1. Flowering-time differences within and among protandrous, hermaphroditic plants shift the floral sex ratio from male- to female-dominated during a population’s flowering season. This dynamic should induce negative frequency-dependent selection favouring relatively greater investment of resources and time in female function by early flowers and early-flowering plants than by late flowers and late-flowering individuals. In contrast, selection for floral integration to facilitate pollen export and import should limit floral variation.
2. We assessed these contrasting expectations for the protandrous Delphinium glaucum by considering the relation of variation and covariation in the lengths of perianth segments, anther and ovule number, and male- and female-phase duration for flowers at three positions within inflorescences to the first flowering date, mean gender and size of 34 plants.
3. Reproductive phenotypes varied among and within plants in association with a shift in population floral sex ratio from male- to female-biased, but not with plant size. Early-flowering plants had female-biased phenotypic gender and smaller flowers with fewer anthers per ovule and shorter male phases than late-flowering plants. Within individuals, earlier (lower) flowers were larger with higher female effort, in terms of both the relative production of ovules and anthers and the relative durations of female and male phases, than later (upper) flowers. Overall, these results are more consistent with the expectations of negative frequency-dependent selection than with proximate resource allocation.
4. Correlations among floral traits indicated significant floral integration. However, these correlations weakened from lower to upper flowers, suggesting that the patterns of intra-individual variation limit the extent of integration.
5.Synthesis. Reproductive phenotypes that incorporate systematic among-flower variation and vary consistently among individuals with flowering time, such as those exhibited by D. glaucum, may be typical of dichogamous species with multi-flowered inflorescences. Such diversity within and among plants should result from a combination of density-dependent selection against simultaneous flowering of individuals and frequency-dependent selection favouring emphasis on the least common sex role during specific stages of the flowering season, both facilitated by the positive assortative mating that accompanies asynchronous flowering among individuals.
Modular structure and hermaphroditism enable flowering plants to produce complex reproductive phenotypes. Because of modularity, plants typically produce multiple flowers, so individual flowers and fruits cannot represent their reproductive phenotypes fully. Instead, components of these phenotypes involved in pollination include the timing and duration of flowering, the number of flowers displayed simultaneously and produced overall, the mean traits of individual flowers, and their variation within and among a plant’s inflorescences (Herrera 2009). Hermaphroditism allows for additional within-plant variety, to the extent that flowers differ in the relative investments in female and male functions (Brunet & Charlesworth 1995; Diggle 2003). Furthermore, when flowers are displayed sequentially, both the mean and variation of a plant’s reproductive phenotype can be dynamic, if the number of flowers it displays and their potential roles as female and male organs (sex roles) vary, owing to shifts in anthesis rate and floral longevity, and/or differences in morphology and sex allocation among early and late flowers (e.g. Thomson & Barrett 1981; Ishii & Sakai 2001, 2002; Harder & Johnson 2005). Each of these components can also vary among plants and, if this variation is genetically determined (e.g. Baker, Burd & Climie 2005; Ashman & Majetic 2006; Conner 2006), can be subject to selection.
The more recent perspective offers adaptive explanations for widespread, systematic variation within inflorescences, especially involving flower size and sex allocation (Brunet & Charlesworth 1995; Diggle 1995, 2003; Herrera 2009). Such patterns cannot arise from developmental instability, but instead must be generated by parallel variation in gene expression and associated hormone gradients, and/or the environment for floral development, including the distribution of resources within inflorescences (Diggle 1995, 2003; Herrera 2009). In species with staggered opening of flowers with separate female and male phases, systematic patterns of sex allocation have been attributed to negative frequency-dependent selection, which favours emphasis on whichever sex role (female or male) is least common during a specific period of a population’s flowering phenology (Brunet & Charlesworth 1995). A recently recognized, and so poorly examined, corollary of this hypothesis proposes a similar explanation for among-plant variation in relative allocation to female and male functions, or phenotypic gender (Brookes & Jesson 2010). Specifically, a shift in the ratio of female- to male-phase flowers in the population (floral sex ratio) during the flowering season should impose negative frequency-dependent selection for the opposite pattern of gender variation between early- than late-flowering plants. Generally omitted from such allocation studies are considerations of temporal investment in female- and male-phases of individual flowers, which can alter the gender expected from gamete production alone.
