Responses to selection on male-phase duration in Chamerion angustifolium


Matthew B. Routley, Department of Biological Sciences, University of Calgary, Calgary, AB, Canada T2N 1N4.
Tel.: +1-403–2203557; fax: +1-403–2899311;


Protandry (when male function precedes female) can enhance fitness by reducing selfing and increasing pollen export and outcrossed siring success. However, responses to selection on protandry may be constrained by genetic variation and correlations among floral traits. We examined these potential constraints in protandrous Chamerion angustifolium (Onagraceae) by estimating genetic variation in male-phase duration and associated floral traits using a paternal half-sib design and selection experiment. Narrow-sense heritability of male-phase duration was estimated as 0.23 (SE ± 0.04) and was positively correlated with floral display. The selection experiment shortened male-phase duration 0.8 SD from the parental average of 17.0 h and lengthened it by 2.0 SD. Furthermore, fixed floral longevity caused a negative association between male- and female-phase durations. These results suggest that selection on male-phase duration is not limited by genetic variation. However, changes in male-phase duration may influence pollinators through correlated changes in floral display and reduced opportunities for pollen receipt during female phase.


Many hermaphroditic plants stagger their genders in time (dichogamy). Dichogamy comprises protandry, in which male function precedes female function, and protogyny, where female phase is first. Lloyd & Webb (1986) suggested that this separation of gender reduces interference between stamens and pistils in the same flower and the resulting conflict between pollen import and export (see Barrett, 2002 for a review of pollen–pistil interference). Harder & Barrett (1995,1996) further suggested that dichogamy reduces between flower interference by minimizing within-inflorescence pollen transfer and the detrimental impacts of geitonogamy and pollen discounting. Indeed, experimental studies have demonstrated that protandrous plants have significant siring advantages relative to adichogamous plants (Harder et al., 2000; Routley & Husband, 2003). However, in natural populations, the evolutionary response to selection for protandry, or the duration of male phase, can be mitigated by many factors.

An important consideration for any study of adaptation is the genetic architecture underlying the character of interest. One basic feature of genetic architecture is the magnitude of available genetic variation. In the absence of heritable genetic variation for the character there can be no evolutionary response to selection (Fisher, 1930). Furthermore characters are often inter-related by genetic or phenotypic correlations (Antonovics, 1976; Lande & Arnold, 1983), creating constraints on the character combinations that can be produced. If selection acts in opposition to these correlations, then the response to selection may be slowed and the most fit phenotype may represent a compromise among trait values (Dickerson, 1955; Lande & Arnold, 1983; Roff, 1996).

Several aspects of the genetic basis of gender have been studied in plants, including sex allocation (Elle, 1998), the evolution of separate sexes (Dorken & Barrett, 2004), floral display (Caruso, 2004), and the relationship between sexual dimorphism and attributes of floral display (Delph et al., 2004). However, very few studies are available on the genetic architecture of male-phase duration and the potential for the constraints described above to operate (Schoen, 1982; Campbell, 1996; Vogler et al., 1999). Presumably, male-phase duration is heritable to some extent. However, the magnitude of heritability has a direct effect on the tempo of evolutionary response to selection for changes in male-phase duration. Furthermore, correlated responses may be common in flowers (Delaguerie et al., 1991; O'Neil & Schmitt, 1993; Ågren & Schemske, 1995) due to the closely coordinated development of floral structures (Jürgens, 1997), the requirement that anthers and stigmas be presented in close physical proximity (Berg, 1960; Armbruster et al., 1999), and the finite resources available to male and female function (Charnov, 1982). As a consequence, changes to the duration of male phase may have important effects on allocation to female function. Moreover, pollinators are influenced by floral size and inflorescence display (reviewed in Harder & Barrett, 1996). Consequently, any genetic correlation between male-phase duration and these traits may influence pollinator attraction and the fitness of the selected phenotype.

In this study we investigated constraints on the evolution of protandry by estimating the genetic basis of variation in male-phase duration in the protandrous plant Chamerion angustifolium. We conducted this investigation with both a paternal half-sib breeding design and an artificial selection experiment with variance components estimates. Specifically, we addressed three questions: (1) How variable is male-phase duration within populations and is this variation heritable?, (2) Does selection on male phase cause correlated changes in floral size and display?, and (3) Is there a correlation between male- and female-phase durations? We find that male-phase duration has a genetic basis and responds rapidly to selection. However, this evolutionary response may be constrained by a correlation with aspects of display size and with female-phase duration.

