FLORAL EVOLUTION IN MILKWEEDS: 
EVIDENCE FOR SELECTION PAST AND PRESENT
 ByRaffica Jeanne La Rosa
A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of 
Plant Biology - Doctor of Philosophy
Ecology, Evolutionary Biology and Behavior - Dual Major
20151ABSTRACT
FLORAL EVOLUTION IN MILKWEEDS: 
EVIDENCE FOR SELECTION PAST AND PRESENT
ByRaffica Jeanne La Rosa
 Adaptation is an important process that allows species to utilize new habitats and to avoid 
extinction, contributing to the biodiversity we find on the planet. 
Many flowering plants rely on 
animals for pollination; the floral traits that are adaptive for pollination are those that influence 
attraction, rewards, or efficiency. Milkweeds (
Asclepias
) have unusual floral structures that 
consist of a gynostegium surrounded by five nectar-filled hoods. In many species, a horn 
develops from the inside base of each hood. Pollen grains are packaged into waxy packets 
(pollinia) and are positioned in the wall of the gynostegium. For fertilization, pollinia must be 
removed by pollinators and inserted into stigmatic openings in the wall of the gynostegium 
between adjacent hoods; pollination is carried out passively by a wide variety of pollinators that 
are almost entirely all insects. Milkweeds are hermaphroditic, so it is possible for the floral traits 
to be adaptive for male fitness, female fitness, or both.
 The floral diversity across 
Asclepias
 is astounding. My dissertation investigates if and 
how the floral structures are adaptive, and 
if the variation among species is the result of natural 
selection. I used two complementary approaches. For the first approach, I used contemporary 
measures of selection and functional studies to focus on the process of adaptation in five species 
of North American 
Asclepias
. Selection works on intraspecific variation within traits and the 
effect of that variation on fitness. I also utilized paternity analyses in two species to measure 
selection through male fitness. For the second approach, I used phylogenetic methods to find 
2signatures of past selection on traits across more than one hundred North American 
Asclepias 
species. Phylogenetic comparative methods focus on patterns of interspecific variation. I used 
tests of correlated evolution between pairs of traits, or traits and pollinators, to investigate 
functional relationships and possible selective agents. I also tested for convergent evolution, 
which can demonstrate adaptive evolution in response to a similar selective regime.
 The six floral traits I studied had an effect on fitness, suggesting they are adaptive. 
I found that many of the floral traits were under significant selection through only one gender, but 
that the direction of selection was similar across genders, showing little conflict between male 
and female function or between male and female fitness. I predicted that the size of the hood and 
gynostegium would influence pollinator attraction, but they were instead more likely to influence 
the efficiency of pollination. I also found no significant link between female pollination success 
and female reproductive success in four of the five species, so traits that increased pollen receipt 
did not in turn affect female fitness, which is consistent with fruit production not being limited 
by pollen receipt.
 Using measures of viable seeds produced (annual female fitness) and viable 
seeds sired (annual male fitness), I determined that selection estimated using total fruit number is 
a good estimate of selection through viable seeds produced; however, total pollinia removed per 
plant, a common estimate of male function and fitness in milkweeds is not a good predictor of 
viable seeds sired. My studies of the gynostegia, hoods, and horns across the phylogeny showed 
that horn loss likely followed the closure of hoods, suggesting a possible loss of horn function. 
There were also three convergent floral phenotypes; each may have evolved in response to 
similar selective regimes. The convergent species provide an excellent starting point for future 
investigations of possible selective agents.
3Copyright by 
RAFFICA JEANNE LA ROSA 
20154I dedicate this dissertation to my parents, Rosemary Warner and Mike LaRosa,
who made the decision to leave the city for the rolling hills of southwest Wisconsin, 
and to Ben Gollmer, Sue Schmidt, and the rest of my family, near and far, 
for their love and support. 
vACKNOWLEDGEMENTS
 Being part of the Conner Lab at the Kellogg Biological Station has been a wonderful 
experience. I owe much of that to Jeff for being a great advisor and a great person to know. I 
appreciate his knowledge, enthusiasm for science, sense of humor, patience, and dedication. I am 
also grateful for his help in making me a better scientist and a better writer, and for his enduring 
and infectious positive attitude. Anne Royer was the best lab mate a person could ask for; she is 
both an exemplary scientist and friend. I am forever thankful to Cindy Mills for teaching me 
every single lab skill I have, for always being willing to answer my questions, no matter how big 
or small, and for being an overall great person. Lastly, many thanks to my campus labbies, Sam 
P”rez and Amanda Charbonneau, and to the many people that helped me every summer in the 
field and in the lab: Debi Kilmartin, Morgan Lucke, Joan La Rosa, Latarsha Porcher, J. Stephen 
Williams, Laura Maihofer, T. Toria Davis, Krystal Harrison, Jasmine George, Natalya Dungee, 
Andre Hall, G. Alex Page, and Kola George. 
 I am also grateful to my committee, Doug Schemske, Jen Lau, and Ian Dworkin, for 
providing insightful feedback and suggestions over the years. I am especially appreciative of 
their time and dedication to mentoring students.
 I would like to thank the many funding 
opportunities at Michigan State and KBS for supporting me and providing funding to attend 
scientific meetings each year. In particular, I would like to recognize the GK-12 program, 
BEACON, MSU Plant Sciences, Paul Taylor, EEBB, the College of Natural Science, and the 
Graduate School.
vi Many thanks to Owen Boyle, Kendra Millam, Tom Rooney, Dave Rogers, and Don 
Waller for playing pivotal roles early on when I was an undergrad, and to Jonathan Coop, Tom 
Givnish, Jen Hauxwell, and Lawrence Harder for providing impactful experiences, and 
continued support. 
 The Kellogg Biological Station was a wonderful place to spend a few years of my life, 
and was great because of the diversity of people and science. Some of the people that have been 
especially important to me are: Jean J
ohnson Magle and Tomomi Suwa for being great 
housemates, the rest of my cohort including Lauren Kinsman and Leila Desotelle, and fellow 
KBSers (past and present): Emily Grman, Liz Schultheis, Melissa Kjelvik, Brook and Emily 
Wilke, Merry Coder, Kane Keller, Jarad Mellard, Lindsey Walters, and campus buddies: Doro 
and Ryan Taylor, Jay Sobel, Grace Chen, Carina Baskett, and Emily Dittmar. Many thanks to my 
writing buddies: Sarah Fitzpatrick, Melissa Kjelvik, Sarah Evans, and Christine Sprunger; spring 
break has never been so productive! 
 Lastly, IÕll miss the wonderful people I have met playing volleyball in Kalamazoo. In 
particular, I will miss my team: Craig, Greg, Melissa, Joe, Brandi, and honorary member Missy. 
It was a great way to decompress. Thanks for the laughs!
vii
TABLE OF CONTENTS
 LIST OF TABLES  x
LIST OF FIGURES  xii
CHAPTER 1  1
INTRODUCTION  1
 Background  1
 Organization of Dissertation  2
LITERATURE CITED  5
CHAPTER 2   7
SELECTION ON FLORAL TRAITS THROUGH MALE AND FEMALE   7
POLLINATION SUCCESS AND FEMALE FITNESS IN THREE SPECIES 
OF MILKWEEDS (
ASCLEPIAS
)  Introduction  7
 Methods  9
 Results   14
 Discussion  18
 Acknowledgements  24
APPENDIX  25
LITERATURE CITED  34
CHAPTER 3   40
FUNCTIONAL STUDIES AND ESTIMATES OF NATURAL SELECTION   40
THROUGH MALE AND FEMALE FITNESS IN MILKWEEDS (
ASCLEPIAS
) Introduction  40
 Methods  43
  Field methods  43
  Pollinator observations and pollen limitation  47
  Fitness measures  49
  Paternity  49
  Analyses  51
 Results  54
  Functional effects of floral traits  54
  Selection on floral traits  55
  Selection through seed production and seed siring success in 2010  56
  Pollinators and pollen limitation  58
 Discussion  58
  Function of floral traits  59
viii
  Selection on floral traits  60
  Function, fitness, and limitations  62
  Flower number and fitness  63
  Conclusions  64
 Acknowledgements  64
APPENDICES  66
 APPENDIX A: Tables and Figures  67
 APPENDIX B: Selection gradient analyses: effect sizes, standard errors, and  84
 model fit 
LITERATURE CITED  93
CHAPTER 4   100
PATTERNS OF ADAPTIVE TRAIT EVOLUTION ACROSS THE   100
NORTH AMERICAN 
ASCLEPIAS
 Introduction  100
 Methods  103
  Traits  103
  Pollinators  104
  Phylogeny  105
  Analyses  106
 Results   108
  Correlated evolution of traits  109
  Correlated evolution of traits and environment  109
  Convergent evolution of floral traits  110
 Discussion  111
  Correlated evolution of traits  112
  Correlated and convergent evolution of traits in response to their environment 113
 Acknowledgements  115
APPENDIX   116
LITERATURE CITED  133
 ixLIST OF TABLES
Table 1. Trait repeatabilities. 26
Table 2. 
Descriptive statistics for the six traits (mm), display size, flower number, and 27
fitness estimates.
Table 3. 
Pollinator visits to 
Asclepias 
populations.  28
Table 4. 
Pearson product-moment correlations among six floral traits and display size in 29
A. syriaca 
(n=45), A. viridiflora 
(n=212), and 
A. tuberosa 
(n=51).Table 5
. Standardized selection gradient estimates (SE) for floral traits and total flower 30
production in three 
Asclepias
 species.
Table 6. Trait repeatabilities. 68
Table 7. 
Sampling effort in each year for 
A. incarnata 
and 
A. exaltata. 
69Table 8. Percent pollinator composition in the 
A. incarnata 
population in 2008 and 2010
. 70
Table 9. Pairwise correlation coefficients for female fitness measures in 2010 for 
A. incarnata
 71
(below the diagonal) and 
A. exaltata 
(above the diagonal and in italics).
Table 10. 
Primer names, annealing temperature, exclusion probabilities and allele 72
frequencies of microsatellite markers used on 
A. exaltata 
and 
A. incarnata
, with genet 
sample sizes of n
.Table 11. 
Pairwise correlation coefficients between estimated total flower number and 73
ramet number for 
A. incarnata 
and 
A. exaltata.
Table 12. Pairwise correlation coefficients for 
A. exaltata 
(a) and 
A. incarnata 
(b) combined 74
across years.
Table 13. 
A. exaltata 
descriptive statistics for the six traits (cm), total flower number, and 75
fitness estimates.
Table 14. 
A. incarnata 
descriptive statistics for the six traits (cm), flower number (estimated 76
as lateral branches in 2007, 2011; ramets in 2008; estimated flowers in 2009, and umbels in 
2010), and fitness estimates.
xTable 15. Selection through insertions per flower in 
A. exaltata.
  85
Table 16. Selection through removals per flower in 
A. exaltata.
  86
Table 17. Selection through fruit number in 
A. exaltata.
   87
Table 18. Selection through viable seeds produced and sired, and total viable seeds in  88
A. exaltata.
 Table 19. Selection through insertions per flower in 
A. incarnata.
  89
Table 20. Selection through removals per flower in 
A. incarnata.
  90
Table 21
. Selection through fruit number in 
A. incarnata. 
91Table 22. Selection through viable seeds produced and sired, and total viable seeds in  92
A. incarnata.
Table 23. Phylogenetic signal of continuous and discrete traits. 117
Table 24. 
Model comparisons of trait evolution towards different adaptive optima  117
using AIC.c (a size corrected AIC for small sample sizes)
xiLIST OF FIGURES
Figure 1. 
Top and side views of 
A. syriaca 
(A), 
A. viridiflora 
(B), and 
A. tuberosa
 (C) 31flowers showing the six floral traits: gynostegium width (
gy), hood length (
hl), hood height (
hh), horn reach (
hr), slit length (
sl), gap width (
gw).Figure 2. 
Path analysis relating traits, pollinator visitation over ten minutes, and fitness 32
components to total yearly fruit production.
Figure 3. 
Top and side views of 
A. exaltata 
(A) and 
A. incarnata 
(B) flowers showing 77
the six floral traits and landmarks: gynostegium width (
g; gr), hood length (
hl), hood height (
hh), horn reach (
hr), slit length (
sl), gap width (
gw).Figure 4. 
The effect of traits and flower number on female (a) and male (b) floral function  78
across three years in 
A. exaltata
.Figure 5. 
The effect of traits and plant size on female (a) and male (b) floral function across  79
three years in 
A. incarnata
.Figure 6. 
Standardized selection gradient estimates for floral traits and flower number on  80
female fitness (fruit number) across three years in 
A. exaltata
.Figure 7. 
Standardized selection gradient estimates for floral traits and flower number on  81
female fitness (fruit number) across three years in 
A. incarnata
.Figure 8. 
Standardized selection gradient estimates for floral traits and flower number  82
through seeds produced, seeds sired, and total seed number for 
A. exaltata
 (a) and 
A. incarnata 
(b).Figure 9. 
Distribution of functional gender (f/(m+f)) for individuals in a population of  83
A. exaltata 
(a) and 
A. incarnata 
(b).Figure 10. 
A photograph of an 
Asclepias exaltata
 flower from the side and top. 118
Figure 11. 
Discrete binary floral traits: (a) horns present, (b) horns absent, (c) hoods 119 
conspicuously open, (d) hoods closed, (e) hood rim flared, (f) hood rim domed.
Figure 12. 
Distribution of the proportion of hymenopteran out of the total number of 120
visitors across 27 
Asclepias 
species.
xii
Figure 13. 
Maximum likelihood ancestral state reconstruction of character traits: hood  121
presence/absence (a) and conspicuous hood opening (b) mapped onto a mirrored phylogram 
of 107 Asclepias 
species.
Figure 14. 
Correlated evolution between hood shape and hymenopteran pollinators. 122
Figure 15. 
Evolution of hood length. 124
Figure 16. 
Evolution of gynostegium width. 125
Figure 17. 
Correlation between phylogenetic independent contrasts for gynostegium width  126
and hood length.
Figure 18. 
Test of correlated evolution of hood length and the proportion of pollinators  127
that are hymenopterans.
Figure 19. 
Test of correlated evolution of the estimated gynostegium width and the  128
proportion of pollinators that are hymenopterans.
Figure 20. 
Results of a SURFACE analysis of gynostegium and hood evolution supporting  129
four adaptive optima of these traits in the North American 
Asclepias.
xiii
CHAPTER 1
INTRODUCTION
Background
 Adaptation is a fundamental process in biology that allows species to cope with their 
environment. Many flowering plants rely on animals for pollination; the floral traits that are 
adaptive for pollination are those that influence attraction (e.g., Conner and Rush 1996; 
Schemske and Bradshaw 1999; Johnson et al. 2003; Hansen et al. 2012), rewards (e.g., Real and 
Rathcke 1991; Silva and Dean 2000), or efficiency (e.g. Nilsson et al. 1987; Fulton and Hodges 
1999; Conner et al. 2009). Adaptive traits can be studied utilizing complementary approaches 
including phylogenetic comparative methods 
(Larson and Losos 1996) 
to study patterns of 
interspecific variation to identify signatures of past selection (e.g., Mahler et al. 2013), and 
measures of contemporary selection that rely on intraspecific variation--a prerequisite for 
evolution. My dissertation combines these two approaches to study adaptive floral traits in North 
American milkweeds (
Asclepias
).  Milkweeds have highly unusual flowers that each consist of a fused male and female 
floral whorl that forms a cylindrical gynostegium surrounded by five hoods. The hoods contain 
nectar, and in many species, they also contain horns. Pollen grains are packaged into waxy 
packets called pollinia and are positioned in the wall of the gynostegium. For fertilization, 
pollinia must be removed by a pollinator and then inserted into a stigmatic opening between 
adjacent hoods. Milkweeds are hermaphroditic, so it is possible for the floral traits to be adaptive 
for male fitness, female fitness, or both.
1 My dissertation includes present-day studies of adaptation on six continuous floral traits: 
gynostegium width, hood length, hood height, horn reach, stigmatic slit length, and the gap width 
between adjacent hoods. I examined the function of these floral traits for male and female 
pollination success, and their effect on reproductive success in five species of 
Asclepias
. I also 
used genetic paternity analyses to measure selection on these six traits through male fitness in 
two species. These results were paired with phylogenetic comparative methods that focused on 
patterns in the evolution of hood and gynostegium related traits, and predictive methods to find 
convergent phenotypes that are indicative of selection. 
Organization of the dissertation
 Chapter 2:
 Floral traits can interact with pollinators to affect both pollination success 
(pollen removal and deposition) and reproductive success (seed siring and production). 
In collaboration with Jeffrey Conner,
 I measured selection on six floral traits in naturally occurring 
populations of 
Asclepias syriaca, A. viridiflora
, and 
A. tuberosa
 to understand if the unusual 
floral traits of 
Asclepias
 may be adaptive
. To test the effects of these traits on male and female 
pollination success and female reproductive success, I used separate selection gradient analyses 
using the fitness components pollinia removed per flower, pollinia inserted per flower, and fruit 
number as dependent variables. I used path analyses to understand the sources of selection, 
including pollinator visitation and the relationships among the three fitness components. I found 
that most traits were under selection through only one gender, and many affected the efficiency 
of pollination rather than pollinator visitation. I also found no significant link between female 
pollination success and female reproductive success in any of the three species, so traits that 
2increased pollen receipt did not affect female fitness, which is consistent with fruit production 
not being limited by pollen receipt.
 Chapter 3: 
In collaboration with Jeffrey Conner, I studied one naturally occurring 
population each of 
A. exaltata 
and 
A. incarnata 
over multiple years. I examined the function of 
six floral traits using estimates of pollinia inserted per flower (female function) and pollinia 
removed per flower (male function). I also measured natural selection on these traits
 through 
fruit number (female fitness) across multiple years. For a single year I genotyped all viable fruits 
and possible parental plants to assign paternity to all offspring in both populations. I then 
measured selection through seeds produced, seeds sired, and total seed number. Only slit length 
functioned to influence pollinia removals per flower in 
A. exaltata, 
and gynostegium width, hood 
length, slit length, and gap width all influenced pollinia insertions and/or removals per flower in 
A. incarnata. 
Gynostegium width was under selection through both genders, but hood height and 
gap width only affected female fitness.
 In a majority of cases, the direction of selection is the 
same between genders, showing little evidence for conflict. I found that selection estimated using 
fruit number is a good estimate of selection through seeds produced, but that total pollinia 
removed per plant, a common estimate of male function and fitness is not a good predictor of 
viable seeds sired.
 Chapter 4: 
In collaboration with Jeffrey Conner and Mark Fishbein, I complemented the 
contemporary studies of selection from Chapters 1 and 2, with phylogenetic comparative 
methods to ask: (1) Was the evolution of floral traits correlated with other floral traits or 
pollinator composition? and (2) Is there convergence of floral traits suggesting a common 
selective environment? I mapped gynostegium, hood, and horn traits of 107 
Asclepias 
species 
3onto a phylogeny of all 125 North American species. Results using PagelÕs correlation method 
suggested horns have been lost a number of times following the closure of the hoods, and that 
flared hood openings are marginally correlated with high proportions of hymenopteran 
pollinators. I found that hood length and gynostegium width were positively correlated across 
species after correcting for phylogenetic relatedness, but that they were not correlated with the 
proportion of hymenopteran pollinators, and that 13 clades or species have converged towards 
three phenotypic optima (4-5 clades each), suggesting convergent evolution. Interspecific 
patterns of evolution suggest that 
Asclepias 
flowers are adaptive and likely evolved in response 
to selective agents.
4LITERATURE CITED
5LITERATURE CITED
Conner, J. K., and S. Rush. 1996. Effects of flower size and number on pollinator visitation to 
wild radish, 
Raphanus raphanistrum
. Oecologia 105:509Ð516.
Conner, J. K., H. F. Sahli, and K. Karoly. 2009. Tests of adaptation: functional studies of pollen 
removal and estimates of natural selection on anther position in wild radish. Annals of 
Botany 103:1547Ð1556.
Fulton, M., and S. A. Hodges. 1999. Floral isolation between 
Aquilegia formosa
 and 
Aquilegia 
pubescens
. Proc. R. Soc. Lond. B 266:2247Ð2252.
Hansen, D. M., T. Van der Niet, and S. D. Johnson. 2012. Floral signposts: testing the 
significance of visual Ònectar guidesÓ for pollinator behaviour and plant fitness. Proc. R. 
Soc. Lond. B 279:634Ð639.
Johnson, S. D., C. I. Peter, L. A. Nilsson, and J. †gren. 2003. Pollination success in a deceptive 
orchid is enhanced by co-occurring rewarding magnet plants. Ecology 84:2919Ð2927.
Larson, A., and J. B. Losos. 1996. Phylogenetic systematics of adaptation. Pp. 187Ð220 
in M. R. 
Rose and G. V. Lauder, eds. Adaptation. Academic Press, San Diego.
Mahler, D. L., T. Ingram, L. J. Revell, and J. B. Losos. 2013. Exceptional Convergence on the 
Macroevolutionary Landscape in Island Lizard Radiations. Science 341:292Ð295.
Nilsson, L. A., L. Johnsson, L. Ralison, and E. Randrianjohany. 1987. Angraecoid orchids and 
hawkmoths in central Madagascar: Specialized pollination systems and generalist 
foragers. Biotropica 19:310Ð318.
Real, L. A., and B. J. Rathcke. 1991. Individual variation in nectar production and its effect on 
fitness in 
Kalmia latifolia
. Ecology 72:149Ð155. 
Schemske, D. W., and H. D. Bradshaw Jr. 1999. Pollinator preference and the evolution of floral 
traits in monkeyflowers (
Mimulus
). Proc Natl Acad Sci USA 96:11910Ð11915.
Silva, E. M., and B. B. Dean. 2000. Effect of nectar composition and nectar concentration on 
honey bee (Hymenoptera: Apidae) visitations to hybrid onion flowers. J. Econ. Entomol. 
93:1216Ð1221.
6  !"CHAPTER 2
 SELECTION ON FLORAL TRAITS THROUGH MALE AND FEMALE POLLINATION 
SUCCESS AND FEMALE FITNESS IN THREE SPECIES OF MILKWEEDS (
ASCLEPIAS) Introduction
  Flowers of animal pollinated plants function to manipulate pollinators into transporting 
pollen to s
tigmas. The floral traits that are adaptive for pollination are those that influence 
attraction (e.g., Conner 
and Rush
 1996; Schemske 
and Bradshaw
 1999; Johnson
 et al. 2003; 
Hansen 
et al. 2012), rewards (e.g., 
Real 
and Rathcke
 1991; Si
lva 
and Dean
 2000), or efficiency 
(e.g., Nilsson
 et al. 1987; Fulton 
and Hodges
 1999; Conner 
et al. 2009). If increased pollination 
success results in increased 
fitness
, selection can act on the floral traits
 that influence pollinator 
behavior
. Most flowering plants are hermaphroditic, so selection on floral traits can occur 
through both male and female fitness. Floral traits can interact with pollinators to increase 
effect
ive pollen removal or receipt, or may influence fruit and seed production 
more 
directly
 through 
pollen 
germination
 or fertilization
; uncovering such relationships helps us understand the functional role of individual floral traits. 
  Seed production (femal
e fitness) is often limited more by abiotic resources such as 
carbon, nitrogen, and water than by pollen receipt, and Bateman 
(1948) predicted that this resource limitation 
will result in
 greater selection on floral traits through male than through 
female fitness. When pollinators are scarce and resources plentiful, 
seed production may be 
limited by pollen receipt rather than abiotic
 resources. In this case we expect a tighter 
relationship between pollen receipt (female pollination success) and 
seed production
, translating into a greater possibility of selection on floral traits through female fitness
 (Wilson
 et al. 1994).  Most measures of selection on floral traits are solely through female fitness. Only a 
  !"handful of studies have measured selection in plants through male (seed siring) success
 (e.g., Meagher
 1994; Morgan 
and Conner
 2001; Hodgins 
and Barrett
 2008; Sahli 
and Conner
 2011; 
Kulbaba 
and Worley
 2012). Milkweeds (
Asclepias
 L.) have a long history of providing insight 
into male function in pl
ants (Willson 
and Rathcke
 1974; Willson 
and Price
 1977; Queller
 1983; 
Wilson
 et al. 1994; Wyatt 
and Broyles
 1994; Fishbein 
and Venable
 1996b), because male 
pollination
 success is easily 
estimated
 as the number of 
exported pollen packets (pollinia). To our knowledge, only two studies 
(Morgan 
and Schoen
 1997; Caruso 
et al. 2005) have measured 
select
ion on floral morphology in 
milkweeds; both studied
 Asclepias syriaca L. and used pollinia removals as their estimate of male fitness.
 Several others have studied the relationship between 
milkweed reproductive traits, including total flower number and infl
orescence size and number, 
and fitness 
components 
such as pollinia removal and 
insertion
 and female fruit (and seed
) production
 (Willson 
and Rathcke
 1974; Willson 
and Price
 1977; Wyatt
 1980; Shannon 
and Wyatt
 1986; Broyles 
and Wyatt
 1990).  Milkweed flowers
 have unusual floral traits (Fig. 1), including two features that 
evolved independently
 in the orchid family (Orchidaceae)
: pollen clustered into pollinia and 
the 
gynostegium
. The latter is a cylindri
cal structure consisting of fused male tissue 
surrounding the 
two 
ovaries
, with five stigmatic openings in 
the 
wall leading to the ovaries
 (Woodson
 1941). The pollinia lay to either side of each stigmatic slit 
with
in the wall of the gynostegium. 