Are these perspectives mutually incompatible, or are they complementary views of largely independent components of the reproductive phenotype? That both perspectives have been supported by results from Stylidium (Armbruster et al. 2009; Brookes & Jesson 2010) seems more consistent with the latter interpretation. However, to date, among- and within-plant variation in floral traits, gender variation within flowering seasons and floral integration have generally been examined independently, and relevant studies have focused on morphological traits (including sex allocation). In contrast, in this study we assess their inter-relations and include sex-phase duration to consider its contribution to gender variation and daily reproductive effort, determine whether morphological and temporal sex allocation vary independently within and among plants, and examine the extent to which among-flower variation dilutes an individual’s floral integration.
To this end, we studied Delphinium glaucum S. Watson, which produces racemes with up to 80 protandrous, spurred, zygomorphic flowers that open from bottom to top within an inflorescence during 2–3 weeks, providing scope for among-flower variation in traits and gender variation during the flowering season. Because of protandry, the floral sex ratio in the population should shift from male- to female-dominated, so that negative frequency-dependent selection should favour relatively greater investment of resources and time in female function by early (lower) flowers and early-flowering plants than by late (upper) flowers and late-flowering individuals (Brunet & Charlesworth 1995; Brookes & Jesson 2010). If flower size also varies within and among plants, then its variation should primarily involve floral traits that act as pollinator signals, with greater consistency among flowers and plants for traits that govern pollen exchange with pollinators, such as the length of the nectar spur. Although zygomorphic flowers tend to be highly integrated (Armbruster et al. 1999), such size variation should weaken the extent of integration. Similarly, correlation of perianth and inflorescence traits with gynoecium and androecium traits (including sex-phase durations) should signal the relative importance of the former to female versus male function. Our results reveal pervasive, systematic variation within and among plants that may occur widely in protandrous species, especially those with staggered flowering among individuals.
Materials and methods
Study species and site
Delphinium glaucum is a perennial, hermaphroditic herb with purple, zygomorphic, spurred flowers. A flowering D. glaucum ramet produces 10–60 flowers on the main inflorescence axis (raceme), and inflorescences of some large plants also include a few basal, lateral branches with 1–10 flowers. Flowers on a main inflorescence axis open sequentially beginning with the basal flowers, and flowers on lateral branches open after most flowers on the main axis have wilted. Thus, flowering position corresponds to flowering order in the main axis. Each flower has five showy petaloid sepals and four less-showy petals (Fig. 1). The upper sepal envelopes two spurs, each formed by a nectar-producing petal. A flower typically presents 14–30 anthers, which dehisce sequentially during an approximately 4-day long male phase. The stigmas of a flower’s two or three carpels then become receptive for an approximately 2-day long female phase. Overall, the gynoecium produces 15–60 ovules.
We studied a large D. glaucum population in a meadow at Sibbald Flat, Alberta, Canada (51°02′ N; 114°49′ W) during June–July, 2005. During this period, the 9-day (the mean longevity of a D. glaucum flower) average temperature increased from 11.5 to 16.0 °C (see Fig. S1 in Supporting Information). Worker bumble bees, primarily Bombus californicus F. Smith and B. flavifrons Cresson, mainly pollinated this population during this study.
Dynamics of plant gender and the population sex ratio
To quantify variation in plant gender and the floral sex ratio (female flowers/all flowers) during the population’s flowering season, we recorded the flowering phenologies of 34 reproductive ramets separated by >3 m, which were chosen before the flowering season. For each plant, first flowering date and flower production were recorded to characterize phenology and plant size (in this population, flower number correlates positively with both basal stem area and stem height, r >0.5, P <0.001 in both cases, n =140; T.Y. Ida, A. Internicola, W.-J. Liao, L.A. Garibaldi & L.D. Harder, unpublished data). Approximately every 24 h, we counted the male- and female-phase flowers on each plant. We considered a flower as being in male phase if at least one anther remained undehisced because anther dehiscence usually starts soon after flower opening (anthesis) and each anther was depleted of pollen within a day of dehiscing. Stigma expansion generally coincided with the completion of male phase. Accordingly, the end of the male phase was considered as the start of the female phase. From this information, we extrapolated the daily population sex ratio (number of female flowers/total flowers) in the population.
