Oecologia

, Volume 146, Issue 1, pp 68–76

Variation in resource limitation of plant reproduction influences natural selection on floral traits of Asclepias syriaca

Authors

    • Department of BiologyGrinnell College
    • Department of MathematicsGrinnell College
    • Department of Integrative BiologyUniversity of Guelph
  • Davin L. D. Remington
    • Department of BiologyGrinnell College
    •  
  • Kate E. Ostergren
    • Department of BiologyGrinnell College
    •  
Plant Animal Interactions

DOI: 10.1007/s00442-005-0183-4

Cite this article as:
Caruso, C.M., Remington, D.L.D. & Ostergren, K.E. Oecologia (2005) 146: 68. doi:10.1007/s00442-005-0183-4
  • 153 Views

Abstract

The availability of both pollen and resources can influence natural selection on floral traits, but their relative importance in shaping floral evolution is unclear. We experimentally manipulated pollinator and resource (fertilizer and water) availability in the perennial wildflower Asclepias syriaca L. Nine floral traits, one male fitness component (number of pollinia removed), and two female fitness components (number of pollinia inserted and number of fruits initiated) were measured for plants in each of three treatments (unmanipulated control, decreased pollinator access, and resource supplementation). Although decreasing pollinators’ access to flowers did result in fewer pollinia inserted and removed, fruit set and phenotypic selection on floral traits via female and male fitness did not differ from the control. In contrast, resource supplementation increased fruit set, and phenotypic selection on seven out of nine floral traits was stronger via female than male fitness, consistent with the prediction that selection via female fitness would be greater when reproduction was less resource-limited. Our results support the hypothesis that abiotic resource availability can influence floral evolution by altering gender-specific selection.

Keywords

Female fitnessMale fitnessPhenotypic selectionPollen limitation

Introduction

Seed production in plants can be limited by the availability of pollen and abiotic resources. Pollen limitation of seed production can be the result of fewer pollinator visits or a decrease in the quantity or quality of pollen deposited per visit (Ashman et al. 2004). Resource limitation of seed production can be the result of inadequate soil nutrients, water, or light (reviewed by Willson and Burley 1983). Although pollen- and resource-limitation are often considered to be alternative hypotheses for inadequate seed production in plants, theory suggests that at equilibrium plants should be limited by both pollen and resources (Haig and Westoby 1988).

Understanding the relative importance of pollen and resources in limiting seed production of hermaphroditic plants is important because they have different effects on the strength of natural selection on floral traits. As reproduction becomes more pollen-limited, floral traits should be under stronger selection via female fitness because plants with characters that enhance pollen receipt should produce more fruits (Johnston 1991a, 1991b; Ashman and Morgan 2004). Pollen-limitation is also hypothesized to influence the strength of natural selection on floral traits via male fitness. Assuming that pollen-limited plants receive fewer pollinator visits, floral traits that enhance pollen export should be under stronger selection via male fitness as reproduction becomes more pollen-limited (e.g., Stanton and Preston 1988). In contrast, as reproduction becomes more resource-limited, floral traits should be under weaker selection via female fitness because plants with characters that enhance pollen receipt will not have the resources to take advantage of high pollen loads (e.g., Totland 2001, 2004). Based on this hypothesized effect of resources on natural selection, stronger resource-limitation of reproduction should result in stronger selection via male than female fitness (e.g., Maad and Alexandersson 2004).

In a recent review, Ashman and Morgan (2004) found strong evidence for a positive relationship between the strength of selection on floral traits via female fitness and the degree of pollen-limitation. In contrast, only two studies have examined whether resource limitation influences gender-specific selection on floral traits (Totland 2001; Maad and Alexandersson 2004). Selection via female fitness on flower size of Ranunculus acris was weaker for plants growing at low temperatures, where resources for seed production were more limited (Totland 2001). Selection on flower number of Platanthera bifolia was stronger via male than female fitness in a drought year when resources for reproduction were limited, but not in a non-drought year (Maad and Alexandersson 2004). To our knowledge, there are no empirical data on the effect of pollen-limitation on the strength of selection on floral traits via male fitness.

We studied the effect of decreasing pollinator access and supplementing resources on selection on floral traits of the common milkweed, Asclepias syriaca. Unlike most plant species, members of the Asclepiadaceae such as A. syriaca package their pollen grains into discrete packets known as pollinia (Woodson 1954) whose removal can be used as a proxy for male fitness (Broyles and Wyatt 1990). In addition, significant selection on floral traits of A. syriaca via both male and female fitness has been documented (Morgan and Schoen 1997). Finally, the reproduction of Asclepias spp. can be pollen- and resource-limited (Wyatt and Broyles 1994), suggesting that the evolution of its floral traits could be influenced by variation in either pollination, resource availability, or both. By experimentally manipulating a natural population of A. syriaca and measuring selection on floral traits, we tested the following predictions:
  1. 1.