Each pair of 
pollinia are 
connected by a corpusculum, which has a small groove that catches the hairs or 
claws of pollinators, aiding in removal; 
a pollinium 
needs to be
 be inserted into another stigmatic 
slit for 
fertilization to occur 
(Wyatt
 1978). Milkweeds also have a corona, consisting o
f a ring of 
hoods that surround the gynostegium. The hoods fill with nectar as a 
pollinator 
reward, but their 
size and volume vary between species. In many species, a horn forms within each hood, and 
  !"sometimes protrudes from the hood
, arching over the gynostegium (Fig. 1A, C). Many of the 
milkweed species studied to date, including 
A. syriaca 
and A. tuberosa
 L., are pollinated by a host of generalist pollinators 
(Kephart
 1983; 
Betz
 et al. 1994; Fishbein 
and Venable
 1996a; 
Ollerton 
and Cranmer
 2002; Ivey
 et al. 2003), including native 
bees, wasps, and butterflies
. Non-native honey bees 
are also a common pollinator for some 
Asclepias 
species, but
 have only been present in the United States
 for
 a few hundred years 
(Brown
 1898); thus
, Asclepias 
floral traits 
may be maladapted to honey bee pollination and may 
be under stronger selection 
where this 
pollinator is common
.   Our goal was to identify adaptive
 floral traits using selection gradient analysis 
(Lande and Arnold
 1983) in three species of 
Asclepias
ÑA. syriaca, A. viridiflora Raf.
, and A. tuberosa
. Flowers of t
hese species are diverse in size, shape, and color, yet populations can be found in 
close proximity and may overlap in th
eir pollinator composition.
 We identified six floral traits 
that may affect male and female 
fitness components
: gynostegium width, hood length, hood 
height, horn reach, slit length, and gap width (Fig. 1). We predicted that gynostegium width and 
the hood t
raits function to attract pollinators, because they affect the visibility of the flowers, and 
that horn reach, slit length, and gap width function to affect the efficiency of pollinators in 
removing and inserting pollinia. We tested these predictions using
 path analysis to complement 
the selection gradient analysis, as well as to determine the relationships among male and female 
pollination success and female 
reproductive success.  Methods
   We studied one naturally occurring population of each of three 
perennial milkweed 
species
 in southwest Michigan
; the maximum distance between the populations was 12.5 km
. We sampled 45 
Asclepias syriaca 
plants at the Pond Lab Facility of Michigan State University
Õs   !"#Kellogg Biological Station (42.410¡ N, 85.392¡ W). We se
lected individuals that were likely to 
be genetically distinct by only choosing one ramet (stalk) per cluster of likely clones; 
Asclepias 
syriaca 
has the ability to spread vegetatively by rhizomes, and Kabat 
(2010) found genets 
(genetic individuals) were made up of ramets that covered 1m
2 to 30m2. We chose ramets from 
clusters that were separated 
by two or more meters and had floral traits that looked distinctively 
different, both in shape and color; ramets within a cluster typically had visibly similar floral 
characteristics (La Rosa pers. obs.). We sampled all 51 
A. tuberosa
 plants in a clearing at Fort 
Custer Recreation Area (42.326¡ N, 
85.331¡ W), and sampled 212 out of more than 
one thousand 
A. viridiflora
 plants at the Fort Custer Training Center (42.306¡ N, 85.333¡ W) by walking 
transects spaced 6m apart, and flagging the nearest plant within
 a 3m radius every 6m along each 
transect. Genets in these last two species are well
-defined because they arise from a single root 
crown 
(Wilbur
 1976). Data on 
A. syriaca
 and A. tuberosa 
were collected
 in 2008, and data on A. viridiflora in 2009.    For trait measurements we photographed 2
-3 fresh flowers from each plant from the top 
and side (with 2.5 hoods removed as in Fig. 1) using a digital SLR camera with a 60mm macro 
lens. We first determined lan
dmarks that were identifiable across the range of floral variability 
(Fig. 1), and then used these landmarks to make calibrated linear measurements from the digital 
photographs using ImageJ 
(Rasband
 1997). Measurement error was quantified by measuring the 
traits on ten
 haphazardly chosen flowers from different plants, and then repeating the trait 
measurements on this same set of ten flowers for a total of three times. We calculated the 
variance of the three replicate measurements for each trait on each flower. For each 
trait, the 
mean of the ten variances (mean measurement variance within individual flowers) was divided 
by the variance in the ten means (variance among flowers) to estimate the measurement error as 
  !!"a proportion of the variance among flowers from different 
plants. The measurement error of each 
trait was less than five percent, with the exception of gynostegium width in 
A. tuberosa 
(7%) and 
A. syriaca 
(11%), and slit length in 
A. viridiflora 
(41%). 
  Using the 2
-3 flowers measured per plant in the full datase
t, we estimated trait 
repeatability, the percent of the total variance in floral traits explained by the variance among 
plants, using ANOVA with plant as the random predictor variable. Among
-plant variation was 
greater than within
-plant variation for all t
raits except gap width in 
A. tuberosa 
and A. viridiflora
 (Table 1
); note that gap width is the only trait measured that is a distance between two floral 
structures (i.e. adjacent hoods) rather than a dimension of a single floral structure. Also note that 
repeatability of slit length in 
A. viridiflora 
is almost
 66% despite the high measurement error for 
this trait.
  Annual fitness component estimates were taken from a single ramet per genet, because it 
is difficult to identify complete genets in 
A. syriaca
, and we wanted sampling to be similar across 
species. 
A. viridiflora
 genets typically had only one ramet (mean = 1.19, s.d. = 0.43) and 
A. tuberosa
 often had several ramets in a single well
-defined cluster (mean = 6.7, s.d. = 6.73, 
median = 5). 
Because f
low
ers open asynchronously within umbels
 and remain 
continuously open for several days 
(Wyatt
 1981; Kephart
 1987), we counted pollinia 
insertions and removals on 
every flower from umbels 
that had al
l flowers open
 and with some flowers were beginning to 
senesce, maximizing the number of days flowers within the umbel were accessible 
to 
pollinators
. To sample the greatest
 number
 of flowers 
per plant
, we
 recorded pollinia insertions, pollinia removals, a
nd display size 
of each ramet 
on a day when the maximum number of umbels were 
mature
 on that ramet
. We sampled 100 (s.d. = 50.1) flowers per 
A. syriaca 
ramet, 52 (s.d. = 17.3) 
per A. viridiflora 
ramet, and 58 (s.d. = 36.8) per 
A. tuberosa 
ramet; thus, we s
ampled pollination 
  !"#success for an average of 53% of all 
A. syriaca 
flowers and 28
-32% of all flowers on the other 
two species (
Table 2
).  We estimated male and female pollination success on a per
-flower basis as the average 
number of pollinia removed per f
lower and average number of pollinia inserted per flower, 
respectively; these measures ranged from 0
-5 because each flower had five pairs of pollinia and 
five stigmatic slits where pollinia could be deposited. We used a hand lens to count pollinia 
insertio
ns and removals in the field. We measured display size as all of the flowers open on the 
ramet on that day, which included every flower sampled for pollinia insertions and removals, 
plus all open flowers from umbels that were not yet mature enough to be sa
mpled; thus, display 
size was always equal to or greater than the number of flowers sampled. 
  We estimated total flower number by counting 
every umbel at the end of the 
season and multiplying by the mean number of flowers per sampled umbel for each indivi
dual. We 
estimated
 the total number of pollinia removed and inserted
 as the product of per
-flower pollinia 
removals 
and insertions 
and total ramet flower number. We measured
 annual female 
reproductive success as the number of 
mature 
fruit
s on each focal ra
met
; we sub
-sampled 203 
fruits from 39 
A. syriaca
 plants, 
to confirm
 that fruit and seed number were 
significantly
 correlated (r = 0.96, p < 0.001). Fruit number was counted in September, once the fruits had 
matured; a few plants from each species died or 
were damaged by deer or rodents before the 
fruits could be counted. 
  To relate floral traits to pollinator 
visitation
, we
 collected pollinator data from each 
A. syriaca 
and A. tuberosa 
individ
ual using Canon VIXIA HF10 high
-definition digital video 
camera
s. Ten or more minutes of video was taken of all open flowers on each focal ramet 
immediately after the pollinia insertions and removals had been counted. From the video we 
  !"#recorded the number of 
pollinator 
visits, the duration of each visit, and pollinato
r taxon 
to 
order (subdivided further
 for Hymenoptera
 and Coleoptera
). An insect taxon was considered to be a 
potential pollinator if at least one individual of that taxon had been captured or seen with pollinia 
on its body. Visitation to the 
A. viridiflora
 population was later observed over 3.5 hours in 2014 
by walking through the population and recording all insect visitors that had pollinia on their 
bodies; results confirmed what we casually observed in 2009 when we collected the trait and 
fitness data.
  Each of our selection gradient analyses regressed fitness 
component 
estimates 
(pollination success or reproductive success
) onto floral traits and display size; the selection 
gradients for the floral traits were robust to the exclusion of display size in a
ll of our models. 
Fitness 
component 
estimates were relativized by dividing by the mean, and the traits were 
standardized to a mean of zero and standard deviation of one. Each measure of pollination or 
reproductive success was regressed separately on
to all of the floral traits and display size, 
totaling three linear multiple regressions per species. Selection gradient analyses were performed 
using JMP 10.0 
(SAS Institute Inc.
 2012).  To help interpret and integrate the results of the s
election gradient analysis, we used 
structural equation models (SEMs) to visualize the relationships between the floral traits, 
pollinator visitation (
A. syriaca 
and A. tuberosa 
only), male and female pollination success, and 
female reproductive success.
 The pollinator data for 
A. syriaca 
included several orders of insects 
combined
, whereas
 A. tuberosa
 included only honey bees, as they were the dominant pollinator 
and the only one to visit the focal ramets in the videos. We fit hypothesized causal 
relationships 
through simple 
paths from each floral trait and display size to pollinia removed per fl
ower (male 
pollination success) and
 pollinia inserted per flower (female pollination success), and 
through 
  !"#compound paths leading to
 pollination success, but passing through 
pollinator visits and duration 
for 
A. syriaca 
and A. tuberosa
. We hypothesized that display size influenced total flower 
number, and that flower number influenced the total number of pollinia removed and inserted on 
the ramet, as well as fruit 
number; additionally, we fit a compound 
path from flower number to 
fruit number 
passing 
through the total number of pollinia insertions. 
  Initially the model for each species had a significant chi
-square goodness-of-fit value, 
indicating they were not accep
table fits to our data and that something was likely missing from 
our model. We used modification index values to identify influential relationships that were not 
included in our original model
 (Grace
 2006). From these values, we chose to add a correlational 
(double-headed) arrow between 
the residual variation of 
total removal
s and insertions and causal 
(single-headed) arrows from gynostegium width, hood length, horn reach, and insertions per 
flower leading to fruit number. These were added to all three models, with the exception of horn 
reach in the A. viridiflora 
model. Two o
f the three final SEM models had non
-significant chi
-square goodness-of-fit values, indicating that they were acceptable fits to the data. The full 
model for 
A. tuberosa
 had a significant chi
-square goodness-of-fit value (P < 0.001), but when 
we split the 
model into male
-only and female
-only models, both reduced models fit acceptably (p 
> 0.05). The coefficients and their significance did not qualitatively change between the full 
model and the 
within
-gender models. All SEM analyses were performed in R v. 3.
1.1 (R Core Team 2014) using the 
lavaan
 (Rosseel
 2012) package. Results
   A. viridiflora
 and A. tuberosa
 were each visited almost exclusively by a single pollinator 
taxon, bumble bees (
Bombus 
spp.) and honey bees (Apis mellifera) 
respectively, while the 
pollinators of 
A. syriaca 
were more diverse (
Table 3
). Still, two
-thirds of the visitors to 
A.   !"#syriaca
 were honey bees, with the next most common being soldier beetles (
Cantharidae
) at 13%. Both honey bees and bumble bees have been shown to insert pollinia i
n A. tuberosa
 in Arizona 
(Fishbein 
and Venable
 1996a). Native bumble bee and non
-native honey bee visitors to 
these 
Asclepias
 species consistently had pollinia on their bodies, perhaps due to their abundance of hairs and spines, as well as the frequency at which they gripped 
the flowers between the hoods 
as they foraged for nectar. The other pollinators with smoother appendages, such as flies and 
soldier beetles, tended to have fewer pollinia (La Rosa pers. obs.) that were only clipped to the 
ends of their tarsi, and never on 
their tibia or femur leg segments (unpub. data from
 A. incarnata
).  Gynostegium width, hood height, and slit length were the least variable floral traits, with 
phenotypic coefficients of variation (CV
P) between three and nine, while hood length, horn 
reach, and gap width were more variable, with CV
P values from 13 to 23 (
Table 2
). A. syriaca had the largest flowers and 
A. viridiflora 
the smallest based on the geometric mean size of the 
floral traits (excluding gap width and horn reach). Correlations among t
he six floral traits were 
low and generally positive, except for some 
weakly 
negative correlations with gap width in 
A. viridiflora and A. tuberosa
 (Table 4
). Display size was not significantly correlated with the six 
floral traits with one exception
 (positive correlation
 with 
hood height in 
A. syriaca
). Fruit number per flower was extremely low, as is common among milkweeds. Each 
species produced a mean of about 185 flowers per ramet, with each flower containing two 
ovaries for a potential of approximately
 370 fruits, yet each ramet produced three to six fruits on 
average, less than two percent of their potential fruit production (
Table 2
). Fruit set did not 
appear to be limited by pollen receipt, as plants had an average of 14 to 24 pollinia inserted for 
every fruit they produced (
Table 2
), though the proportion of 
deposited 
incompatible 
self
-pollen
,   !"#or perhaps improperly inserted pollinia,
 is unknown
.  We found significant directional selection for increased hood length and hood height in 
A. syriaca 
and A. viridiflora (Table 5
). Three of four significant selection gradients on the hood 
were through male pollination success (pollinia removed per flower), and the fourth was through 
female reproductive success (fruit number) in 
A. viridiflora
. We hypothesized 
that the hoods 
affected fitness by attracting pollinators to the flowers, and the path analysis for 
A. syriaca 
(Fig. 
2A) confirms that plants with larger hoods had more visits. However, increased visitation rate 
did not significantly increase pollinia remo
vals; instead,
 the number of pollinia removed per 
flower was 
direc
tly 
affected by hood height
 rather than through increased pollinator visitation, suggesting
 that hoods affect the efficiency
 of 
pollinia removal
. The path analysis also suggests 
that 
the 
positive selection on hood length through 
fruit number
 in A. viridiflora
 was
 not mediated 
through pollinia insertions
 (Fig. 2B)
. We found significant directional selection to increase stigmatic slit length through 
increased female pollination success (pollini
a inserted per flower) in 
A. viridiflora
 and A. tuberosa 
(Table 5
). We hypothesized that slit length was an efficiency trait and the path analyses 
confirmed that it 
directly 
influenced pollinia insertions per flower in both 
species
, although this path was 
only marginally significant in 
A. viridiflora
; furthermore, slit length
 did not affect 
honey bee visitation in 
A. tuberosa 
(Fig. 2C). The path analysis showed no relationship between 
pollinia insertions and fruit number in 
A. tuberosa
, so the effect of sli
t length on insertions per 
flower and total insertions did not translate into increased female reproductive success, 
consistent with the selection gradient analysis based on fruit number 
(Table 5
) and with fruit 
number not being limited by insertions (
Table 2). Horn reach, which we hypothesized would affect the efficiency of 
pollination
, had no   !"#effect on pollinia removed or inserted in either the selection gradient results (
Table 5
) or the path analyses (Fig. 2). Instead, it was under selection through frui
t number in both species that have 
horns. In A. syriaca, 
where horns only extend over a portion of the gynostegium (see Fig. 1A), 
there was marginally significant selection for longer horns through fruit number (this was the 
strongest standardized selectio
n gradient i
n our study), and a significant
 single causal 
path 
from 
horn reach to fruit number
 (Fig. 2A). Functionally, pollinators were attracted 
to smaller horns, 
and an increase in pollinator visitation increased the number of insertions per flower,
 consistent 
with the large negative (but not significant) selection gradients for insertions and removals.
 Because 
insertions did not affect fruit set (Fig. 2A)
, this negative functional relationship did not 
affect the direct positive effect of horn reach on 
fruit number. In A. tuberosa, 
where horns often 
overlap
 each other across the top of the 
gynostegium (see Fig. 1C), there was selection for 
shorter horns through fruit number, but the causal path from horn reach to fruits was only 
marginally significant (Fi
g. 2C).  There was significant directional selection to increase floral display size through fruit 
number in 
A. syriaca 
and A. viridiflora
, and through pollinia removed per flower
 in A. viridiflora 
(Table 5
). The selection through fruit number in both spec
ies was 
likely due to
 an effect of 
display size on total flower number, which in turn was a strong determinant of fruit number
 (Fig. 
2). Because such a small percentage of flowers set fruit, these relationships 
may be due to 
overall 
ramet size. In 
A. virid
iflora
 plants with larger displays had significantly more removals per 
flower
, consistent with the selection gradient analysis
, which translated into more total removals 
(Fig. 2B). In 
A. tuberosa,
 larger displays increased both the number and duration of h
oney bee visits, and longer visits caused increased
 pollinia removals per flower; t
his positive effect was
 partially
 counteracted by a significant negative effect of display size 
directly 
on removals per 
  !"#flower, which may explain the lack of significant se
lection on display size through pollinia 
removed per flower (
Table 5
). Total flower number was the strongest determinant of both male 
and female fitness components, affecting male and female 
total 
pollination success
 in all three species
, and female reprod
uctive success in 
A. syriaca 
and A. viridiflora
. However, across the 
three species
 the models explained
 only 28-48% of the variation in fruit number.
 We 
tested for differences among the three species in the strength of selection on floral 
traits by compari
ng the magnitude of all 18
-21 standardized 
selection gradients between pairs of 
species using 
three 
paired t
-tests
, where each selection gradient was paired with the selection 
gradient from the same fitness component and trait in the other species
. A. syriaca and A. tuberosa 
each had significantly stronger 
average selection
 than
 A. viridiflora (t17 = 3.17, P = 0.003; 
t17 = 5.03, P < 0.0001, respectively
), but average selection did not differ between 
A. syriaca 
and A. tuberosa (t20 = 0.91, P = 0.373). Thus, selection on floral traits was strongest in 
A. syriaca 
and A. tuberosa
, species that were heavily visited by non
-native honey bees. 
Note
 that the
se two species 
had much smaller 
sample sizes 
than 
A. viridiflora, raising the possibility that the
 stronger selection gradients were 
bias
ed upward
, akin to the ÒBeavis effectÓ
 (Beavis
 1998); however,
 the proportio
n of individuals sampled in these two populations was much greater
 (~75-100%)
 than in A. viridiflora
 where, despite the
 large sample size, 
we sampled
 approximately 
20% of the total population
.  Discussion
  Floral traits of hermaphroditic plants can affect 
male or female fitness exclusively or 
simultaneously; we found that only hood length had significant effects through both genders. 
Because selection in hermaphrodites can act through each gender separately, there is an 
opportunity for conflicting selection
 on individual traits through male and female fitness. 
  !"#Nevertheless, we found no evidence of conflicting selection between genders on any trait within 
the three species, consistent with previous studies
 on other taxa (e.g., Sahli 
and Conner
 2011; 
reviewed by Delph 
and Ashman
 2006). Hood height influenced only pollinia removal by 
pollinators, while slit length and horn re
ach affected only female fitness components. Conflicting 
selection through the two genders also seems less likely, because pollinia removals
 per flower
 and insertions per flower 
were always positively correlated, albeit not significantly so in 
A. tuberosa
 (Table 2
).  Our estimate of male pollination success (pollinia removed per flower) was most affected 
by hood dimensions, but not through pollinator visitation as predicted. Instead, hoods seemed to 
influence the effectiveness of pollinators in removing pol
linia. Caruso et al. 
(2005) also found 
positive direc
tional selection on hood length, but in the population of 
A. syriaca 
they studied, 
selection was through fruit initiation, their measure of female fitness. Milkweeds abort many of 
the fruits they initiate, so fruit initiation 
might
 better
 reflect female po
llination success
 (pollinia insertions per flower)
 than reproductive success.  The only significant selection gradients through female pollination success were for 
increased slit length in 
A. viridiflora 
and A. tuberosa
. The path analysis confirmed that s
lit length 
directly affected pollinia insertions, and it was not mediated through pollinator visitation in 
A. tuberosa 
(Fig 3C), supporting our prediction that this is an efficiency trait. 
A longer slit length 
may function to increase the chance that pollin
ator legs, claws, and hairs will slide between the 
adjacent anther wings that form the stigmatic opening, resulting in an inserted pollinium. We did 
not find selection on slit length in 
A. syriaca, 
but Morgan and Schoen 
(1997) found selection for 
longer slits through pollinia insertions in their study. 
 Selection through female reprod
uctive success (fruit number) was for increased display 
  !"#size in two of the species, horn reac
h in the two species with horns,
 and hood length in the one species without horns. The selection on horn reach was 
due to its
 effect on fruit production
 directly,
 without having an effect on
 pollinia insertions or pollinator visitation (Fig. 2). Selection 
on hood length in 
A. viridiflora, 
where the hood openings are extremely reduced and there are no 
horns, was positive and also 
affected
 fruit number
 without influen
cing the number of pollinia 
insertions
; hood length 
in 
A. viridiflora 
may be acting similarly to horn reach in the other two 
species. By maneuvering pollinators, the horns and hoods may be altering the ÒqualityÓ of an 
insertion, such that it is more or les
s likely to result in a fruit.
  The strong selection on horn reach in populations dominated by honey bee pollinators 
suggests that horns may function to maneuver this relative newcomer into effectively depositing 
pollinia. The first recorded arrival of dom
esticated honey bees into what is now the United States 
was recorded in 1622 
(Brown
 1898), and different pollinator species can exert differential 
selection 
(Galen
 1985; Schemske 
and Horvitz
 1989; Conner 
et al. 2009; Sahli 
and Conner
 2011); consequently, North American 
Asclepias 
species have had less time to adapt to 
honey bees relative to the native pollinators. This may also explain why
 A. syriaca 
and A. tuberosa
 are experiencing stronger selection than in A. viridiflora
 overall
Ñthey
 are not yet optimized for 
efficient pollination by honey bees. The population of 
A. syriaca 
that C
aruso et al. (2005) studied had a mean horn reach very similar to our population, though they found significant selection for 
a shorter horn reach. Their study did not describe the pollinator composition, so the degree to 
which honey bees contributed to the selection they measured is unknown.
 Alternatively, 
horns and hoods may be correlated with an unmeasured trait that facilitates 
fertilization once pollinia have been deposited. For example, hood length may be correlated with 
a trait such as nectar volume. Nectar is produced within the gynostegium, flows fre
ely between 
  !"#the gynostegium and hoods in 
A. curassavica
 (Galil 
and Zeroni 1965), and is required for pollen 
germination in 
A. syriaca
 (Kevan 
et al. 1989). Larger hoods may allow the plant to hold more 
nectar within the gynostegium to incr
ease pollen germination. Future studies should examine the 
functional relationships between hood dimensions, nectar volume, and fruit number. 
 We never found selection on a trait through both insertions per flower and fruit number
; in other words
, floral t
raits influencing female pollination success were distinct from those 
influencing female reproductive success. In parallel with this, there were no significantly positive 
relationships between total insertions and fruit number for any of the species (Fig. 
2); in 
A. tuberosa,
 the strong path between total insertions and fruit number had a large standard error and 
was therefore not statistically significant, and the significant path in 
A. viridiflora
 was actually 
negative. A disconnect between pollen received
 and fruit production suggests some combination 
of improper placement of pollinia by pollinators, self
-incompatibility, and resource limitation of 
fruit production; we discuss each of these in turn, while noting that the our data cannot 
distinguish between
 these three mechanisms.
 Pollinia that are improperly inserted (low quality insertions
, above) may restrict pollen 
tubes from reaching the ovaries within the gynostegium. A pollinium has the advantage over 
loose pollen of potentially delivering a saturatin
g amount of pollen in a single pollinator visit 
(e.g., Bookman
 1984; Queller
 1985). However, the pollen tubes grow from only one edge of the 
pollinium 
(Galil 
and Zeroni 1969; 
Wyatt
 1976), so if the pollinium is inserted 
in reverse
, and the edge where the pollen tubes emerge
 is not encased in nectar, the 
pollen 
grains within the 
pollinium 
will not germinate
, preventing fertilizat
ion 
(Kevan
 et al. 1989; Sage 
and Williams
 1995). The inserted pollinia that we counted on 
A. syriaca 
and A. tuberosa 
may 
often have been 
  !!"selfed
 due to the way honey bees forage. In 
A. syriaca
, Howard and Barrows 
(2014) found that 
88% of pollinia 
inserted 
by honey bees was self
ed and Pleasants 
(1991) calculated that 71% of 
the pollinia that honey bees inserted came from within one meter of the recipient flower. Given 
that 
A. syriaca 
spreads vegetatively, ramets 
within one meter are often from the same genet. This 
level of geitonogamous selfing could be occurring in 
A. tuberosa 
populations as well
, given that most ramets within a meter were from the same root crown. A possible mechanism that may 
reduce the occurrence of geitonogamous selfing is the rotation of pollinia upon removal; the 
arms connecting the pollinia to the corpusculum rotate the pollinia approximately 90 degrees 
over several minutes in 
A. sullivantii 
(Robertson
 1886) and 107 seconds in A. exaltata (Queller
 1983). The average duration of honey bee visits to the focal ramets of 
A. syriaca 
and A. tuberosa
 was 67 seconds and 40 seconds, respectively, but some honey bees remained on ramets for more 
than five minutes in both sp
ecies, which would circumvent this mechanism. With very low levels 
of self
-compatibility in 
A. tuberosa 
(Wyatt
 1976), large numbers of pollinia inserted would not 
necessarily
 translate to high fruit set, and could even have a negative effect (as in 
A. viridiflora
, Fig. 2B
) by wasting resources on the initiation of fruits that are then aborted due to a late
-acting self
-incompatibility system 
(Gibbs
 2014).  Even if pollinia are inserted properly and compatible, fruit production could be 
resource-limited. Under conditions of resource limitation, variation in female pollination success should 
have little effect on fruit number. Without a link to reproductive su
ccess, the floral traits that 
function in female pollination success will not be under selection from pollinators. Caruso et al. 