From a functional perspective, a plant’s daily floral sex ratio depicts its gender incompletely, because it incorporates no information about the mating opportunities for female and male success, which depend on the population floral sex ratio (Lloyd 1980). Therefore, we estimated the phenotypic gender of plant i on day t according to
where fi,t and mi,t are its numbers of female- and male-phase flowers, respectively, and
(Lloyd 1980). The equivalence factor, Et, accounts for the relative opportunity of a plant’s male-phase flowers to sire seeds, given the prevailing female : male ratio within the population. Thus, plants with the same floral sex ratio could differ in gender, depending on the population context that determined their mating opportunities. We estimated a plant’s mean gender during the flowering season according to
where ni,t is the number of flowers displayed by plant i on day t. Note, that our estimates of phenotypic gender are probably conservative, because they do not incorporate within- and among-plant patterns of ovule and anther deployment that emphasize the minority sex role (see Results). As indicated below, we identified these patterns based on samples of flowers per inflorescence, whereas incorporation of ovule and anther counts in estimates of phenotypic gender requires these counts for all flowers.
We assessed each plant’s floral characteristics based on flowers collected from three positions within the inflorescence; at the middles of the lower, central and upper thirds of the main raceme. During each plant’s flowering period, we collected one flower from each of these positions, referred to as the LOW, MID and TOP flowers, respectively, to measure floral morphology (see below). We also monitored the phenology of an adjacent flower that opened on the same date (LOW′, MID′ and TOP′ flowers, respectively). Adjacent flowers share very similar characteristics, as is illustrated by strong correlations for stamen and ovule number between both sets of flowers (r >0.85 for all flower positions, n = 34). For the three collected flowers, we measured the lengths of the flattened petals and sepals with digital calipers and counted the anthers and ovules. As racemes bloom acropetally, LOW flowers opened before MID flowers, which in turn opened before TOP flowers. For paired perianth components, we measured the right-hand member, when possible.
To quantify flowering phenology, we counted the dehisced and pre-dehiscent anthers in LOW′, MID′ and TOP′ flowers approximately every 24 h from anthesis, until wilting. We measured each flower’s average rate of anther dehiscence, d, by the slope of the regression of the cumulative number of dehisced anthers on days since anthesis (see Fig. 2a). Given this rate, and a flower’s total anther number, a, the duration of its male phase was estimated as a/d. Female phase was assumed to last from the end of a flower’s male phase, until it wilted. We regarded a flower as wilting when more than two sepals had fallen, at which time other perianth parts fell easily with a slight touch. After a plant’s flowering season, we harvested the capsules of the LOW′, MID′, and TOP′ flowers. Seed set could not be counted because of widespread damage by dipteran seed predators; however, ovule number could be estimated for these flowers by counting ovule traces within capsules (this information was used to assess the correlation in characteristics among adjacent flowers within inflorescences).
We quantified variation within and among plants with the coefficient of variation (CV). Following Herrera (2009), we calculated these coefficients by partitioning the total variance for the LOW, MID and TOP flowers of the 34 study plants into within- and among-plant components, using restricted maximum likelihood. The square roots of these variance components were divided by the overall mean and multiplied by 100 to calculate the CVs. We used bootstrap techniques (Manly 1997) to estimate the 95% confidence intervals for the CVs based on 1000 re-samplings (with replacement) of the original data for individual plants (rather than individual flowers). Thus, whenever plant i was drawn for a pseudosample the observations for all three of its sampled flowers were included to retain the observed within-plant variation.
The effects of a plant’s first flowering date and size (flower number), and the positions of individual flowers (LOW, MID and TOP) on each floral and phenological trait were analysed with repeated-measure ancovas. The analysis for anther dehiscence rate also included anther number as a continuous independent variable. Plant size was removed from the final models, because it rarely had significant effects (there was only a marginally significant interaction between flowering date and size for female-phase duration).
We used correlation analyses to identify relations among morphological and phenological traits. Pearson correlations were used to measure associations among the lengths of perianth segments and stamen and ovule numbers for LOW, MID and TOP flowers as well as flower production per plant. For phenological associations, we assessed Pearson correlations for anthesis date and the durations of male- and female-phases for LOW′, MID′ and TOP′ flowers. We also used Pearson correlations to measure associations between all morphological and phenological traits. To illustrate the overall relation between plants’ average flower sizes and their positions in the population phenology, we regressed the first principal component (PC1) of each plant’s average floral traits for LOW, MID and TOP flowers against its first flowering date.