Materials and methods

Study species

Chamerion angustifolium (L.) Holub (Onagraceae) is a multi-flowered, herbaceous perennial with protandrous flowers. Individual plants are approximately 2 m tall, with indeterminate inflorescences bearing from 10 to 15 open flowers at a time, each with four, 10–15 mm long, pink petals (Husband & Schemske, 2000). Each flower has eight anthers, which shed blue-green or yellow pollen, held together with viscine threads (Myerscough, 1980). The period from seed germination to first flower has a duration of approximately 45 days (Husband & Schemske, 1997) and pollination to seed set of approximately 14–21 days (personal observation). During the initial male phase, when the anthers have dehisced, the style is strongly deflexed away from the anthers and the four stigma lobes are closed. The stigma lobes then spread apart becoming receptive and the style straightens, positioning the stigma near the centre of the flower. This protandry and the acropetalous (maturing from base to tip) development of the inflorescence, produces an inflorescence with female phase flowers at the base and male phase flowers toward the apex (Galen & Plowright, 1988). In this greenhouse study, pollinators were excluded and pollen remained in the anthers throughout anthesis. Consequently, the female phase is in fact a hermaphroditic phase. However, in the presence of pollinators, pollen is rapidly removed from the anthers (Galen & Plowright, 1988; personal observation), so that the genders do not overlap.

Paternal half-sib design

We created paternal half-sib families in the summer of 2001 with seed collected from population D2 in Montana (Husband & Schemske, 1997). We germinated the seed from open-pollinated families and transferred them to 4 L pots. From these plants we randomly crossed 42 sires from different families with 53 unrelated dams to produce 304 fruits with an average of 7.2 dams per sire. One plant from each of the 304 fruits was grown to flowering. To estimate male- and female-phase durations, we divided anthesis (described above) into seven developmental stages (Table 1). Male phase was bounded by the dehiscence of the first anthers and the appearance of the receptive stigma. Female phase was identified by the opening of stigma lobes and ended once the petals closed. We classified male phase as stages 3–5 and female phase as stages 6–7. We classified the third to seventh open flowers on each plant per half-sibship by stage twice daily (at approximately 9:00 and 16:00 h). This documentation began when the first flower of the sequence reached stage 3 and ended when all five flowers reached stage 7. We estimated male-phase duration as the average number of hours elapsed between stages 3 and 6 for the five flowers. Our resolution of male-phase duration was limited by the two sampling periods per day. For the purposes of this study, we assumed that if a flower changed stages between sampling periods the change occurred at the end of the earlier period. For example, if at 9:00 h a flower was in stage 5 and then in stage 6 at 16:00 h, male phase ended at 9:00 h and 8 h had elapsed in female phase. Alternate assumptions (e.g. the switch occurring just before the second sampling period) would effectively shift all duration estimates by approximately 8 h. Since we are interested in relative allocations and relative changes in phase durations, this assumption does not bias our results. We estimated female-phase duration in a similar way as the average number of hours elapsed from stage 6 to 7.

Table 1.  The floral stages used to describe anthesis in C. angustifolium. Stages 3–5 represent male phase while 6 and 7 are female phase.
1Closed flower bud
2Open flower, no dehisced anthers
3Open flower, ≤4 dehisced anthers
4Open flower, >4 dehisced anthers
5Open flower, all anthers dehisced, stigma closed
6Open flower, all anthers dehisced, stigma open, style reflexed
7Open flower, all anthers dehisced, stigma open, style fully extended
8Closed flower