    Natural selection on floral traits via male fitness is stronger when pollinators are limiting.

     
  2. 2.

    Natural selection on floral traits via female fitness is stronger when pollinators are limiting and when resources are abundant.

     
  3. 3.

    When resources are abundant, natural selection on floral traits via female fitness should be stronger than selection via male fitness.

     

Methods

Study system

A. syriaca L. (Asclepiadaceae) is a perennial wildflower common to disturbed habitats throughout northeastern North America (Woodson 1954). Plants in central Iowa bloom from late May to early July and produce 1–10 inflorescences per plant (Caruso, unpublished data). Although A. syriaca has an indeterminate growth form (Wilbur 1977), the number of inflorescences appeared to be predetermined at our study site (Caruso, personal observation). An inflorescence contains 1–250 (Caruso, unpublished data) self-incompatible, insect-pollinated flowers (Willson and Bertin 1979), each of which has a corona consisting of five nectar-bearing hoods and associated horns (Fig. 1). A milkweed flower produces five pollinaria, each of which consists of two pollinia connected by translator arms to a corpusculum.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-005-0183-4/MediaObjects/442_2005_183_Fig1_HTML.gif
Fig. 1

Top view of an A. syriaca flower. Flowers are pentamerous, with five nectar filled hoods (outlined by points 1–5), each subtended by a single horn (from point 6 to 7). The corpusculum is at point 8, whereas the stigmatic slit where pollinia are inserted is at point 9. Hood length was measured as the distance from point 1 to the midpoint of points 4 and 5. Hood width was measured as the distance from point 2 to 3. Hood separation was measured as the distance from point I-3 to II-2, etc. Horn length was measured as the distance from point 6 to 7. Floral landmarks used to measure corpusculum length and stigmatic slit width are not shown. All floral landmarks follow Morgan and Schoen (1997)

When an insect visits, it brushes up against a corpusculum, which is located between two of the hoods, and inadvertently pulls a polliniarium off the flower. As the translator arms dry out, the pollinia are reoriented into a position that allows them to be inserted into one of five stigmatic slits (Wyatt and Broyles 1994). A. syriaca flowers have an average lifespan of five days (Wyatt and Broyles 1994) and all flowers within an inflorescence do not open simultaneously (Caruso, personal observation). In central Iowa, 0–10 mature fruits per plant are produced from late July to early September (Caruso, unpublished data).

Experimental design

We studied a large population of A. syriaca growing at the Jacob Krumm Nature Preserve, Jasper Co., Iowa, USA (41°42.3′N, 92°47.2′W). Because A. syriaca can be highly clonal (Wyatt and Broyles 1994), we systematically sampled plants throughout the population to maximize the probability that individuals included in the study were independent of each other. On 5 June 2000, we ran 15, 50 m long transects, each spaced 2 m apart, into the study population. We then systematically sampled 450 A. syriaca for inclusion in the study by walking each transect and flagging one plant with buds approximately every meter, up to a maximum of 30 per transect. We only selected plants that had initiated inflorescences, but not yet flowered.

To ensure interspersion of treatments, we systematically assigned equal numbers of marked A. syriaca plants to the control (C), decreased pollinator access treatment (−P), and resource supplementation treatment (+R). This single-factor design does not allow us to test for interactive effects of pollinator and resource limitation on selection on floral traits. Control plants were not manipulated. Inflorescences of plants in the −P treatment were covered with bridal veil bags (as in Queller 1985) for two days and then left uncovered for two days for the duration of the experiment (19 June to 30 July 2000), reducing the amount of time that flowers were available to pollinators by 50%. Given that individual flowers are open for an average of five days (Wyatt and Broyles 1994), four days does not cover the entire life of an inflorescence. In addition, we acknowledge that reducing the amount of time that flowers were available to pollinators by 50% is not identical to reducing the overall abundance of pollinators in the population. Bagging can increase nectar volume and decrease nectar concentration in A. syriaca flowers (Wyatt et al. 1992), but it is not known whether bagging influences floral morphology.