Floral integration was quantified as I =100V(λ)/Vmax, where V(λ) is the population variance of all eigenvalues from a principal-components analysis of floral traits and Vmax is the maximum possible variance (equal to the number of traits: Wagner 1984). This integration index ranges from 0 when traits are not correlated to 100 when all traits are perfectly correlated. We used bootstrap techniques (Manly 1997) to estimate the 95% confidence intervals for I based on 1000 re-samplings (with replacement) of the original data for individual plants. The integration index was compared with the value expected owing to sampling error alone, Irand = 100(m−1)/mn, where m is the number of sampled traits and n is the number of observations (Wagner 1984; also see Chevrud, Wagner & Dow 1989).
Pollination dependence of floral phenology
To test whether sex-phase duration and floral longevity varied independently of pollinator visitation we applied artificial pollination, artificial pollen removal and pollinator exclusion. We marked three flowers from the middle of the main inflorescences of 26 plants and assigned them randomly to three treatments: (i) the anthers of one flower were brushed gently with a small paintbrush during male phase to simulate pollen removal by pollinators; (ii) one flower was artificially cross-pollinated repeatedly every day during female phase; and (iii) one flower was covered with a small mesh bag to prevent pollinator visitation. Pollen used for cross-pollination was gathered (and mixed) from more than three plants growing >5 m from the recipient plants. For each flower, the durations of the male and female phases were calculated as described in Fig. 2a. The effects of pollination treatment and flowering phenology on floral longevity and male- and female-phase durations were analysed with repeated-measures ancovas, with treatment as a within-plant categorical factor and the plant’s date of first flowering as a covariate.
Flowering phenology and gender dynamics
Despite growing in a relatively uniform, flat meadow, plants differed extensively in their flowering phenology, creating seasonal variation in floral sex ratio within the population. Plants began flowering from late June to mid-July (Fig. 3a, horizontal lines) and flowering peaked in the population (Fig. 3a, histogram) a few days after the peak in the number of male-phase flowers (Fig. 3a, black stepped line). Because of this pattern and the strong protandry of D. glaucum flowers, the population floral sex ratio was male-biased early during the season and became increasingly female-biased (Fig. 3b, dashed line).
Variation in the phenology of individual flowers depended on both plant and floral characteristics. Overall, male phase lasted about 60% longer than female phase (F1,165 = 267.98, P <0.001); however, the difference in phase duration increased with flower position (phase × position; F2,165 = 18.19, P <0.001), because male duration increased with flower position (F2,165 = 14.97, P <0.001), whereas female duration decreased (F2,165 = 5.08, P <0.01; Fig. 2b). The contrasting effects of position on female- and male-phase duration counterbalanced each other, so that overall floral longevity did not differ significantly among positions (F2,66 = 1.88, P > 0.1). In general, flowers on plants that began flowering late lasted longer than those on early flowering plants (F1,32 = 6.82, P <0.025), with a 1-day increase in longevity for every 7.5 days that a plant’s initial flowering was delayed. This effect occurred consistently, regardless of flower position (first flowering date × position; F2,163 = 2.22, P >0.1). Durations of both sexual phases and floral longevity did not differ significantly between flowers that were excluded from pollinators, repeatedly hand pollinated, or subject to repeated pollen removal (P >0.05 in all cases). This consistency indicates that the effects of a plant’s date of first flowering on floral longevity (and male-phase duration) and of a flower’s position on male- and female-phase durations occurred independently of their direct effects on pollinator activity.
The interplay between plant and population dynamics of floral sex ratio and sex-phase duration created systematic patterns of phenotypic gender within plants, as the two plants depicted in Fig. 3b illustrate. Of the sampled plants, Plant 14 (grey dashed line) flowered first, so its initial flowers had no mates during their male phases. These flowers became disproportionately valuable when they shifted to female phase in an environment dominated by >60 male-phase flowers, so Plant 14’s phenotypic gender on its first day of functional flowering was female biased. Correspondingly, as Plant 14 flowered while the population sex ratio was primarily male biased, its average phenotypic gender was female dominant. In contrast, Plant 7 (Fig. 3b, grey dash-dotted line) began flowering last among the sample plants, so its initial male-phase flowers opened in an environment with many female-phase flowers, resulting in male-biased phenotypic gender, which persisted during most of its flowering period. Overall, such contrasting gender dynamics created a negative relation between a plant’s mean phenotypic gender and its date of first flowering (Fig. 4a). In contrast, mean phenotypic gender did not vary significantly with plant size, as measured by flower number (r = −0.183, P >0.25).