Selection experiment

In addition to the paternal half-sib breeding design, we also investigated the genetic relationship between male- and female-phase durations and other floral characters (described below) with an artificial selection experiment. Beginning with the paternal half-sib plants as the base population, we implemented the selection design depicted in Fig. 1. Plants were ranked according to their male-phase duration and for the parental generation the shortest 20% were assigned to one of two short selection-lines and the longest 20% to one of two long selection-lines. Two control lines were created by randomly selecting plants from the entire distribution of male-phase duration in the base population. Each selected plant was randomly mated to two sires from the same selection line and replicate to produce the first selected generation. Flowers were pollinated by removing two anthers from a sire with forceps and applying the pollen to all four lobes of the recipient stigma. For the second generation of selection the upper and lower 50% of male-phase duration from the first selected generation were chosen as parents for the long and short selection-lines, respectively. Due to the intense inbreeding depression in C. angustifolium (inbred offspring have a relative fitness of 5%, Husband & Schemske, 1995) all crosses were made between plants from different half-sib families to generate the first generation, and half-sib plants were not crossed for the second generation (plants may have shared one grand-parent, but not parents). Although this crossing design reduced the influence of inbreeding on our trait measures, the number of suitable mates and population sizes declined with each generation. Consequently, only two rounds of selection were possible to yield three generations (parental, first, and second selected generations).

Figure 1.

A schematic representation of the selection design used. The numbers represent sample sizes. A replicate of the long male-phase duration was lost to greenhouse pests.

Correlates of male-phase duration

In addition to male- and female-phase durations, we measured several other floral traits for each plant in each generation. Floral width, measured as the greatest diameter across the flower, was taken for one randomly chosen flower. Floral separation was measured as the distance from the pedicel of the randomly chosen flower to the pedicel of the next highest flower. Finally, the display length occupied by the censused flowers was measured as the distance from the pedicel of the first measured flower to the pedicel of the last measured flower of the five. Display size was measured only for the final generation of the selection experiment as the average number of open flowers on a plant for three randomly chosen days the week after the census.

Data analysis

We estimated the heritability of male-phase duration from the paternal half-sib design following the methods of Lynch & Walsh (1998). After correcting for unbalanced progeny numbers in each paternal half-sib family, heritability was estimated from the intraclass correlation (tPHS) as:


where Var(s) is the variation among sires and Var(e) is the error variance. The 95% confidence interval of this heritability estimate was calculated from the F distribution with 41 and 262 d.f.

In addition to the estimate from the paternal half-sib design, we estimated genetic variances and covariances from the parental, paternal half-sib plants and two selected generations with restricted maximum-likelihood variance components estimates (remlvce, ver. 4.2, Neumaier & Groeneveld, 1988). This approach is well suited for unbalanced designs (Shaw, 1987). In addition, since the genealogical relationships of the plants are incorporated into this analysis, the use of selected lines does not bias the genetic estimates obtained (Shaw, 1987). This approach does not explicitly use the selection line or generation treatments. Rather, it incorporates a pedigree of the plants used to estimate the genetic architecture of male-phase duration and associated traits. Since the analysis is based on the phenotypes of individuals, rather than their offspring, this approach is often termed an ‘animal’ model (Lynch & Walsh, 1998). We included replicate selection lines as a random effect in the model. vce also readily generates standard errors for all heritability and genetic correlation estimates. Based on these standard errors, we constructed 95% confidence intervals for each estimate. If an individual estimate's confidence limit was greater than zero and less than one it was considered statistically significant.

We analysed the responses of male- and female-phase durations, and total floral longevity to selection on male phase by anova using R software (R Development Core Team, 2004) and means were compared with the multcomp package (Bretz et al., 2004). Unfortunately, this analysis is complicated by the attrition of plants in the second generation of selection due to greenhouse pests and severe inbreeding depression. In particular, one of the long selection-line replicates was completely lost and a short selection-line replicate was reduced to two individuals. Consequently, we analysed changes in duration for the first and second generation of selection separately. For the first selected generation we analysed changes in duration with selection treatment as a fixed effect and replicate lines within treatment as a random effect. We then used statistical contrasts to compare the means of low and high selection lines. For the second selected generation we analysed the effect of line only and used statistical contrasts to compare the mean of the high selection line with the two low selection lines. We analyzed the responses to selection of floral width, floral separation, inflorescence size, and floral display for the second generation of selection plants by anova with selection treatment as a fixed effect and values pooled across replicate treatments.