(2005) supplemented water and fertilizer in 
A. syriaca
; despite a 46% increase in the numbers of 
fruits initiated, 
demonstrating a reduction in resource limitation, 
there were actu
ally fewer 
significant selection gradients through female fitness (fruits initiated) in the supplemented 
  !"#treatment compared to the unmanipulated control.
 Additionally, despite evidence for resource 
limitation of female fitness in the
ir unmanipulated contro
l, there was not greater selection 
through male fitness 
compared to female fitness 
as predicted by BatemanÕs principle
. However
, it could be that full fruit production, not initiation, is where the impact of resource limitation on 
female fitness occurs.
 In our study, despite evidence of adequate pollinator availability that hints 
at possible resource limitation, we also did not find greater selection through male fitness. 
Instead, there were equal instances of significant selection through pollinia removed 
flower 
(male) and fruit number (female) across the three species
 (Table 5
). Milkweed flowers undoubtedly interact with pollinators to influence the movement of 
pollinia, but our data show that there is a disconnect between pollinia insertions and fruit num
ber that we still do not fully understand. Some proportion of insertions are likely ineffective due to 
placement or 
incompatibility
, and may be masking a positive relationship between effective 
pollinia insertions and fruit number. However, 
resource limita
tion
 will 
also undermine this 
relationship, reducing selection on floral 
traits 
through female fitness
.  We also know very little about the relationship between pollinia removals and seeds 
sired, apart from one study in 
A. exaltata
, where they were positiv
ely, but weakly, correlated 
(R2 = 22; Broyles 
and Wyatt
 1990). Because
 resource limitation 
is not expected to 
impact male 
fitness 
to the extent that
 it affects female fitness, the relative strength of selection on fl
oral traits 
through male and female fitness could be impacted by the resource limitation of female fitness as 
predicted by Bateman 
(1948). Bateman's hypothesis has been invoked in milkweed studies (e.g., 
Queller
 1983; Caru
so et al. 2005) because of the relative ease of estimating male fitness through 
pollinia removed. However
, without a clearer understanding o
f the relationship between pollinia 
removed and fruits sired, it is difficult to make a robust comparison of selection through male 
  !"#and female reproductive success. We found that several floral traits function to increase pollen 
removal and deposition and 
are likely adaptive for pollination success in these species, but we 
have yet to uncover what factors strengthen the relationship between pollination success and 
reproductive success to cause adaptive evolution of floral traits in 
Asclepias
. Acknowledgemen
ts We thank 
J. Lau, D. Sc
hemske, and I. Dworkin
 for 
insightful comments
, and D. Kilmartin, M. 
Lucke, and L. Porcher for help 
with fieldwork. This research was 
funded
 in part by the National 
Science Foundation Graduate Research Fellowship Program, a Michiga
n State University Plant 
Sciences Fellowship, 
a grant from Prairie Biotic Research, Inc., 
a T. Wayne and Katherine Porter 
Fellowship, a G.
H. 
Lauff Research Award
, the W.K. Kellogg Foundation, 
and National 
Science
 Foundation Grant DBI
-0638591. We also thank
 the Michigan DNR and the Fort Custer 
Training Center
 for access to field sites
.   !"#                      APPENDIX
   !"#Table 1.
 Trait repeatabilities
 The percent of total variation in floral traits explained by the 
difference among plants calculated from nested random
-effects ANOVA. 
  A. syriaca
 A. viridiflora
 A. tuberosa
 Gynostegium width
 66.1
 81.2
 59.2
 Hood 
length
 68.9
 76.7
 72.6
 Hood height
 83.1
 81.8
 78.4
 Horn reach
 70.2
 - 52.8
 Slit length
 61.4
 65.8
 62.6
 Gap width
 82.8
 36.5
 32.8
  27 Table 2
. Descriptive statistics for the six traits (mm), display size, flower number, and fitness estimates.
 Means and phenotypic 
coefficients of variation (CV
P) were calculated for each species. The geometric mean of the floral traits (excluding gap width and horn 
reach) is an estimate of overall flower size. Relative size was calculated within each species by 
dividing each floral trait by the 
geometric mean. Flowers, pollinia removals, pollinia insertions, and fruits were counted on a single ramet. 
Pearson product
-moment 
correlations are given between pollinia removals per flower and insertions per flower. 
Poll
inia insertions per fruit were calculated by 
dividing the number of pollinia insertions per flower sampled by the number of fruits per flower over the entire ramet.
    A. syriaca 
(n = 42-45)  A. viridiflora 
(n = 205-212)  A. tuberosa 
(n = 40-46)    Mean 
(SD)
 Rel. size
 CVP Mean (SD)
 Rel. size
 CVP Mean (SD)
 Rel. size
 CVP Gynostegium width
  2.19 (0.12)
 0.76
 5.4
  2.74 (0.12)
 1.67
 4.3
  1.70 (0.06)
 0.64
 3.5
  Hood length
  3.24 (0.42)
 1.12
 12.9
  0.40 (0.09)
 0.24
 23.1
  2.58 (0.37)
 0.98
 14.1
  Hood height
  5.08 
(0.44)
 1.76
 8.9
  4.74 (0.31)
 2.89
 6.5
  5.93 (0.36)
 2.25
 6  Horn reach
  2.04 (0.34)
 0.71
 16.5
  - - -  1.74 (0.25)
 0.66
 14.3
  Slit length
  1.90 (0.12)
 0.66
 6.1
  1.40 (0.09)
 0.85
 6.2
  1.88 (0.08)
 0.71
 4.2
  Gap width
  0.61 (0.12)
 0.21
 19.2
  0.62 (0.08)
 0.38
 13.4
  0.43 (0.09)
 0.16
 21  Geometric mean
  2.88
    1.64
    2.64
    Display size
  109.51 (68.70)
  62.7
  110.46 (52.56)
  47.6
  66.04 (43.76)
  66.3
  Total flower number
  189.03 (111.21)
  58.8
  185.11 (107.51)
  58  182.27 (179.52)
  98.5
  Pollinia
 removals per flower
  2.02 (1.21)
  59.8
  2.32 (0.75)
  32.3
  1.11 (0.67)
  60.2
  Pollinia insertions per flower
  0.42 (0.23)
  56  0.30 (0.17)
  57.2
  0.18 (0.11)
  62.5
  Removal
-insertion correlation
  r = 
0.23*
    r = 
0.37*
    r = 
0.06
    Fruit number
  5.95
 (5.60)
  94.2
  3.42 (2.52)
  73.6
  3.18 (2.27)
  71.6
  Pollinia insertions per fruit
  24 (5.1)
    20 (1.3)
    14 (3.5)
    * P < 0.05.
               28 Table 3
. Pollinator visits to 
Asclepias 
populations.
 Visits to 
A. tuberosa 
were recorded from 8 
hours and 10 minutes of video from 2008, and visits to 
A. syriaca 
were recorded from 9 hours 
and 29 minutes of video from 2008. Visitation to 
A. viridiflora 
was surveyed in the population in 
2014 (see Methods). The mean (SEM) 
number of visits per minute 
and cumulative 
(total) duration
 per minute
 for 
A. syriaca 
and A. tuberosa 
is calculated from ten 
or more minutes of 
video of 
each focal ramet
.    Number of visits (%)
 Pollinator
 A. syriaca
 A. viridiflora
 A. tuberosa
 A. mellifera 
(honey 
bee)
 86 (64.7
%)     74 (93.7
%)  Bombus spp. 5 (3.8
%)  30 (100
%)  1 (1.3%)  Diptera
 6 (4.5
%)     0   Cantharidae
 sp.
 17 (12.8
%)     0   Lepidoptera
 6 (4.5
%)     4 (5.1%)  Wasps
 9 (6.8
%)     0   Hemiptera
 4 (3.0
%)     0   Total number of visits
 133   30   79   Mean no. visits per 
min
 0.25
 (0.28)
     0.17
 (0.24)
€  Mean total duration per min
 0.28
 (0.35)
     0.19
 (0.30)
€  €Honey bee visits only.
  29 Table 4
. Pearson product-moment correlations among six floral traits and display size in 
A. 
syriaca 
(n=45), 
A. viridiflora 
(n=212), and 
A. tuberosa 
(n=51).
   A. syriaca
  A. viridiflora
  A. tuberosa
  Gynost. 
width
 Hood 
length
 Hood 
height
 Horn 
reach
 Slit 
length
 Gap 
width
  Gynost
. width
 Hood 
length
 Hood 
height
 Slit 
length
 Gap 
width
  Gynost. 
width
 Hood 
length
 Hood 
height
 Horn 
reach
 Slit 
length
 Gap 
width
 Hood length
 0.30
 *            0.13
           0.20
            Hood height
 0.28
  0.42
 *          0.17
 * 0.15
 *        0.04
  0.25
          Horn reach
 0.23
  0.63
 * 0.61
 *                   0.17
  0.18
  0.22
        Slit length
 0.39
 * 0.26
  0.34
 * 0.35
 *      0.24
 * 0.08
  0.22
 *      0.26
  0.14
  0.28
 * 0.00
      Gap width
 0.04
  0.16
  0.25
  0.25
  0.01
     0.25
 * -0.04
  -0.05
  0.01
     0.31
 * 0.22
  0.06
  -0.26
  0.08
    Display size
 0.01
  0.20
  0.34
 * 0.29
  0.07
  -0.14
   0.09
  0.08
  0.01
  -0.05
  -0.02
   -0.26
  -0.17
  -0.02
  -0.18
  -0.05
  -0.04
  * P < 0.05.
   30 Table 5
. Standardized selection gradient estimates (SE) for floral traits and total flower production in three 
Asclepias
 species.
 R2 is the percent variance in that fitness component explained by all the traits plus the significance of the whole model. 
    A. syria
ca  A. viridiflora
  A. tuberosa
   Pollinia 
removed per 
flower
 Pollinia
 inserted per 
flower
 Fruit
 number
  Pollinia 
removed per 
flower
 Pollinia 
inserted per 
flower
 Fruit
 number
  Pollinia 
removed per 
flower
 Pollinia 
inserted per 
flower
 Fruit
 number
 Gynost.
 width
  -0.10 (0.09)
  -0.03 (0.10)
  0.24 (0.14)
   -0.01 (0.02)
  -0.02 (0.04)
  0.03 (0.05)
   -0.09 (0.10)
  -0.16 (0.10)
 € -0.16 (0.11)
  Hood length
  0.27 (0.10)
 * 0.18 (0.11)
  -0.07 (0.17)
   0.01 (0.02)
  0.00 (0.04)
  0.16 (0.05)
 **  0.14 (0.09)
  0.12 (0.09)
  0.07 (0.10)
  Hood height
  0.28 (0.11)
 * 0.19 (0.12)
  -0.04 (0.17)
   0.05 (0.02)
 * 0.03 (0.04)
  -0.01 (0.05)
   -0.16 (0.09)
 € -0.16 (0.09)
 € 0.10 (0.11)
  Horn reach
  -0.20 (0.12)
  -0.21 (0.13)
  0.39 (0.20)
 €  ---  ---  ---   0.06 (0.10)
  -0.04 (0.09)
  -0.23 (0.11)
 * Slit length
  -0.09 (0.09)
  -0.01 (0.10)
  0.03 (0.14)
   -0.01 (0.02)
  0.09 (0.04)
 * 0.04 (0.05)
   0.05 (0.09)
  0.22 (0.09)
 * -0.14 (0.10)
  Gap width
  -0.14 (0.09)
  -0.09 (0.09)
  -0.11 (0.14)
   -0.02 (0.02)
  -0.06 (0.04)
  -0.01 (0.05)
   0.13 (0.10)
  0.07 (0.09)
  0.17 (0.10)
 € Display size
  0.07 (0.09)
  -0.01 (0.10)
  0.34 (0.14)
 *  0.07 (0.02)
 ** 0.05 (0.04)
  0.19 (0.05)
 **  -0.06 (0.09)
  0.13 (0.09)
  0.17 (0.10)
  n  45  45  42   212  212  206   51  51  44  Total R
2  0.35
 * 0.1
  0.4
 *  0.08
 ** 0.05
  0.13
 **  0.15
  0.26
 € 0.37
 * ** P < 0.01, * 
P < 0.05, 
€ P < 0.10.
 slhhgyhlhrgw2 mm
gyhlgwslhhgyhlhrgwslhhCBAFigure 1. (submitted to JEB)
Figure 1.
 Top and side views of 
A. syriaca 
(A), 
A. viridiflora 
(B), and 
A. tuberosa 
(C)
 flowers showing the six floral traits: gynostegium width (
gy), hood length 
(hl), hood height (
hh), horn reach (
hr), slit length (
sl), gap width (
gw). Note that 
hoods are greatly reduced and horns absent in 
A. viridiflora. 
Two and a half hoods 
were removed for the side views.
31Hood HHood LDisplay sizeTotal 
Flower #Horn R
Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals********************Number of poll. visits**************Gyn WGap WTotal poll. 
duration**A. syriacaHood HHood LDisplay sizeTotal 
Flower #Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals******************************Gyn WGap W*A. viridißora***<0.20 0.20 - 0.39 0.40 - 0.59 0.60 - 0.79 0.80 - 0.99(a)
(b)
*******n = 42
n = 205
Figure 2. 
(submitted to JEB)
Figure 2.
 Path analysis relating traits, pollinator visitation over ten minutes, and 
fitness components to total yearly fruit production.
 Blue arrows are paths influencing 
male function (pollinia removals per flower), red arrows are paths influencing female 
function (pollinia insertions per flower) and fitness (fruit number), and black arrows 
affect both or show variation unexplained by the model when there is no trait at their 
origin. Blue, red, and black arrows are P < 0.10 and gray are P 
! 0.10. Single-headed 
arrows are hypothesized causal relationships and double-headed arrows are correlations. 
Arrow thickness represents the magnitude of the path coefficient, and dashed lines 
represent negative coefficients. White boxes represent variables included in the selection 
gradient analyses. Correlations between residual variation of endogenous variables are not 
shown for simplicity. *P < 0.05, ** P < 0.01, *** P < 0.001.
32(c)
A. tuberosa
Hood HHood LDisplay sizeTotal 
Flower #Horn R
Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals************************Number of HB visits*************Gyn WGap WTotal HB 
duration**n = 37
Figure 2 cont.
 (submitted to JEB)
Hood HHood LDisplay sizeTotal 
Flower #Horn R
Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals********************Number of poll. visits**************Gyn WGap WTotal poll. 
duration**A. syriacaHood HHood LDisplay sizeTotal 
Flower #Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals******************************Gyn WGap W*A. viridißora***<0.20 0.20 - 0.39 0.40 - 0.59 0.60 - 0.79 0.80 - 0.99(a)
(b)
*******n = 42
n = 205
Figure 2. 
(submitted to JEB)
Hood HHood LDisplay sizeTotal 
Flower #Horn R
Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals********************Number of poll. visits**************Gyn WGap WTotal poll. 
duration**A. syriacaHood HHood LDisplay sizeTotal 
Flower #Slit LTotal 
Insertions
Insertions 
per ßowerFruit #Removals per ßowerTotal 
Removals******************************Gyn WGap W*A. viridißora***<0.20 0.20 - 0.39 0.40 - 0.59 0.60 - 0.79 0.80 - 0.99(a)
(b)
*******n = 42
n = 205
Figure 2. 
(submitted to JEB)
Figure 2 (contÕd)
33 34             LITERATURE CITED
  35  LITERATURE CITED
  Bateman, A. J. 1948. Intra
-sexual selection in 
Drosophila
. Heredity 2:349
Ð368. Beavis, W. D. 
1998. QTL Analyses: Power, Precision, and Accuracy. Pp. 145
Ð162 in A. H. 
Paterson, ed. Molecular dissection of complex traits. CRC Press, Boca Raton.
 Betz, R. F., R. D. Struven, J. E. Wall, and F. B. Heitler. 1994. Insect pollinators of 12 milkweed 
(Asclep
ias
) species. Thirteenth North American Prairie Conference 45
Ð60. Bookman, S. S. 1984. Evidence for selective fruit production in 
Asclepias
. Evolution 38:72Ð86. Brown, A. 1898. The First Republic in America: An Account of the Origin of this Nation, 
Written
 from the Records Then (1624) Concealed by the Council, Rather Than from the 
Histories Then Licensed by the Crown. Houghton, Mifflin and Company, Boston and 
New York.
 Broyles, S. B., and R. Wyatt. 1990. Paternity analysis in a natural population of 
Asclepi
as exaltata: multiple paternity, functional gender, and the ÒPollen
-Donation Hypothesis.Ó 
Evolution 44:1454
Ð1468. Caruso, C. M., D. L. D. Remington, and K. E. Ostergren. 2005. Variation in resource limitation 
of plant reproduction influences natural select
ion on floral traits of 
Asclepias syriaca
. Oecologia 146:68
Ð76. Conner, J. K., and S. Rush. 1996. Effects of flower size and number on pollinator visitation to 
wild radish, 
Raphanus raphanistrum
. Oecologia 105:509
Ð516. Conner, J. K., H. F. Sahli, and K. Ka
roly. 2009. Tests of adaptation: functional studies of pollen 
removal and estimates of natural selection on anther position in wild radish. Annals of 
Botany 103:1547
Ð1556. Conner, J. K., S. Rush, S. Kercher, and P. Jennetten. 1996. Measurements of natural 
selection on 
floral traits in wild radish (
Raphanus raphanistrum
). II. Selection through lifetime male 
and total fitness. Evolution 50:1137
Ð1146.  Delph, L. F., and T.
-L. Ashman. 2006. Trait selection in flowering plants: how does sexual 
selection contribu
te? Integrative and Comparative Biology 46:465
Ð472. Fishbein, M., and D. L. Venable. 1996a. Diversity and temporal change in the effective 
pollinators of 
Asclepias tuberosa
. Ecology 77:1061Ð1073. Fishbein, M., and D. L. Venable. 1996b. Evolution of inflore
scence design: theory and data. 
Evolution 2165
Ð2177.  36  Fulton, M., and S. A. Hodges. 1999. Floral isolation between 
Aquilegia formosa
 and Aquilegia 
pubescens. Proc. R. Soc. Lond. B 266:2247
Ð2252. Galen, C. 1985. Regulation of seed
-set in 
Polemonium viscosum
: floral scents, pollination, and 
resources. Ecology 66:792Ð797. Galil, J., and M. Zeroni. 1965. Nectar system of
 Asclepias curassavica. Botanical Gazette 
126:144
Ð148.  Galil, J. 
and M. Zeroni.
 1969. On the or
ganization of the pollinium in 
Asclepias curassavica
. Bot. Gaz.
 130: 1Ð4. Gibbs, P. E. 2014. Late
-acting self-incompatibility
--the pariah breeding system in flowering 
plants. The New Phytologist 203:717
Ð734. Grace, J. B. 2006. Part IV Applications and 
illustrations. Pp. 206
Ð288 in Structural equation 
modeling and natural systems. Cambridge University Press.
 Hansen, D. M., T. Van der Niet, and S. D. Johnson. 2012. Floral signposts: testing the 
significance of visual Ònectar guidesÓ for pollinator 
behaviour and plant fitness. Proc. R. 
Soc. Lond. B 279:634
Ð639. Hodgins, K. A., and S. C. H. Barrett. 2008. Natural selection on floral traits through male and 
female function in wild populations of the heterostylous daffodil 
Narcissus triandrus
. Evolution
 62:1751Ð1763. Howard, A. F., and E. M. Barrows. 2014. Self
-pollination rate and floral
-display size in 
Asclepias syriaca
 (Common Milkweed) with regard to floral
-visitor taxa. BMC Evol. 
Biol. 14:144.
 Ivey, C. T., P. Martinez, and R. Wyatt. 2003. Variation 
in pollinator effectiveness in swamp 
milkweed, 
Asclepias incarnata 
(Apocynaceae). American Journal of Botany 90:214
Ð225. Johnson, S. D., C. I. Peter, L. A. Nilsson, and J. †gren. 2003. Pollination success in a deceptive 
orchid is enhanced by co
-occurring r
ewarding magnet plants. Ecology 84:2919
Ð2927. Kabat, S.M. 2010. Genetic structure of the common milkweed, 
Asclepias syriaca
 L. (Doctoral 
dissertation). Retrieved from deepblue.lib.umich.edu.
 Kephart, S. 1983. The partitioning of pollinators among three spe
cies of 
Asclepias
. Ecology 64:120
Ð133. Kephart, S. R. 1987. Phenological variation in flowering and fruiting of 
Asclepias
. American 
Midland Naturalist 118:64
Ð76. Kevan, P. G., D. Eisikowitch, and B. Rathwell. 1989. The role of nectar in the germination of 
pollen in 
Asclepias syriaca
 L. Botanical Gazette 150:266
Ð270.  37  Kingsolver, J. G., H. E. Hoekstra, J. M. Hoekstra, D. Berrigan, S. N. Vignieri, C. E. Hill, A. 
Hoang, P. Gibert, and P. Beerli. 2001. The strength of phenotypic selection in natural 
populations.
 The American Naturalist 157:245
Ð261. Kulbaba, M. W., and A. C. Worley. 2012. Selection on floral design in 
Polemonium brandegeei 
(Polemoniaceae): female and male fitness under hawkmoth pollination. Evolution 
66:1344
Ð1359.  Lande, R., and S. Arnold. 1983. 
The measurement of selection on correlated characters. 
Evolution 37:1210
Ð1226. Meagher, T. R. 1994. The quantitative genetics of sexual dimorphism in 
Silene latifolia
 (Caryophyllaceae). II. Response to sex
-specific selection. Evolution 48:939
Ð951.  Morgan,
 M. T., and D. Schoen. 1997. Selection on reproductive characters: floral morphology in 
Asclepias syriaca
. Heredity 79:433
Ð441. Morgan, M. T., and J. K. Conner. 2001. Using genetic markers to directly estimate male 
selection gradients. Evolution 55:272
Ð281. Nilsson, L. A., L. Johnsson, L. Ralison, and E. Randrianjohany. 1987. Angraecoid orchids and 
hawkmoths in central Madagascar: Specialized pollination systems and generalist 
foragers. Biotropica 19:310
Ð318. Ollerton, J., and L. Cranmer. 2002. Latitudinal 
trends in plant
!pollinator interactions: are tropical 
plants more specialised? Oikos 98:340
Ð350.  Pleasants, J. M. 1991. Evidence for short
-distance dispersal of pollinia in 
Asclepias syriaca
 L. Funct Ecol 5:75
Ð82. Queller, D. C. 1985. 
Proximate and ultimate causes of low fruit production in 
Asclepias exaltata
. Oikos 44:373
Ð381.  Queller, D. C. 1983. Sexual selection in a hermaphroditic plant. Nature 305:706
Ð707. R Core Team.
 2014. R: A language and environment for statistical computing. R Foundation for 
Statistical Computing, Vienna, Austria URL http://www.R
-project.org/.
 Rasband, W. S. 1997. ImageJ. U. S. National Institutes of Health, Bethesda, MD, USA.
 Real, L. A., and B. 
J. Rathcke. 1991. Individual variation in nectar production and its effect on 
fitness in 
Kalmia latifolia
. Ecology 72:149Ð155.  Robertson, C. 1886. Notes on the mode of pollination of 
Asclepias
. Botanical Gazette 11:262
Ð269.  Rosseel, Y
. 2012. lavaan: an R package for structural equation modeling. J Stat Softw. 
48: 1-36.  38  URL http://www.jstatsoft.org/v48/i02.
  Sage, T. L., and E. G. Williams. 1995. Structure, ultrastructure, and histochemistry of the pollen 
tube pathway in the milkweed 
Asclepias exaltata
 L. Sexual Plant Reproduction 8:257
Ð265. Sahli, H. F., and J. K. Conner. 2011. Testing for conflicting and nonadditive selection: floral 
adaptation to multiple pollinators through male and female fitness. Evolution 65:1457
Ð1473. SAS Institute Inc. 2012. JMP. C
ary, NC, USA.
 Schemske, D. W., and C. Horvitz. 1989. Temporal variation in selection on a floral character. 
Evolution 43:461
Ð465. Schemske, D. W., and H. D. Bradshaw Jr. 1999. Pollinator preference and the evolution of floral 
traits in monkeyflowers (
Mimul
us). Proc Natl Acad Sci USA 96:11910
Ð11915. Shannon, T. R., and R. Wyatt. 1986. Reproductive biology of 
Asclepias exaltata
. American 
Journal of Botany 73:11
Ð20. Silva, E. M., and B. B. Dean. 2000. Effect of nectar composition and nectar concentration on 
honey bee (Hymenoptera: Apidae) visitations to hybrid onion flowers. J. Econ. Entomol. 
