EXPERIMENTAL STUDIES OF ADAPTATION AND SPECIATION IN TWO
NEOTROPICAL COSTUS SPECIES
By
Grace Fu-chun Chen

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Plant Biology
Ecology, Evolutionary Biology and Behavior
2011

ABSTRACT
EXPERIMENTAL STUDIES OF ADAPTATION AND SPECIATION IN TWO
NEOTROPICAL COSTUS SPECIES
By
Grace Fu-chun Chen
To understand the extraordinary diversity and complexity of organisms requires
comprehensive studies of the mechanisms by which populations diverge to become different
species. The importance of ecology in speciation has long been recognized; however, there is
little direct evidence that ecological factors are the principal isolating barriers between closely
related species. To examine how adaptation to ecological factors may contribute to reproductive
isolation and lead to speciation, I studied the isolating mechanisms between two closely related
Neotropical herbs, Costus allenii and C. villosissimus. These perennial species occur in the same
geographic region, but occupy distinct habitats. Costus allenii is located along shady ravines in
mature forests, while C. villosissimus is found in drier, open sites along forest edges. Both
species are pollinated by euglossine bees and can be crossed to produce fully fertile hybrids. I
conducted field studies in central Panama, where the two species co-occur, to quantify the
strength of multiple isolating barriers and total reproductive isolation. I also conducted reciprocal
transplant experiments in the field that were coupled with greenhouse studies to determine
whether the two species are locally adapted to ecological factors in their respective habitats, and
to identify putative adaptive traits that may contribute to local adaptation and speciation.
There are four chapters in my dissertation. Chapter 1 discusses the mechanisms
contributing to sexual isolation in these species. I found that pollinator-mediated barriers and
gametic isolation restrict heterospecific gene flow asymmetrically between C. allenii and C.
villosissimus. Chapter 2 describes the parapatric distribution of the two species along a water

availability gradient. The results of reciprocal transplant experiments at two life stages, seeds and
juveniles, suggest that local adaptation contributes to strong microhabitat isolation and
asymmetrical extrinsic postzygotic isolation between C. allenii and C. villosissimus. As habitat
isolation has been found to be strong in this system, the environmental factors contributing to
this form of isolation are of great interest. Chapter 3 summarizes comparisons of the two parental
habitats and shows differences in putative adaptations to these habitats among C. allenii, C.
villosissimus, and their F1 hybrids in the greenhouse. Higher leaf mass per area was found in C.
allenii, which occupies habitats with lower light availability, while higher drought tolerance was
found in C. villosissimus, which occupies habitats with lower soil moisture. The F1 hybrids had
leaf mass per area similar to that of C. villosissimus, although hybrid fitness was not reduced in
C. allenii habitats compared to pure C. allenii transplants. The F1 hybrids had intermediate
drought tolerance, which is consistent with their lower seedling survival in C. villosissimus
habitats. Chapter 4 presents an examination of multiple isolating barriers and comparisons of
their relative contribution to speciation in C. allenii and C. villosissimus. Total reproductive
isolation was found to be high between the two species, and the major isolating barriers are
ecogeographic and microhabitat isolation. Because ecogeographic isolation represents spatial
isolation based on genetic differences between species, presumably due to local adaptation, and
microhabitat isolation is found to be a consequence of local adaptation (Chapter 2), I conclude
that local adaptation is the primary mechanism of speciation in C. allenii and C. villosissimus.
In this research, I used a comprehensive approach to the study of speciation by linking
adaptation to the origin of reproductive isolation. My dissertation provides empirical evidence
for how local adaptation to different environmental conditions contributes to reproductive
isolation and lead to speciation.

Copyright by
GRACE FU-CHUN CHEN
2011

DEDICATION

This dissertation is dedicated to my parents, who have always been supportive of me.

v

ACKNOWLEDGMENTS

This dissertation would not have been possible without the support of many people and
programs. I would like to thank the members of my guidance committee, Doug Schemske, Jeff
Conner, Kay Gross, Jen Lau and Frank Ewers for providing helpful suggestions and guidance
through this project. I would like to especially thank my major advisor, Doug Schemske, for
giving me the opportunity to learn from the best, for providing valuable guidance and resources,
for being patient with me but pushing me to my limits, and for encouraging and supporting me in
every possible way every step throughout this research. Doug has taught me many things, both
scientific and nonscientific, and it is the greatest honor for me to have him as my mentor. I would
also like to express my appreciation to other professors at MSU for their support, especially to
Andy Jarosz, Barb Sears, and Stephen Thomas.
I am indebted to many colleagues for their help in developing and conducting this
research. All the current and previous graduate students and post-doctoral fellows in the
Schemske lab have provided helpful critical discussions on this research, including Kathleen Kay,
Amy Angert, Changbao Li, Jeff Evans, Jay Sobel, Lorna Watt, Emily Dittmar, and Chris Oakley.
Special thanks go to Jay for his significant contribution to my research ideas and for his friendly
support. Jay and I have been collaborating on the unification of methods for estimating the
strength of reproductive isolation and on the application of ecological niche modeling for
estimating ecogeographic isolation. It is my great honor and pleasure to work and share the
office with Jay. I also thank Lorna for editing the chapters at the last minutes. Fellow graduate
students in other labs have also been supportive and helpful, especially Mike Grillo, Anne Royer,

vi

Raffica LaRosa, Colin Kremer, Emily Grman, Todd Robinson, Chin-mei Lee, Brad Chang, and
Tomomi Suwa.
I would also like to thank Mark Hammond for all of his assistance. My greenhouse and
field work would never have been possible without Mark’s advice and support. I also have
received valuable assistant from the undergrads in the Schemske lab including, Darin Ellair, Eric
Minar, Rachael Fuller, Stephanie Carpenter, Krystal Richardson, Lindsay Petroff, Alexandra
Beels, Nick Batora, Daniel Greiner, and Nate Rowland. They not only did a fantastic job keeping
my tropical plants healthy in the greenhouse in Michigan, but also helped me with everything
that I needed in this project. I am particularly grateful to Jan Szyren for providing a microscope
when I needed one to measure pollen viability.
I would like to extend my heartfelt gratitude to the scientists who have provided
insightful advice and encouragement when I was working in Panama: Allen Herre, Harilaos
Lessios, Egbert Leigh, Jr., Sunshine Van Beal, Phillis (Lissy) Coley, Thomas Kursar, Ingrid
Parker, Greg Gilbert, Annette Aiello, Richard Condit, Bettina Engelnbrecht, Jeneene Touchton,
Patrick Kelley, Corey Tarwater, Ioana Chivez, Adam Smith, Gogi Kalka, and Joe Wright. I
would like to especially thank Lizzie Leigh, who passed away in 2008, for her kindness and
hospitality.
My field work relied heavily on the support of Smithsonian Tropical Research Institute,
especially from the people in the Visitor Service Office, in Gamboa, and on Barro Colorado
Island. In particular, I would like to thank Orelis Arosemena for preparing all the permits for me,
Raineldo Urriola for solving all my vehicle and permit problems, Marcela Paz and Yara Clemons
for arranging my lodging and transportation, Laurie Seid for saving my life after I got stung by a
bullet ant, Oris Acevedo and Belkys Jimenez for supporting my research on BCI and Gigante,