On 5 June 2000, each plant in the +R treatment received 120 g of 180-day 14-14-14 NPK slow release fertilizer (Osmocote) and approximately 1.5 l of water. The same volume of water was applied to +R plants on 13, 21, and 30 June 2000. Because there was above-average precipitation during July 2000 (Herzmann 2004), further water was not applied to the +R plants. Because we applied the +R treatment after inflorescence initiation (Caruso, personal observation), our resource supplementation treatment was unlikely to influence natural selection by increasing flower number or changing flowering phenology.

Data collection

We measured date of first flower, length of flowering period, and total flower number of all 450 plants included in the experiment. These traits were measured because they are under significant selection in other plant species (e.g., Johnston 1991a; O’Connell and Johnston 1998; Caruso 2000). Plants were checked every day from 19 June until 30 July 2000 to determine if they were flowering. We used this information to calculate the date of first flower (relative to 19 June 2000) and the length of the flowering period. Total flower number was estimated as the total of number of open flowers and unopened buds on all inflorescences. For 101 of the plants, all of the flowers on one or more inflorescences senesced before they could be counted. In these cases, we estimated the number of flowers on a senesced inflorescence as the mean number of flowers on the other inflorescences on that plant.

The six floral morphology traits were measured on approximately 65 plants per treatment group. We collected flowers for morphology measurements from one of the last two inflorescences to flower on each plant. If the inflorescence contained ≥50 flowers, we collected 15 for later analysis. Only ten flowers were collected if the inflorescence contained <50 flowers. All flowers were preserved in jars that contained a sponge soaked in 25 ml of a 1:3 solution of acetic acid and ethanol (as in Morgan and Schoen 1997). Because A. syriaca flowers are small, all floral morphology traits were measured using a digital imaging system (NIH Image for the Macintosh and a Leica MZ8 microscope w/digital camera). We captured lateral and top views of five flowers per plant. Hood length, hood width, hood separation, horn length, stigmatic slit width, and corpusculum length were measured for each flower (see Fig. 1 for floral landmarks used for these measurements). These traits were chosen because they are under significant selection in other populations of A. syriaca (Morgan and Schoen 1997). We averaged the five measurements of hood, horn, and stigmatic slit characters for each flower. When more than one corpusculum remained on a flower, these measurements were also averaged.

We estimated both male and female fitness of A. syriaca. Because each flower produces ten pollinia that are attached in pairs to five highly visible corpuscula, male fitness was estimated as twice the number of corpuscula removed (range=0–10). Number of pollinia removed is strongly correlated with the number of seeds sired in Asclepias exaltata (Broyles and Wyatt 1990), suggesting that it is a reasonable index of male fitness. Female fitness was estimated by probing the five stigmatic slits and counting the number of slits (range=0–5) that contained a pollinium. We estimated pollinia removed and inserted for 10–15 flowers/plant (depending on how many were collected) and approximately 65 plants per treatment. An additional measure of female fitness was estimated by counting the number of fruits initiated on plants in all treatments as of early August 2000. Although the abortion of fertilized fruits is common in Asclepias (Wyatt and Broyles 1994), the number of initiated and mature fruits is correlated in A. syriaca (e.g., Willson and Rathcke 1974). We did not measure the number of seeds per fruit, but seeds per fruit and fruits per plant are uncorrelated in A. syriaca (Lipow and Wyatt 1999). The two published studies that have estimated maternal and paternal effects on fruit set of Asclepias spp. find no evidence for female choice of pollinia via abortion of fertilized fruits (Morse and Schmitt 1991; Lipow and Wyatt 1999).

Data analysis

We used one-way analysis of variance (ANOVA) to determine if floral traits differed among treatments. When the ANOVA for a floral trait was significant, we used Tukey’s HSD test to detect pairwise differences between treatments. The assumptions of normality and homogeneity of variance for all ANOVAs presented in this paper were tested using Lilliefors’ (Wilkinson 1997) and Levene’s (Underwood 1997) tests, respectively.

We used one-way ANOVA to determine if fitness differed among treatments. Because we had a priori hypotheses as to how fitness would differ between the three treatments, we used planned 1-df contrasts to test whether (1) plants in the −P treatment had fewer pollinia removed and inserted than those in the control, (2) plants in the −P treatment initiated fewer fruits than those in the control, and (3) plants in the +R treatment initiated more fruits than those in the control. In addition, we used a 1-df contrast to test the assumption that the +R treatment did not influence the number of pollinia removed or inserted. Although some of our predictions about fitness variation among treatments were directional, the P-values associated with F-tests are inherently many-tailed (e.g., Motulsky 1995).