To quantify the magnitude of selection on male-phase duration, we calculated selection differentials for each generation as the weighted average of the difference in male-phase duration of plants chosen for crossing from the average of all plants for the specific generation and replicate population (Falconer & Mackay, 1996). We then summed these selection differentials to calculate cumulative selection differentials for each selection line across the two generations of selection. We measured the response to selection as the average response of the replicates of each selection line for the first and second generations.

We tested our control and selection lines for evidence of drift using the methods of Lande (1976, 1977, also see Sarkissian & Harder, 2001). This approach calculates a test statistic:


where Ne is the effective population size (taken as the harmonic mean of the first and second generation), z2 is the squared divergence from parental to final generation in SDs, h2 is the heritability (estimated from remlvce), t is the number of generations (2), and inline image is the overall phenotypic variance (the MSerror from the anova of selection on male-phase duration). D is distributed as a standard normal curve and significant P values show that drift cannot account for the observed changes.


Paternal half-sib design

We found extensive variation in male- and female-phase durations across the paternal half-sib families. Based on all 304 plants, male phase lasted for 17.0 ± 0.5 (mean ±SE) h and female phase for 62.9 ± 1.0 h. Average male-phase duration ranged from 9.1 to 27.5 h across the paternal half-sib families. The coefficient of variation (SD/x) for male-phase duration was 0.5. The pattern of variation in male-phase duration for these families (Table 2) leads to a heritability estimate of h2 = 0.27 with a 95% confidence limit of 0.05–0.54. Flowers were 31.5 ± 0.3 mm wide, with an average of 18.0 ± 0.5 mm between flowers. Display length was 38.2 ± 0.1 cm.

Table 2. anova of male-phase duration in C. angustifolium from the paternal half-sib design. These values were used for estimating heritability and for evaluating whether genetic drift can account for the observed changes in the selection experiment.
Sourced.f.F ratioP

Response to selection

We observed significant treatment responses after two generations of selection on male-phase duration (Table 3a). When compared to the control lines, the long and short selection-lines rapidly diverged from each other (Fig. 2), although the absolute amount of divergence was three times greater for long selection lines. At the end of the selection experiment, male-phase duration had decreased by 0.8 SD from the base population and increased by 2.0 SD (Fig. 3). The first generation of selection had a significant effect on male-phase duration and the long and short selection-lines differed significantly (t = −2.5, Padj < 0.05). In the second generation of selection we detected a significant line effect (F4,49 = 145.2, P < 0.05) with the one high and two low lines significantly different from each other (t = −3.4, Padj < 0.01). However, due to the loss of a replicate high line, this must be interpreted as a line effect and not necessarily due to selection. The plotted relationship between cumulative selection differentials and cumulative responses (Fig. 4) showed that for the first generation of selection the selection differentials and responses were very similar among selection treatments and replicate lines. One of the short selection replicates then diverged from the remaining lines by showing a reduced response to a high differential.

Table 3. anovas of evolutionary responses in (a) male-phase duration, (b) female-phase duration, and (c) total floral longevity for the first generation of selection on male-phase duration in Chamerion angustifolium. The divergence from controls for male-phase duration is presented in Fig. 2. Treatment means for male- and female-phase durations after both generations of selection are presented in Figs 3 and 5.
(a) Male-phase duration
Selection treatment2333.145.6<0.0001
Replicate/selection treatment37.30.2>0.50
(b) Female-phase duration
Selection treatment2127.815.4<0.05
Replicate/selection treatment38.30.1>0.70
(c) Floral longevity
Selection treatment291.62.9>0.10
Replicate/selection treatment331.10.2>0.50
Figure 2.

The divergence (±SE) from control lines (standardized to 0) for male-phase duration after two generations of selection in C.angustifolium. To simplify the presentation, the average of the two replicate lines is plotted. Triangles with points up are long selection-lines and triangles with points down short selection-lines. Statistical details are presented in Table 3a and the text.

Figure 3.

Mean (±SE) male-phase duration for the final generation of selected plants in C. angustifolium. Triangles with points up are long selection-lines, triangles with points down short selection-lines, and circles control lines. Closed and open symbols represent the first and second replicate-lines, respectively. See Table 3a and the text for statistical details.

Figure 4.