Among sampled plants, median phenotypic gender (Fig. 3b, black solid line) usually exceeded the overall proportion of female-phase flowers, which were in the minority during most of the flowering period (Fig. 3b, black dashed line), because flowers spent more time in male phase. This difference reflects the frequency dependence of mating, as represented by the equivalence factor in eqn 1. During peak flowering, phenotypic gender was generally balanced (i.e. Gp≈ 0.5; Fig. 3b), indicating equal opportunities for genetic contributions through male and female function. During early flowering, median phenotypic gender was generally male biased, because although plants with female-phase flowers had female-biased gender, most plants displayed only male-phase flowers. The opposite pattern occurred during late flowering, when most plants displayed only female-phase flowers.
Coefficients of within-plant floral variation were generally smaller than among-plant coefficients (Table 1). Among plants, the lengths of perianth segments varied less than the numbers of stamens and ovules and the durations of male and female phases. In contrast, within plants the CV for stamen number was almost as small as those for perianth lengths. Among all floral characteristics examined, female-phase duration varied most and spur length varied least both within and among plants.
Table 1. Means (±SD) and among-plant and within-plant coefficients of variation (95% CI) for aspects of the morphology and phenology of flowers from 34 Delphinium glaucum plants. See Fig. 1 for an explanation of the sepals and petals
*Data from LOW, MID and TOP flowers.
†Data from LOW′, MID′ and TOP′ flowers.
Length of sepal A* (mm)
13.33 ± 1.12
Length of sepal B* (mm)
13.05 ± 0.98
Length of sepal C* (mm)
20.93 ± 1.66
Length of petal A* (mm)
10.72 ± 0.82
Length of petal B (spur)* (mm)
17.02 ± 1.01
24.59 ± 3.17
24.02 ± 3.42
39.42 ± 8.15
39.24 ± 7.61
Male phase duration† (days)
4.13 ± 0.79
Female phase duration† (days)
2.57 ± 0.80
Floral longevity† (days)
7.92 ± 1.01
Except for spur length, late-blooming plants produced larger flowers (see Fig. 4b) with more stamens per flower than early-blooming plants (Table 2), even though plant size, as measured by flower number, did not vary significantly with date of first flowering (r = −0.183, 32 d.f., P >0.25). As the average phenotypic gender of plants declined with their first-flowering date (Fig. 4a), the more-male plants in the sample generally had larger flowers on average than the more-female plants (Fig. 4c; r = −0.473, 32 d.f., P <0.005). In contrast to perianth lengths and stamen number, a plant’s flowering time did not affect its average ovule number per flower, nectar-spur length, or anther-dehiscence rate (Table 2). Flower size (i.e. PC1) did not vary significantly among plants with total flower number (r =0.275, 32 d.f., P >0.1).
Table 2. Results of repeated-measures ancovas (F-tests) of the effects of first flowering date (partial regression coefficient, b ± SE) and flower position (adjusted means ± SE) on aspects of the morphology and phenology of Delphinium glaucum flowers
First flowering datea
b ± SE
Adjusted means ± SE
Letters accompanying adjusted means for each flower position indicate the outcomes of Tukey’s multiple comparisons; means that share the same letter do not differ significantly (α = 0.05). The analysis of anther dehiscence rate also included anther number as a continuous independent variable (see text). All models initially also included flower number as a measure of plant size and the flowering date × position interaction, but they were generally non-significant and so were excluded from the results presented here (see text for exceptions). See Fig. 1 for an explanation of the sepals and petals.
†P < 0.1; *P < 0.05; **P < 0.01; ***P < 0.001 for increase and ###P < 0.001 for decrease with order or season.
ad.f.N = 1, d.f.D = 32 for all analyses.
bd.f.N = 2, d.f.D = 66 for all analyses except anther dehiscence rate, for which d.f.N = 2, d.f.D = 65.
cData from LOW, MID and TOP flowers.
dData from LOW′, MID′ and TOP′ flowers.