We calculated standardized directional phenotypic selection differentials for each combination of floral trait, treatment, and fitness measure using a series of simple linear regression models (reviewed in Conner 2001; Kingsolver et al. 2001; Conner and Hartl 2004). Phenotypic selection differentials include both direct selection on a trait and indirect selection on the trait via correlated characters. We calculated relative fitness (the dependent variable) by dividing by mean fitness (Lande and Arnold 1983) and standardized each floral trait (the independent variable) to a mean=0 and variance=1 (Sokal and Rohlf 1995). When fitness is regressed on a trait, the slope of the simple linear regression line is equivalent to the directional phenotypic selection differential for that trait (Conner 1988). The directional selection differential calculated using this regression approach is mathematically equivalent to the differential calculated by subtracting the mean trait value for selected individuals from the mean trait value for the whole population (Lande and Arnold 1983; Conner 1988). Because of sample size variation, our power to detect significant selection differentials varied among floral traits. To control for inflated Type I error rates associated with multiple comparisons among correlated traits, we adjusted P-values for the selection differentials using the sequential Bonferroni correction by the Dunn–Sidak method (Sokal and Rohlf 1995). P-values were adjusted within each combination of treatment and fitness measure. The assumption of normality of residual variance for all regression models presented in this paper was tested using Lilliefors’ test (Wilkinson 1997). We tested the assumption of homogeneity of residual variance for all of our regression models by calculating the Spearman rank correlation between the residuals and relative fitness (Neter et al. 1990).

We calculated standardized directional phenotypic selection gradients for each treatment and fitness measure using a series of multiple regression models (reviewed in Conner 2001; Kingsolver et al. 2001; Conner and Hartl 2004). Phenotypic selection gradients estimate direct selection on each floral trait, controlling for indirect selection via correlated traits. Fitness measures were relativized and floral traits standardized separately for each treatment. When a fitness measure is regressed on all of the floral traits, the partial regression coefficients are equivalent to the directional phenotypic selection gradients for those traits (Lande and Arnold 1983). We ran a total of three treatments × three fitness measures=nine multiple regression models. Only four out of 26 correlations among floral traits were significantly >0.70 (data not shown), suggesting that our selection gradients were not biased by multicollinearity.

Analysis of covariance (ANCOVA) was used to determine if selection differentials and gradients differed between treatment groups (e.g., Dudley 1996). Data from all treatment groups were combined and analyzed with a model that included continuous terms for the floral traits, a categorical term coding for treatment, and floral trait × treatment terms. Relative fitness was the dependent variable. We calculated relative fitness and standardized floral traits separately within each treatment. We then used 1-df contrasts on the floral trait × treatment term to test a priori hypotheses that selection on floral traits via all three fitness measures will be stronger for plants in the −P treatment relative to the control and that selection on floral traits via number of pollinia inserted and number of fruits initiated will be stronger for plants in the +R treatment relative to the control. Single degree of freedom contrasts were also used to test whether selection on floral traits via number of pollinia removed differs between the control and +R treatment.

We used t-tests (Zar 1999) to determine whether phenotypic selection on floral traits via female fitness was stronger than selection via male fitness when resources were more abundant (the +R treatment). Two-tailed t-tests were used to compare (1) selection differentials via pollinia inserted vs pollinia removed and (2) selection differentials via pollinia removed vs fruits initiated, within the +R treatment.

This selection study, like many of those reviewed by Kingsolver et al. (2001), has relatively small sample sizes and thus likely low statistical power to detect small but significant selection differentials and gradients. However, our sample sizes were similar across treatments, suggesting that our power to detect the effects of pollen and resource availability on selection should be similar. In general, the power to detect significant selection gradients is less than the power to detect significant selection differentials, even if the sample size is identical (Conner 1988).

Results

Floral traits

Six out of nine floral traits, including length of flowering period, date of first flower, total flower number, hood length, hood separation, and horn length, did not vary among treatment groups. Plants in the control treatment produced flowers with approximately 3% wider hoods than those in the −P treatment. In contrast, plants in the +R treatment produced flowers with 1.5% wider hoods than those in the control. Plants in the −P and +R treatments produced flowers with narrower stigmatic slits and shorter corpuscula than those in the control (Table 1).
Table 1

Average floral traits and fitness measures (±1 SE) for A. syriaca plants in each of three experimental treatments

Trait

Treatment

Unmanipulated control (C)

Decreased pollinator access (−P)

Resource supplementation (+R)

Length of flowering period (d)

14.8±0.4a

15.7±0.4a

14.7±0.4a

Date of first flower

3.0±0.2a

3.3±0.3a

3.6±0.3a

Total flower number

263.5±15.3a

280.8±17.0a

243.3±13.7a

Hood length (mm)