The cumulative selection differentials and cumulative responses for male-phase duration in the two generations of short and long selection-lines in C. angustifolium. Triangles with points up are long selection-lines and triangles with points down short selection-lines. Closed and open symbols represent the first and second replicate lines, respectively. The dashed line represents complete heritability. The first long selection-line was lost after the first generation of selection.

Female-phase duration changed in the opposite direction to male-phase duration, increasing by 1.1 SD in the low selection-line and decreasing by 0.5 SD in the high selection-line (Fig. 5 and Table 3b). In the second generation, there was a significant effect of line on female-phase duration (F4,49 = 4.8, P < 0.01) with the high line differing from the low selection-lines (t = −4.3, Padj < 0.01). In the analysis of total anthesis duration, we found no significant effect of selection after the first generation of selection (Table 3c). We also detected no significant effect of selection line on total anthesis duration in the second generation of selection (F4,49 = 1.9, n.s.).

Figure 5.

Female-phase duration (±SE) for the final generation of selected plants in C. angustifolium. Triangles with points up are long selection-lines, triangles with points down short selection-lines, and circles control lines. Closed and open symbols represent the first and second replicate lines, respectively. See Table 3b and the text for statistical details.

Overall, the maximum inbreeding coefficient in our selection experiment was 0.13 and, among the inbred plants, the average inbreeding coefficient was 0.10. Four hundred and eighty plants had inbreeding coefficients of zero. The observed changes in selection lines were not due to drift (P < 0.001) as indicated by the consistency of responses for the first generation of selection (Fig. 4). The changes observed in control lines also were not due to drift (P < 0.001) and are likely due to inadvertent selection or changing environmental conditions.

Despite the large changes in male- and female-phase duration, the anova of the additional floral characters showed no significant differences among selection lines for the final generation (Tables 4 and 5). Flowers were 31.5 ± 0.3 (mean ± SE) mm wide, with an average distance of 18.0 ± 0.5 mm between flowers. Display length was 38.2 ± 0.1 cm and display size 6.9 ± 4.5 flowers.

Table 4. anova of the effect of selection for male-phase duration on four other floral characters in C. angustifolium.
Sourced.f.F ratioP
Floral width2, 450.71>0.45
Floral separation2, 450.02>0.95
Inflorescence size2, 440.73>0.45
Display size2, 511.31>0.25
Table 5.  Summary of floral traits measured in C. angustifolium from all three generations. Display size was measured in the second selected generation only.
Male-phase duration (h)49818.88.1
Female-phase duration (h)49867.617.3
Floral width (mm)45331.94.6
Floral separation (mm)44717.48.8
Display length (mm)36039.213.8
Display size (no. of flowers)546.94.5

Genetic variances and covariances

Based on all measured generations, each of the floral traits measured had significant heritabilities (Tables 6 and 7). In particular, the narrow-sense heritability of male-phase duration was estimated as 0.23 ± 0.04 (mean ± SE). We detected significant positive genetic correlations between male-phase duration and both floral separation distance and display length. In addition, floral separation, display length, and display size had positive genetic correlations (Table 6). The vce analysis had a final likelihood of 4 330 and reached status 1 (a unique solution was found) after 103 iterations.

Table 6.  Heritabilities (on the diagonal) of and genetic correlations (above diagonal) among male- and female-phase durations, floral width, floral separation, display length, and display size in C. angustifolium as estimated from all three generations.
Male phaseFemale phaseFloral widthFloral separationDisplay lengthDisplay size
  1. Each estimate is ±SE and estimates significantly greater than zero and less than one, based on 95% CI, are indicated in bold.

0.23 ± 0.04−0.17 ± 0.14−0.13 ± 0.120.49 ± 0.220.68 ± 0.120.22 ± 0.32
 0.17 ± 0.040.04 ± 0.11−0.26 ± 0.22−0.18 ± 0.150.38 ± 0.29
  0.19 ± 0.040.26 ± 0.210.47 ± 0.16−0.13 ± 0.10
   0.07 ± 0.030.89 ± 0.130.68 ± 0.24
    0.18 ± 0.040.46 ± 0.23
     0.22 ± 0.32
Table 7.  Variances (on the diagonal) and covariances (above diagonal) of male- and female-phase durations, floral width, floral separation, display length and display size in C. angustifolium as estimated from all three generations.
Male phaseFemale phaseFloral widthFloral separationDisplay lengthDisplay size