vii

and Argelis Ruiz for taking care of me after my bike accident and for finding field assistants for
me. My field work in Panama could not have been conducted without extensive help from many
local assistants, William Jacome, Dixon, Nicolas Fabbroni, Julia Sarco, Lisssette Jimenez,
Anyuri Gonzalez, Jilma Rios, Jose (Santos) Lopez, and Jorge Murillo. Doug Schemske, Lorna
Watt and Emily Dittmar also contributed to my field work. A big thank you goes to these great
people who have worked with me in the rain.
I am also grateful for the insight I gained through discussion with researchers developing
aster model for estimating fitness including, Charles Geyer, Ruth Shaw, and John StantonGeddes.
Funding for this research is gratefully acknowledged. Department of Plant Biology at
MSU provided six Paul Taylor Endowment Funds, which supplied the travel to field sites and
conferences. College of Natural Science at MSU provided Dissertation Continuous and
Completion Fellowships to support the field work as well as my living expenses. Smithsonian
Tropical Research Institute also provided a Short Term Fellowship for my field work in Panama
in 2007. I would also like to thank my advisor, Doug Schemske, for financially supporting me
and allowing me to hire assistants and to purchase everything I needed for the research through
his discretionary account.
I would like to thank Authoridad del Canal de Panamá (ACP) and Authoridad Nacional
del Ambiente (ANAM) for issuing the research, observation, import, and export permits. I am
also grateful to Missouri Botanical Garden for allowing me to access the herbarium specimens
and data and to Genomics Technology Support Facility at MSU for performing Genescan when I
measured sexual isolation.

viii

I would never have pursued a Ph.D. degree in the U.S. without the support of many of my
former teachers, professors, and senior colleagues. Many thanks go to Chang-Sheng Kuoh, Lucia
Liu Severinghaus, and Liang-Li (Leo) Liu for their encouragement and guidance before and
during my life in grad school. I thank Bob Marquis for being the advisor for my Master degree
and suggesting me to apply for Ph.D. in the Schemske lab.
Finally, I am sincerely grateful to my family and friends. My parents have always
encouraged and supported my education. This dissertation is dedicated to them. I thank my
sisters, Edelgard, Kate, and Phoenix, for their understanding and for taking great care of my
parents. I also thank Hsi-Chien Shih for his love and support in our seven year relationship. I
would also like to thank my old friends in TFG, NCKU, and UMSL, as well as my basketball
teammates in the “Chopsticks” for their support. In particular, I want to thank Chu-Yin (Evelyn)
Yeh for her endless love, encouragement (with great food), understanding and patience in the last
year of this project. I also thank BaJiao for accompanying me throughout my Ph.D. program.

ix

TABLE OF CONTENTS
LIST OF TABLES………………………………………………………………………….……xii
LIST OF FIGURES…………………………………………………………..………..………..xiii
CHAPTER 1
THE MAGNITUDE AND MECHANISMS OF SEXUAL ISOLATION IN TWO BEEPOLLINATED COSTUS (COSTACESE)…………………………………………….…………..1
Abstract………………………………………………………………………....................2
Introduction……………………………………………………………….………….……4
Materials and Methods………………………………………………………….................8
Results……………………………………………………….……………………….…..23
Discussion………………………………………………….…………………….………29
Tables……………………………………………………………………….……………42
Figures…………………………………………………….………………….…………..45
Appendix…………………………………………………….…………………………...49
Literature Cited……………………………………………………………..……………58
CHAPTER 2
LOCAL ADAPTATION TO DIFFERENT MICROHABITATS IN NEOTROPICAL
COSTUS…………………………………………….……………………………………………66
Abstract…………………….……………………………………………….....................67
Introduction……………………………….……………………………………………...69
Materials and Methods……………………………………….…………………………..73
Results…………………………….…………………………………….………………..85
Discussion………………………….……………………………………………….……92
Figures……………………………….………………………………………………….102
Literature Cited………………………………………………………………..………..112
CHAPTER 3
ECOLOGICAL FACTORS CONTRIBUTING TO LOCAL ADAPTATION IN TWO
NEOTROPICAL COSTUS………………………………….…………………….……………118
Abstract…………………………………………………………….…………………...119
Introduction………………………………………………………….…………………121
Materials and Methods………………………………………………………….……...127
Results………………………………………………………………………………….134
Discussion……………………………………………………….……………………..136
Tables…………………………………………………………….……………………..147
Figures………………………………………………………….………………………148
Literature Cited……………………………………………………………..…………..153
CHAPTER 4
REPRODUCTIVE ISOLATION BETWEEN TWO CLOSELY RELATED NEOTROPICAL
HERBS………………………………….……………………….…………………...…………159
Abstract…………………………………………….…………………………………...160
 

x

Introduction………………………………………….…….……………………………161
Materials and Methods…………………………………….…….……………………...165
Results…………………………………………………….……………….……………181
Discussion………………………………………………….…………………………...187
Tables……………………………………….…………………………………………..199
Figures…………………………………………………..………………………………201
Literature Cited……………………………………………………..…………………..206

