we studied (Eckert and Herlihy 2004) suggest that selfed progeny rarely survive to reproduce, leaving little opportunity for identity disequilibrium to develop.

Genetic variation for herkogamy might be maintained in populations if the positive effect of increased herkogamy on fitness via reduced selfing is balanced by negative effects on other components of the mating system. Because herkogamy is a composite trait, arising from the relative positioning of anthers and stigmas, alterations in herkogamy may affect female or male outcrossing success. The effect of herkogamy on pollen export and the receipt of outcross pollen might, in turn, depend on the extent to which variation in herkogamy is caused by variation in stigma position versus variation in anther position. Our results support the general expectation that, in species like A. canadensis, where stigmas are exserted beyond the anthers, herkogamy should covary positively with pistil length and negatively with stamen length (Webb and Lloyd 1986), although the latter was only observed after we statistically controlled for the positive covariance between pistil and stamen length that arises due to overall variation in flower size. We also detected consistent positive correlations between herkogamy and pistil length for 18 of the 19 population × year combinations in which we measured one flower per plant (mean r=+0.54, range = 0.23–0.79), within the five northern populations in which we portioned floral variation into among- and within-plant components (mean r=+0.44, range = 0.23–0.69), and among maternal plants within the populations for which we investigated the effect of herkogamy on self-fertilization (mean r=+0.47, range = 0.17–0.78). Finally, our analysis of variance among open-pollinated families and parent–offspring regressions for floral traits revealed positive genetic correlations between herkogamy and pistil length for two northern populations.

The increased exsertion of stigmas in the most herkogamous flowers might possibly reduce the deposition of outcross pollen (e.g., Eckert and Barrett 1995) and consequently the production of outcrossed seeds. However, a re-analysis of data from Herlihy and Eckert (2002) does not support this hypothesis. The number of seeds produced by flowers of A. canadensis that have been emasculated to remove the confounding effect of self-pollination covaries positively, although weakly, with pistil length (r=+0.16, P= 0.01, n= 229 flowers, variation in pistil length among populations removed). Increased seed production in flowers with longer pistils is not due to a larger number of ovules, as ovule number does not covary with pistil length or any other component of flower size (reanalysis of data from Kliber and Eckert 2004).

Herkogamy might also influence outcrossed siring success. Most theoretical models anticipate that self-pollination reduces outcrossed male fitness simply because it reduces the amount of pollen available for export (pollen discounting; Lloyd 1992; Holsinger 1996, but see Johnston 1998). Hence, increased herkogamy, because it reduces self-pollination, might increase outcrossed siring. Within the northern populations of A. canadensis we studied, pistil length, the main determinant of herkogamy, correlates positively with pollen production per flower (reanalysis of data from Kliber and Eckert 2004). This suggests an additional positive influence of herkogamy on outcrossed siring. It remains possible that increased stigma exsertion (and hence herkogamy) may negatively influence the pick-up or positioning of pollen onto the bodies of hummingbird or bumblebee pollinators. For example, using paternity analysis in a natural population of M. guttatus, van Kleunen and Ritland (2004) found that outcrossed siring correlated negatively with herkogamy, although the underlying mechanism was not investigated. In this case, a direct negative selection on herkogamy was balanced by indirect positive selection via genetic correlations between herkogamy and other aspects of floral morphology. Although the results of this and previous studies on A. canadensis predict selection for reduced self-fertilization, probably through increased herkogamy (Eckert and Herlihy 2004), testing this hypothesis will require a more complete analysis of how variation in the floral structures that determine and are correlated with herkogamy also influence male and female fitness through outcrossing.

Associate Editor: J. Shykoff


We thank C. Griffin, M. Bhardwaj, J. Brown, and F. Thompson for help in the field, laboratory, and greenhouse; J. Busch, J. K. Kelly, and J. Shykoff for helpful comments on the manuscript; The Queen's University, Mountain Lake, and Highlands Biological stations for logistical support in the field; Queen's University for a Dean's Doctoral Field Travel Grant to CRH; Highlands and Mountain Lake Biological Stations for grant-in-aid scholarships to CRH; and the Natural Sciences and Engineering Council of Canada for a discovery grant to CGE.


Table Appendix 1..  Locations of populations of Aquilegia canadensis used in this study, and the type of data collected in each year of study: F, floral morphology on one flower per plant; V, variation in floral morphology within- and among-plants; M, morphological basis for herkogamy; S, effect of herkogamy on self-fertilization; A, effect of herkogamy on autogamy; and G, quantitative genetics of herkogamy. The first two letters of population codes for central populations indicate state: NC = North Carolina, VA = Virginia. Northern populations were all in Ontario (Q = Queen's University Biological Station).
PopulationCountyLocationLat. (N)Long. (W)199920002001
  QCA2FrontenacCamelot 244°33′03″76°21′04″ VM
  QFP1FrontenacFrank Phelan's44°29′42 ″76°25′11″FSAFVSGM
  QLL1FrontenacLindsay Lake 144°32′09 ″76°22′43″FSA  
  QLL2FrontenacLindsay Lake 244°32′16 ″76°22′56″FSA  
  QLL3FrontenacLindsay Lake 344°32″18 ″76°23′13″ FVSGM
  QMT1Leeds and GrenvilleMoore's Tract44°35′12 ″76°20′13″ FVSM
  QOR1FrontenacOpinicon Road44°28′55 ″76°28′01″FSAFVSGM
  QPG1FrontenacPost Office Gate Lane44°31′50 ″76°21′55″FS  
  QSP1FrontenacSpar Marsh44°30′48 ″76°23′36″FSA  
  QTF1FrontenacTurid Forsythe's44°32′36 ″76°22′26″FSA  
  NCSMR1PolkSkyuka Mountain Road35°15′53 ″82°14′07″ FS 
  VACBW1MontgomeryChristiansburg Wayside37°08′33 ″80°18′36″ FSG 
  VAELL3MontgomeryEllett37°11′40 ″80°22′26″ FS 
  VAFSR1MontgomeryFriendship Road37°11′30 ″80°16′49″ FS 
  VAHUF1CraigHuffman37°21′22 ″80°23′04″ FS 
  VANPT1GilesNewport37°18′08 ″80°29′33″ FSG 
  VAPFR1MontgomeryPrices Fork37°11′46 ″80°30′31″ FS 
  VAZMR1GilesZells Mill Road37°18′41 ″80°30′23″ FS 
Table Appendix 2..  Mean and range of herkogamy for groups of flowers used in the categorical analysis of the effect of herkogamy on total self-fertilization and autogamous self-fertilization. nfam is the number of maternal families sampled for each group. The number of families from emasculated flowers (Emasc) is also provided for populations in which autogamy was estimated (see methods).
YearRegionPopulationHigh herkogamy groupLow herkogamy groupEmasc nfam
nfamMean (mm)Range (mm)nfamMean (mm)Range (mm)
QLL2114.92.8–8.4 91.80.0–2.532