Length of sepal Ac (mm)
0.189 ± 0.049
13.71 ± 0.16A
13.40 ± 0.16A
12.87 ± 0.16B
Length of sepal Bc (mm)
0.172 ± 0.042
13.33 ± 0.14A
13.07 ± 0.14A
12.75 ± 0.14B
Length of sepal Cc (mm)
0.249 ± 0.077
21.32 ± 0.26A
20.99 ± 0.26AB
20.48 ± 0.26B
Length of petal Ac (mm)
0.115 ± 0.037
10.80 ± 0.13A
10.74 ± 0.13A
10.60 ± 0.13A
Length of petal B (spur)c (mm)
0.099 ± 0.053
17.04 ± 0.17A
17.04 ± 0.17A
16.99 ± 0.17A
0.690 ± 0.136
24.12 ± 0.43A
24.85 ± 0.43A
24.82 ± 0.43A
0.661 ± 0.166
23.50 ± 0.50B
23.97 ± 0.50AB
24.59 ± 0.50A
0.814 ± 0.406
43.44 ± 1.25A
39.21 ± 1.25B
35.62 ± 1.25C
0.359 ± 0.409
43.20 ± 1.20A
39.06 ± 1.20B
35.44 ± 1.20C
Anther dehiscence rate (anthers/day)d
−0.359 ± 0.041
6.46 ± 0.15A
5.91 ± 0.15B
5.33 ± 0.15C
Male-phase durationd (days)
0.079 ± 0.029
3.68 ± 0.12C
4.12 ± 0.12B
4.57 ± 0.12A
Female-phase durationd (days)
0.032 ± 0.039
2.76 ± 0.13A
2.68 ± 0.13A
2.27 ± 0.13B
Floral longevityd (days)
0.133 ± 0.051
7.76 ± 0.16A
8.00 ± 0.16A
8.00 ± 0.16A
Except for the lengths of the petals, one of which forms the nectar spur, floral morphology and phenology varied significantly among flower positions within inflorescences. Ovule number, female-phase duration, anther-dehiscence rate and sepal lengths decreased significantly between LOW, MID and TOP flowers, whereas stamen number and male-phase duration increased significantly (or marginally: Table 2). In addition to the effect of flower position, anthers dehisced faster in flowers with many stamens (F1,65 = 12.32, P <0.001; partial regression coefficient ± SE = 0.115 ± 0.032: Fig. 2c). The contrasting influences of flower position on stamen number and dehiscence rate caused the positive association between flower position and male-phase duration described above (Fig. 2b). The effects of first flowering date and flower position interacted significantly only for stamen number per flower (F2,64 = 5.19, P <0.01 for LOW, MID and TOP flowers and F2,64 = 20.14, P <0.001 for LOW′, MID′ and TOP′ flowers, respectively) and the lengths of sepal B (F2,64 = 3.29, P <0.05) and petal A (F2,64 = 10.94, P < 0.001). These interactions occurred because the positive influence of a plant’s date of first flowering was stronger for lower flowers than for middle or upper flowers.
Aspects of floral morphology other than ovule number correlated positively with each other among plants and more strongly than with phenological traits; however, all correlations weakened with increasing flower position within inflorescences (Fig. 5). The most consistent morphological correlations among positions involved the lengths of the showy sepals with each other and with stamen number. Because of these correlation patterns, floral morphology was significantly integrated overall (I =20.9%, LCL = 12.8%, UCL = 28.8%; Irand = 2.52%), despite a slight decline in integration with increasing flower position (LOW flowers, I = 24.2%, LCL = 13.8%, UCL = 34.6%; MID flowers, I = 21.9%, LCL = 13.1%, UCL = 31.0%; TOP flowers, I = 19.9%, LCL = 13.0%, UCL = 28.1%; Irand = 2.52%).