93:1216
Ð1221. Wilbur, H. M. 1976. Life history evolution in seven milkweeds of the genus 
Asclepias
. The Journal of Ecology 64:223
Ð240.  Willson, M. F., and B. J. Rathcke. 
1974. Adaptive design of the floral display in 
Asclepias 
syriaca
 L. American Midland Naturalist 92:47
Ð57. Willson, M. F., and P. W. Price. 1977. The evolution of inflorescence size in 
Asclepias
 (Asclepiadaceae). Evolution 31:495
Ð511.  Wilson, P., J. D. Tho
mson, M. L. Stanton, and L. P. Rigney. 1994. Beyond floral Batemania: 
gender biases in selection for pollination success. The American Naturalist 143:283
Ð296. Woodson, R. E., Jr. 1941. The North American Asclepiadaceae. I. Perspective of the genera. 
Annals
 of the Missouri Botanical Garden 28:193
Ð244. Wyatt, R. 1976. Pollination and fruit
-set in 
Asclepias
: A reappraisal. American Journal of Botany 
63:845
Ð851. Wyatt, R. 1978. Experimental evidence concerning the role of the corpusculum in 
Asclepias
 pollination. Systematic Botany 3:313
Ð321.  Wyatt, R. 1980. The reproductive biology of 
Asclepias tuberosa
: I. Flower number, arrangement, 
 39  and fruit
-set. New Phytologist 85:119
Ð131.  Wyatt, R. 1981. The reproductive biology of 
Asclepias tuberosa
. II. Facto
rs determining fruit
-set. New Phytologist 88:375
Ð385.  Wyatt, R., and S. B. Broyles. 1994. Ecology and evolution of reproduction in milkweeds. Annual 
Review of Ecology and Systematics 25:423
Ð441. CHAPTER 3
FUNCTIONAL STUDIES AND ESTIMATES OF NATURAL SELECTION THROUGH 
MALE AND FEMALE FITNESS IN MILKWEEDS (
ASCLEPIAS
)Introduction
 Diversity in floral form has enticed biologists and non-biologists alike. Many flowering 
species rely on animals for pollination, which requires floral traits that can attract suitable 
pollinators and maneuver them efficiently. Some plant species have converged on suites of traits 
that may facilitate pollination by particular pollinator taxa (Fenster et al. 2004; Faegri and Van 
Der Pijl 2013), but many species appear to be more general in their pollination (Ollerton 1996; 
Waser et al. 1996; Zamora 2000), possibly because they experience selection from a dynamic 
network of pollinators. Regardless, the floral traits that directly increase the fitness of the plant 
are adaptive, and so identifying these adaptive traits and understanding how they function gives 
us insight into the diversity of floral forms. 
 Understanding these interspecific interactions can be achieved by studying how the 
flowers function to increase pollen receipt or removalÑthe effect of floral traits on pollination 
success (e.g. Galen and Stanton 1989; Murcia 1990; Young and Stanton 1990; Harder and Barrett 
1993; Conner et al. 1995). Functional studies can then be followed by studies of natural selection 
showing which floral traits affect male and/or female fitnessÑthe reproductive success of the 
plant (e.g. Johnston 1991; Conner et al. 1996; Caruso et al. 2003). 
 Available resources and pollen receipt are both important for plant fitness, and limitations 
in one or the other can have ecological and evolutionary consequences (Ashman et al. 2004; 
Knight et al. 2005). Female fitness is considered to often be resource limited; however, pollen 
40limitation cannot be overlooked, particularly in perennial outcrossing species (Charlesworth 
1989) that rely on pollinators for adequate pollination (Burd 1994). Following Bateman (1948), 
male fitness is not as likely as female fitness to be limited by resources, and so pollinator 
visitation and effectiveness can strongly influence male fitness. These factors limiting female and 
male fitness can in turn affect selection on floral traits (Bateman 1948; Wilson et al. 1994; 
Ashman and Morgan 2004; Knight et al. 2005); in populations where female fitness is limited by 
resources, male fitness may be more variable, allowing for selection on traits affecting 
pollination, and in populations where both female fitness and male fitness are limited by 
pollinators, there may be selection through both (Wilson 1994).
 Measures of selection in plants are often only through female fitness (Conner 2006), even 
though more than 90% of flowering species are hermaphrodites (Yampolsky and Yampolsky 
1922). Male fitness is best estimated by the number of viable seeds sired. This estimate requires 
paternity analyses of offspring to identify likely sires. To date, there are but a handful of studies 
that have measured selection on plant traits using fruits sired or seeds sired as their measure of 
fitness (Meagher 1994; Morgan and Conner 2001; van Kleunen and Burczyk 2007; Hodgins and 
Barrett 2008; Sahli and Conner 2011; Kulbaba and Worley 2012). 
 For decades, 
Asclepias
 has been a common plant taxon for studying male function. The 
pollen packets, or pollinia, make it easy to estimate male pollination success, by scoring pollinia 
removals per flower or per plant. Much of this earlier work focused on investigating the very low 
fruit to flower ratio in 
A. syriaca
, finding that only male function (estimated as the number of 
pollinia removed) suffered when flower number was reduced by half (Queller 1983), and that 
only male function was positively affected when flower number reached above a threshold 
41(Willson and Rathcke 1974; Willson and Price 1977). However, none of these studies addressed 
individual floral traits. Since then, two studies have measured selection on floral traits, including 
selection through male fitness components, using pollinia removals (Morgan and Schoen 1997; 
Caruso et al. 2005), and two studies have used paternity analysis or genotyped deposited pollinia 
in 
Asclepias
 to answer questions about seed siring within fruits, functional gender, and self-
pollination rates (Broyles and Wyatt 1990; Howard and Barrows 2014), but none have used 
paternity analyses to measure selection through male fitness using seeds sired.
 In addition to being an excellent system for asking questions related to male fitness, 
milkweeds are also known for their unusual floral structures. The flowers have a calyx of five 
sepals and a corolla of five petals, but are otherwise atypical (Fig. 3). The pistils are surrounded 
by a gynostegium that has five stigmatic slits leading to the central chamber. The gynostegium 
has ten pollen packets (pollinia) tucked into the wall; every two pollinia are attached by a small 
clip (corpusculum) that sits at the end of the stigmatic slit where it can catch onto pollinators. 
Surrounding the gynostegium is the corona, made up of five hoods that hold nectar, the only 
reward for pollinators. The hoods of many species contain horns, that arise from the inside base 
of the hood and extend up and out of the hood opening. 
 Our study focused on traits that we hypothesize affect pollinator attraction and efficiency. 
Traits such as gynostegium width, hood length, and hood height contribute to the overall size of 
the flower, making them likely to play a role in pollinator attraction, as well as helping position 
pollinator bodies, and legs in particular, between the hoods to contact the corpuscula and 
stigmatic slits. We also hypothesize that less conspicuous traits, such as slit length, horn reach, 
and the gap between hoods increase pollinator effectiveness. We predict that selection is more 
42likely to be directional (linear) rather than stabilizing (negative non-linear), because both species 
are generalists and are visited by pollinators of many shapes and sizes.
 We investigated these traits from a functional perspective, evaluating their effect on 
pollination success (pollinia insertions and removals). We also examined their effect on 
reproductive success (fruit number, viable seeds produced, and viable seeds sired) in two 
milkweed species, 
Asclepias incarnata 
and 
A. exaltata
. We asked: (1) Which floral traits function 
to increase the insertion and removal of pollinia and do they function similarly for both genders? 
(2) Which floral traits are under selection? Are they primarily adaptations for male fitness, 
female fitness, or both? and (3) Is there inter-annual variation in selection? 
Methods
Field methods
We studied single naturally occurring populations of 
A. exaltata 
(poke milkweed) and 
A. 
incarnata
 (swamp milkweed) in southwest Michigan between 2007 and 2011. The population of 
A. incarnata 
was located at the Pond Lab Facility of Michigan State UniversityÕs Kellogg 
Biological Station (42.410¡ N, 85.392¡ W) and flowered throughout July and early August. 
A. 
incarnata
 is a long-lived perennial with an unknown lifespan; this population was naturally 
established at some point after ponds were renovated in 2000, so was no more than six years old 
when our study began in 2007. Our study population of 
A. exaltata
 was located at the Fort Custer 
Training Center (42.298¡ N, 85.323¡ W). 
A. exaltata 
is also long-lived and flowers for two to 
three weeks in June; the age of plants in this population was unknown. 
Asclepias exaltata
 and 
A. incarnata 
are not thought to spread from rhizomes (Wilbur 
1976; but see Queller 1985), so we identified genetic individuals (genets) as clusters of ramets 
43that appeared to be growing from the same origin (i.e. growing from a single root crown). In 
2007 and 2008 we identified genets by sight, and in 2010 (as well as 2009 for 
A. exaltata
) we 
genotyped each ramet to check our assignments, especially for genets that were growing very 
close together; only about 4.5% of ramets had been wrongly assigned. In 2011, we used the 
assignments from 2010, and any new ramets were assigned to genets by sight. 
A. exaltata
 genets 
typically had just one ramet (mean = 1.14, s.d. = 0.40) and 
A. incarnata
 sometimes had two or 
more ramets per genet (mean = 1.89, s.d. = 1.42). 
A. exaltata 
is typically self-incompatible 
(Broyles and Wyatt 1993; Wyatt and Broyles 1994; but see Lipow 1999), and self-fertile plants 
suffer high inbreeding depression (Himes and Wyatt 2005). 
A. incarnata 
can produce hundreds 
to thousands of flowers and is sometimes self-compatible (Kephart 1981); however, despite the 
possibility of high rates of geitonogamy, populations typically have high rates of outcrossing 
(Ivey et al. 1999).
We collected trait and fitness data from the 
A. incarnata 
population for five years 
(2007-2011). In 2007 and 2008 we sampled the same 50 plants haphazardly chosen from around 
four ponds, and in 2009-2011 we sampled every flowering plant in the population; however, 40 
of the 50 
A. incarnata
 plants from 2007 and 2008, and a number of plants from 2009 were 
destroyed before 2010 due to pond renovations. We collected data from the 
A. exaltata 
population for three years (2009-2011), collecting data from every flowering plant in the 
population each of those years. In every year we measured floral traits, scored pollinia removals 
to assess male function of the floral traits, and counted fruits to measure selection through female 
fitness. In 2007-2009, we also scored pollinia insertions to assess female function. We collected 
trait, pollination, and fitness data from every ramet per plant, except in 2008, when we collected 
44these data from just one ramet per 
A. incarnata 
genet. 
We measured floral traits by photographing two fresh flowers from each plant in 2007, 
three in 2008, and 1-4 (mean = 1.8-2.3) in each year between 2009 and 2011. We photographed 
each flower from the top and side (with 2.5 hoods removed to allow traits to be clearly seen) 
using a digital SLR camera with a 60mm macro lens. We first determined landmarks that were 
identifiable across the range of floral variability (Fig. 3), and then used these landmarks to make 
calibrated linear measurements from the digital photographs, that each included calipers set at 
5mm, using ImageJ (Rasband 1997). The traits we measured were: gynostegium width, hood 
length, hood height, horn reach, slit length, and gap width. We did not measure gynostegium 
width or gap width in 2007, and gynostegium width was measured differently in 2011, because 
we changed to landmarks that could be identified more accurately (Fig. 3, 
gr) to estimate 
gynostegium width.
Using a subset of plants that had two or three flowers measured within a year, we 
calculated trait repeatabilities across plants within year as the percent of the total variance in 
floral traits in that year explained by the variance among plants using ANOVA with plant as the 
random predictor variable; we repeated this for each year in both species. Among-plant variation 
explained more than 60% of the variation for 20 of 28 traits across years in 
A. incarnata
, and 
more than 70% of the variation for 16 of 18 traits across years in 
A. exaltata. 
Among-plant 
variation explained less than 50% of the variation in gap width in 2010 and 2011, and several 
traits in 2007Ñall in 
A. incarnata
 (Table 6); note that gap width is the only trait measured that is 
a distance between two floral structures (i.e. adjacent hoods) rather than a dimension of a single 
floral structure.
45We estimated total flower number to account for effects of resources, age, and plant size 
on fitness. We treated flower number both as a predictor variable that may directly affect fitness, 
as well as a measure of plant condition that is correlated with environmental variables that could 
also affect fruit production (Scheiner et al. 2002; Stinchcombe et al. 2002). For 
A. exaltata
 we 
counted the total number of flowers on every plant. For 
A. incarnata
, which can produce 
thousands of flowers, we estimated flower number in 2007 and 2011 as the total number of 
flowering branches across all ramets, in 2008 the flower number estimate was the number of 
ramets, and in 2009 and 2010 we estimated total flower number by counting all umbels and 
multiplying this by the mean number of flowers per umbel per plant. The correlations among 
ramet number, branch number, and umbel number ranged between 0.35 and 0.75. Resources are 
not likely to be affecting the individual floral traits as the magnitude of the correlation between 
flower number and the six floral traits was less than 0.35 in 
A. incarnata 
and less than 0.20 in 
A. 
exaltata.
To assess floral trait function, we scored pollinia removals and insertions. Using a hand 
lens in the field, we scored pollinia removed per flower, indicated by a missing corpusculum 
(arrow; Fig.1) (range: 0-5), and examined the five stigmatic slits for inserted pollinia (range: 
0-5); we discontinued collecting data on pollinia insertions after 2009, because they were a poor 
predictor of fruit number (see results). Because flowers within an umbel open asynchronously, 
then remain continuously open for several days (Wyatt 1981; Kephart 1987), we scored pollinia 
removals and insertions on every flower only from umbels that had fully bloomed and were 
mature enough that some flowers were beginning to senesce; note that some flowers remained 
open for several days after sampling and likely continued to receive and donate pollinia. The 
46only exception was that in 2007 we sampled just three flowers from three mature umbels on each 
ramet. 
A. incarnata 
plants tended to have many umbels (in 2010: mean = 56.9, s.d. = 64.1), so 
we sampled 1-4 umbels per ramet across one to several days with the goal of sampling 
approximately 100 flowers per genet (Table 7). 
A. exaltata
 plants had fewer umbels (mean = 2.3, 
s.d. = 1.3), so
 we scored removals and insertions from nearly every flower from every umbel in 
the population in 2009-2010, so we also have estimates for total pollinia inserted and removed 
per plant; in 2011, we were unable to score pollinia removals and insertions from every plant 
before flowers senesced.
We conducted paternity analyses for offspring in 2010 using microsatellite markers to 
genotype parents and offspring. We collected leaf tissue from all flowering ramets in both 
populations and stored the leaf samples at -80 C. We collected tissue from a total of 216 
A. 
incarnata 
ramets and 155 
A. exaltata 
ramets. To genotype offspring, we allowed fruits to ripen 
on the plants, then collected every fruit prior to dehiscence, noting from which maternal plant 
and ramet it was collected; we collected 132 fruits from 
A. exaltata 
and 1971 fruits from 
A. 
incarnata
.Pollinator observations and pollen limitation
We conducted extensive pollinator observations over the duration of the 
A. incarnata 
flowering seasons in 2008 and 2010 and more limited observations late in 2009 that did not 
represent the full breadth of insect visitors. In 2008, we made video recordings of each of the 50 
plants we sampled for at least ten minutes each for a total observation time of 9h. We used video 
so we could observe pollinators on the same day that we collected data on pollination success. 
We also wanted to document pollinators so their behavior, contact with corpuscula and stigmatic 
47slits, and pollinia loads could be observed on a frame by frame basis. In 2010, we observed 113 
plants for a total of 9h 25m. In each year we calculated the percent pollinator composition by 
dividing the number of individuals in each pollinator group by the total number of pollinators 
observed. We also captured 119 insects, chilled and photographed them, scored the mean number 
of pollinia they carried, and divided that by the total pollinia carried by all pollinator groups to 
calculate the percent total pollinia. Pollinator importance was calculated by multiplying the mean 
pollinia carried for a taxon by the number of individuals of that taxon in a given year, and 
dividing by the grand total for all taxa (Table 8). 
We attempted pollinator observations of the population of 
A. exaltata,
 including dusk and 
dawn observations, in addition to the many days we were in the population during peak 
flowering, but we saw fewer than five individuals carrying pollinia over three years.
 There is 
little published data on nocturnal pollinators in 
A. exaltata
, but nectar is produced throughout the 
night and increases in sugar content throughout the day, reaching concentrations that are ideal for 
bee pollination (Wyatt and Shannon 1986)
. Bumble bees (
Bombus 
sp.)
 and skippers (
Epargyreus
 sp.) were the only two taxa that we saw with pollinia on their bodies; another study of 
A. exaltata
 recorded pollination by 
Bombus 
and 
Apis
 (Betz et al. 1994), though we never saw honey bees 
(Apis 
sp.) at this site
. Additional visitors we recorded were: fritillary (
Speyeria 
sp.), skipper 
(Thorybes 
sp.), and black swallowtail (
Papilio 
sp.)
 butterflies, one ruby-throated hummingbird, 
and single Lepidopteran legs stuck in different corpuscula. Despite these other studies finding 
little to no nocturnal pollination, we cannot rule out the importance of nighttime pollinators in 
our study population given that the few pollinators we saw during the day cannot account for the 
number of fruits that were produced each year. For night flying insects, attraction, or the contrast 
48of flowers against their forest background, may be the most important factor influencing 
pollination.
In 2012, we conducted a pollen addition experiment to test for evidence of pollen 
limitation in the population of 
A. exaltata
 because of the scarcity of observed pollinators. We 
paired plants (n = 22 pairs) based on umbel number, flower number, and flowering time. We then 
randomly assigned one plant per pair to receive the treatment of one hand-inserted pollinium per 
flower from a mix of pollen donors. This was an average increase of 60 times the pollinia 
naturally received. We then recorded fruit initiation and final fruit number and compared the 
manipulated plants to the unmanipulated plants using a paired t-test.
Fitness measures
We estimated male and female fitness for both species. Because both of these species are 
perennial, we were unable to measure lifetime fitness, and instead estimated annual fitness across 
multiple years. We used total fruit number per plant as our estimate of annual female fitness; 
fruits per plant is highly correlated with total seeds per plant for both species (r > 0.97; Table 9). 
All fruits were counted in September once they had matured. In 2008, we counted fruits on just 
one ramet and then multiplied them by the total number of ramets. We estimated female and 
male fitness in 2010 using viable seeds produced and viable seeds sired (see below), and then 
summed these for a measure of total annual fitness.
Paternity
Milkweed fruits are commonly sired by a single father, because a pollinium has a 
sufficient number of pollen grains to fertilize all ovules within an ovary (Wyatt 1976; Kephart 
1981). Flowers have five stigmatic openings; two openings lead to one ovary and three lead to 
49the other ovary, so there is the possibility of two or three sires if multiple pollinia are deposited 
in close succession. We measured the occurrence of multiple paternity for both populations by 
genotyping ten randomly chosen offspring per fruit for each of ten fruits in 
A. exaltata
, and 8-19 
randomly chosen offspring per fruit for each of 18 fruits in 
A. incarnata
. We used GERUD2 
(Jones 2001; 2005)
 to identify the combination of genotypes of likely fathers, given the maternal 
genotype. We found no evidence of multiple paternity in our population of 
A. exaltata
; however, 
we found that one of the 18 
A. incarnata 
fruits was sired by more than one father, yet the primary 
father sired up to 70% of the 19 seeds genotyped from that fruit. Given these results, we treated 
seeds within each fruit as full siblings for our paternity analyses.
We attempted to germinate 2-15 seeds per fruit for all fruits that contained at least one 
fully developed seed. Initially, 15 seeds from each fruit were nicked and sterilized in 10% bleach 
water, then were germinated in small petri plates between damp paper towel and transplanted 
into 2.25Ó pots. Later, 2-6 seeds were sown directly into moist soil in 72-cell trays without 
nicking or sterilization, then thinned to one seed per cell after germination. The latter technique 
had similar success germinating at least one seed per fruit (82% of fruits compared to 85% for 
the first technique). In both cases, seeds were kept moist and were cold-stratified for 4-5 weeks 
at 4-6C, then placed at ~27C to germinate where they received 14-16h daylight at around 24C 
for 4-6 weeks either in the greenhouse or a growth chamber. We collected tissue from one 
seedling per fruit for 
A. incarnata 
(1303 seedlings) and 1-3 seedlings per fruit for 
A. exaltata, 
(271 seedlings from 112 fruits)
. We extracted DNA from 0.1g of parental and offspring tissue using FastDNA Kits from 
MP Biomedicals. We tested 41 microsatellite markers developed for 
A. syriaca 
(O'Quinn and 
50Fishbein 2009; Kabat et al. 2010; Straub et al. 2011) and nine of those amplified and had 
sufficient allelic variation in one or both of our populations. We genotyped 
A. incarnata 
individuals at six loci and 
A. exaltata 
individuals at five loci (Table 10). We genotyped each 
parent twice and each offspring once. The first set of parental
 PCR products was run on 5% 
acrylamide gels, visualized on an FMBIO; alleles were scored using FMBIO Analysis 8.0 
(Hitachi Software Engineering 1991Ð1999) and placed in bins using Allelogram version 2.2 
(Manaster 2002). The second set of the parents and all offspring were visualized using an 
Applied Biosystems 3730
xl DNA Analyzer at YaleÕs DNA Analysis Facility and genotyped using 
Geneious version 8 (www.geneious.com; Kearse et al. 2012). 
Analyses
We used maximum likelihood paternity analyses to assign fractional paternity to the 
seeds within each fruit in both population. Based on the population allele frequencies, the 
exclusion probability, or chance of excluding individuals that were not the true sire, given the 
maternal genotype, was 91% and 93% for 
A. incarnata 
and 
A. exaltata
, respectively (Table 10). 
For 
A. incarnata 
we used Cervus (Marshall et al. 1998) to determine fractional paternity for all 
of the viable fruits. We assumed that all seeds within a fruit were full-siblings (see above), so 
possible sires were assigned seeds from that fruit proportional to their LOD scores for the single 
offspring from each fruit. Based on the exclusion probabilities and the number of parental plants, 
we expected an average of 8.9 sires per 
A. exaltata 
offspring and 12.5 sires per 
A. incarnata 
offspring. For each parent plant, their assigned seeds were summed across fruits to determine 
their total seeds sired in 2010, or annual male fitness.
Because we genotyped multiple seeds per fruit in 83% of 
A. exaltata 
fruits, we used 
51Colony2 (Wang 2004) to determine paternity. Similar to Cervus, Colony2 uses maximum 
likelihood to determine parent-offspring relationships and the output contains possible sires and 
their probability of paternity. Unlike Cervus, Colony2 can consider both paternal and maternal 
sibship among full siblings to increase the power to determine the paternal genotype of the 
unknown sire and exclude additional non-sires. The input file can be found in the supplementary 
materials.
We assigned paternity to 6,639 
A. exaltata 
seeds and 65,941 
A. incarnata 
seeds. We were 
able to assign a unique sire to only 3% of the 
A. incarnata 
fruits, and seeds were proportionally 
divided among fewer than 13 possible sires for 61% of the fruits (note that 12.5 was predicted 
average number of sires with positive LOD scores). On average, seeds within each fruit were 
divided among 12.5 (± 8.98) possible sires, exactly matching the exclusion probability. In 
contrast, we were able to assign a unique sire to 79% of the 
A. exaltata 
fruits, and the remaining 
21% of fruits had five or fewer possible sires, which is much lower than the exclusion 
probability predicted; this was likely due to the inclusion of paternal sibship data.
 Each of our selection gradient analyses (Lande and Arnold 1983) regressed male and 
female pollination success (pollinia inserted or removed per flower) or fitness estimates (fruit 
number, seeds produced, seeds sired, and total seeds) onto the six floral traits and flower number. 
We considered including ramet number as a covariate as an estimate of plant size, but its 
inclusion in the models had no effect on the outcome of the analyses for either species; ramets 
vary greatly in size (La Rosa pers. obs.) and so flower number may be a better estimate of the 
vegetative size of plants. Correlations between ramet number and flower number ranged from 
0.35 - 0.65 (Table 11). All response variables were relativized by dividing by the mean, and all 
52predictor variables were standardized to a mean of zero and standard deviation of one. In total, 
we performed 2-5 linear multiple regressions per species per year. The distribution of all floral 
traits was normal in each year for both species. The residuals of some models were 
heteroscedastic; however, even though log transforming female fitness and flower number did 
improve the distribution of the residuals, it did not improve the fit of the models. Selection 
gradients, standard errors, and significance are thus from untransformed variables. Between 2009 
and 2011 only 51 
A. exaltata 
plants and 70 
A. incarnata 
plants
 bloomed in all three years. In 
2009, 41% of the 
A. incarnata
 ramets were destroyed by voles before the fruits had matured; 
these were excluded from analyses of selection through female fitness. This should not bias the 
selection gradients unless the vole damage was correlated with the floral traits. In all other years, 
fewer than 2% of plants died or were damaged by rodents before the fruits could be counted. 