 

xi

LIST OF TABLES
Table 1.1. Comparisons of floral traits between C. allenii and C. villosissimus…………….…..42
Table 1.2. Reproductive isolation index values and the corresponding results of repeated G tests
of goodness of fit on floral mechanical isolation, gametic isolation, and the cumulative
estimate of sexual isolation…………………………………………………………...43
Table 1.3. Components of reproductive isolation and absolute contributions to sexual isolation
for the reproductive barriers studied………………………………………………….44
Table A.1. Results of repeated G tests of goodness of fit on floral mechanical isolation, gametic
isolation, and the overall strength of sexual isolation………………………………...50
Table 3.1. Summary of ANOVA results for soil moisture measured in habitats of Costus allenii
and C. villosissimus in two dry and wet seasons…………………………….………147
Table 4.1. Climatic variables used in ecological niche modeling……………………….….…..199
Table 4.2. Estimates of the studied reproductive isolating barriers between C. allenii and C.
villosissimus………………………………………………………………….………200

 

xii

LIST OF FIGURES
Figure 1.1. Flowers of (A) Costus allenii and (B) C. villosissimus with the pollinator Euglossa
imperialis and graphic illustrations of floral structures and measurements of the
morphological traits in a front view (C) and a longitudinal section view (D)………..45
Figure 1.2. Map of the study site in Central Panama (modified from
http://www.stri.si.edu./images/Mapas/Pipeline_BarroColorado.jpg)…..................…..46
Figure 1.3. Pollinator transitions between C. allenii and C. villosissimus in the pollination array
in 2007 (A), 2008 (B), and for the two years combined
(C)………………………………………………………………………….…….……47
Figure 1.4. Dye deposition scores measured on stigmas of C. allenii and C. villosissimus in the
pollination array in 2007 (A), 2008 (B), and for the two years combined
(C)………………………………………………………………………………....…..48
Figure 2.1. Map of the study sites in Central Panama (modified from
http://www.stri.si.edu./images/Mapas/Pipeline_BarroColorado.jpg)..........................102
Figure 2.2. Comparisons of monthly precipitation at Frijolito and Gamboa in the wet and dry
seasons…………………………………………………………………...……....…..103
Figure 2.3. Soil moisture and species’ distribution along Pipeline Road (PLR)…………….…104
Figure 2.4. Fitness components of seed reciprocal transplant experiment for the 2007 and 2008
cohorts combined……………………………………………………………….…...105
Figure 2.5. Comparisons of the changes in seed germination and seedling mortality through time
between reciprocal transplanting sites and among C. allenii, F1allenii, F1villo, and C.
villosissimus for the 2007 and 2008 cohorts combined……………..………….……106
Figure 2.6. Fitness components of cutting reciprocal transplant experiment……………….….107
Figure 2.7. Comparisons of cutting survival among C. allenii, F1allenii, F1villo, and C.
villosissimus and among transplanting sites AG (A), AP (B). VG (C), and VP (D)...108
Figure 2.8. Comparisons of cutting growth among C. allenii, F1allenii, F1villo, and C.
villosissimus and among transplanting sites AG (A), AP (B). VG (C), and VP (D)...109
Figure 2.9. Comparison of relative total fitness among C. allenii, F1allenii, F1villo, and C.
villosissimus and between parental habitats………………………………..………..110
Figure 2.10. Proportion of pest damage on new leaves of C. allenii, F1allenii, F1villo, and C.
 

xiii

villosissimus in the cutting reciprocal transplant experiment…………………….….111
Figure 3.1. Experimental design of the rehydration treatments in the greenhouse drought
tolerance experiment………………………………………………………...………148
Figure 3.2. Soil moisture measured in habitats of C. allenii and C. villosissimus in wet and dry
seasons……………………………………………………………………..…….…..149
Figure 3.3. Comparisons among C. allenii, C. villosissimus and their F1 hybrids in Leaf Mass per
Area (LMA)……………………………………………………………...…………..150
Figure 3.4. Changes in plant appearance and size in the drought tolerance experiment……….151
Figure 3.5. Comparisons among C. allenii, C. villosissimus and their F1 hybrids in drought
tolerance (DTI) measured as the number of days until the plant is stemless………..152
Figure 4.1. Overlay map showing ecogeographic isolation between C. allenii and C.
villosissimus……………………………………………………………….….……...201
Figure 4.2. Flowering phenology of C. allenii and C. villosissimus in 2007 (A) and 2008
(B)……………………………………………………………………………...…….202
Figure 4.3. Comparisons of seed set among C. allenii, F1allenii, F1villo, and C. villosissimus...203
Figure 4.4. Comparisons of proportion of pollen germinating, representing male fertility, among
C. allenii, F1allenii, F1villo, and C. villosissimus………………………………....….204
Figure 4.5. Relative contribution of isolating barriers to total isolation in C. allenii and C.
villosissimus…………………………………………………………..……………...205

 

xiv

Chapter 1

The Magnitude and Mechanisms of Sexual Isolation in Two Bee-Pollinated Costus (Costaceae)

1

ABSTRACT
Examining different reproductive isolating barriers and comparing their importance in
gene flow reduction are essential for understanding processes of speciation. Sexual isolation has
been found to contribute to speciation in many sympatric taxa, while its role in parapatric taxa
where interspecific gene flow is plausible is poorly understood. I investigated sexual isolation in
Costus allenii and C. villosissimus, two closely related Neotropical understory species which
occupy different and often adjacent habitats, yet are within cruising range of their shared insect
pollinators. Individual barriers investigated include pollinator isolation, floral mechanical
isolation, and gametic isolation. A pollination array was used to test whether individual
pollinators travel between species, to estimate the proportion of hetero- and conspecific pollen
deposited on the stigmas, and to examine the proportion of hybrid progeny. In comparison to C.
villosissimus, C. allenii produces flowers with smaller labella, shorter stamen-labellum distances,
greater stigma-to-corolla-aperture ratios, and shorter styles. Pollinators preferred C. villosissimus,
but visited both species, and displayed low constancy. In C. allenii, heterospecific pollinator
transitions were more common than conspecific transitions, but floral mechanical isolation
greatly reduced the likelihood of heterospecific gene flow. Costus villosissimus, however, was
isolated from C. allenii largely because it is preferred by pollinators. The contribution of gametic
isolation was weak in C. allenii and moderate in C. villosissimus. Combining estimates for
pollinator isolation, floral mechanical isolation, and gametic isolation, I conclude that sexual
isolation is weak in C. allenii, restricting heterospecific gene flow by 12%, but moderate in C.
villosissimus, where gene flow from C. allenii is reduced by 58%. To understand the relative
contribution of sexual isolation to total isolation in this parapatric species pair requires

2

estimating the importance of other prezygotic isolating barriers such as ecogeographic,
phenological and habitat isolation, and both extrinsic and intrinsic postzygotic barriers.