Delphinum glaucum plants have highly dynamic reproductive phenotypes that differ among individuals, depending on their flowering time and associated mean gender. Within individuals, flower size decreased from lower to upper flowers, in concert with increased male effort in terms of both the relative production of anthers and ovules and the relative durations of male and female phases (Table 2). Among individuals, late-flowering individuals produced larger flowers with more anthers per ovule and spent longer in male phase than early-flowering plants, on average (Table 2). Both patterns are consistent with the expectations of negative frequency-dependent selection in a population in which floral sex ratio shifts from male- to female-biased because the constituent plants are protandrous. In general, these within- and among-plant patterns varied independently. Furthermore, despite significant intra-individual variation in floral morphology, floral integration was an order of magnitude greater than expected from random variation, although it was only a fifth as large as expected with perfect correlations among traits. We now consider key features of such within- and among-plant variation and their implications, as illustrated by our observations of D. glaucum.
The prevalence of dichogamy (Lloyd & Webb 1986) and extensive interspecific variation in floral longevity among angiosperms (Primack 1985) demonstrates that their reproductive phenotypes commonly include key temporal components. Our observations of D. glaucum reveal that this component varies systematically among flowers within inflorescences. Intriguingly, the relative time spent in male versus female phases increased from lower to upper flowers (Fig. 2b), indicating a temporal emphasis on male function as inflorescences age, which parallels the changes in stamen and ovule number. Changes in sex-phase duration among a plant’s flowers influence its gender both directly, by governing its instantaneous exposure of pollen and ovules, and indirectly, by affecting the population floral sex ratio. Given that protandry necessarily shifts the floral sex ratio from male- to female-biased, the benefits of prolonged male phase and truncated female phase should increase during the flowering season as male-phase flowers become relatively more valuable to individual plants. This reasoning also provides a consistent explanation for the prolonged male phase of flowers on late-flowering plants (Table 2), which had male-biased genders, compared to that of early-flowering plants, which had female-biased genders.
This increased emphasis on male phase within plants did not arise simply because upper flowers produced more stamens, as they also dehisced anthers slower, even though average temperature generally increased during the flowering season. This slowing of anther dehiscence is consistent with pollen-presentation theory, which predicts that: (i) limited pollen removal by individual pollinators enhances pollen dispersal when the proportion of removed pollen that is exported to other plants declines with increasing removal; and (ii) restricted removal is most beneficial when pollinators are abundant (Harder & Thomson 1989). Delphinium glaucum flowers effectively restrict pollen removal through the staggered dehiscence of an average of 24 anthers (Table 1) during 3–5 days (Fig. 2b). In this population, pollinator visitation is moderate (median = 0.16 probes per flower per hour: Ishii & Harder 2006) and increases during the population flowering period as new bees recruit (L.D. Harder pers. obs.). The latter effect would favour increased restriction of pollen removal by later (upper) flowers on inflorescences, as reflected in slower anther dehiscence. Sargent (2003) also reported a positive relation of male-phase duration to pollinator abundance for Chamerion angustifolium, although she observed seasonal shortening of male phase in association with declining pollinator abundance.
Flower size and integration
Like sex-phase duration, the size of most perianth segments varied with flowering time; however, the patterns differed among traits within and among plants, as sepal lengths (but not petal lengths) decreased within plants, whereas the average lengths of the sepals and Petal A increased among plants (Table 2). The declining relations within plants are consistent with spatial proximity and temporal priority to resources transported from below the inflorescence (reviewed by Herrera 2009), as well as with intrinsic (architectural) limitations associated with floral development (Diggle 1995, 2003 and references therein). However, this explanation should apply equally to the showy sepals and petals, whereas petal lengths did not vary significantly among flower positions and they varied least among the traits we measured (Tables 1 and 2). These differences and the generally weaker correlations of the two petal types with each other and with the sepals than among the sepals themselves (Fig. 5) instead suggest explanations related to their pollination functions and functional specialization.