2.53±0.054a

2.49±0.058a

2.52±0.058a

Hood width (mm)

1.96±0.023a

1.90±0.028a,b

1.99±0.023a,c

Hood separation (mm)

1.18±0.030a

1.21±0.034a

1.14±0.033a

Horn length (mm)

1.90±0.042a

1.84±0.042a

1.80±0.041a

Stigmatic slit width (mm)

0.60±0.004a

0.55±0.006b,d

0.56±0.005c,d

Corpusculum length (mm)

0.53±0.007a

0.50±0.009b,d

0.49±0.005c,d

Number of pollinia inserted

0.57±0.05

0.34±0.05†

0.68±0.08

Number of pollinia removed

4.0±0.2

2.5±0.3†

3.8±0.2

Number of fruits initiated

4.1±0.2

4.3±0.3

6.0±0.3†

Each floral trait was compared among treatments using univariate ANOVA followed by a Tukey post hoc test. Fitness in the -P and +R treatments were compared to the unmanipulated control using univariate ANOVA followed by 1-df contrasts. Contrasting superscripts indicate that a floral trait differed significantly (P< 0.05) between treatments. A dagger (†) indicates that a fitness measure differed significantly (P<0.05) from the unmanipulated control. N=140–150 for length of flowering period, date of first flower, total flower number, and number of fruits initiated. N=60–62 for all other floral traits and fitness measures

Phenotypic selection on A. syriaca: general patterns

There was significant directional phenotypic selection on all floral traits. Approximately 30% of the selection differentials and 40% of the selection gradients were negative (i.e., selection for smaller trait values), although only a few of these were significantly different from zero (Tables 2 and 3). Phenotypic selection via our two female fitness measures (pollinia inserted and fruits initiated) was congruent for only one floral trait (corpusculum length; Table 3). Selection via male and female fitness (pollinia removed and inserted) was congruent for one out of nine floral traits. Specifically, resource supplemented plants with longer hoods had more pollinia both removed and inserted (Table 2).
Table 2

Standardized directional selection differentials (±1 SE) for nine floral traits and three fitness measures of A. syriaca in each of three experimental treatments. These differentials were estimated as the slope of the simple linear regression line between each trait and each fitness measure

Fitness measure

Floral trait

Treatment

Unmanipulated control (C)

Decreased pollinator access (−P)

Resource supplementation (+R)

Pollinia removed

 

Length of flowering period (d)

−0.046±0.060

0.086±0.111

−0.060±0.048

Date of first flower

0.097±0.059

0.083±0.111

0.053±0.049

Total flower number

0.009±0.061

0.051±0.112

−0.016±0.051

Hood length (mm)

0.010±0.061

0.195±0.109

0.188±0.045*

Hood width (mm)

0.115±0.059

−0.061±0.112

0.121±0.048

Hood separation (mm)

−0.048±0.060

0.164±0.110

0.109±0.049

Horn length (mm)

−0.088±0.059

0.068±0.111

0.100±0.049

Stigmatic slit width (mm)

0.090±0.059

−0.010±0.112

−0.063±0.050

Corpusculum length (mm)

0.034±0.060

−0.027±0.112

0.015±0.051

Pollinia inserted

 

Length of flowering period (d)

0.017±0.086

−0.058±0.158

−0.173±0.120

Date of first flower

0.028±0.086

−0.055±0.158

−0.132±0.121

Total flower number

0.063±0.086

−0.046±0.145

−0.043±0.121

Hood length (mm)

0.094±0.085

0.165±0.144

0.405±0.110*

Hood width (mm)

0.219±0.081

−0.218±0.143†

0.188±0.119

Hood separation (mm)

0.027±0.086

0.180±0.144

0.318±0.114*

Horn length (mm)

0.002±0.086

0.118±0.145

0.352±0.113*†

Stigmatic slit width (mm)

0.048±0.086

0.029±0.146

0.024±0.121

Corpusculum length (mm)

0.171±0.083

0.177±0.144

−0.126±0.120†

Fruits initiated

 

Length of flowering period (d)

0.413±0.049*

0.466±0.051*

0.402±0.045*

Date of first flower

−0.215±0.057*

−0.184±0.062*

−0.135±0.055

Total flower number

0.371±0.051*

0.433±0.055*

0.306±0.050*

Hood length (mm)

0.053±0.089

0.222±0.096

0.134±0.088

Hood width (mm)

0.092±0.089

0.129±0.099

0.169±0.087

Hood separation (mm)