3

INTRODUCTION
In accordance with the biological species concept (Mayr, 1942), species are formed as a
consequence of accumulated reproductive isolation limiting gene flow between populations. A
number of barriers have been identified as components of reproductive isolation (Dobzhansky,
1937; Mayr, 1947, 1963; Schluter, 2000; Coyne and Orr, 2004; Nosil et al., 2005) and their
strength has been estimated in a variety of organisms (e.g., Palumbi and Metz, 1991; Hatfield
and Schulter, 1999; Hurt et al., 2005; Whiteman and Semlitsch, 2005; Mallet, 2006; Nosil, 2007;
see review in Lowry et al., 2008 for examples in flowering plants). To assess the overall strength
of isolation and to estimate the relative importance of different barriers, the strength of multiple
isolating barriers is estimated sequentially, and summed over the life history of the organism
(Ramsey et al., 2003; Coyne and Orr, 2004; Husband and Sabara, 2004; Kay, 2006;
Matsubayashi and Katakura, 2009; Dopman et al., 2010; Schemske, 2010; Sobel et al., 2010).
The relative importance of pre- vs. postzygotic isolating barriers has been extensively
debated. Because the strength of intrinsic postzygotic isolation increases with time since
divergence (Coyne and Orr, 1989, 1997; Presgraves, 2002; Price and Bouvier, 2002; Bolnick and
Near, 2005), it is unclear whether the barriers which eliminate interspecific gene flow today are
the same as those which were important during speciation. Schemske (2010) has suggested that
prezygotic barriers such as ecogeographic, habitat and sexual isolation may often play the major
role in speciation, while intrinsic postzygotic isolation typically evolves after speciation is
complete.
Sexual isolation has received considerable attention due to its direct relationship to
reproduction. Sexual isolation (i.e., nonecological isolation in Coyne and Orr, 2004; chapter 6)
involves prezygotic isolating barriers such as mating preference, mechanical isolation, and

4

gametic isolation. Because reproduction between specific individuals can be better
experimentally controlled when the subjects are plants, this study focuses on studying the
barriers in plants rather than in animals. Mating preference contributes to reproductive isolation
when conspecifics mate more often than is expected by chance. Assortative mating has been
observed in systems including sticklebacks (e.g., Nagel and Schulter, 1998; McKinnon et al.,
2004), fruit flies (e.g., Coyne and Orr, 1989, 1997; Rundle et al., 2005), African cichlids (e.g.,
Seehausen et al., 1999; Couldridge and Alexander, 2002), birds (Slabbekoorn and Smith, 2002),
and plants (Grant, 1994; Ramsey et al., 2003). Although plants do not directly choose their mates,
the process of specialized pollination is similar to mating preference in animal systems. When
flowering synchronously, plant species with distinct flower color, shape, odor, and/or nectar
production may utilize different pollinators, thereby reducing heterospecific gene flow (e.g.,
Ramsey et al., 2003; Campbell, 2008; Peakall et al., 2010; but see Wesselingh and Arnold, 2000).
Even if different plant species are visited by the same pollinator assemblage, fidelity of
individual pollinators to a single plant species will increase the degree of reproductive isolation
between taxa (Fulton and Hodges, 1999; Husband and Sabara, 2003; Yang et al., 2007; Smith et
al., 2008).
When two individuals from different species attempt to mate with each other, mechanical
isolation may hinder heterospecific gene flow if their reproductive structures are mismatched
(e.g., Niklas and Buchmann, 1987; Sota and Kubota, 1998). In plants, individual pollinators
traveling interspecifically may not efficiently transfer pollen from one to the other. This may be
due to differences among plant species in the locations of stigmas and anthers, and as a
consequence, differences among pollinators in the sites of pollen deposition. Such mechanical

5

isolation has been shown to prevent hybridization in a number of taxa (Grant, 1994; Fulton and
Hodges, 1999; Kay, 2006; Yang et al., 2007).
Gametic isolation occurs when heterospecific gametes have a reduced fertilization rate
(e.g., Palumbi, 1998; Price et al., 2000, 2001). There are relatively few studies of gametic
isolation, perhaps because this barrier is difficult to study, especially in internally fertilizing
animals and plants, in which the gametic interactions are cryptic. In plant systems, the
differences between con- and heterospecific pollen-stigma (Heslop-Harrison, 1982; Pellegrino et
al., 2010), pollen tube-style (Levin, 1978), and/or pollen tube-ovule interactions (EscobarRestrepo et al., 2007) may cause gametic isolation and prevent heterospecific zygote formation.
The relative importance of pollinator isolation, mechanical isolation, and gametic
isolation varies across study systems. Ramsey et al. (2003) found that Mimulus cardinalis and M.
lewisii were almost completely isolated by their distinct pollinators, hummingbirds and bees,
respectively. Floral mechanical isolation was not estimated in this system, but gametic isolation
was strong. Fulton and Hodges (1999) studied reproductive isolation in Aquilegia formosa and A.
pubescens, and found evidence of strong floral isolation. Although both plant species were
visited by multiple pollinator guilds, the fidelity of individual pollinators paired with their
differential pollination effectiveness caused substantial sexual isolation (Fulton and Hodges,
1999). Kay (2006) found that the Neotropical herb species Costus scaber and C. pulverulentus
shared the same pollinator, the hummingbird Phaethornis superciliosus, and that individual
pollinators showed no constancy. Nevertheless, floral mechanical isolation was almost complete:
pollinators carried pollen of C. scaber on the distal half of their bills and that of C. pulverulentus
on their forehead. Gametic isolation was also strong between these two Costus species (Kay,