For the selection gradient analyses using seeds produced and seeds sired, the total number 
of seeds was summed from fruits that had at least one seed germinate (viable fruits), eliminating 
almost one third of the fruits, and the seeds they contained, from these analyses. The seeds 
assigned to possible sires also came from viable fruits, as those are the only fruits from which we 
could collect DNA and genotype to perform paternity analyses. However, there is a high 
correlation between total fruit number per plant (viable plus inviable fruits) and seed number 
from viable fruits; the correlation is 0.95 for 
A. incarnata 
and 0.87 for 
A. exaltata 
(Table 9). A 
majority of the fruits from eight plants were not genotyped and thus were not included in the 
paternity analysis for 
A. incarnata
. We also found no negative relationship between total flower 
number and the proportion of the fruits that were viable that would have suggested that the 
inviable fruits were due to higher rates of geitonogamy and inbreeding depression on larger 
53plants. Using seeds produced and seeds sired, we also measured functional gender (Lloyd 1980) 
that ranges from 0 (male; plant fitness is entirely from seeds sired) to 1 (female; plant fitness is 
entirely from seeds produced). Measures of functional gender allow us to test the prediction that 
plants with more flowers function more as male. All selection gradient analyses were conducted 
using JMP 10.0 (SAS Institute Inc. 2012). 
Results
The highest correlations among the six floral traits that we measured were between horn 
reach and hood height in 
A. incarnata 
(r = 0.52; Table 12a), and between horn reach and hood 
length in 
A. exaltata 
(r = 0.47; Table 12b). In both 
A. exaltata 
and 
A. incarnata
, the trait means 
were similar across years and the phenotypic coefficient of variation for each trait remained 
relatively constant (Tables 13 and 14). 
Functional effects of floral traits
 The composition and quantity of hymenopteran pollinators visiting these two species 
differed, which may have resulted in differences in selection. 
A. incarnata 
had wasp and bee 
visitors of many sizes (Table 8), while we only observed large bees and lepidopteran pollinators 
visiting 
A. exaltata
. Slit length was the only trait to affect pollinia insertions or removals per 
flower in 
A. exaltata
; shorter slits increased pollinia removals in 2010 (Fig. 4b). In 
A. incarnata
, longer stigmatic slits increased the number of pollinia inserted per flower in 2009 (Fig. 5a) and 
also increased the number of pollinia removed per flower in 2011 (Fig. 5b). We also found that 
increased gynostegium width (in 2 years), hood length (in 3 years), and gap width all 
significantly increased pollinia removals per flower in 
A. incarnata
. Across 
A. incarnata
, the 
direction of the selection gradients matched between male and female function in 14 of 16 
54instances, suggesting very little conflict between genders; in 
A. exaltata 
they only matched two 
out of six times.
Finally, flower number in 
A. exaltata
 had a significant effect on both pollinia removals 
per flower and insertions per flower in 2009. This was not so in 
A. incarnata
, where flower 
number was often 30 or more times that of 
A. exaltata
.Selection on floral traits 
We detected selection on hoods and gynostegia (Figs. 6 and 7), which make up a majority 
of the flower. There was selection on hood height through fruit number in both species. Selection 
gradients for hood height were consistently negative across years in 
A. exaltata
; in contrast, 
selection fluctuated between years in 
A. incarnata
. Hood length was also under selection through 
fruit number, and selection was nearly always positive for both species, except in two years in 
A. 
incarnata.
 Selection on gynostegium width through fruit number in 
A. incarnata, 
was positive 
three out of four years.
There was selection on the less conspicuous trait slit length through fruit number in both 
species (Figs. 6 and 7). Selection on slit length was positive in all years but one in 
A. incarnata
, and consistently positive in 
A. exaltata. 
There was also fluctuating selection on gap width in 
A. 
exaltata
; there was selection to decrease it in 2009, then selection to increase it in 2010.
Comparing the results from pollinia inserted per flower (female pollination success) to 
the selection gradients for fruit number (female fitness), there was a very weak and negative 
correlation between the functional effects and fitness effects on the six floral traits (r < |-0.17|, p 
> 0.73 for both species). Pollinia insertions did not appear to affect fruit set, which could be due 
to a variety of mechanisms including resource limitation, genetic incompatibility, and improper 
55placement of pollinia. 
To check for evidence of stabilizing or disruptive selection on the floral traits, we re-ran 
the selection models and included non-linear terms for each predictor variable in the selection 
gradient analyses. We found stabilizing selection on horn reach through fruit number (
! = -0.18, P < 0.05) in 2011 in 
A. incarnata
; horn reach was the only trait that was never under significant 
directional selection in either species
. Flower number had the greatest effect on female fitness. Plants with more flowers, 
flowering branches, or ramets produced more fruits (Figs. 6 and 7). Recall that flower number 
did not affect floral function in terms of per flower removal and insertion rates, except for 
A. 
exaltata
 pollinia removals and insertions
 in 2009 (Fig. 4a, b). 
Selection through seed production and seed siring success in 2010
In 2010, the only instance of selection through viable seeds produced was through gap 
width in 
A. exaltataÑ
larger spaces between hoods results in higher male fitness
. And 
gynostegium width was the only trait under selection through female fitness, but was under 
negative directional selection in both species. Overall, the direction of the selection through 
seeds produced and sired was never in conflict (Fig. 8). 
In A. exaltata
, the selection gradients for fruit number and viable seeds produced were 
consistently in the same direction, but were of different magnitudes for gynostegium width and 
hood height between the fitness estimates (Figs. 6 and 8a). There was significant selection on 
hood height through fruit number (
" = -0.23, p = 0.01; Fig. 6), but there was not significant, or 
even marginally significant, selection through seeds produced (
" = -0.11, p = 0.25; Fig. 8a). In 
A. 
incarnata, 
there was also significant selection on hood height through fruit number in 2010 (
" = 560.18, p = 0.03; Fig. 7), but selection through seeds produced was identical, yet not significant (
! = 0.18, p = 0.10; Fig. 8b). Significant selection on gap width in 
A. exaltata 
remained almost 
identical through fruits and seeds produced. The major difference between fruit number and 
seeds produced is the exclusion of the non-viable fruits.  
The effect of flower number on fitness in 2010 was strong and significant for both species 
and through both genders (Fig. 8). We found that in 
A. exaltata,
 the effect of flower number was 
similar through male and female fitness, but in 
A. incarnata 
the effect was much stronger 
through female fitness than through male fitness. We suspect that the weaker effect on flower 
number through male fitness may be an artifact of fractional paternity, and that if we had more 
power to detect the single sire, the variance in male fitness would increase and the effect of 
flower number would be stronger, perhaps becoming similar to the effect through female fitness, 
as is the case for 
A. exaltata.
Flower number may play a role in the functional gender of individuals if additional 
flowers are in fact more important for male fitness. Our paternity analyses allowed us to quantify 
the functional gender of individuals and of each population (Fig. 9) by comparing identical 
fitness currencies for each gender. In the 
A. exaltata
 population, there were some plants that were 
exclusively female, (i.e., they produced seeds but did not sire seeds), but zero plants that were 
exclusively male. Of the 130 genets, 77 produced viable fruits, and 57 of those maternal genets 
sired all of the offspring. Thus, 53 plants had zero total fitness in 2010 and are not included in the 
functional gender plot. We found that plants within this population had a mean functional gender 
of 0.61 (Fig. 9a), making them more female than male on average. In 
A. incarnata
 plants were on 
average more male (functional gender = 0.35; Fig. 9b), but this is also likely an artifact of the 
57way paternity was assigned in 
A. incarnata 
that spread out the distribution of paternity.
Pollinators and Pollen limitation
 There were substantial differences in pollinator composition between 2008 and 2010 in 
A. 
incarnata 
(Table 8)
. In all years, pollinators appeared to be plentiful, and in the three years 
between 2007-2009 
A. incarnata 
plants produced one fruit for every 34 pollinia inserted on 
average, suggesting their fitness was not limited by pollen. 
A. exaltata
, on the other hand, 
produced on average one fruit for each pollinium that was inserted. Adding supplemental pollen 
to 
A. exaltata 
plants in 2012 had a strong effect that resulted in a fruit increase of more than 50%, 
which was a marginally significant increase in fruits per plant (0.86 ± 0.14 vs. 0.54 ± 0.15, 
respectively; t = 1.69, df = 25, p = 0.0517), suggesting pollen limitation. If pollen was limited in 
2009-2011 as well, we should find that there will be selection through female fitness on traits 
that interact with pollinators.
Discussion
 In this study, we asked questions about the function of six floral traits for pollination and 
the effect of those traits on selection through both male and female components. We used pollinia 
inserted and removed per flower to find that slit length, gynostegium width, hood length, and gap 
width functioned to affected male and female pollination success. We then found that all six traits 
had some effect on fruit number, a common measure of female fitness in milkweeds (e.g., 
Willson and Rathcke 1974; Wyatt 1976; Morgan and Schoen 1997; Caruso et al. 2005), in one or 
both species. To quantify selection through male fitness, we calculated seeds sired, a measure 
that is more accurate than pollinia removed, but that is not often estimated in plants. We found 
that gynostegium width was important for male fitness in both species. 
58Function of floral traits
 Four of the six floral traits functioned to increase the insertion and removal of pollinia on a 
per-flower basis, but only slit length had an effect in 
A. exaltata
. Slit length is unlikely to affect 
pollination through attraction, so shorter slits increased pollinia removals per flower by possibly 
increasing contact with the corpusculum, the structure that attaches pollinia to pollinators. The 
end of the slit near the base of the gynostegium is raised away from the gynostegium and then 
tapers towards the corpusculum that is positioned at the upper end of the stigmatic slit near the 
top of the gynostegium. 
A. exaltata 
is one of the larger flowered 
Asclepias 
species in the region, 
and so individuals with longer slits may have more area for pollinators to grip the flower closer 
to the base, missing the region where the corpusculum is positioned, and reducing pollinia 
removals. This population of 
A. exaltata 
may also be visited by pollinators primarily at night; 
however, our lack of evidence of day pollinators is not proof of important nocturnal pollinators. 
If this population relies on nocturnal pollinators, then floral display may be more important for 
attracting pollinators and only traits that contribute to pollinator efficiency, such as slit length, 
would have an effect on pollinia insertions and removals. 
 In 
A. incarnata, 
slit length and gynostegium width were important for male and female 
floral function. In almost every year, both traits acted to increase pollinia removals and insertions 
as they increased in size and length, so there was rarely any conflict between genders. Pollinators  
typically clasp the flowers between the hoods on either side of the gynostegium or across several 
flowers in this species, which results in their tarsi often contacting the corpuscula and the 
stigmatic slits (Macior 1965; La Rosa pers. obs.). If the width of the gynostegium correctly 
positioned pollinators, and the slit length in this small-flowered species allowed for increased 
59probability of pollinator legs sliding between the adjacent anther wings (the structures that form 
the stigmatic slit), then both male and female pollination success could be increased 
simultaneously. Two other traits, hood length and gap width functioned to increase pollinia 
removals per flower in 
A. incarnata, 
but had no effect on pollinia insertions per flower
. Hood 
length showed the most consistent pattern of all the traits, having a significant positive effect on 
pollinia removals per flower for three consecutive years, but we do not know if this effect was do 
to an increase in pollinator visitation or an increase in efficient pollination. 
Selection on floral traits
 Hood height was consistently important for female fitness (fruit number) and was 
significant in 2010; however, selection on hood height was not significant through seeds 
produced in 2010. In that year, hood height contributed to total fruit production, but about one 
third of the fruits produced seeds that did not germinate. Fruits in milkweeds are known to be 
aborted after they have begun to develop, but typically this abortion happens while the fruits are 
still quite small, and all of the fruits that we collected were fully mature. Himes et al. (2005) 
found that in 
A. exaltata
, plants that can mature fruits from self pollen, will have low seed 
germination; this could also be true of 
A. incarnata.
 The high inviability of seeds within mature 
fruits might indicate high levels of inbreeding depression due to geitonogamous selfing. Given 
that plants with large floral displays have been shown to have higher rates of geitonogamous 
selfing (Harder and Barrett 1995), we were surprised to find that plants with more total flowers 
did not have higher proportions of inviable fruits. Other factors, such as resource availability or 
disease, may have affected seeds as they matured. 
 Only hood height showed inter-annual variation in significant selection between 2009, 
60when there was selection for shorter hoods, and 2010, when selection was for taller hoods. Our 
pollinator observations and others have shown that pollinator composition and importance can 
change drastically between years (Table 8) (Herrera 1988; Price et al. 2005; Wiggam and 
Ferguson 2005), although we did not have sufficient data to directly compare the pollinator 
composition in 2009 to the other two years. Pollinator composition can also change dramatically 
within a season (Herrera 1988; Ashman and Stanton 1991; Wiggam and Ferguson 2005). We do 
know that plants flowered 11-14 days later in 2009 than they did in 2010, and this may have 
caused a shift towards later season pollinators, which could affect fitness (Rafferty and Ives 
2012). If pollinators were also very different in 2009, it may explain the fluctuations in 
significant selection in hood height, since we would expect the pollinator types to differ in the 
amount or direction of selection they impose as others have found in a generalist species 
(Schemske and Horvitz 1989; Sahli and Conner 2011). 
 Estimates of annual fitness for 
Asclepias 
that others have used in the past (fruit number and 
pollinia removals) differ in their ability to predict selection through viable seeds produced and 
sired. We found that measuring selection through fruit number is a good way to estimate the 
direction of selection, and in some cases the magnitude, but does not necessarily predict which 
traits are under significant selection through female fitness. By summing the pollinia removed 
per flower in 
A. exaltata 
and multiplying removals per flower by the estimated flower number in 
A. incarnata 
we estimated total pollinia removed and found that its correlation with seeds sired 
was 0.44 and 0.45 for these two species, which is nearly identical to the value calculated by 
Broyles and Wyatt (1990) for 
A. exaltata
. Given that the R
2 is only 0.19, total removals are a 
rather poor predictor annual male fitness. 
61Function, fitness, and limitations
 Gynostegium width and slit length were the only traits to have a similar effect on female 
pollination success and reproductive success; otherwise, the overall disconnect between how the 
traits function to affect pollinia insertion and removal and how they affect female and male 
fitness may depend a lot on the availability of resources or pollinators. In populations where 
resources limit female fitness, individuals receive adequate pollen, so that the only differences in 
fitness are due entirely to available resources. Under such circumstances, we would not expect to 
find selection on a trait that differentially affects pollination success. Alternatively, if fitness is 
limited by the abundance or effectiveness of the pollinators, the traits that differentially affect 
pollinia insertions and removals may differentially affect female and male fitness, respectively. 
 The population of 
A. incarnata 
appeared to be resource limited due to its high number of 
pollinia insertions per fruit, yet we found traits that were similarly important for pollen receipt 
and fruit number in 2009. One possible explanation is that female fitness in the population of 
A. 
incarnata 
could have been limited by access to 
effective pollinators in 2009. It is pollinated by a 
variety of generalist pollinators,
 so it is likely that some pollinators are more effective than others 
and that the pollinia inserted into the stigmatic slits will not always result in fertilization. Studies 
have demonstrated that different insects are not equally effective at removing and inserting 
pollinia in 
A. incarnata 
(Ivey et al. 2003; Rafferty and Ives 2012) or in 
A. exaltata 
(Stoepler et al. 
2012), and studies on other species have shown that different generalist pollinators may not be 
equally effective (Herrera 1988; Fishbein and Venable 1996; Sahli and Conner 2007; Rocca and 
Sazima 2012). 
62Flower number and fitness
 Flower number appears to be overwhelmingly important for male and female fitness. Many 
people have studied milkweeds because of their puzzling low fruit to flower ratio and concluded 
that the large floral displays benefit male fitness components (Willson and Rathcke 1974; 
Willson and Price 1977; Chaplin and Walker 1982; Queller 1983). However, Broyles and Wyatt 
(1990) found that there was no negative correlation between functional gender and flower 
number in plants that had both male and female fitness (n = 17, r = 0.35) demonstrating that 
plants with more flowers did not tend to have greater male fitness relative to female fitness. We 
found this to be similarly true in both of our species. The correlation between functional gender 
and flower number was 0.11 (n = 57) for 
A. exaltata 
and 0.54 (n = 104) for 
A. incarnata
; positive 
correlations indicate that plants with more flowers tended to be more female than male. 
 Flower number appeared to have a strong positive effect on female fitness in both species, 
but it may be that flower number and fruit number are both correlated with resources, thus 
resources may actually be the primary cause of this strong relationship (Scheiner et al. 2002; 
Stinchcombe et al. 2002). Male fitness is not likely to be influenced by resources in the way that 
female fitness is (Bateman 1948), and so is interesting that the effect of flower number on male 
fitness in 
A. exaltata 
is nearly as strong as the effect on female fitness, showing again that total 
flower number in milkweed is indeed important for male fitness (Willson and Rathcke 1974; 
Willson and Price 1977; Chaplin and Walker 1982; Queller 1983). In 
A. incarnata
, we found that 
the effect of flower number on male fitness is much weaker than on female fitness. We suspect 
that this to be an artifact from assigning fractional paternity to many sires, even though each seed 
had just one sire. By assigning paternity to many sires (mean = 12.5), we distributed seeds from 
63each fruit to many possible sires across the population, and thus reduced the variation in male 
fitness. This reduction in variation potentially reduced the estimates of selection through male 
fitness. Additionally, we do not know if the difference in functional gender between the two 
species is biological or if they also differ as a result of the two fractional paternity methods. In 
the future, we will genotype additional seeds per fruit as we did for 
A. exaltata 
and rerun the 
selection analyses, using the additional paternal sibship information.
 Conclusions
 We found significant directional selection on all of the floral traits except horn reach, 
which we found to be under stabilizing selection for one year in 
A. incarnata. 
These results 
suggest that the floral traits are likely to be adaptive, and gynostegium width and slit length are 
adaptive by increasing pollinia insertions in 
A. incarnata 
in years when pollinators could be 
limiting. The population of 
A. exaltata 
was pollen limited, and so we did not expect more 
selection through male fitness (Bateman 1948), and our results from 2010 match this. We found 
that flower number was very important for total seeds produced and seeds sired; however, we do 
not know to what extent available resources contributed to female fitness or the production of 
flowers. Both species are visited by pollinators that differ in their behavior and effectiveness, and 
possibly the selection they impose. We have identified floral traits that are likely adaptive, but 
additional studies that separate the effects of different pollinators, remove external selective 
agents such as resources as Caruso et al. (2005) did in 
A. syriaca
, while estimating male fitness 
using seeds sired, can help us further understand how the traits are adaptive in these species.
 Acknowledgements
 We would like to thank D. Kilmartin, M. Lucke, L. Porcher, J. S. Williams, L. Maihofer, J. 
64George, K. Harrison, T. Davis, and A. Hall for helping collect many years of field and lab data, 
and N. Dungee for persevering to count and weigh every last 
A. incarnata 
seed. We also thank I. 
Dworkin, J. Lau, and D. Schemske for comments and discussion, M. Fishbein, A. Liston, and A. 
Agrawal for sharing resources for microsatellite markers and tips on growing 
Asclepias
, and the 
Fort Custer Training Center and the Kellogg Biological Station for access to field sites. This 
research was funded in part by the National Science Foundation Graduate Research Fellowship 
Program, a Michigan State University Plant Sciences Fellowship, a grant from Prairie Biotic 
Research, Inc., a T. Wayne and Katherine Porter Fellowship, a G.H. Lauff Research Award, the 
W. K. Kellogg Foundation, a microMORPH travel grant, and National Science Foundation 
Grants DBI-0638591 and 
DEB-0919452
.65APPENDICES
66APPENDIX A
Tables and Figures
67Table 6.
 Trait repeatability. 
Among plant variation from plants with two or three photographed flowers. Plants sampled ranged
 from 
50-236.  A. incarnata
A. exaltata
Year
20072008200920102011
200920102011
Gynostegium width
-78.27069.569.782.48783.9Hood height
80.981.98167.582.185.882.990.1Hood length
4758.362.753.467.973.780.873.6Horn reach
30.863.859.463.166.874.970.778.5Slit length
43.670.763.16960.158.779.276.3Gap width
-6064.435.23965.170.173.4Plants sampled
505023612713295144127Flowers photographed
232-322-42-322!68Table 7
. Sampling effort in each year for 
A. incarnata 
and 
A. exaltata.
  A. incarnata
A. exaltata
Year
20072008200920102011
200920102011
Mean flowers sampled per plant
12.1100.753.392.1117.3
22.636.330.3Std. dev.
10.553.63043.568.315.92619.4Median
9925385.588.5193625.5Plants sampled
505030313213491130122!69Table 8.
 Percent pollinator composition in the 
A. incarnata 
population in 2008 and 2010
. Pollinia carried on pollinators gives the 
number of insects collected, the mean pollinia attached to each individual, and the standard error of the mean (SEM). The percent 
pollinia are the proportion of total pollinia carried by that pollinator type. Percent importance is calculated by multiplying the raw 
number of pollinators observed by the mean number of pollinia carried and dividing by the total of all pollinators.
 !70% Composition
Pollinia carried on pollinators
% Importance
20082010nMean 
pollinia
SEM% pollinia
2008 poll. 
importance
2010 poll. 
importance
Butterflies
311015.212.15.61.90.7Flies2326120.30.20.10.30.6Bees (sm/med)
517216.00.05.93.719.0Honey bees
4131534.410.512.661.97.1Bees (large)
511143.816.716.010.04.0Wasps (small)
14382811.0
3.74.07.028.4Wasps (medium)
9143139.97.614.615.237.7Wasps (large)
0010112.8
12.641.30.02.6n173600Table 9. Pairwise correlation coefficients for female fitness measures in 2010 for 
A. incarnata 
(below the diagonal) and 
A. exaltata 
(above the diagonal and in italics).
 Seeds 
produced, seeds sired, and the sum of the two (p+s) come from viable fruits only. These two 
measures exclude fruits that could not be included in the paternity analysis because none of the 
seeds that were planted germinated. P < 0.0
001 for all. 
!71Total 
fruits
Viable 
fruits
Total 
seeds
Seeds 
produced
Seeds 
sired
Seeds     
(p+s)
Total fruits
-0.860.970.870.470.73Viable fruits
0.94-0.820.960.530.81Total seeds
0.990.95-0.890.470.74Seeds produced
0.950.980.97-0.570.85Seeds sired
0.370.360.380.38-0.91Seeds (p+s)
0.930.950.950.970.58-Table 10
. Primer names, annealing temperature, exclusion probabilities and allele frequencies of microsatellite markers used 
on 
A. exaltata 
and 
A. incarnata
, with genet sample sizes of n
. A106 is showing expected allele frequencies with a null allele; the 
expected null allele frequency is in parentheses. Primers were labeled with 6-FAM, HEX, or a HEX-labeled M13 tail to distinguish 
between loci with overlapping allele sizes. 
!72AllelesLocus
TAP{excl.}
1234567A. exaltata
 (n=129)0.93Asyr-B121
HEX580.560.280.260.200.190.07< 0.01Asyr-B2
HEX580.400.390.340.230.020.02Asyr-C102
HEX480.340.470.380.11
0.04Asyr-C103
HEX480.430.350.320.280.020.020.01As71725
M13-HEX
480.250.700.170.13A. incarnata
 (n=142)0.91Asyr-A106
FAM
560.340.460.330.09(0.08)0.03< 0.01< 0.01ASF9HEX560.380.370.280.26< 0.01< 0.01Asyr-B102
FAM
560.280.620.220.14Asyr-B121
HEX560.300.660.230.18Asyr-C124
M13-HEX
560.350.490.340.11
0.03< 0.01As71725
M13-HEX
480.360.460.290.27< 0.01Table 11. Pairwise correlation coefficients between estimated total flower number and 
ramet number for 
A. incarnata 
and 
A. exaltata.
   !73A. incarnata
A. exaltata
Year
20072008200920102011
200920102011
r0.64-0.64350.750.360.650.35Table 12. Pairwise correlation coefficients for 
A. exaltata 
(a) and 
A. incarnata 
(b) combined 
across years
. Bold is P < 0.05.
 !74(a) 
A. exaltata 
Gyn W
Hood H
Hood L
Horn R
Slit L
Gap W
Ramets
Hood H
-0.33Hood L
-0.290.32Horn R
-0.040.300.47Slit L
0.130.300.090.14Gap W
0.08-0.030.00-0.16-0.00Ramets
-0.040.020.04-0.04-0.02-0.01Flower number
-0.030.01-0.20-0.13-0.11
-0.170.48(b) A. incarnata
Gyn W
Hood H
Hood L
Horn R
Slit L
Gap W
Ramets
Hood H
0.45Hood L
0.090.42Horn R
0.270.520.15Slit L
0.510.340.120.14Gap W
-0.010.10.10.030.09Ramets
0.080.070.020.06-0.070.08Est. flower number
0.350.09-0.010.060.270.11
0.41Table 13.
 A. exaltata 
descriptive statistics for the six traits (cm), total flower number, and fitness estimates.
 Mean, standard 
deviation, and phenotypic coefficients of variation (CV
P) were calculated for each year. Pollinia removals, pollinia insertions, and 
fruits were measured on every flower, and fruits were counted on all ramets. Pearson product-moment correlations are given between 
total pollinia removals and total insertions in 2007-09, and between seeds produced and sired in 2010. Gynostegium width was 
measured differently in 2011. 
 !75200920102011
Aexa traits
mean(s.d.)
CVPmean(s.d.)
CVPmean(s.d.)