Male-phase duration in C. angustifolium is highly variable. Before selection, the average male-phase duration among half-sib families ranged from 9.1 to 27.5 h with a coefficient of variation equal to 0.5. The capacity for selection to operate on this variation was demonstrated by the change in mean duration in just two generations of selection to 22 h (0.8 SDs from the parental average) in the short selection lines and 31 h (2 SDs) in the long selection-lines. These responses produced an estimated narrow-sense heritability of ∼0.25 for male-phase duration. Moreover, female-phase duration responds negatively to selection on male-phase duration and floral longevity appears to be fixed in C. angustifolium producing a negative correlation between male- and female-phase durations. Finally, despite the genetic correlations observed between male-phase duration and floral attributes, responses to selection in male-phase duration were not associated with any correlated phenotypic changes in floral size or number.

Our estimates of phenotypic variation in male-phase duration are high compared to previous studies of floral variation. In fact, variation in male-phase duration was equivalent to variation measured for nectar production, the most variable trait in Cresswell's (1998) survey of floral traits. Our estimates of variation in correlated floral characters are based on single measurements from each individual. Substantial variation within individuals may also contribute to the total variation of characters. Although this wide variation could limit the precision of our trait estimates, our experimental design was sufficient to detect significant genetic variation and genetic correlations with other traits.

Consequences of variation

Variation in male-phase duration is evolutionarily significant only if it has a genetic basis. We estimated narrow-sense heritability for male-phase duration with two methods and obtained consistent results of h2 = 0.25 (0.27 from the paternal half-sib design and 0.23 from the remlvce analysis). However, these estimates should be considered within the context of heritability estimates from greenhouse studies (Weigensberg & Roff, 1996; Conner et al., 2003). Due to the reduced environmental variance inherent to greenhouse studies, our estimated heritabilities may be inflated relative to field conditions. Conner et al. (2003) found that heritability estimates in Raphanus from the greenhouse were substantially higher than field estimates for six floral traits. Furthermore, the variance–covariance matrixes of these wild radish traits were not proportional to each other. Consequently, extrapolating greenhouse quantitative genetic estimates to field conditions should be done with caution. However, Weigensberg & Roff (1996) surveyed close to 350 heritability estimates from laboratory and field studies and concluded that laboratory studies provide reasonable estimates of field-measured heritabilities.

Heritability estimates of male-phase duration are rare. Vogler et al. (1999) estimated broad-sense heritability for male-phase duration in three discrete environments as 0.27 in Campanula rapunculoides (Campanulaceae) based on daily observations of 96 plants. Two previous studies of between flower estimates of male-phase duration have found similar heritabilities. Schoen (1982) calculated a protandry index (the proportion of male phase flowers relative to hermaphroditic flowers) in Gilia achilleifolia (Polemoniaceae) grown in the greenhouse. Based on variation within and among plants from a fruit, Schoen estimated the heritability of the protandry index as 0.47. Campbell (1996) estimated the proportion of time spent by Ipomopsisaggregata (Polemoniaceae) flowers in female phase by censusing paternal half-sibs twice a week in the field and estimated heritability near 0.3. Considering surveys of heritabilities in the laboratory and the wild (Mousseau & Roff, 1987; Weigensberg & Roff, 1996; Conner et al., 2003), these estimates of gender-phase duration combined with ours show that protandry is moderately heritable. Consequently, selection for changes in male-phase duration could produce rapid results.

Consequences of correlations

Due to the importance of floral display for plant reproduction (Harder & Barrett, 1996), a correlation with floral display could significantly affect the evolution of male-phase duration. Our results suggest that selection for increased male-phase duration will increase the separation between flowers (r = 0.49) which will increase floral display size (r = 0.68). The average genetic correlation among morphological traits in plants and animals is 0.47 (Roff, 1996). This places the correlations between male-phase duration and floral display at the higher end of the range of values measured to date. Large displays can increase both male and female reproductive success (Campbell, 1989; Emms & Arnold, 1997). However, larger displays may also experience negative consequences as a result of increased geitonogamy and pollen discounting (Klinkhamer & de Jong, 1993; Harder & Barrett, 1996). Consequently, selection to increase pollen removal in pollinator-limited environments by increasing male-phase duration may also increase pollinator attraction by increasing floral display. The equilibrium male-phase duration would then be a compromise between the gains in siring success due to this increased attraction and the losses due to within-plant pollen transfer (Klinkhamer & de Jong, 1993).