6

2006). Pellegrino et al. (2010) found weak pollinator and mechanical isolation between the
orchids Orchis italica and O. papilionacea, but gametic isolation was complete.
All of these studies were conducted with species pairs which are currently sympatric--the
importance of sexual isolation between parapatric species located within the cruising range of
their pollinators is largely unknown. In general, the distribution of a species refers to its cruising
range where the organisms can reproduce. Thus, only individuals of sympatric species have the
opportunity of encountering heterospecific mates. Unsuitable habitats beyond the cruising range
of pollinators inhibit dispersal and cause geographic isolation between allopatric species (Mayr,
1947). Sexual isolation is irrelevant to contemporary gene flow between allopatric species that
are geographically isolated at a large spatial scale. However, many species exhibit parapatric
distributions, and although largely isolated by ecogeographic barriers, such taxa may still have
an opportunity for heterospecific gene flow. Furthermore, although different plant species may
occupy distinct, nonoverlapping habitats, the cruising range of their pollinators may be greater
than that of the plants. If two plant species share the same assemblage of pollinators and the
pollinators are able to carry pollen beyond the geographic boundaries of the parents,
ecogeographic barriers may not be sufficient to eliminate all heterospecific gene flow. In such
parapatric species, studying sexual isolation becomes important for understanding whether
sexual isolation may further reduce heterospecific gene flow. This is especially of interest when
hybrids are found in nature, as this proves that heterospecific gene flow is possible.
Here I investigated the strength of sexual isolation in Costus allenii and C. villosissimus,
a pair of closely related, parapatric species that in Central Panama use the same pollinators and
grow within the pollinators’ cruising range. Flowering individuals of these two species are in
close contact (within 1 km), but are not sympatric, and there is no hybrid zone (see Ch. 2).

7

Nevertheless, hybrids are observed within the parental habitats on rare occasions (see Ch. 2),
suggesting that a small amount of gene flow may permeate the strong isolating barrier of habitat
differentiation. The pollinators of both C. allenii and C. villosissimus, Euglossine bees, can fly
long distances (Janzen, 1971; Dressler, 1982; Wikelski et al., 2010), and may be capable of
carrying pollen beyond the distribution of the parental species. Studying sexual isolation in such
a system becomes a critical step towards understanding the pattern of reproductive isolation in
parapatry. The objectives of this study are 1) to measure the strength of sexual isolation, and 2)
to determine the relative contribution of pollinator isolation, floral mechanical isolation, and
gametic isolation to sexual isolation. Sexual isolation has been shown to play an important role
in many sympatric systems. Here I show that sexual isolation can also contribute to reproductive
isolation in parapatric species.
MATERIALS AND METHODS
Study System
The pantropical family Costaceae, also known as the spiral gingers, comprises four
genera and about 100 species. Costus is the largest genus (~60 species), with its greatest
diversity centered in the Neotropics (Maas, 1972; Specht et al. 2001; Kay et al., 2005). Recent
and rapid speciation in the genus coincided with climatic fluctuations and geographic uplift in
central and northern South America (Kay et al., 2005). In general, Costus are perennial herbs
which are found in the understory of primary forests. Their large, entire leaves are spirally
arranged along the deciduous stems. Most species are clonal, with several stems sprouting from
an underground rhizome. A terminal inflorescence produces morphologically complex flowers,
and species are pollinated either by euglossine bees or hummingbirds.

8

A molecular phylogeny of Neotropical Costus (Kay et al., 2005) shows that the two study
species are very closely related, and are probably sister taxa. They occur in the same geographic
region (Maas, 1972) but occupy distinct habitats. Costus allenii is located in shady habitats along
ravines in old forests, while C. villosissimus inhabits drier, open sites along forest edges. These
habitat differences result in a parapatric distribution. Both species flower approximately from
May to August, which is the early wet season in Central America, and the fruits mature in late
September through early October. Individuals of both species typically produce one flower per
day and may flower for several months. Although their flowers are morphologically distinct,
both species are pollinated by female euglossine bees. C. allenii produces cream-colored flowers
with a small, red-striped labellum (Fig. 1.1A) and C. villosissimus has larger, yellow flowers
with a broad labellum (Fig. 1.1B). Flowering individuals of the two species have not been
observed in sympatry, but individuals on the edges of their respective distributions are
sometimes within pollinator flight distances (Chen, pers. observ.). The two species can be easily
crossed to produce fully fertile F1 and F2 hybrids, suggesting weak intrinsic postzygotic
isolation. Natural hybrids between the two species are rare in the study region, representing 2%
of the natural populations of the two species (see Ch. 2).
Study Sites
The study was conducted in central Panama, the center of distribution of Neotropical
Costus (Maas, 1972; Kay et al., 2005). In this region there is a sharp rainfall gradient across the
Isthmus of Panama, from the wet Atlantic side to the north to the much drier Pacific side to the
south. According to the Panama Canal Authority Meteorological and Hydrological Service
weather station network, the average annual precipitation is 3234 mm at Cristobal, which is close
to the Caribbean Sea, and 1798 mm at Balboa Heights, which is close to the Pacific Ocean. The

9

species composition of the plant community differs substantially across this gradient
(Engelbrecht et al., 2007), as does the distribution of C. allenii and C. villosissimus. Costus
allenii occupies wetter, northern regions, while C. villosissimus is found in southern, drier
regions (Ch. 2). The two species exhibit a parapatric distribution in the region.
The study was conducted largely along Pipeline Road (PLR), which runs south to north
through primary forest in Soberania National Park (Fig. 1.2). The southern end of PLR is wide
and the canopy is open. As the road progresses northward, the canopy closes gradually and the
road narrows until becoming nearly inaccessible. Natural populations of both species are found
in the vicinity of the road. Most plants included in the study were located within approximately
10 meters of the road, and the sampled individuals occupied microhabitats which are typical of
the overall populations of each species.
Most (> 95%) of the flowering C. villosissimus included in the study were found in the
southern stretch of PLR, from the entrance near Gamboa to approximately 5.8 km northwards. In
contrast, most flowering C. allenii were found in the northern stretch beyond the 7.8 km point of
PLR. The closest flowering heterospecific individuals were approximately 75 m apart, while
more than 80% of the flowering plants of the two species were at least 500 m apart.
A pollination array was established on PLR in a region closest to the edges of the natural
distributions between the two species (approximately 6.9 km north of the road entrance near
Gamboa). No flowering individuals of the two species were found in the immediate vicinity, and
the experimental plants were transplanted from the natural populations in the vicinity of PLR.
The location of the pollination array thus represents the geographic scale at which gene flow by
pollinators might be expected if flowering individuals of the two species did coexist in the area.