0.027±0.089

0.159±0.098

−0.031±0.090

Horn length (mm)

−0.035±0.089

0.086±0.099

0.074±0.089

Stigmatic slit width (mm)

−0.033±0.089

−0.335±0.090*†

0.115±0.098

Corpusculum length (mm)

0.064±0.089

−0.180±0.097

0.130±0.088

Asterisks (*) indicate whether selection within a treatment was significantly (P < 0.05) different from zero after sequential Bonferroni correction by the Dunn–Sidak method. A dagger (†) indicates that selection differed significantly (P< 0.05) from the unmanipulated control by a 1-df contrast. N=139–150 for selection on length of flowering period, date of first flower, and total flower number via fruits initiated. N=60–62 for all other combinations of floral traits and fitness measures

Table 3

Standardized directional selection gradients (±1 SE) for nine floral traits and three fitness measures of A. syriaca in each of three experimental treatments. These gradients were estimated as partial regression coefficients from multiple regressions of all floral traits on each fitness measure

Fitness measure

Floral trait

Treatment

Unmanipulated control (C)

Decreased pollinator access (−P)

Resource supplementation (+R)

Pollinia removed

 

Length of flowering period (d)

−0.088±0.099

0.072±0.175

−0.075±0.070

Date of first flower

0.036±0.083

0.129±0.122

0.033±0.059

Total flower number

0.092±0.104

0.005±0.176

0.005±0.063

Hood length (mm)

0.236±0.160

0.826±0.315*†

0.219±0.117

Hood width (mm)

0.067±0.107

−0.345±0.145*†

0.004±0.075

Hood separation (mm)

−0.005±0.114

−0.114±0.199

0.021±0.088

Horn length (mm)

−0.291±0.119*

−0.553±0.235*

−0.034±0.093

Stigmatic slit width (mm)

0.046±0.086

0.072±0.139

−0.070±0.048

Corpusculum length (mm)

0.026±0.062

−0.079±0.144

0.091±0.054

Pollinia inserted

 

Length of flowering period (d)

−0.062±0.135

−0.082±0.233

−0.533±0.173**

Date of first flower

−0.091±0.114

−0.114±0.162

−0.295±0.146*

Total flower number

−0.023±0.142

−0.203±0.233

0.057±0.157

Hood length (mm)

0.071±0.218

0.524±0.418

0.149±0.290

Hood width (mm)

0.409±0.147**

−0.310±0.192†

0.105±0.187

Hood separation (mm)

0.117±0.155

0.220±0.264

0.123±0.218

Horn length (mm)

−0.243±0.162

−0.337±0.312

0.192±0.230

Stigmatic slit width (mm)

−0.193±0.117

−0.097±0.184

0.004±0.119

Corpusculum length (mm)

0.240±0.085**

0.213±0.191

0.061±0.135

Fruits initiated

 

Length of flowering period (d)

0.308±0.097**

0.287±0.122*

0.362±0.113**

Date of first flower

0.001±0.082

0.044±0.085

0.108±0.094

Total flower number

0.306±0.102**

0.189±0.122

0.130±0.104

Hood length (mm)

0.462±0.156**

0.542±0.219*

0.463±0.194*

Hood width (mm)

−0.148±0.105

−0.073±0.101

−0.143±0.123

Hood separation (mm)

−0.162±0.111

−0.323±0.138*

−0.282±0.147

Horn length (mm)

−0.366±0.116**

−0.274±0.163

−0.141±0.151

Stigmatic slit width (mm)

0.059±0.084

−0.193±0.097*

0.113±0.079

Corpusculum length (mm)

0.136±0.061*

−0.061±0.100

0.070±0.091

N

60

62

61

Asterisks (*) indicate whether selection within a treatment was significantly different from zero. A dagger (†) indicates that selection differed significantly (P< 0.05) from the unmanipulated control by a 1-df contrast

*P<0.05, **P<0.01, ***P<0.001

Effect of decreased pollinator access

As would be expected if they received fewer pollinator visits, plants in the −P treatment had 41% fewer pollinia inserted and 38% fewer pollinia removed relative to the control. However, this reduction in number of pollinia inserted did not translate to lower fruit set, as the number of fruits initiated did not differ between the C and −P treatment groups (Table 1). Although we predicted that phenotypic selection on floral traits should be stronger for plants in the −P treatment relative to the control, this was true for only five out of 27 combinations of floral traits and fitness measures. Selection differentials via pollinia removed on hood separation was stronger in the −P treatment relative to the control (Table 2). Selection gradients on hood length and width via the same fitness measure were also stronger in the −P treatment (Table 3). Finally, selection differentials on slit width and corpusculum length via fruits initiated were stronger in the −P treatment relative to the control (Table 2). In addition, both selection gradients and differentials for hood width differed in direction but not magnitude between the C and −P treatment groups (Tables 2 and 3).