CVPGynostegium width
0.25(0.02)8.10.24(0.01)5.70.13(0.01)6.1Hood height
0.37(0.04)9.90.38(0.03)8.60.41(0.03)8.1Hood length
0.22(0.03)12.10.17(0.02)11.7
0.21(0.02)10.2Horn reach
0.25(0.05)20.10.20(0.05)26.10.23(0.04)18.5Slit length
0.25(0.01)5.50.24(0.01)4.90.24(0.01)4.6Gap width
0.06(0.01)22.10.06(0.01)18.60.06(0.01)20.1Estimated flower number
25.7(18.4)70.536.3(26.0)71.430.3(19.4)64.0Sample size
n = 91n = 130n = 153Insertions per flower
0.02(0.03)182.7Removals per flower
0.13(0.15)112.6
0.29(0.15)52.00.27(0.19)68.8Pollinia (male-female corr.)
r = 0.60*Fruit quantity
0.73(0.86)117.1
1.02(1.12)110.4
1.01(0.96)95.0Seeds produced
51.1(57.3)112.2
Seeds sired
51.1(72.8)142.6Total seeds
102.1(115.40)
112.9
Seeds (male-female corr.)
r = 0.57* Table 14. 
A. incarnata 
descriptive statistics for the six traits (cm), flower number (estimated as lateral branches in 2007, 2011; 
ramets in 2008; estimated flowers in 2009, and umbels in 2010), and fitness estimates.
 Mean, standard deviation, and phenotypic 
coefficients of variation (CV
P) were calculated for each year. Except in 2008, pollinia removals, pollinia insertions were sampled 
across genets, and fruits were counted on all ramets. Pearson product-moment correlations are given between total pollinia removals 
and total insertions in 2007-09, and between seeds produced and sired in 2010. Gynostegium width was measured differently in 2011. 
 (b) A. incarnata
20072008200920102011
mean(s.d.)
CVPmean(s.d.)
CVPmean(s.d.)
CVPmean(s.d.)
CVPmean(s.d.)
CVPGynostegium width
0.14(0.01)4.40.14(0.01)5.10.13(0.01)4.40.07(0.00)4.9Hood height
0.22(0.02)7.70.23(0.02)8.50.22(0.02)9.00.21(0.02)7.80.20(0.02)8.2Hood length
0.14(0.02)12.80.14(0.02)10.70.14(0.02)12.30.13(0.02)11.0
0.14(0.02)11.2
Horn reach
0.10(0.02)16.90.09(0.02)22.20.09(0.02)21.50.08(0.02)23.00.07(0.02)20.4Slit length
0.16(0.01)4.20.16(0.01)3.60.17(0.01)4.60.16(0.00)3.70.16(0.01)3.6Gap width
0.04(0.01)15.00.05(0.01)22.70.05(0.01)15.80.05(0.01)11.8
Estimated flower number
9.9(7.7)77.73.1(2.3)72.6818.2(845.4)103.31663.6(2000.2)120.29.4(7.6)80.6Sample size
n = 50n = 50n = 305n = 129n = 134Insertions per flower
0.38(0.26)68.70.17(0.10)56.50.25(0.17)68.0Removals per flower
3.41(0.73)21.31.94(0.64)32.92.62(0.73)27.91.22(0.48)39.61.87(0.71)37.8Pollinia (male-female corr.)
r = 0.16r = 0.64*r = 0.61*Fruit quantity
28.1(24.3)86.321(24.5)117.0
10.1(11.8)
117.3
15.2(18.6)122.611.1
(11.7)
105.3Seeds produced
556(902)162.2Seeds sired
511
(258)50.5Total seeds
102(115)
112.9
Seeds (male-female corr.)
r = 0.38*!76slhhghlhrgr2 mm
ghlgrhrslhh(b)(a)
2 mm
gwgwFigure 3. Top and side views of 
A. exaltata 
(A) and 
A. incarnata 
(B) flowers showing 
the six floral traits and landmarks: gynostegium width (
g; gr), hood length (
hl), hood height (
hh), horn reach (
hr), slit length (
sl), gap width (
gw). The arrow points to 
the dark corpusculum, which is attached to a pair of pollinia within the wall of the 
gynostegium. 
77Selection gradient (!)*****Selection gradient (!)  n(91)
(130)
(122)
78Gynost. 
widthHood heightHood lengthHorn 
reach
Slit 
lengthGap widthFlower 
number
(b) removals per ßower (male function)
(a) insertions per ßower (female function)
Figure 4. The effect of traits and flower number on female (a) and 
male (b) floral function across three years in 
A. exaltata
. Bars 
show standardized selection gradients for the six floral traits and 
flower number estimates on pollinia inserted per flower (a) and 
pollinia removed per flower (b). Error bars represent two standard 
errors. 
€P<0.10, *P<0.05, **P<0.01, ***P<0.001
A. exaltata
****€*********Selection gradient (!)***€€*Selection gradient (!)Figure 5. The effect of traits and plant size on female (a) and 
male (b) floral function across three years in 
A. incarnata
. Bars 
show standardized selection gradients for the six floral traits and 
flower number estimates on pollinia inserted per flower (a) and 
pollinia removed per flower (b). Error bars represent two standard 
errors. 
€P<0.10, *P<0.05, **P<0.01, ***P<0.001
  n(50)
(50)
(292)
(126)
(131)
Gynost. 
widthHood heightHood lengthHorn 
reach
Slit 
lengthGap widthFlower 
number
(b) removals per ßower (male function)
(a) insertions per ßower (female function)
A. incarnata
79Selection gradient (!)€€€*************total fruits (female Þtness)Gynost. 
widthHood heightHood lengthHorn 
reach
Slit 
lengthGap widthFlower 
number
  n(90)
(130)
(121)
80Figure 6. Standardized selection gradient estimates for floral 
traits and flower number on female fitness (fruit number) across 
three years in 
A. exaltata
. Bars show standardized selection 
gradients for the six floral traits and flower number estimates. Error 
bars represent two standard errors. 
€P<0.10, *P<0.05, **P<0.01, 
***P<0.001
A. exaltata
Selection gradient (!)****€***************€€total fruits (female Þtness)  n(50)
(49)
(180)
(126)
(131)
Gynost. 
widthHood heightHood lengthHorn 
reach
Slit 
lengthGap widthFlower 
number
81Figure 7. Standardized selection gradient estimates for floral 
traits and flower number on female fitness (fruit number) across 
three years in 
A. incarnata
. Bars show standardized selection 
gradients for the six floral traits and flower number estimates. Error 
bars represent two standard errors. 
€P<0.10, *P<0.05, **P<0.01, 
***P<0.001
A. incarnata
Selection gradient (!)***********€€(a) A. exaltata (n = 130)€*********Selection gradient (!)(b) A. incarnata 
(n = 126)Figure 8.
 Standardized selection gradient estimates for floral 
traits and flower number through seeds produced, seeds sired, 
and total seed number for 
A. exaltata
 (a) and 
A. incarnata 
(b).
 Bars show standardized selection gradients for the six floral traits 
and flower number estimates, and error bars represent two 
standard errors. 
€P<0.10, *P<0.05, **P<0.01, ***P<0.001
Gynost. 
widthHood heightHood lengthHorn 
reach
Slit 
lengthGap widthFlower 
number
8205101520253000.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1Frequency
Functional gender
05101520253000.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1Frequency
Functional gender
(a) A. exaltata(b) A. incarnata
Figure 9.
 Distribution of functional gender (f/(m+f)) for individuals in a population 
of A. exaltata 
(a) and 
A. incarnata 
(b).
 Individuals with zero fitness are excluded.
 83Functional genderFunctional genderFrequency
Frequency
APPENDIX B
Selection gradient analyses: effect sizes, standard errors, and model fit
84Table 15. Selection through insertions per flower in 
A. exaltata. 
Standardized selection 
gradient estimates for floral traits and estimated flower number in 2009 in 
A. exaltata.
 2009 (total ins)
2009Gynostegium width
0.16(0.22)0.02(0.25)Hood height
0.08(0.22)0.14(0.25)Hood length
-0.05(0.20)0.03(0.23)Horn reach
-0.08(0.22)-0.04(0.25)Slit length
-0.24(0.20)-0.17(0.22)Gap width
-0.01(0.18)-0.08(0.20)Est. flower number
0.95(0.18)***0.66(0.21)*Total R
20.300.14Adj. R
20.240.07P<0.0001***0.07n9191€P<0.10, *P<0.05, **P<0.01, ***P<0.001
!85Table 16. Selection through removals per flower in 
A. exaltata. 
Standardized selection 
gradient estimates for floral traits and estimated flower number across three years in 
A. exaltata.
 200920102011
Gynostegium width
-0.07(0.15)-0.03(0.05)-0.08(0.08)Hood height
-0.05(0.15)0.04(0.05)-0.04(0.08)Hood length
0.12(0.14)0.02(0.05)-0.01(0.07)Horn reach
-0.04(0.15)-0.01(0.05)0.12(0.07)Slit length
-0.02(0.14)-0.14(0.05)**-0.04(0.07)Gap width
-0.17(0.13)-0.00(0.05)0.06(0.07)Est. flower number
0.30(0.13)*0.05(0.05)0.09(0.07)Total R
20.130.090.06Adj. R
20.050.04-0.001P0.11
0.09€0.45n91130122€P<0.10, *P<0.05, **P<0.01, ***P<0.001
!86Table 17. Selection through fruit number in 
A. exaltata.
 Standardized selection gradient 
estimates for floral traits and estimated flower number across three years in 
A. exaltata.
 200920102011
Gynostegium width
0.24(0.13)€-0.02(0.09)0.01(0.09)Hood height
-0.20(0.13)-0.23(0.09)*-0.08(0.10)Hood length
0.27(0.11)
*0.13(0.09)0.10(0.10)Horn reach
-0.19(0.13)0.03(0.09)0.02(0.10)Slit length
0.19(0.11)
€0.06(0.08)0.05(0.10)Gap width
-0.19(0.10)€0.24(0.09)**0.13(0.09)Est. flower number
0.74(0.11)
***0.66(0.08)***0.45(0.09)***Total R
20.460.370.20Adj. R
20.410.340.15P<0.0001***<0.0001***0.0005***n90130121€P<0.10, *P<0.05, **P<0.01, ***P<0.001
!87Table 18. Selection through viable seeds produced and sired, and total viable seeds in 
A. exaltata.
 Standardized selection gradient estimates for floral traits and estimated flower number 
in 2010 in 
A. exaltata.
 2010Seeds set 
!Seeds sired 
"Total fitness
Gynostegium width
-0.08(0.09)-0.25(0.13)€-0.16(0.09)€Hood height
-0.11
(0.09)-0.02(0.14)-0.07(0.10)Hood length
0.13(0.09)0.15(0.13)0.14(0.09)Horn reach
0.01(0.09)0.05(0.13)0.03(0.09)Slit length
0.06(0.08)0.18(0.12)0.12(0.09)Gap width
0.22(0.09)*0.16(0.12)0.19(0.09)*Est. flower number
0.70(0.08)***0.62(0.12)***0.66(0.09)***Total R
20.380.210.34Adj. R
20.340.160.30P<0.0001***0.0002***<0.0001***n130130130€P<0.10, *P<0.05, **P<0.01, ***P<0.001
#88Table 19. Selection through insertions per flower in 
A. incarnata.
 Standardized selection 
gradient estimates for floral traits and estimated flower number across three years in 
A. 
incarnata.
 200720082009Gynostegium width
0.11
(0.10)0.21(0.05)***Hood height
0.21(0.12)0.19(0.11)
€-0.08(0.05)Hood length
-0.08(0.11)
-0.10(0.10)0.01(0.04)Horn reach
-0.03(0.10)-0.15(0.10)0.02(0.04)Slit length
0.18(0.10)€0.07(0.09)0.08(0.04)€Gap width
-0.03(0.08)0.02(0.04)Est. flower number
-0.02(0.11)
0.09(0.08)-0.06(0.04)Total R
20.150.150.14Adj. R
20.050.010.12P0.210.41<0.0001n5050292€P<0.10, *P<0.05, **P<0.01, ***P<0.001
!89Table 20. Selection through removals per flower in 
A. incarnata. 
Standardized selection 
gradient estimates for floral traits and estimated flower number across five years in 
A. incarnata.
 20072008200920102011
Gynostegium width
0.07(0.06)0.07(0.02)***0.02(0.04)0.07(0.03)*Hood height
0.004(0.04)0.06(0.07)-0.01(0.02)0.05(0.05)-0.03(0.04)Hood length
-0.04(0.04)-0.02(0.06)0.05(0.02)***0.09(0.04)*0.15(0.03)***Horn reach
-0.02(0.03)-0.02(0.06)0.01(0.02)-0.01(0.04)-0.03(0.03)Slit length
0.05(0.03)-0.03(0.05)0.03(0.02)€-0.01(0.04)0.07(0.03)*Gap width
-0.004(0.05)0.03(0.02)*0.00(0.04)0.03(0.03)Est. flower number
0.02(0.04)0.04(0.05)-0.01(0.02)0.05(0.04)-0.04(0.03)Total R
20.090.080.170.11
0.31Adj. R
2-0.01-0.070.150.080.27P0.490.81<0.00010.05<0.0001n5050292126131€P<0.10, *P<0.05, **P<0.01, ***P<0.001
!90Table 21. Selection through fruit number in 
A. incarnata.
 Standardized selection gradient 
estimates for floral traits and estimated flower number across five years in 
A. incarnata.
 20072008200920102011
Gynostegium width
-0.15(0.20)0.16(0.08)*-0.05(0.07)0.04(0.07)Hood height
0.06(0.09)-0.11
(0.23)-0.23(0.08)**0.18(0.08)*0.13(0.09)Hood length
0.11
(0.08)-0.25(0.21)0.14(0.07)*0.09(0.07)-0.01(0.08)Horn reach
0.01(0.08)0.06(0.21)0.12(0.07)€-0.03(0.07)0.01(0.08)Slit length
0.15(0.08)€0.11
(0.17)0.14(0.07)€-0.06(0.07)0.06(0.07)Gap width
-0.18(0.16)-0.01(0.06)0.03(0.07)0.01(0.07)Est. flower number
0.79(0.08)***0.46(0.17)**0.85(0.06)***1.01(0.06)***0.74(0.07)***Total R
20.700.220.540.700.52Adj. R
20.660.090.520.680.50P<0.00010.14<0.0001<0.0001<0.0001n5049180126131€P<0.10, *P<0.05, **P<0.01, ***P<0.001
!91Table 22. Selection through viable seeds produced and sired, and total viable seeds in 
A. incarnata.
 Standardized selection gradient estimates for floral traits and estimated flower 
number in 2010 in 
A. incarnata.
 Seeds produced 
!Seeds sired 
"Total fitness
Gynostegium width
-0.03(0.09)-0.08(0.05)€-0.05(0.06)Hood height
0.18(0.11)
0.01(0.05)0.10(0.07)Hood length
0.14(0.10)-0.00(0.05)0.08(0.06)Horn reach
-0.08(0.10)-0.04(0.05)-0.06(0.06)Slit length
-0.07(0.09)0.00(0.04)-0.04(0.05)Gap width
0.02(0.09)0.05(0.04)0.04(0.05)Est. flower number
1.19(0.09)***0.25(0.04)***0.77(0.05)***Total R
20.640.240.66Adj. R
20.610.200.64P<0.0001<0.0001<0.0001n126126126€P<0.10, *P<0.05, **P<0.01, ***P<0.001
#92LITERATURE CITED
93LITERATURE CITED
Ashman, T.-L., and M. L. Stanton. 1991. Seasonal Variation in Pollination Dynamics of Sexually 
Dimorphic 
Sidalcea Oregana ssp. Spicata
 (Malvaceae). Ecology 72:993Ð1003.
Ashman, T.-L., and M. T. Morgan. 2004. Explaining phenotypic selection on plant attractive 
characters: male function, gender balance or ecological context? Proc. R. Soc. Lond. B 
271:553Ð559.
Ashman, T.-L., T. M. Knight, J. A. Steets, and P. Amarasekare. 2004. Pollen limitation of plant 
reproduction: ecological and evolutionary causes and consequences. Ecology 85:2408Ð
2421.Bateman, A. J. 1948. Intra-sexual selection in 
Drosophila
. Heredity 2:349Ð368.
Betz, R. F., R. D. Struven, J. E. Wall, and F. B. Heitler. 1994. Insect pollinators of 12 milkweed 
(Asclepias
) species. Thirteenth North American Prairie Conference 45Ð60.
Broyles, S. B., and R. Wyatt. 1990. Paternity analysis in a natural population of 
Asclepias 
exaltata
: multiple paternity, functional gender, and the ÒPollen-Donation Hypothesis.Ó 
Evolution 44:1454Ð1468.
Broyles, S. B., and R. Wyatt. 1993. The consequences of self-pollination in 
Asclepias exaltata
, a 
self-incompatible milkweed. American Journal of Botany 80:41Ð44.
Burd, M. 1994. Bateman's principle and plant reproduction: the role of pollen limitation in fruit 
and seed set. The Botanical Review 60:83Ð139.
Caruso, C. M., D. L. D. Remington, and K. E. Ostergren. 2005. Variation in resource limitation 
of plant reproduction influences natural selection on floral traits of 
Asclepias syriaca
. Oecologia 146:68Ð76.
Caruso, C. M., S. B. Peterson, and C. E. Ridley. 2003. Natural selection on floral traits of 
Lobelia
 (Lobeliaceae): spatial and temporal variation. American Journal of Botany 90:1333Ð
1340. Chaplin, S. J., and J. L. Walker. 1982. Energetic Constraints and Adaptive Significance of the 
Floral Display of a Forest Milkweed. Ecology 63:1857Ð1870.
Charlesworth, D. 1989. Evolution of low female fertility in plants: Pollen limitation, resource 
allocation and genetic load. Trends Ecol Evol 4:289Ð292.
94Conner, J. K. 2006. Ecological genetics of floral evolution. Pp. 260Ð277 
in L. D. Harder and S. 
C. H. Barrett, eds. Ecology and Evolution of Flowers. Oxford University Press, New 
York.
Conner, J. K., R. Davis, and S. Rush. 1995. The effect of wild radish floral morphology on 
pollination efficiency by four taxa of pollinators. Oecologia 104:234Ð245.
Conner, J. K., S. Rush, S. Kercher, and P. Jennetten. 1996. Measurements of natural selection on 
floral traits in wild radish (
Raphanus raphanistrum
). II. Selection through lifetime male 
and total fitness. Evolution 50:1137Ð1146.
Faegri, K., and L. Van Der Pijl. 2013. Principles of Pollination Ecology. 3rd ed. Elsevier.
Fenster, C. B., W. S. Armbruster, P. Wilson, M. R. Dudash, and J. D. Thomson. 2004. Pollination 
syndromes and floral specialization. Annu. Rev. Ecol. Evol. Syst. 35:375Ð403.
Fishbein, M., and D. L. Venable. 1996. Diversity and temporal change in the effective pollinators 
of Asclepias tuberosa
. Ecology 77:1061Ð1073.
Galen, C., and M. L. Stanton. 1989. Bumble Bee Pollination and Floral Morphology: Factors 
Influencing Pollen Dispersal in the Alpine Sky Pilot, 
Polemonium viscosum
 (Polemoniaceae). American Journal of Botany 76:419Ð426. 
Harder, L. D., and S. C. H. Barrett. 1993. Pollen Removal From Tristylous 
Pontederia Cordata
: Effects of Anther Position and Pollinator Specialization. Ecology 74:1059Ð1072.
Harder, L. D., and S. C. H. Barrett. 1995. Mating cost of large floral displays in hermaphrodite 
plants. Nature 373:512Ð515. 
Herrera, C. M. 1988. Variation in mutualisms: the spatiotemporal mosaic of a pollinator 
assemblage. Biological Journal of the Linnean Society 35:95Ð125.
Himes, S. L., and R. Wyatt. 2005. Costs and benefits of self-fertility in 
Asclepias exaltata
 (Apocynaceae). The Journal of the Torrey Botanical Society 132:24Ð32. 
Hitachi Software Engineering, C., Ltd. n.d. FMBIO Analysis. Hitachi Software Engineering, Co., 
Ltd.
Hodgins, K. A., and S. C. H. Barrett. 2008. Natural selection on floral traits through male and 
female function in wild populations of the heterostylous daffodil 
Narcissus triandrus
. Evolution 62:1751Ð1763.
Howard, A. F., and E. M. Barrows. 2014. Self-pollination rate and floral-display size in 
Asclepias 
syriaca
 (Common Milkweed) with regard to floral-visitor taxa. BMC Evol. Biol. 14:144.
95Ivey, C. T., P. Martinez, and R. Wyatt. 2003. Variation in pollinator effectiveness in swamp 
milkweed, 
Asclepias incarnata
 (Apocynaceae). American Journal of Botany 90:214Ð225.
Ivey, C. T., S. R. Lipow, and R. Wyatt. 1999. Mating systems and interfertility of swamp 
milkweed (
Asclepias incarnata 
ssp.
 incarnata
 and ssp. 
pulchra
). Heredity 82:25Ð35. 
Johnston, M. O. 1991. Natural Selection on Floral Traits in Two Species of 
Lobelia
 with 
Different Pollinators. Evolution 45:1468Ð1479. 
Jones, A. G. 2005. GERUD 2.0: a computer program for the reconstruction of parental genotypes 
from half
-sib progeny arrays with known or unknown parents. Molecular Ecology Notes 
5:708Ð711.
Jones, A. G. 2001. GERUD 1. 0: a computer program for the reconstruction of parental 
genotypes from progeny arrays using multilocus DNA data. Molecular Ecology Notes 
1:215Ð218.
Kabat, S. M., C. W. Dick, and M. D. Hunter. 2010. Isolation and characterization of 
microsatellite loci in the common milkweed, 
Asclepias syriaca
 (Apocynaceae). American 
Journal of Botany 97:e37Ðe38.
Kearse, M., R. Moir, A. Wilson, S. Stones-Havas, M. Cheung, S. Sturrock, S. Buxton, A. Cooper, 
S. Markowitz, C. Duran, T. Thierer, B. Ashton, P. Meintjes, and A. Drummond. 2012. 
Geneious Basic: An integrated and extendable desktop software platform for the 
organization and analysis of sequence data. Bioinformatics 28:1647Ð1649.
Kephart, S. R. 1981. Breeding Systems in 
Asclepias incarnata
 L., 
A. syriaca
 L., and 
A. 
verticillata
 L. American Journal of Botany 68:226Ð232.
Kephart, S. R. 1987. Phenological variation in flowering and fruiting of 
Asclepias
. American 
Midland Naturalist 118:64Ð76.
Knight, T. M., J. A. Steets, J. Vamosi, S. J. Mazer, M. Burd, D. R. Campbell, M. R. Dudash, M. 
O. Johnston, R. Mitchell, and T.-L. Ashman. 2005. Pollen Limitation of Plant 
Reproduction: Pattern and Process. Annu. Rev. Ecol. Evol. Syst. 36:467Ð497.
Kulbaba, M. W., and A. C. Worley. 2012. Selection on floral design in 
Polemonium brandegeei
 (Polemoniaceae): female and male fitness under hawkmoth pollination. Evolution 
66:1344Ð1359. 
Lande, R., and S. Arnold. 1983. The measurement of selection on correlated characters. 
Evolution 37:1210Ð1226.
96Lipow, S. R., S. B. Broyles, and R. Wyatt. 1999. Population differences in self-fertility in the 
Òself-incompatibleÓ milkweed 
Asclepias exaltata
 (Asclepiadaceae). American Journal of 
Botany 86:1114Ð1120.
Lloyd, D. G. 1980. Sexual strategies in plants III. A quantitative method for describing the 
gender of plants. New Zealand Journal of Botany 18:103Ð108.
Macior, L. W. 1965. Insect Adaptation and Behavior in 
Asclepias
 pollination. Bulletin of the 
Torrey Botanical Club 92:114Ð126. 
Manaster, C. 2002. Allelogram: a program for normalizing and binning microsatellite genotypes. 
https://code.google.com/p/allelogram/.
Marshall, T. C., J. Slate, L. E. B. Kruuk, and J. M. Pemberton. 1998. Statistical confidence for 
likelihood-based paternity inference in natural populations. Molecular Ecology 7:639Ð
655.Meagher, T. R. 1994. The quantitative genetics of sexual dimorphism in 
Silene latifolia
 (Caryophyllaceae). II. Response to sex-specific selection. Evolution 48:939Ð951.
Morgan, M. T., and D. Schoen. 1997. Selection on reproductive characters: floral morphology in 
Asclepias syriaca
. Heredity 79:433Ð441.
Morgan, M. T., and J. K. Conner. 2001. Using genetic markers to directly estimate male selection 
gradients. Evolution 55:272Ð281.
Murcia, C. 1990. Effect of Floral Morphology and temperature on pollen receipt and removal in 
Ipomoea trichocarpa
. Ecology 71:1098Ð1109.
O'Quinn, R. L., and M. Fishbein. 2009. Isolation, characterization and cross-species 
amplification of polymorphic microsatellite loci in 
Asclepias
 (Apocynaceae). 
Conservation Genetics 10:1437Ð1440.
Ollerton, J. 1996. Reconciling Ecological Processes with Phylogenetic Patterns: The Apparent 
Paradox of PlantÑPollinator Systems. Journal of Ecology 84:767Ð769. 
Price, M. V., N. M. Waser, R. E. Irwin, D. R. Campbell, and A. K. Brody. 2005. Temporal and 
spatial variation in pollination of a montane herb: a seven-year study. Ecology 86:2106Ð
2116.
Queller, D. C. 1985. Proximate and ultimate causes of low fruit production in 
Asclepias exaltata
. Oikos 44:373Ð381.
Queller, D. C. 1983. Sexual selection in a hermaphroditic plant. Nature 305:706Ð707.