10

Species Differences in Floral Traits
To explore the potential mechanisms of pollinator isolation, floral mechanical isolation,
and gametic isolation, four floral traits were compared between natural populations of the two
species (Fig. 1.1C and 1.1D). Labellum width (LW) was measured as an indicator of flower size,
as it may be relevant to pollinator attraction. The ratio of the width of the stigma to the width of
the corolla aperture (SA) was calculated to represent the likelihood of a pollinator making
contact with the stigma while entering the flower. The shortest distance between the petaloid
stamen and the limb of the labellum (PL) was measured to determine the likelihood of antherpollinator contact as a pollinator enters the corolla tube. Style length (SL) was measured to
estimate the distance that pollen tubes must grow in order to reach the ovule. LW, SA and SL
were measured, with a few exceptions, in 16 C. allenii and 18 C. villosissimus flowers in 2007.
PL was measured in another 34 C. allenii and 65 C. villosissimus flowers in 2008. All four traits
were compared between species using two-tailed t tests.
In addition to these morphological traits, nectar production may also contribute to sexual
isolation if pollinators prefer flowers containing more nectar. Because nectar production can
have a large environmental component (Zimmerman and Pyke, 1988; Boose, 1997), and is
inducible upon removal (Pyke, 1991), an indirect but noninvasive measurement was taken on the
flowers used in the pollination and floral mechanical isolation experiment described in later
sections. Handling time of a pollinator has been shown to be positively correlated with nectar
production in nectar-rich flowers (Harder, 1982, 1983) and was used here to estimate the
standing crop of nectar available to pollinators. The duration of a bee’s visit to a flower, from the
time the bee entered the corolla tube until it left, was recorded for the first visit to each flower on
each observation day. Because the flowers were bagged before the observation period on each

11

day, nectar accumulated in the flower until it was removed by the first visitor. Therefore,
pollinator handling time for the first visit to each flower on each observation day was measured
as an indirect assessment of nectar production. For each species, the mean pollinator handling
time was calculated on each day (mean of two flowers/day), and a paired t-test was used to
compare the difference in handling time between species.
To determine if pollinators visiting the two species differed in the mechanics of pollen
transfer or deposition, video observations were made on modified flowers in the pollinator array.
An elliptical window, approximately 3 cm long and 1 cm wide, was cut from one side of the
corolla tube. The lobe of the labellum was left intact so that the image of the flower as viewed by
an approaching pollinator did not differ between manipulated and un-manipulated flowers. The
corolla tube was then wrapped with transparent plastic to prevent pollinators from exiting
through the hole in the corolla and to provide an unobstructed view for video observation. Video
recordings were made only on days when no other experiments were conducted in the arrays.
Pollinator Isolation
The pollination array consisted of 5 C. allenii and 9 C. villosissimus in 2007, and 4 of
each species in 2008. These plants were originally collected from natural populations along PLR
and they were chosen mainly due to their accessibility, i.e., to allow transportation without
damaging the flowering stems. The densities of both species are very low in natural populations-rarely are there more than a few individuals in flower in a local area. To mimic natural densities,
I used two flowers per species per day in the pollination array to estimate pollinator isolation.
This number of flowers was small enough to be monitored in an area where pollinator activities
could be closely observed but not disturbed. Since plants often do not flower every day, multiple
plants of each species were required to provide enough flowers for daily observation. Flowers

12

which were not included in the pollinator isolation experiment were removed before the daily
observation or were bagged to prevent natural pollination. The focal inflorescences were
approximately two meters apart, and the locations of the inflorescences were randomly
reassigned daily by moving the pots to prevent pollinators from memorizing the locations. The
inflorescences were bagged before the daily observations began to ensure that all pollination
events were documented.
The two Costus species have very similar pollinator assemblages and thus sexual
isolation caused by differential pollinator specialization is negligible in this system. Orchid bees
in the tribe Euglossini are the primary pollinators. Video observation in the natural populations
of the two species on Barro Colorado Island (BCI) (n = 6 flowers for each species) and in the
pollination array on PLR (n = 3 C. allenii flowers and n = 7 C. villosissimus flowers) revealed
that 97% and 89% of the pollinator visits on C. allenii and C. villosissimus, respectively, were
bees in the genus Euglossa. Most of these visits were by Euglossa imperialis. The remaining 3%
of visits to C. allenii were made by bees in the genus Eulaema, while 3% of the visits on C.
villosissimus were by Eulaema, 7% by Exaerete, and 1% by the hummingbird, Phaethornis
superciliosus (Chen unpublished data). Schemske (1981) also found that Euglossa imperialis
was the only visitor to C. allenii flowers on BCI.
Pollinator observations were conducted in June and July in 2008 and 2009, from 7 A.M.
to 11 A.M., the period with the highest pollinator visitation. Each pollination bout consisted of a
pollinator visiting one or more flowers. The pollinator was recorded and followed from its first
visit until it left the array. For each bout, I recorded the plant species visited flowers in sequence
to determine whether individual pollinators visit randomly or exhibit constancy to one or the
other species (Waser, 1986; Jones, 1997; Smith et al., 2008). Pollinator preference was

13

determined by comparing the number of bouts in which the pollinators visited C. allenii first to
those in which they visited C. villosissimus first using a G-test of goodness of fit.
To estimate flower constancy, i.e., the tendency of a pollinator to move sequentially
between different flowers of the same species, only bouts with more than two visits and only the
first two visits of those bouts were analyzed. These criteria were implemented because with just
two flowers of each species in the array, any pollinator visiting the two flowers of a single
species in sequence during a bout must switch to a different species for a third or fourth visit. In
this case, analyzing pollinator transitions from the second to the third visits and the subsequent
transitions would unrealistically reduce the estimated flower constancy. This would not be an
issue if there were many flowers of each species presented in the pollination array, but as
discussed above, flower density was kept low to mimic the natural conditions. Thus, only the
transition from the first to the second visited flower of each bout was included in the analysis of
flower constancy. G-tests of independence were performed to determine whether the plant
species of the succeeding visit was independent of the species of the preceding visit (Flanagan et
al., 2009).
A constancy index (CI) was calculated for each species as suggested by Gegear and
Laverty (2005) to represent the extent to which pollinators traveled between conspecific flowers.
The values of CI range from -1 (complete inconstancy) to 0 (random transition) to 1 (complete
constancy). Because the second-visited flowers served as the maternal parent of the potential
hybridization, a CI was calculated with pollinator transitions of which the second-visited flowers
belong to a given species. This approach allows pollinator isolation to be analyzed in the
direction of gene flow from male to female, as in the analyses of floral mechanical and gametic
isolation. In the calculation of CI, the observed numbers of transitions were compared to the