Effect of resource supplementation

The number of pollinia inserted and removed did not differ between the C and +R treatments, but +R plants initiated 32% more fruits (Table 1). Despite this difference in fruit set, phenotypic selection on floral traits via fruits initiated and pollinia inserted was not stronger for plants in the +R treatment relative to the control. None of the selection gradients or differentials via fruits initiated differed between the C and +R treatments (Tables 2 and 3). Horn length was the only trait for which selection via pollinia inserted was stronger in the +R treatment relative to the control (Table 2).

Relative strength of selection via male vs female fitness

As predicted, the strength of phenotypic selection on floral traits was stronger via female than male fitness when reproduction was less resource-limited (the +R treatment). Selection differentials for length of flowering period (t=2.33; df=120; P<0.05), hood length (t=2.35; df=120; P<0.05), hood separation (t=2.15; df=120; P<0.05), and horn length (t = 2.62; df = 120; P < 0.01) were stronger via pollinia inserted than pollinia removed. Selection differentials on length of flowering period (t=7.12; df=203; P<0.001), date of first flower (t=2.42; df=203; P<0.02), total flower number (t=4.49; df=202; P<0.001), and stigmatic slit width (t=2.31; df=119; P<0.05) were stronger via fruits initiated than pollinia removed (Table 2).

Discussion

We found little support for our prediction that pollinator availability influences the strength or direction of selection on floral traits of A. syriaca. Although plants in the −P treatment received fewer pollinia than those in the control, they did not set fewer fruits per plant (Table 1), suggesting that reproduction of A. syriaca was not pollen-limited in this population. Not surprisingly given the lack of pollen-limitation, phenotypic selection was stronger for plants in the −P treatment relative to the control for only five out of 27 combinations of floral trait and fitness measure (Tables 2 and 3). Pollen limitation may influence phenotypic selection in smaller, more ephemeral A. syriaca populations (such as those studied by Morse and Fritz 1983) whose reproduction is limited by the receipt of compatible pollinia.

In contrast, our data support the prediction that resource limitation influences the strength of selection on floral traits of A. syriaca. When resources were more abundant (+R), phenotypic selection differentials on seven out of nine floral traits were stronger via female than male fitness (Tables 2 and 3). These results, along with those of Totland (2001) and Maad and Alexandersson (2004), suggest that variation in abiotic resources such as water, nutrients, and light can influence pollinator-mediated selection on floral traits. Seven out of nine floral traits did not differ between the C and +R treatments (Table 1), indicating that resource supplementation did not influence natural selection by changing the phenotypic distribution of these traits. However, the precise mechanism by which resource availability interacts with A. syriaca’s pollinators to influence phenotypic selection is unclear.

Higher fruit production in the +R treatment and lower pollinia removal and insertion rates in the −P treatment (Table 1) suggest that our manipulations were effective, but they may have had unintended effects on the unmeasured trait of nectar production. Supplemental watering, a component of the +R treatment, can increase nectar production of milkweeds (Wyatt et al. 1992). Nectar accumulates and is not reabsorbed in bagged milkweed flowers (Willson and Bertin 1979), suggesting that plants in the −P treatment might have had a higher standing crop when they were unbagged. Our results should be robust to these potential unintended treatment effects. Plants in the +R treatment did not receive or export more pollen relative to the control (Table 1), indicating that any increase in nectar production due to watering did not influence pollination. In addition, an increase in nectar standing crop in the −P treatment should increase pollinia import and export, which would make any treatment effect more conservative.

Our data also support the hypothesis that female fitness of flowering plants is limited more by resources than pollen (Stephenson 1981; Willson and Burley 1983). A. syriaca in the +R treatment set 31% more fruits than those in the control, but even though flowers on plants in the −P treatment received 41% fewer pollinia than those on control plants, they did not set fewer fruits (Table 1). Fertilizer also increased fruit set in a population of A. syriaca in Illinois (Willson and Price 1980), as well as in its congeners A. verticillata (Willson and Price 1980) and A. exaltata (Queller 1985). Our data reinforce the hypothesis that the very low fruit set observed throughout the Asclepiadaceae is the result of resource limitation (Wyatt and Broyles 1994).