97Rafferty, N. E., and A. R. Ives. 2012. Pollinator effectiveness varies with experimental shifts in 
flowering time. Ecology 93:803Ð814.
Rasband, W. S. 1997. ImageJ. U. S. National Institutes of Health, Bethesda, MD, USA.
Rocca, M. A., and M. Sazima. 2012. Quantity versus quality: identifying the most effective 
pollinators of the hummingbird-pollinated 
Vriesea rodigasiana
 (Bromeliaceae). Plant 
Systematics and Evolution 299:97Ð105.
Sahli, H. F., and J. K. Conner. 2011. Testing for conflicting and nonadditive selection: floral 
adaptation to multiple pollinators through male and female fitness. Evolution 65:1457Ð
1473.Sahli, H. F., and J. K. Conner. 2007. Visitation, effectiveness, and efficiency of 15 genera of 
visitors to wild radish, 
Raphanus raphanistrum
 (Brassicaceae). American Journal of 
Botany 94:203Ð209.
SAS Institute Inc. 2012. JMP. Cary, NC, USA.
Scheiner, S., K. Donohue, L. Dorn, S. J. Mazer, and L. Wolfe. 2002. Reducing Environmental 
Bias When Measuring Natural Selection. Evolution 56:2156Ð2167.
Schemske, D. W., and C. Horvitz. 1989. Temporal variation in selection on a floral character. 
Evolution 43:461Ð465.
Stinchcombe, J. R., M. T. Rutter, D. S. Burdick, P. Tiffin, M. D. Rausher, and R. Mauricio. 2002. 
Testing for Environmentally Induced Bias in Phenotypic Estimates of Natural Selection: 
Theory and Practice. The American Naturalist 160:511Ð523.
Stoepler, T. M., A. Edge, A. Steel, R. L. O'Quinn, and M. Fishbein. 2012. Differential pollinator 
effectiveness and importance in a milkweed (
Asclepias
, Apocynaceae) hybrid zone. 
American Journal of Botany 99:448Ð458.
Straub, S. C. K., M. Fishbein, T. Livshultz, Z. Foster, M. Parks, K. Weitemier, R. C. Cronn, and 
A. Liston. 2011. Building a model: developing genomic resources for common milkweed 
(Asclepias syriaca
) with low coverage genome sequencing. BMC Genomics 12:211.
van Kleunen, M., and J. Burczyk. 2007. Selection on floral traits through male fertility in a 
natural plant population. Evolutionary Ecology 22:39Ð54.
Wang, J. 2004. Sibship reconstruction from genetic data with typing errors. Genetics 166:1963Ð
1979.Waser, N. M., L. Chittka, M. V. Price, N. M. Williams, and J. Ollerton. 1996. Generalization in 
Pollination Systems, and Why it Matters. Ecology 77:1043Ð1060.
98Wiggam, S., and C. J. Ferguson. 2005. Pollinator Importance and Temporal Variation in a 
Population of 
Phlox divaricata
 L. (Polemoniaceae). The American Midland Naturalist 
154:42Ð54.
Wilbur, H. M. 1976. Life history evolution in seven milkweeds of the genus 
Asclepias
. The 
Journal of Ecology 64:223Ð240.
Willson, M. F., and B. J. Rathcke. 1974. Adaptive design of the floral display in 
Asclepias 
syriaca
 L. American Midland Naturalist 92:47Ð57.
Willson, M. F., and P. W. Price. 1977. The evolution of inflorescence size in 
Asclepias
 (Asclepiadaceae). Evolution 31:495Ð511.
Wilson, P., J. D. Thomson, M. L. Stanton, and L. P. Rigney. 1994. Beyond floral Batemania: 
gender biases in selection for pollination success. The American Naturalist 143:283Ð296.
Wyatt, R. 1976. Pollination and fruit-set in 
Asclepias
: A reappraisal. American Journal of Botany 
63:845Ð851.
Wyatt, R. 1981. The reproductive biology of 
Asclepias tuberosa
. II. Factors determining fruit-set. 
New Phytologist 88:375Ð385.
Wyatt, R., and S. B. Broyles. 1994. Ecology and evolution of reproduction in milkweeds. Annual 
Review of Ecology and Systematics 25:423Ð441.
Wyatt, R., and T. R. Shannon. 1986. Nectar production and pollination of 
Asclepias exaltata
. Systematic Botany 11:326Ð334.
Yampolsky, C., and H. Yampolsky. 1922. Distribution of sex forms in the phanerogamic flora. 
Ger Borntraeger.
Young, H. J., and M. L. Stanton. 1990. Influences of Floral Variation on Pollen Removal and 
Seed Production in Wild Radish. Ecology 71:536Ð547.
Zamora, R. 2000. Functional equivalence in plant-animal interactions: ecological and 
evolutionary consequences. Oikos 88:442Ð447.
99CHAPTER 4
PATTERNS OF ADAPTIVE TRAIT EVOLUTION ACROSS 
THE NORTH AMERICAN 
ASCLEPIAS
Introduction
 Adaptation is a fundamental process in biology that allows species to cope with their 
biotic and abiotic environment. Studies of adaptation on extant species can identify which traits 
are currently adaptive (e.g. Andersson 1982; Conner 1988; Grant and Grant 2002), and 
phylogenetic comparative methods offer complementary approaches to build and test adaptive 
hypotheses by studying interspecific patterns using contemporary trait measurements (Doughty 
1996). Two patterns that suggest evolution in response to selection are trait convergence and 
correlations with selective agents (Larson and Losos 1996). Convergence of a trait across clades 
suggests it evolved in response to a shared selective agent, or similar suite of agents
Ña selective 
regime (Coddington 1994; discussed by Pagel 1994b). A phylogenetically corrected correlation 
between traits suggests pairs of traits may be functionally related and correlations between a trait 
and a possible selective agent suggests the trait is an adaptation to that particular environment 
(Fenster et al. 2004; Weiblen 2004; Knudsen and Tollsten 2008).
 The array of phylogenetic tests that are currently available complement each other to 
provide a multidimensional understanding of an adaptive trait
Ñwe can model the evolution of a 
trait and predict its ancestral form (ancestral state reconstruction), identify other traits that it 
evolved in conjunction with or selective agents it was responding to (correlated evolution), and 
predict the adaptive peaks it evolved towards (convergent evolution). For example, phylogenetic 
comparative methods using data from Caribbean 
Anolis 
lizards show that adaptive radiations on 
100four Caribbean islands resulted in phenotypic convergence of ecomorphs (Mahler et al. 
2013). 
 Angiosperm diversity exploded 90 million years ago during the Cretaceous period. Prior 
to that time, plant reproduction was more dependent on abiotic factors (e.g., wind pollination), 
but a reliance on animals for reproduction introduced a new diversity of selective agents on floral 
traits. We expect that pollinators played an important role in the evolution of many floral traits 
(e.g. GŠmez et al. 2015), yet it can be difficult to prove the link between an adaptive trait and its 
selective agent(s) (but see Bradshaw and Schemske 2003). In many cases we assume that the 
pollinators we see today are the pollinators that were important in the past, and for plants that are 
not specialized, those pollinators can be composed of an array of taxa.  
 The milkweed genus 
Asclepias
 has over 125 North American species (Fishbein et al. 
2011), which all share the same floral structures that are diagnostic for the family, but show great 
diversity in their size and shape. In addition to five petals, the unusual 
Asclepias
 floral traits 
consist of a gynostegium housing two ovaries, the stigmatic chamber and pollinia (pollen 
packets), and a ring of five hoods that hold nectar (Fig. 10). Aside from the petals, the 
gynostegium and hoods together make up the bulk of the flower, with considerable variation in 
hood length and gynostegium width (5-6 standard deviations, see methods). The hoods also have 
discrete features; most of the 125 species have hoods with conspicuous openings rather than 
covered openings, and at least three quarters of the species have horns, which develop from the 
inside surface of the hoods (Kunze 1990) (Fig. 10). Further, 
Asclepias 
rely completely on animal 
pollination for sexual reproduction, because the pollen and stigmatic surface are isolated from 
each other within a flower. 
Asclepias 
tend to be generalists that are commonly pollinated by 
101insects such as bumble bees, honey bees, butterflies, and wasps (Woodson 1941; Pleasants 1991; 
Betz et al. 1994; Wyatt and Broyles 1994; Fishbein and Venable 1996; Ivey et al. 2003; Rafferty 
and Ives 2012). 
 Studies of contemporary selection have found that the width of the gynostegium, hood 
dimensions, and horn length can influence the fitness of several 
Asclepias
 species (Morgan and 
Schoen 1997; Caruso et al. 2005; Chapters 1 and 2), suggesting these floral traits may be 
currently adaptive. Gynostegium width had an effect on pollination success in 
A. tuberosa
 and 
A. 
incarnata 
(Chapters 1 and 2), longer hoods increased pollinia removals in two species and 
increased fruit production in three species (Caruso et al. 2005; Chapters 1 and 2), and the degree 
to which horns overlapped the gynostegium affected fruit number in 
A. tuberosa
 (Chapter 1).  
 We performed phylogenetic comparative analyses on hood, horn, and gynostegium traits 
to look for patterns of trait evolution that could indicate selection and complement the studies of 
contemporary selection (Morgan and Schoen 1997; Caruso et al. 2005; Chapters 1 and 2). We 
made several hypotheses regarding these three traits and their patterns of evolution across the 
genus. First, we hypothesized that if the hoods become closed, the horns cannot function. The 
function of the horns is unknown, yet they may interact with pollinators to guide their bodies 
between the hoods towards the reproductive structures or affect how pollinators approach the 
hoods to drink nectar. If the horns become enclosed by the hoods, they likely cannot do either of 
these proposed functions. Second, we hypothesized that different pollinator taxa are attracted to 
different hood shapes, as pollinators have been shown to preferentially visit flowers based on 
shape (Hegland and Totland 2005; GŠmez et al. 2008). Third, different environments, biotic or 
abiotic, select for particular combinations of hood and gynostegium size. If the 
Asclepias 
hoods 
102and gynostegia are adaptive they likely evolved through natural selection to attract or to be 
effectively pollinated by particular pollinators. Some may have also evolved in response to harsh 
environments that can make it challenging to maintain nectar rewards for pollinatorsÑmany 
Asclepias 
species live in hot, dry, and sunny habitats (Woodson 1954). 
 Following these hypotheses, we made several predictions. We predicted that horn loss 
had evolved in conjunction with hood closure. We also predicted species with hoods that have a 
flared rim, giving the appearance of easy access to the nectar reward, were more attractive to 
hymenopteran pollinators and hoods that appear to have smaller and less conspicuous openings 
attracted more non-hymenopteran pollinators. We do not have specific predictions for the 
different phenotypes made from combinations of short and long hoods and wide and narrow 
gynostegia, but we do predict that particular environments select for similar phenotypes, and so 
we tested for convergent evolution of these two traits.
Methods
Traits
 To study intraspecific patterns of trait evolution of the hoods, the horns that develop 
within the hoods, and gynostegia, we scored and measured floral traits on 107 of approximately 
125 North American 
Asclepias
 species, two African species from the genus 
Gomphocarpus
, and 
two more distantly related outgroups, 
Pergularia daemia
 and 
Calotropis procera,
 all from the 
tribe Asclepiadeae within Apocynaceae (Fishbein et al. 2011). Specimens were primarily 
collected by M. Fishbein, and supplemented with specimens from a number of other collectors 
between 1973 and 2010; all had been preserved in formalin-acetic acid (FAA) or ethanol. Using 
a Canon digital SLR with a 60mm macro lens, we photographed 1-3 flowers per species (mean = 
1031.6) from the top, side, and again from the side with 2.5 hoods removed, exposing the horn in 
order to identify horned species. For species that were sampled from more than one locale, we 
photographed one flower per locale and averaged the measurements within species. 
 We chose landmarks that would be visible across the diversity of floral morphology in the 
family, and took measurements using ImageJ (Rasband 1997). Our discrete variables were open 
(conspicuous) or closed (inconspicuous) hood openings, presence or absence of horns, and flared 
or domed hood openings
Ñthese are hoods that the opening does not flare out, but instead is 
straight or arcs inward (Fig. 11). The continuous traits we measured were hood length and 
gynostegium width, both measured from the top view of the flower (Fig. 10). Hood length was 
the distance from the outer most tip of the hood to its point of contact with the gynostegium (
h, Fig. 10), and gynostegium width was estimated by taking the distance between adjacent 
corpuscula (
g, Fig. 10) which can be divided by 0.588 to calculate the diameter. Using ANOVA 
with species as a random factor on a subset of 57 species that had 2-3 flowers measured, we 
found that 95% of the variation in hood length and 98% of the variation in gynostegium width 
was explained by differences between species. Variation explained changed to 97% for hood 
length and 98% for gynostegium width when we used only the 10 species for which we had three 
flowers measured.
Pollinators
 We obtained pollinator data from ASCLEPOL (Ollerton and Liede 1997), a publicly 
available list of flower visitors to species in the former Asclepiadaceae family. ASCLEPOL is 
compiled from publications and personal communications. Visitors are coded to indicate if they 
were: (1) observed depositing pollinia into a stigmatic opening, (2) carrying pollinia, or (3) 
104contacting the flower; we considered all three categories to be possible pollinators. There was 
pollinator data for 27 North American 
Asclepias 
species plus the four outgroup species. The 
number of observed pollinators ranged from 1-157 (mean = 25) records per 
Asclepias
 species. 
We grouped insect genera and species by order (Hymenoptera, Lepidoptera, Diptera, Coleoptera, 
and Hemiptera) and calculated the proportion of total visits that were from hymenopterans. We 
also placed each species into the binary categories of less than 50% or greater than or equal to 
50% Hymenoptera to allow comparisons to binary floral traits. Hymenopterans were by far the 
most common visitors, visiting 26 of the 27 
Asclepias
 species and accounting for 57% of the 683 
pollinators observed across all 27 species. 
Pollinator observations on five 
Asclepias
 species 
suggest that hymenopteran pollinators walk across the flowers in an inflorescence, while 
lepidopteran and dipteran pollinators tend to fly between flowers (La Rosa pers. obs.). Fishbein 
and Venable (1996)
 and 
Ivey et al. (2003)
 found that hymenopterans consistently carry and 
transfer more pollinia than lepidopterans in 
two species of 
Asclepias
. Phylogeny
 We use
d a maximum likelihood phylogram of 143 taxa made from three non-coding 
regions of the plastid genome (Fishbein et al. 2011). Branch lengths were proportional to 
substitutions and the tree was made ultrametric using the R package ÔapeÕ (Paradis et al. 2004; 
Paradis 2012) in R (R Core Team 2014). We pruned species to produce a tree made of the 111 
species (Tree111) for which we had floral trait data, which included four outgroup species from 
three genera. We pruned this tree down to the 31 species that we also had pollinator data for 
(Tree31). From Tree111, we pruned the two outgroup species leaving two species within the 
genus 
Gomphocarpus 
to produce a tree with 109 species (Tree109). Next we pruned all 
105outgroups from Tree109 and Tree31, as their clades are underrepresented and they may influence 
adaptive models of trait evolution; this resulted in trees with 107 species (Tree107) and 27 
species (Tree27).
Analyses
 Reconstructing the ancestral history allowed us to make predictions of the trait value of 
the ancestor to the 
Asclepias 
genus, which establishes the direction of evolution, and what the 
new possibly adaptive phenotype is. We identified the ancestral form for each discrete trait 
(hoods open/closed and horns present/absent) using maximum likelihood (ML) (Pagel 1994a) 
with equal rates of change using Ôape,Õ which provides likelihoods of the trait character at all 
nodes across Tree111. For the continuous traits, we also used ML to estimate ancestral states 
(Felsenstein 1985) using the R package ÔphytoolsÕ (Revell 2012), which displays trait values as a 
heat map ranging from small trait values (red) to large trait values (blue) across Tree109. 
 We assessed the pattern of trait evolution to determine if there was strong phylogenetic 
signal (evolving through random Brownian motion (BM)) or if traits had likely evolved towards 
optima. We estimated phylogenetic signal of discrete traits using Purvis and FritzÕs D (Fritz and 
Purvis 2010) using the phylo
.d function in the R package ÔcaperÕ (Orme et al. 2013). D ranges 
from zero to one; values near zero indicate a phylogenetic signal. For continuous traits, we used 
the phylosig function in the R package ÔphytoolsÕ (Revell 2012) to calculate PagelÕs lambda (
!) as an estimate of phylogenetic signal (Pagel 1999). Lambda ranges from zero to one; values near 
one indicate a phylogenetic signal, and values near zero indicate a lack of signal. 
 We tested two predictions of correlated trait evolution. We predicted that horns only have 
a function when the hoods are open and they can contact the pollinators. Depending on the 
106ancestral state of hoods and horns, we predicted either a gain in horns would be correlated with a 
shift to open hoods, or a loss of horns with a shift to closed hoods. We used PagelÕs correlation 
method (Pagel 1994a) in Mesquite (Maddison and Maddison 2011), which tests for correlation 
between two discrete traits and takes branch lengths into account. Next we tested our prediction 
of a positive correlation between gynostegium width and hood length, two major size 
components of the flower. We tested for correlations between phylogenetic independent contrasts 
(PIC) (Felsenstein 1985) of hood length and gynostegium width using the R package 
ÔapeÕ (Paradis et al. 2004). 
 To test for pollinator-mediated selection on the traits, we first tested the association 
between the discrete values for the shape of the hood opening and two pollinator categories: less 
than 50% hymenopteran pollinators and greater than or equal to 50% hymenopteran pollinators; 
a distribution of the proportion of hymenopterans can be seen in Figure 12. We again used 
PagelÕs (1994a) correlation method in Mesquite (Maddison and Maddison 2011) and our 31-
species tree (Tree31). We 
also used the subset of 27 
Asclepias 
species to test for correlations 
between the phylogenetically corrected continuous traits hood length and gynostegium width and 
the proportion of recorded visitors that were hymenopterans using PIC (Felsenstein 1985). 
 Expanding out to our larger tree and using our full trait data from 107 species, we tested 
our prediction of selection towards two adaptive peaks using the two hypothesized pollinator 
categories and each of the continuous traits, hood length and gynostegium width. 
We conducted 
these analyses using the R package ÔouchÕ (Butler and King 2004, King and Butler 2009). OUCH 
allowed us to compare models including a null Brownian-motion model (representing no 
adaptive peaks), an Ornstein-Uhlenbeck (OU) model with just a single adaptive peak, an OU 
107model with two adaptive peaks representing our two pollinator categories with their ancestral 
nodes having been reconstructed using maximum likelihood, and an OU model with three 
adaptive peaks representing an unknown ancestral group and the two pollinator groups of the 
extant species. 
 We then used a reverse method; rather than testing hypotheses of a predicted number of 
selective agents each selecting for an optimum phenotype, we used trait values of the 107 species 
to predict the number of convergent adaptive optima, or peaks
Ñthe interpretation being that 
each adaptive optimum was the result of a similar selective regime. We used the R package 
ÔsurfaceÕ (Ingram and Mahler 2013) to conduct this analysis. SURFACE is a two-phase analysis 
that uses stepwise Akaike information criterion (AIC) to find convergent phenotypes. In the 
forward phase it used gynostegium width and hood length together, accounting for relatedness, to 
find trait combinations of individuals or clades that were divergent from close relatives. Then in 
the reverse phase of the analysis it compared the locally divergent species and clades across the 
phylogeny to find ones that converged on the same phenotype. The outcome was the number of 
likely convergent adaptive optima of the trait valuesÑeach optimum corresponded to a 
potentially similar selective regime. An advantage to this method is that we can utilize data from 
the larger phylogeny (Tree107) to identify convergent phenotypes, allowing us to make 
predictions of selective regimes that can later be tested.    
Results
 We found that there was variation in horn presence, hood characters, hood length, and 
gynostegium width across the phylogeny. Of the 107 extant 
Asclepias
 species we sampled, 85% 
of them had horns and 70% of them had open hoods (Fig. 13). The base of the 
Asclepias 
clade 
108shows a very high likelihood that the ancestral state had open hoods containing horns (solid blue 
nodes; Fig. 13). Over half of the 107 species (56%) had flared hoods, but the hood shape of the 
ancestor of 
Asclepias 
was not clear (Fig 5A); of the subset of 27 species for which we had 
pollinator data, 63% had flared hoods. Hood length and the estimated gynostegium width were 
variable and ranged from 0.33mm to 7.35mm and from 0.74mm to 3.14mm, respectively.  
Correlated evolution of traits
 The closure of hoods and the loss of horns was highly correlated (P < 0.01), and hood 
closure likely proceeded horn loss as demonstrated by the shift in likelihood shown by the pie 
charts at shallower nodes in Figure 13A. Hood openings have closed at least four times, and were 
often followed by horn loss in at least one species; horns have been independently lost at least 
six times (Fig. 13A). This has left 16 extant 
Asclepias 
species without horns and 32 without 
conspicuous hood openings. Note that there are also some instances of species with open hoods 
with no horns, and vice versa. 
 We found that hood length had a weak phylogenetic signal (
! = 0.14; Table 15), 
indicating that it has evolved more than we would expect based on its phylogenetic history (Fig. 
15). Gynostegium width had a much higher phylogenetic signal (
! = 0.72; Table 15); its 
evolution more closely resembled a BM model of evolution (Fig. 16). Despite the difference in 
phylogenetic signal, gynostegium width and hood length were positively correlated after 
correcting for phylogenetic non-independence using PIC (r = 0.56, P < 0.001; Fig. 17).
   Correlated evolution of traits and environment
 Hood shape (flared vs. domed) appeared to be rather labile. Species with flared hoods are 
scattered across the phylogeny, so it is not surprising that hood opening shape shows very little 
109phylogenetic signal (D = 0.74, Table 15). This evolutionary lability may be caused by pollinator-
mediated selection, as
 species in which the majority of pollinators are hymenopterans 
were 
marginally more likely to have hood openings that were flared (Fig. 14B). As this association 
was only marginal, we also looked for associations between the pollinator categories and our 
continuous variables to bolster our finding.
 By comparing OU models across the 27 
Asclepias 
species, we were unable to find 
additional support for evolution in response to the two pollinator categories above for either of 
our continuous traits, hood length or gynostegium width (Table 16). Our two-optimum OU 
model, where hood length evolved towards the two pollinator categories, did not fit significantly 
better than models of one or three optima or random BM trait evolution (Table 16). The same 
was true when we compared OU models of gynostegium width evolution. We also found no 
significant correlation between the evolution of hood length or gynostegium width and the 
continuous proportion of hymenopterans after correcting for phylogenetic non-independence 
using PIC (Figs. 18 and 19, respectively)
 These analyses tested adaptive hypotheses, but could 
only be conducted on the reduced phylogenetic tree as we only had pollinator data for 27 species. 
We followed these analyses with tests that could use all 107 species to form additional testable 
adaptive hypotheses.
Convergent evolution of floral traits
 When we modeled hood length and gynostegium width together, we found that they 
predicted evolution towards multiple adaptive optima. Our results from the forward phase of the 
analysis in SURFACE found that there were 15 clades or species that had phenotypes that 
diverged from their closest relatives (Fig. 20A). The reverse phase of the analysis found that 
110those 15 clades/species clustered into four phenotypic groups based on gynostegium width and 
hood length together (Fig. 20). Three of these phenotypic groups each consisted of 4-5 
convergent clades or individual species (Fig. 20B). The hood length and gynostegium width of 
each species is plotted in Figure 20C, and the species mostly fell along a size axis. The clusters 
also mainly separated along a size axis for these two traits, with the gray (non-convergent) clade 
having the smallest phenotype, the 71 species that were neither convergent nor divergent (black) 
having larger hoods and gynostegia, and the ten species that make up the blue clades having even 
larger hoods and gynostegia. The red and green clades diverged from the size axis, with the red 
clades having exceptionally long hoods relative to their gynostegium width, and those in the 
green clades having hoods that are shorter relative to the gynostegium; four representative 
species from the convergent clusters are shown in Figure 20D. Keep in mind that the position of 
the species in Figure 20C is due solely to their trait values, while the colors of the dots 
incorporate phylogenetic relatedness.
  Unfortunately the pollinator data in ASCLEPOL only contains a few of the species from 
the convergent clusters, and collection locations and species distributions did not strongly 
suggest that species from each of the convergent clusters live in divergent habitats. The four 
species in the red regime live in a wide range of habitats; one species grows in Florida wetlands, 
two others in mountains in the southwest United States, and the fourth lives in a broad range of 
habitats that vary in moisture. Species in the green and blue regimes live in deserts in the 
southwestern US and Mexico. 
Discussion
   Studies of contemporary selection have found evidence for selection on hood traits and 
111gynostegium width in several North American 
Asclepias 
species (Caruso et al. 2005; Chapters 1 
and 2) suggesting these traits are currently adaptive. Interspecific patterns of trait evolution that 
are non-random are suggestive as signatures of selection, keeping in mind that non-random 
evolution of traits does not necessarily refute the presence of selection. In this present study we 
found evidence that the hoods in 
Asclepias
 are likely to be adaptive, because their evolution does 
not follow their phylogenetic relatedness, they are correlated with the evolution of other traits, 
and a subset of species have phenotypes that show associations, albeit weakly, with 
hymenopteran. Our analyses of convergent evolution suggest that the hoods and the gynostegia 
are evolving in response to a common selective regime and provide new adaptive hypotheses that 
require further investigation.  