14

expected values, which were calculated by multiplying the total number of transitions by the
pollinator preference calculated above. Such comparisons eliminate the effects of pollinator
preference on constancy. For example, if pollinators prefer species A to species B with a ratio of
80:20, then before a second visit to species A, the expected probability of the first visit to A is
0.8. By chance, 80% of the pollen received by species A would be from a conspecific flower.
Flower constancy represents the case where the probability of the first flower visited being
species A is significantly greater than 0.8 when species A serves as the maternal parent of the
potential hybridization.
Floral Mechanical Isolation
Following Kay’s (2006) protocol, I marked newly dehiscent anthers on flowers in the
pollination array using radiant color dye (Magruder Color Co., NJ) as a pollen analog and
observed dye deposition on the stigmas to estimate floral mechanical isolation; N = 18 flowers
per species in 2007, and N = 44 flowers per species in 2008. This experiment was conducted on
the same flowers used for pollinator observations and on flowers produced when pollinator
observations were not conducted. On each day, one flower per species was marked as the pollen
donor and the other flower of the same species was marked as the pollen recipient. The donor
flowers of each species were randomly assigned the color pink or blue daily; the stigmas of the
recipient flowers of both species were examined in the afternoon when the flowers were about to
wither. For each species, the proportion of the stigma covered by con- and heterospecific dye
was measured visually in a three-step process: first, the stigmatic surfaces were divided into four
areas; second, each section was assigned either 0 (no dye deposition), 1 (faint coloration), or 2
(strong coloration) for each color; and finally, the numbers assigned to each area for each color
were summed separately, yielding scores ranging from 0 (no dye deposition) to 8 (entire

15

stigmatic surface covered by dye). The amount of con- and heterospecific dye deposition on
stigmas was compared using paired t-tests for each species.
Gametic Isolation
Gametic isolation was measured in 2008 using fruits from hand-pollinated flowers
produced by plants in the pollination array. Only flowers not involved in other experiments were
used for this purpose. On each day of hand pollination, a 50:50 pollen mixture was made in early
morning with pollen collected from one flower of each species. Plants used as pollen donors
were from the natural C. allenii and C. villosissimus populations on PLR, and they were chosen
for their accessibility and flower production on the day when hand pollination was carried out.
All pollen donors and recipients were bagged to prevent natural pollination. A total of 14 C.
allenii and 14 C. villosissimus flowers were hand pollinated. Fruits were collected in late
September and seeds were germinated shortly thereafter (October - November) in the greenhouse
at Michigan State University. The environmental conditions of the greenhouse were set to be
near a maximum of 26°C during the day and a minimum of 15°C during the night, with the
actual temperature being subject to change depending on the conditions outside the greenhouse.
Supplemental light was used from 6 A.M. to 6 P.M to mimic the natural photoperiod in Panama.
Seeds from each fruit were sowed in potting soil (High Porosity Professional Mix, Baccto) in a
4-L pot. Once a seed germinated and the first true leaf was fully expanded, the seedling was
transplanted to a 5 cm x 5 cm pot. All the plants were fully hydrated daily with fertilized water
(18-9-18 pH Reducer Fertilizer, 100 ppm N, PLANTEX®).
F1 hybrids of these two species cannot be distinguished from the parental species by their
morphology in seed or seedling stages, and it would require too much time and space to grow
plants until flowering. Therefore, I assessed the frequency of F1 hybrids by genotyping up to ten

16

seedlings per fruit, using amplified fragment length polymorphisms (AFLP). Fruits that produced
less than three seeds or seedlings were excluded from the analyses. For the remaining fruits,
seedlings were chosen randomly with respect to germination times. In total, 90 seedlings from 12
C. allenii fruits and 96 seedlings from 11 C. villosissimus fruits were genotyped.
Apical leaf tissue was collected into microcentrifuge tubes (FastDNA® kits, MP
Biomedicals) and DNA was extracted following the standard FastPrep® procedures. An initial
marker screen was conducted with 6 EcoR1 and 3 Mse1 primers resulting in 18 primer pairs. Six
plants per species and two F1s from each direction of reciprocal crosses were screened. The three
primer pairs which generated the most species-specific markers were used to distinguish hybrids
from the parental species: AGG + CCG, ACG + CGG, and ACT + CCG. From these three
primer pairs, a total of 12 polymorphic markers, 6 diagnostic for C. allenii and 6 for C.
villosissimus, were used to distinguish hybrids from the parental species. Seedlings with 6
species-specific markers of a given parental species were identified as pure species and those
with all 12 markers of both species were identified as hybrids. Following digestion, ligation, preselective amplification, and final amplification were performed by G. Chen, and Genescan for
AFLPs was performed by the Genomics Technology Support Facility at Michigan State
University.
The relative proportions of hybrids and pure species were compared within each fruit
using repeated G tests of goodness of fit (Husband and Schemske 2000). The expected ratios
were set to be 0.50 for both hybrids and pure species as the pollen mixtures were 50:50
( heterospecific:conspecific). Three G values were reported for each repeated G test:
GHeterogeneity, GPool, and GTotal. GHeterogeneity gives the test of heterogeneity among replicates.
GPool represents the significance of the overall differences between con- and heterospecific gene