Although A. syriaca is a popular study species among evolutionary ecologists (Wyatt and Broyles 1994), phenotypic selection on its floral traits has been measured in only one other population (Morgan and Schoen 1997). Assuming that plants in our C treatment were comparable to the unmanipulated plants studied by Morgan and Schoen (1997), the pattern of selection on floral traits was similar in the two populations. For example, contrary to Bateman’s principle (Bateman 1948), floral traits of A. syriaca in both Quebec (Morgan and Schoen 1997) and Iowa (Tables 2 and 3) were more frequently under selection via female than male fitness. In Iowa, as in Quebec (Morgan and Schoen 1997), the coefficient of variation for female fitness of A. syriaca (CV (pollinia insertion in the C treatment)=0.66; CV (fruits initiated in C)=0.73) was greater than for male fitness (CV (pollinia removal in C)=0.46). This suggests that stronger phenotypic selection via female than male fitness in both populations is caused by the greater opportunity for selection via female fitness (Crow 1958; Arnold and Wade 1984; Morgan and Schoen 1997).

Phenotypic selection on floral traits of A. syriaca via two components of female fitness (pollinia inserted and fruits initiated) was generally not congruent (Tables 2 and 3). In some cases (e.g., selection differential on hood length in the +R treatment (Table 2)), selection was significant via pollinia inserted but not fruits initiated, suggesting that pollinator-mediated selection on floral traits of A. syriaca is substantially modified by post-pollination events. Weaker selection on floral traits via seed and fruit set than via pollination has also been detected in other species (Ipomopsis aggregata (Campbell 1991); Penstemon centranthifolius (Mitchell et al. 1998)), suggesting that pollinator-mediated selection is frequently opposed by selection later in the reproductive cycle. Given that milkweeds mature only 1–5% of their flowers into fruits (Wyatt and Broyles 1994), the link between selection on floral traits during pollination and later stages of the life cycle may be particularly weak for A. syriaca. In other cases (e.g., selection gradient on hood length in the C treatment (Table 3)), phenotypic selection was significant via fruits initiated but not pollinia inserted, indicating that selection on some floral traits of A. syriaca is not pollinator-mediated. Although it is often implicitly assumed that selection on floral traits is pollinator-mediated, this assumption is rarely tested (Ashman and Morgan 2004; Totland 2004). In addition to measuring fitness components at different points in the reproductive cycle, as we did in A. syriaca, pollinator manipulations (as in Galen 1996) can be used to test assumptions about the agent of selection on floral traits.

One limitation of our study is that we were unable to examine the effect of any interactions between pollen- and resource-limitation on phenotypic selection on floral traits. To our knowledge, only three studies have experimentally manipulated pollen- and resource-availability in a crossed design (Campbell and Halama 1993; Juenger and Bergleson 1997; Mattila and Kuitunen 2000). None of these studies detected a significant interactive effect of pollen and resources (nutrients and water) on reproduction. Given that theory suggests that at equilibrium reproduction will be limited by both pollen and resources (Haig and Westoby 1988), future studies that measure phenotypic selection and utilize a crossed design may be particularly informative.

In summary, the availability of abiotic resources, but not pollinators, influences phenotypic selection on floral traits of A. syriaca. Specifically, increasing the resources available to plants resulted in stronger phenotypic selection on floral traits via female than male fitness. More generally, our results support the “context-dependence” hypothesis (Wilson et al. 1994; Ashman and Diefenderfer 2001; Ashman and Morgan 2004) for the evolution of floral traits, which predicts that the relative strength of selection via female vs male fitness will depend on the ecological context. To date, this hypothesis has focused on the effect of variation in the biotic environment, including pollinator identity, pollinator abundance, and the gender of neighboring plants, on gender-specific selection (Ashman and Diefenderfer 2001; Ashman and Morgan 2004). However, our results suggest that variation in the abiotic environment, particularly soil nutrients and water availability, may also generate context-dependent, gender-specific selection on floral traits.

Acknowledgements

We thank B. Casper, H. Maherali, A. Parachnowitsch, and three anonymous reviewers for their comments on this manuscript. D. Black (Jasper County Conservation Board) provided permission to work on public land and J. Brown provided access to a digital imaging system. I. Smith provided the drawing for Fig. 1. This work was supported by a National Science Foundation AIRE grant (awarded to Grinnell College) and the Grinnell College Committee in Support of Faculty Scholarship. During the writing of this manuscript, C. M. Caruso was supported by an operating grant from the Natural Science and Engineering Research Council of Canada.

Copyright information

© Springer-Verlag 2005