Delphinium petals and sepals serve different functions and so may be subject to contrasting selection. Together with the number of open flowers, the showy sepals contribute to a plant’s long-distance signals, and plants with shortened sepals attract fewer pollinators (Ishii & Harder 2006). Large sepals likely contribute most to this role as inflorescences begin flowering, rather than later when inflorescences display more flowers. In addition, male function generally benefits more than female function from multiple pollinator visits (see Bell & Cresswell 1998), a relation that likely holds for D. glaucum, given that flowers spend 60% longer time in male than female phase. Consequently, late-flowering plants, which have male-biased gender, may experience stronger selection for large sepals than early-flowering plants, with female-biased gender. In combination, these benefits would be greatest for initial (lower) flowers on late-flowering plants; an explanation consistent with the interaction between flower position and a plant’s date of first flowering observed for Sepal B. In contrast to the sepals, Petal A acts as a nectar guide and forms the entrance to the nectar spur and Petal B covers the stamens and stigmas externally and forms the nectar spur internally. Because of their roles in physical interaction with pollinators, the petals may be subject to pollinator-mediated stabilizing selection, an interpretation that is supported by both petals exhibiting the smallest coefficients of variation of all measured traits, both within and among plants (see Cresswell 1998). Similar patterns have been observed for the gynostemium in an orchid species (Ushimaru & Nakata 2001), the pollination unit of Iris species (Ishii & Morinaga 2005) and corolla tubes of tubular flowers (e.g. Conner & Via 1993).
Delphinium glaucum flowers are significantly integrated, largely because of the strong correlations among the showy sepals and between them and stamen number (Fig. 5). The >20% integration is in the upper end of the range of observations for other species (Armbruster et al. 2009) and is consistent with the preceding functional interpretations, which seem more likely than developmental explanations (reviewed by Armbruster et al. 2004), given that the sepal and stamen whorls are separated by the petals, with which they correlated weakly (Fig. 5). Nevertheless, the 21% overall integration of D. glaucum flowers is much less than the maximum possible with perfect correlations among organs (100%). That the heterogeneity in organ size and number within and among plants varies significantly with flowering time suggests that it and the associated variation in plant gender limit the extent of floral integration, either because of plasticity or selection for phenotypic diversity.
Gender dynamics and selection
The dynamic nature of floral morphology, sex allocation and gender within and among plants pervades our findings for D. glaucum. Three hierarchical features govern these dynamics; plants differ in their flowering periods; flowers open sequentially, rather than simultaneously within inflorescences; and male function precedes female function within flowers. Together, these temporal features cause the floral sex ratio in the population to shift from male- to female-biased, in turn placing a premium on female function during early flowering and on male function during late flowering, both within and among plants (also see Brunet & Charlesworth 1995; Brookes & Jesson 2010). As indicated above, these dynamics provide consistent explanations for the observed variation in floral morphology, sex allocation and intrafloral phenology, suggesting that variation has been influenced by selection for enhanced reproduction. The flowering dynamics exhibited by D. glaucum are typical of protandrous species (e.g. Brunet & Charlesworth 1995; Sargent & Roitberg 2000; Aizen 2001), so that their consequences should be equally widespread.
Both density- and frequency-dependent selection could shape reproductive phenotypes in dichogamous populations. Devaux & Lande (2010) demonstrated theoretically that adaptive variation in flowering time among individuals balances the benefits of synchronous flowering for pollinator recruitment to the population against the associated dilution of visitation frequency per plant. They also reviewed considerable evidence that flowering time is heritable, so that among-plant variation in flowering time may commonly reflect a history of such density-dependent selection. Staggered anthesis within plants may also commonly be adaptive, as floral display size balances the attractive benefits of large displays against costs owing to self-pollination among flowers and the associated pollen discounting and inbreeding depression (Harder & Barrett 1995); a balance that can be modified by the extent to which dichogamy reduces geitonogamy (Harder, Barrett & Cole 2000). When coupled with dichogamy, flowering dynamics within and among plants consistently shift the population floral sex ratio and thereby induce frequency-dependent selection on sex allocation, intra-floral phenology and floral traits that serve female and male function differentially (Brunet & Charlesworth 1995; Brookes & Jesson 2010; this study). Responses to such selection should generate genetic correlations between flowering phenology, on one hand, and sex allocation and floral morphology, on the other, as well as favouring sex-role specialization among and within plants to the extent allowed by the overall benefits of hermaphroditism. Such selection should be facilitated by the assortative mating that necessarily accompanies non-synchronous flowering (Weis et al. 2005). These expectations await explicit assessment of the dynamics of reproductive performance in populations of dichogamous plants.
We thank S.C.H. Barrett for comments on the manuscript. This research was funded by a Postdoctoral Fellowship (0124) from the Japan Society for the Promotion of Science for Research Abroad (HSI) and a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada (LDH). We have no conflict of interest to declare.