Correlated evolution of traits
 Hoods have independently evolved a closed phenotype multiple times. Open hoods, the 
ancestral state, would have been easily accessible by pollinators and non-pollinating insects 
alike. The vulnerability of losing nectar to thieves, inefficient pollinators, or evaporation may 
have resulted in selection for the closure of hoods in some environments. Changes in pollinators 
can cause changes in multiple traits (e.g. Castellanos et al. 2004). The ancestor likely had horns, 
and with the closure of hoods, horns likely lost their function, removing or relaxing selection 
(Lahti et al. 2009)  to keep them, leading to their loss. Armbruster et al. (2002)
 found that the 
evolution of larger flowers in 
Collinsia
 reduced precision of selfingÑan example of a change in 
a floral trait that corresponded with a loss of function of another trait.
 Comparisons of flowers with open hoods and closed hoods will be needed to understand 
why hood closure evolved. Of the five species with closed hoods and no horns for which we also 
112have pollinator data, four are associated with prairie or grassland habitat, are light green, and are 
predominantly visited by bumble bees (
Bombus 
spp.). Evaporation may have been a strong 
selective force, but closing hoods to reduce visits by inefficient pollinators may also be a 
possible explanation for the phenotype. Determining if other species with closed hoods and no 
horns are also visited primarily by bumble bees, if they generally have fewer pollinator taxa than 
species with open hoods, and if they have lower nectar production could explain some of the 
consequences of hood closure and horn loss, and shed light on the selective regime that lead to 
this current phenotype.
Correlated and convergent evolution of traits in response to their environment
 We have some evidence that a pollinator fauna that is more than half hymenopteran 
corresponds with flared hoods, yet we found no evidence that the two pollinator regimes or the 
range of proportion of hymenopterans had an effect on hood length or gynostegium width. 
Whether or not hymenopterans represent the majority of pollinator visitors does not make for 
distinct selective regimes that select for two phenotypic optima of our other hood and 
gynostegium traits. This two-optimum scenario is not the best fit to explain the evolution of hood 
length and gynostegium separately, but in combination, these traits fit a 4-optimum scenario 
(three convergent optima and one non-convergent optimum) produced by our analysis in 
SURFACE. Pollinator data on most of the 
Asclepias 
species is incomplete, and additional 
observations would improve our ability to test these hypotheses of evolution towards the three 
adaptive optima. The three convergent clusters are excellent starting points to look for 
environmental similarities, as they offer three replicate scenarios, involving multiple independent 
113origins, where we would expect the pollinators and/or habitats to be more similar to each other 
than to the pollinators and habitats of the other 
Asclepias 
species.
 The convergence of species towards three distinct adaptive optima for hood length and 
gynostegium width is very suggestive that these traits are evolving in response to selective agents 
and that they are not evolving randomly. Our use of a large phylogeny that demonstrates 
independent evolution toward a particular phenotype multiple times strengthens our assertion 
that these floral traits are adaptive (Doughty 1996). 
 Studies have used comparative methods utilizing floral traits as a way to understand 
shifts in pollinators (e.g. Perez et al. 2006); however, our focus is on understanding the reasons 
for shifts in the floral traits (e.g. Friedman and Barrett 2008), which we predict is mostly in 
response to pollinators, and is more indicative of the direction that selection occurs. Our study 
has the advantage of having trait data for nearly all of the North American 
Asclepias 
species, a 
rather large clade of 125 species, yet we still lack sufficient pollinator data for most of the 
species. A good starting point for future data collection will be the 25 species in the three 
convergent clusters. Much like the significance of wings in birds, insects, and mammals, there is 
power in the repeated independent evolution of a phenotype to support adaptive hypotheses. In 
our phylogeny of over one hundred 
Asclepias 
species, the three repeated phenotypes, each with 
four or more independent origins, likely evolved in response to common selective regimes. The 
next step of acquiring data on the possible selective agents, including the pollinators that visit 
these species and their abiotic environment, will be important to confirm these adaptive 
hypotheses.  
114Acknowledgements
We would like to thank T. Barkman, N. Swenson and M. Weber for help and advice on 
phylogenetic analyses, and I. Dworkin, J. Lau, and D. Schemske for discussion and comments. 
This research was funding in part by a microMORPH travel grant.
115APPENDIX
116Table 23.
 Phylogenetic signal of continuous and discrete traits.
 Phylogenetic signal for 
continuous traits is measured by 
!; values near 0 have no phylogenetic signal are may have 
evolved in response to selection and values near 1 have a phylogenetic signal representative of 
Brownian-motion. For discrete traits the scale is reversed; when D is near 0 there is high BM 
phylogenetic signal and when D is 1 there is no phylogenetic signal. Sigma squared indicates the 
rate of evolution.
Continuous/Discrete
Trait
!logL
"2DH0: D = 0
H0: D = 1
Continuous
Gynostegium width
0.72-64.9604Continuous
Hood length
0.14-170.3490Discrete
Horn presence
0.71P = 0.025
P = 0.069
Discrete
Hood open
0.55P = 0.035
P = 0.012
Discrete
Hood opening shape
0.74P = 0.004
P = 0.067
Table 24.
 Model comparisons of trait evolution towards different adaptive optima using 
AIC.c (a size corrected AIC for small sample sizes)
. Brownian motion (BM) models a random 
walk with no adaptive optima; OU1 models a single optimum; OUanc models two adaptive 
optima (majority and minority Hymenopteran) with ancestral state reconstruction; OU3 models 
three optima (majority and minority Hymenopteran, and a third unknown ancestral optimum).
Model/Trait
BMOU1OUanc
OU3Hood length
109.8106.0107.7110.2
Gynostegium width
41.142.745.143.7117hgSCFigure 10.
 A photograph of an 
Asclepias exaltata
 flower from the side and top.
 Asclepias
 floral traits include five petals (P), a gynostegium (G), five hoods (Hd), and 
the hoods of some species each contain a horn (Hn). Pollinia are housed within the wall 
of the gynostegium and pairs are connected by a dark clip-like corpusculum (C) that is 
located at the top of the stigmatic slit (S) where the pollinia are deposited for 
fertilization. Our measured traits were hood length (
h) and gynostegium width estimated 
from the distance between adjacent corpuscula (
g). 118
Figure 11.
 Discrete binary floral traits: (a) horns present, (b) horns absent, (c) 
hoods conspicuously open, (d) hoods closed, (e) hood rim flared, (f) hood rim 
domed. 
Horn
Hood opening
Hood
abcdfe119
!"#"$"%"&"'!"'#"'$"!"!('"
!(#"
!()"
!($"
!(*"
!(%"
!(+"
!(&"
!(,"
'"Frequency
Proportion Hymenopteran
Figure 12. Distribution of the proportion of hymenopteran out of the total number 
of visitors across 27 
Asclepias 
species. 
120Figure 13.
 Maximum likelihood ancestral state reconstruction of character traits: 
hood presence/absence (a) and conspicuous hood opening (b) mapped onto a 
mirrored phylogram of 107 
Asclepias 
species. The black circles represent the ancestral 
node to 
Asclepias
. Boxes represent the character state of each species and the pie charts 
represent the maximum likelihood character reconstruction at each node. There was a 
significant correlation between these two traits (p < 0.01) using PagelÕs correlation 
method.
Pergularia
Calotropis
Gfruticosus
Gcancellatus
Acutleri
Aleptopus
Asubulata
Amasonii
Aalbicans
Asubaphylla
Aemoryi
Aoenotheroides
Anyctaginifolia
Aarenaria
Aprostrata
Aengelmanniana
Astenophylla
Ahirtella
Alongifolia
Aviridiflora
Asolanoana
Acalifornica
Agreenei
Aparishii
Avestita
Aerosa
Adavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
Arosea
Aquinquedentata
Azacatecana
Anummularia
Alemmonii
Aotarioides
Aovata
Anotha
Aoaxaca
Asp5175
Abifida
Asimilis3000
Azanthodacryon
Afournieri
Abrachystephana
Apellucida
Aauriculata
Ajorgeana
Apringlei5119
Acircinalis
Apratensis
Amacrotis
Acoulteri
Aelegantula
Ajaliscana
Aboliviensis
Anivea
Acurassavica
Afascicularis
Averticillata
Apumila
Asubverticillata
Atexana
Aperennis
Aincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
Alanceolata
Alatifolia
Aspeciosa
Aovalifolia
Avariegata
Ahallii
Acurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
Amichauxii
Asullivantii
Atomentosa
Aconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
Asanjuanensis
Aelata
Aglaucescens
Amirifica
Alynchiana
Amacroura
Afeayi
Acinerea
Aviridula
Apedicellata
Arubra
Pergularia
Calotropis
Gfruticosus
Gcancellatus
Acutleri
Aleptopus
Asubulata
Amasonii
Aalbicans
Asubaphylla
Aemoryi
Aoenotheroides
Anyctaginifolia
Aarenaria
Aprostrata
Aengelmanniana
Astenophylla
Ahirtella
Alongifolia
Aviridiflora
Asolanoana
Acalifornica
Agreenei
Aparishii
Avestita
Aerosa
Adavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
Arosea
Aquinquedentata
Azacatecana
Anummularia
Alemmonii
Aotarioides
Aovata
Anotha
Aoaxaca
Asp5175
Abifida
Asimilis3000
Azanthodacryon
Afournieri
Abrachystephana
Apellucida
Aauriculata
Ajorgeana
Apringlei5119
Acircinalis
Apratensis
Amacrotis
Acoulteri
Aelegantula
Ajaliscana
Aboliviensis
Anivea
Acurassavica
Afascicularis
Averticillata
Apumila
Asubverticillata
Atexana
Aperennis
Aincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
Alanceolata
Alatifolia
Aspeciosa
Aovalifolia
Avariegata
Ahallii
Acurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
Amichauxii
Asullivantii
Atomentosa
Aconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
Asanjuanensis
Aelata
Aglaucescens
Amirifica
Alynchiana
Amacroura
Afeayi
Acinerea
Aviridula
Apedicellata
Arubra
121Horn present
Horn absent
Hood openHood closed
(a)
(b)
Pergularia
CalotropisGfruticosus
GcancellatusAcutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
(a)
Figure 14.
 Correlated evolution 
between hood shape and 
hymenopteran pollinators. 
(a) A 
phylogram of the maximum likelihood 
ancestral state reconstruction of hood 
opening shape. (b) A phylogram of 31 
species with hood opening shape 
(domed or flared; green and blue) and 
pollinator composition (< or 
! 50% hymenopteran; light green and light 
blue, respectively) mapped to the tips. 
Boxes represent the character state of 
each species and the pie charts represent 
the maximum likelihood character 
reconstruction at each node There was a 
marginally significant association 
between hood opening shape and 
pollinator composition (P
 = 0.087).Hood ßared
Hood domed
122Hood ßared
Hood domed
! 50% Hymenoptera
< 50% Hymenoptera
(b)
123Figure 14 (contÕd)
Gfruticosus
GcancellatusAcutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
0.3327.352trait value
length=0.239Figure 15.
 Evolution of hood length. 
Phylogram using a continuous color gradient to 
show maximum likelihood ancestral trait reconstruction for hood length (see inset). 
The black circle represents the ancestral node to 
Asclepias
. The rate of evolution (
!2) is 490.5 and the phylogenetic signal (
") is 0.14. 
Gfruticosus
GcancellatusAcutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
0.3327.352trait value
length=0.239124Figure 16.
 Evolution of gynostegium width.
 Phylogram using a continuous color 
gradient to show maximum likelihood ancestral trait reconstruction for gynostegium 
width (estimated by the distance between adjacent corpuscula; see inset). The black 
circle represents the ancestral node to 
Asclepias
. The rate of evolution (
!2) of Asclepias 
species is 604.3 and the phylogenetic signal (
") is 0.72.
Gfruticosus
GcancellatusAcutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
0.7443.142trait value
length=0.239125Gfruticosus
GcancellatusAcutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
0.7443.142trait value
length=0.239r = 0.56 P < 0.001Figure 17. Correlation between p
hylogenetic independent contrasts for 
gynostegium width and hood length.  
Est. Gynostegium Width PICs
Hood Length PICs
126!5051015!2!1012Hood Length (PIC)Proportion Hymenopteran (PIC)
Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
01trait value
length=0.161Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
01trait value
length=0.161Percent Hymenoptera
Hood LengthAsubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
0.3327.352trait value
length=0.161Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
0.3327.352trait value
length=0.161Figure 18.
 Test of correlated 
evolution of hood length and the 
proportion of pollinators that are 
hymenopterans.
 (a) Predicted 
evolution of hood length (right) and 
corresponding changes in 
proportion hymenopteran (left) 
visualized using a continuous color 
gradient to show maximum 
likelihood ancestral reconstruction. 
(b) Correlatin between 
phylogenetic independent contrasts 
for hood length
 and percent 
hymenopteran (r = 0.06, P = 0.77).
(b)(a)127!3!2!1012!2!1012Est. Gynostegium Width (PIC)Proportion Hymenopteran (PIC)
Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
01trait value
length=0.161Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
01trait value
length=0.161Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
0.7833.142trait value
length=0.161Asubulata
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAparishii
AerosaAcryptoceras
Ahypoleuca
Acurassavica
Afascicularis
Averticillata
Asubverticillata
Aincarnata
Apulchra
Aamplexicaulis
Alatifolia
AspeciosaAhalliiApurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Asullivantii
AtomentosaAviridis
0.7833.142trait value
length=0.161Proportion Hymenopteran
Est. Gynostegium Width
Figure 19. Test of correlated 
evolution of the estimated 
gynostegium width and the 
proportion of pollinators that are 
hymenopterans. 
(a) Predicted 
evolution of gynostegium width 
(right) and corresponding changes 
in proportion hymenopteran (left) 
visualized using a continuous color 
gradient to show maximum 
likelihood ancestral reconstruction. 
(b) Correlation between 
phylogenetic independent contrasts 
for gynostegium width and percent 
hymenopteran (r = 0.22, P = 0.29)
.128(b)(a)Acutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
14715313265412811109(a)129Figure 20.
 Results of a 
SURFACE analysis of 
gynostegium and hood 
evolution supporting four 
adaptive optima of these traits 
in the North American 
Asclepias.
 (a) Phylogenetic tree 
with divergent clades and 
species in gray. (b) Phylogenetic 
tree with convergent (colored) 
and non-convergent (grey) 
clades painted onto the branches; 
Black clades were not placed in 
any phenotypic optima. 
Numbers in circles represent 
transitions to the phenotypic 
optima. (c) Trait values for each 
species (small circles) and 
estimated optima (large circles) 
calculated as the phenotypic 
mean of the species within each 
regime, with regime colors 
matching those in the tree. (d
) Photos of representative species 
of each regime. Colored frames 
on each box match the regime 
color; gray is on the left, blue in 
the center, green on the right, 
and red on top.
Acutleri
AleptopusAsubulata
AmasoniiAalbicansAsubaphylla
Aemoryi
AoenotheroidesAnyctaginifolia
Aarenaria
Aprostrata
AengelmannianaAstenophylla
Ahirtella
Alongifolia
Aviridiflora
AsolanoanaAcalifornica
AgreeneiAparishii
Avestita
AerosaAdavisii
Acryptoceras
Ahumistrata
Acordifolia
Alinaria
Aatroviolacea
Ahypoleuca
Alaxiflora
AroseaAquinquedentataAzacatecanaAnummularia
AlemmoniiAotarioides
Aovata
AnothaAoaxacaAsp5175AbifidaAsimilis3000Azanthodacryon
Afournieri
Abrachystephana
ApellucidaAauriculata
AjorgeanaApringlei5119
AcircinalisApratensis
AmacrotisAcoulteri
AelegantulaAjaliscanaAboliviensisAnivea
Acurassavica
Afascicularis
Averticillata
ApumilaAsubverticillata
Atexana
AperennisAincarnata
Apulchra
Amexicana
Aangustifolia
Agentryi
Aamplexicaulis
AlanceolataAlatifolia
AspeciosaAovalifolia
Avariegata
AhalliiAcurtissii
Apurpurascens
Aexaltata
Aquadrifolia
Asyriaca
Ainterior2816
Aobovata
AmichauxiiAsullivantii
AtomentosaAconnivens
Awelshii
Amacrosperma
Ainvolucrata
Alabriformis
Acapricornu
Aasperula
Aviridis
Aruthiae
Auncialis
AsanjuanensisAelataAglaucescensAmirifica
AlynchianaAmacroura
Afeayi
AcinereaAviridula
ApedicellataArubra
14715313265412811109(b)130Figure 20 (contÕd)
5 mm
1.01.52.02.53.01234567Gynostegium Side (mm)Hood Length (mm)Est. Gynostegium Width (mm)
131(c)Figure 20 (contÕd)
(d)5 mm
132Figure 20 (contÕd)
LITERATURE CITED
133LITERATURE CITED
Andersson, M. 1982. Female choice selects for extreme tail length in a widowbird. Nature 
299:818Ð820.
Armbruster, W. S., C. P. H. Mulder, B. G. Baldwin, S. Kalisz, B. Wessa, and H. Nute. 2002. 
Comparative analysis of late floral development and mating-system evolution in tribe 
Collinsieae (Scrophulariaceae s.l.). American Journal of Botany 89:37Ð49.
Betz, R. F., R. D. Struven, J. E. Wall, and F. B. Heitler. 1994. Insect pollinators of 12 milkweed 
(Asclepias
) species. Thirteenth North American Prairie Conference 45Ð60.
Bradshaw, H. D., Jr, and D. W. Schemske. 2003. Allele substitution at a flower colour locus 
produces a pollinator shift in monkeyflowers. Nature 426:176Ð178.
Butler, M. A., and A. A. King. 2004. Phylogenetic Comparative Analysis: A Modeling Approach 
for Adaptive Evolution. The American Naturalist 164:683Ð695. 
Caruso, C. M., D. L. D. Remington, and K. E. Ostergren. 2005. Variation in resource limitation 
of plant reproduction influences natural selection on floral traits of 
Asclepias syriaca
. Oecologia 146:68Ð76.
Castellanos, M. C., P. Wilson, and J. D. Thomson. 2004. 'Anti-bee' and Òpro-birdÓ changes 
during the evolution of hummingbird pollination in 
Penstemon
 flowers. J Evol Biol 
17:876Ð885.
Coddington, J. A. 1994. The roles of homology and convergence in studies of adaptation. Pp. 53Ð
78 in P. Eggleton and R. I. Vane-Wright, eds. Phylogenetics and Ecology. Academic 
Press, San Diego.
Conner, J. K. 1988. Field measurements of natural and sexual selection in the fungus beetle, 
Bolitotherus cornutus. Evolution 42:736Ð749.
Doughty, P. 1996. Statistical analysis of natural experiments in evolutionary biology: comments 
on recent criticisms of the use of comparative methods to study adaptation. The American 
Naturalist 148:943Ð956.
Felsenstein, J. 1985. Phylogenies and the Comparative Method. The American Naturalist 125:1Ð
15.Fenster, C. B., W. S. Armbruster, P. Wilson, M. R. Dudash, and J. D. Thomson. 2004. Pollination 
syndromes and floral specialization. Annu. Rev. Ecol. Evol. Syst. 35:375Ð403.
134Fishbein, M., and D. L. Venable. 1996. Diversity and temporal change in the effective pollinators 
of Asclepias tuberosa
. Ecology 77:1061Ð1073.
Fishbein, M., D. Chuba, C. Ellison, R. J. Mason-Gamer, and S. P. Lynch. 2011. Phylogenetic 
relationships of 
Asclepias
 (Apocynaceae) inferred from non-coding chloroplast DNA 
sequences. Systematic Botany 36:1008Ð1023. 
Friedman, J., and S. C. H. Barrett. 2008. A Phylogenetic Analysis of the Evolution of Wind 
Pollination in the Angiosperms. International Journal of Plant Sciences 169:49Ð58.
Fritz, S. A., and A. Purvis. 2010. Selectivity in mammalian extinction risk and threat types: a 
new measure of phylogenetic signal strength in binary traits. Conserv. Biol. 24:1042Ð
1051.GŠmez, J. M., F. Perfectti, and J. Lorite. 2015. The role of pollinators in floral diversification in a 
clade of generalist flowers. Evolution 69:863Ð878.
GŠmez, J. M., J. Bosch, F. Perfectti, J. Fern⁄ndez, M. Abdelaziz, and J. Camacho. 2008. Spatial 
variation in selection on corolla shape in a generalist plant is promoted by the preference 
patterns of its local pollinators. Proc. R. Soc. B 275:2241-2249.
Grant, P. R., and B. R. Grant. 2002. Unpredictable evolution in a 30-year study of Darwin's 
finches. Science 296:707Ð711.
Hegland, S. J., and ¯. Totland. 2005. Relationships between species' floral traits and pollinator 
visitation in a temperate grassland. Oecologia 145:586Ð594.
Ingram, T., and D. L. Mahler. 2013. SURFACE: detecting convergent evolution from 
comparative data by fitting Ornstein
-Uhlenbeck models with stepwise Akaike 
Information Criterion. Methods Ecol Evol. 4: 416-425.
Ivey, C. T., P. Martinez, and R. Wyatt. 2003. Variation in pollinator effectiveness in swamp 
milkweed, 
Asclepias incarnata
 (Apocynaceae). American Journal of Botany 90:214Ð225.
King, A. A. and M. A. Butler. 2009. ouch: Ornstein-Uhlenbeck models for phylogenetic 
comparative hypotheses (R package), http://ouch.r-forge.r-project.org 
Knudsen, J. T., and L. Tollsten. 2008. Floral scent in bat-pollinated plants: a case of convergent 
evolution. Botanical Journal of the Linnean Society 119:45Ð57.
Kunze, H. 1990. Morphology and evolution of the corona in Asclepiadaceae and related families. 
Tropische und subtropische Pflanzenwelt.
135Lahti, D. C., N. A. Johnson, B. C. Ajie, S. P. Otto, A. P. Hendry, D. T. Blumstein, R. G. Coss, K. 
Donohue, and S. A. Foster. 2009. Relaxed selection in the wild. Trends Ecol Evol 
24:487Ð496.
Larson, A., and J. B. Losos. 1996. Phylogenetic systematics of adaptation. Pp. 187Ð220 
in M. R. 
Rose and G. V. Lauder, eds. Adaptation. Academic Press, San Diego.
Maddison, W. P., and D. R. Maddison. 2011. Mesquite: a modular system for evolutionary 
analysis.
Mahler, D. L., T. Ingram, L. J. Revell, and J. B. Losos. 2013. Exceptional Convergence on the 
Macroevolutionary Landscape in Island Lizard Radiations. Science 341:292Ð295.
Morgan, M. T., and D. Schoen. 1997. Selection on reproductive characters: floral morphology in 
Asclepias syriaca
. Heredity 79:433Ð441.
Ollerton, J., and S. Liede. 1997. Pollination systems in the Asclepiadaceae: a survey and 
preliminary analysis. Biological Journal of the Linnean Society 62:593Ð610.
Orme, D., R. Freckleton, G. Thomas, T. Petzoldt, S. Fritz, N. Isaac, and W. Pearse. 2013. caper: 
Comparative Analyses of Phylogenetics and Evolution in R. R package version 0.5.2. 
http://CRAN.R-project.org/package=caper
Pagel, M. 1994a. Detecting Correlated Evolution on Phylogenies: A General Method for the 
Comparative Analysis of Discrete Characters. Proc. R. Soc. Lond. B 255:37Ð45. 
Pagel, M. 1994b. The adaptionist wager. Pp. 29Ð52 
in P. Eggleton and R. I. Vane-Wright, eds. 
Phylogenetics and Ecology. Academic Press, San Diego.
Pagel, M. 1999. Inferring the historical patterns of biological evolution. Nature 401:877Ð884.
Paradis, E. 2012. Analysis of phylogenetics and evolution with R. 2nd ed. Springer, New York.
Paradis, E., J. Claude, and K. Strimmer. 2004. APE: Analyses of Phylogenetics and Evolution in 
R language. Bioinformatics 20:289Ð290.
Perez, F., M. T. K. Arroyo, R. Medel, and M. A. Hershkovitz. 2006. Ancestral reconstruction of 
flower morphology and pollination systems in 
Schizanthus
 (Solanaceae). American 
Journal of Botany 93:1029Ð1038.
Pleasants, J. M. 1991. Evidence for short-distance dispersal of pollinia in 
Asclepias syriaca
 L. 
Funct Ecol 5:75Ð82.
R Core Team. 2014. R: A language and environment for statistical computing. R Foundation for 
Statistical Computing, Vienna, Austria.
136Rafferty, N. E., and A. R. Ives. 2012. Pollinator effectiveness varies with experimental shifts in 
flowering time. Ecology 93:803Ð814.
Rasband, W. S. 1997. ImageJ. U. S. National Institutes of Health, Bethesda, MD, USA.
Revell, L. J. 2012. phytools: an R package for phylogenetic comparative biology (and other 
things). Methods Ecol Evol. 3:217-223.
Weiblen, G. D. 2004. Correlated evolution in fig pollination. Systematic Biology 53:128Ð139.
Woodson, R. E., Jr. 1941. The North American Asclepiadaceae. I. Perspective of the genera. 
Annals of the Missouri Botanical Garden 28:193Ð244.
Woodson, R. E., Jr. 1954. The North American Species of 
Asclepias
 L. Annals of the Missouri 
Botanical Garden 41:1Ð211.
Wyatt, R., and S. B. Broyles. 1994. Ecology and evolution of reproduction in milkweeds. Annual 
Review of Ecology and Systematics 25:423Ð441.
137