Despite our estimated genetic correlations from the remlvce analysis, we detected no correlated response of floral display in our selection experiment. There are several potential ways to reconcile these findings. First, retrospective power analyses of our anova show that display length and size had adjusted powers of 0.16 and 0.27, respectively. Consequently, we cannot rule out the possibility that we simply had insufficient power to detect the changes in display. Second, related to the first, two generations of selection may not have been sufficient to change characteristics of floral display in C. angustifolium. Further generations of selection may have produced the predicted responses. Third, the variance–covariance structure of our measured traits may have changed during the selection experiment. However, given the concordance between the pre-selection, paternal half-sib estimate of heritability and the remlvce estimate from all three generations, this seems unlikely. Furthermore, two generations of selection is not thought to be sufficient to change correlation structures (e.g., Arnold, 1992; Schluter, 1996, but see Stanton & Young, 1994).

In this study, we found a negative phenotypic correlation between male- and female-phase duration (Figs 3 and 5). Furthermore, floral longevity remained fixed while gender-phase duration responded to selection. Despite this realised genetic correlation, the observed negative genetic correlation between gender durations was not statistically distinguishable from zero. Analogous arguments to those made for floral display may be relevant here. Unfortunately, we cannot evaluate the strength of those arguments for reconciling the estimated and observed genetic correlation between male- and female-phase durations and the basis of this trade-off is unclear. Nonetheless, since hermaphroditic plants must maximize fitness gains through both genders (Charnov, 1982; Morgan & Schoen, 1997), this negative correlation can limit the evolution of male-phase duration. Similarly, Devlin & Stephenson (1984) found that female-phase duration decreased with increased male-phase duration in Lobelia cardinalis (Campanulaceae).

Although we have demonstrated substantial variation for gender-phase duration, we do not know whether this continuous variation has any functional significance. Two experimental studies of the functional significance of protandry (Harder et al., 2000; Routley & Husband, 2003) compared dichogamous and adichogamous plants as discrete categories of gender separation. Although both studies demonstrated that dichogamy provides a siring advantage, it is unclear how continuous variation in gender duration would change this advantage. Gender function is often limited by the frequency of pollinator visits. Consequently, selection must optimize each gender-phase duration to satisfy both male and female reproductive-success. If pollinators visit flowers rapidly and consistently, a short male-phase duration may be sufficient to remove all available pollen and saturate male fitness. However, rare or idiosyncratic pollinator activity may favour longer male phases and mechanisms that prolong the removal of pollen (Harder & Thomson, 1989) to maximize siring success (Lloyd & Yates, 1982; Devlin & Stephenson, 1984; Richardson & Stephenson, 1989; Robertson & Lloyd, 1993). In Alstreomeria aurea (Alstroemeriaceae), Aizen & Basilio (1998) found that male function required three times the pollinator activity as female function to maximize reproductive success. In accordance, nectar production was three times greater during male phase to compensate for this asymmetry in pollinator requirements. In C.angustifolium, nectar quantity (volume and percent sugar) does not differ between male and female phases (Komlos, 1999), indicating that male phase in this species is not more pollinator limited than female phase. In contrast, Parker et al. (1995) found that C. angustifolium allocated 1.5 times the resources (measured as mg of dry mass) to male phase relative to female phase. Clearly, more research on the relationship between male and female duration, pollinator attraction, and the consequences for male and female reproductive success are required to explain the current allocation of male and female duration in C.angustifolium.


We thank P. Kron for significant help with the paternal half-sib design and M. Isaac and S. Miller for assistance with the selection design. We appreciate the help of C. Caruso and A. Worley with using vce. Constructive comments from C. Caruso and two anonymous reviewers improved earlier drafts of the manuscript. We also acknowledge financial support from the Natural Sciences and Engineering Research Council of Canada to M. B. R. and B. C. H. and the Premier's Research Excellence Award and Canada Research Chair to B. C. H.