17

flow as compared to the expected proportions. GTotal is the sum of GHeterogeneity and GPool,
and indicates whether the data set conforms to the null expectation as a whole (Sokal and Rohlf,
1981).
Gametic isolation was estimated from the frequencies of F1 hybrids produced in the
hand-pollinated fruits. If there was no gametic isolation, the frequencies of F1 hybrids in the
seedlings should be equal to the frequencies of heterospecific pollen deposited on the stigmas.
The frequency of hybrid formation observed could include the effects of both gametic isolation,
a prezygotic barrier, and early-acting, intrinsic postzygotic barriers such as seed abortion (see
review by Johnson, 2010) and differences between hybrids and parents in seed germination rate,
and/or seedling mortality (Martin and Willis, 2007). However, there is no evidence of F1 genetic
incompatibilities in crosses between these species--fruits which were hand pollinated with
heterospecific or conspecific pollen produce similar seed set, and there are no noticeable
differences in germination and mortality rates between the hybrids and the parental species
(Chen, unpublished data). Therefore, the estimates of gametic isolation in this study are due
mainly to prezygotic mechanisms.
Strength of Individual Barriers
The procedures of Sobel and Chen (in prep) were used to estimate the strengths of
reproductive isolating barriers. In brief, their approach generates isolation indices which have a
biologically and mathematically meaningful range of -1 (complete heterospecific mating or
heterosis) to 0 (no isolation) to 1 (complete isolation). Isolation indices calculated by this
approach are comparable to each other. These indices indicate the relative under-representation
of heterospecific reproduction or the relative over-representation of conspecific reproduction
relative to expectations under random mating. To consider the possibility of asymmetric isolation
18

as suggested by Kay (2006) and Martin and Willis (2007), I calculated the isolation indices for C.
allenii and C. villosissimus separately.
Pollinator isolation (RIpollinator) of each species was calculated using the first pollinator
transitions of all the bouts in which at least two flowers were visited (see above), as

⎛ H ⎞
RI = 1 − 2 * ⎜
⎟
⎝C + H ⎠

(1)

(Sobel and Chen in prep.). The probability of conspecific gene flow (C) was estimated as the
proportion of pollinator transitions between flowers of the same species, and the probability of
heterospecific gene flow (H) was estimated as the proportion of pollinator transitions between
flowers of different species. By calculating RIpollinator in this way, both pollinator preference
and constancy were taken into account simultaneously. If only a small proportion of pollinators
travel between species, isolation would occur whether it is caused by preference or by constancy.
The significance levels of the RIpollinator values were determined with G-tests of goodness of fit
on the number of con- and heterospecific pollinator transitions. The expected values used in the
G-tests were calculated by multiplying the total number of transitions by 0.5, assuming
pollinators visited all flowers randomly.
The strength of floral mechanical isolation (RIflormech) was estimated using the
proportion of con- and heterospecific dye deposition. Because the dye was carried by natural
pollinators, the deposition ratios were dependent on both pollinator preference and constancy. To
eliminate the effects of pollinator isolation on floral mechanical isolation, the isolation index of
each flower was calculated as

19

RI flormech = 1 −

2(observed H expected H )
(observed C expected C ) + (observed H expected H )

(2)

(Sobel and Chen, in prep.). Observed C and H were estimated from dye coverage on the stigmas
(see above). Expected C and H were calculated by multiplying the proportion of con- and
heterospecific pollinator transition (previous barrier) and the total amount of dye observed on
each stigma. The products give the expected amount of the respective dye which would be
deposited on the stigma by the pollinators if there was no floral mechanical isolation. RIflormech
of individual flowers was then averaged within each species. The significance levels of
RIflormech values were determined by repeated G tests of goodness of fit conducted on the
amounts of con- and heterospecific dye deposition on stigmas of the species with the expected
ratios calculated from pollinator isolation, treating each stigma as an independent replicate.
For each species, gametic isolation (RIgametic) was estimated using the proportion of
parental and hybrid seedlings in the hand-pollinated fruits. Because equal proportions of conand heterospecific pollen were deposited on the stigmas, the measurement of gametic isolation
was not affected by floral mechanical isolation. Therefore, I computed RIgametic with equation (1)
as suggested by Sobel and Chen (in prep.). The proportion of conspecific seedlings in each
individual fruit was used to represent the probability of conspecific gene flow (C), while the
probability of heterospecific gene flow (H) was the proportion of heterospecific seedlings in the
same fruit. Estimates of RIgametic from individual fruits were then averaged and the mean
RIgametic was calculated for each species.

20

Sexual Isolation

The three reproductive isolating barriers measured in this study act sequentially -pollinator isolation is followed by floral mechanical isolation, and then by gametic isolation.
Cumulative sexual isolation (SI) between C. allenii and C. villosissimus was computed as a
multiplicative function of the three individual barriers as suggested by Sobel and Chen (in prep):
SI = 1 −

2 × H pollinator × H flormech × H gametic
(3).
C pollinator × C flormech × C gametic + H pollinator × H flormech × H gametic

This approach assumes independence of the proportion of heterospecific gene flow reduced by
each barrier.
To verify the accuracy of the estimates of cumulative sexual isolation, the overall
strength of sexual isolation was measured in the field in 2008 as the proportion of hybrid seed
production in fruits from naturally-pollinated flowers. Fruits from the flowers observed in the
floral mechanical isolation experiment (n = 22 for each species) were used to estimate the overall
strength of sexual isolation. These flowers were naturally pollinated in the pollination array, and
the reduction in the proportion of hybrid seed production in these fruits thus represents the
overall strength of pollinator, mechanical, and gametic isolation. These fruits were collected and
their seeds were germinated and grown in the greenhouse at Michigan State University under the
same conditions as described above (see gametic isolation). Eighty-one seedlings from 15 C.
allenii fruits and 131 seedlings from 19 C. villosissimus fruits were genotyped following the
protocol described above to determine the proportion of hybrid seedlings in each fruit. Equation
(1) was used to calculate the strength of sexual isolation (RIsexual) in which the proportion of
hetero- and conspecific gene flow were represented by the proportion of hybrid and parental
species seedling formation in these naturally-pollinated fruits (Sobel and Chen, in prep.).

21

Therefore, sexual isolation represents the summation of barriers that occur from the point of
heterospecific pollinator transition to the production of hybrid seedlings. To assess the
significance level of the cumulative isolation indices, repeated G tests of goodness of fit were
employed to compare the differences between the observed and expected (in this case, 0.5 for all
directions of gene flow) proportion of hybrids in the naturally pollinated fruits.
To dissect the individual contribution of each isolating barrier to RIsexual, the absolute
contribution (AC) of each barrier was calculated as

ACi =

2 × ∏ H i −1

−

2 × ∏ Hi

∏ Ci −1 + ∏ H i −1 ∏ Ci + ∏ H i

(4)

where the subscripted i denotes the order of an individual barrier: 1 for pollinator isolation, 2 for
floral mechanical isolation, and 3 for gametic isolation (Sobel and Chen, in prep.). Because
pollinator isolation was the first-acting barrier measured in this study, ACpollinator was set to be
the same as RIpollinator as suggested by Ramsey et al. (2003). To calculate ACflormech and
ACgametic, the same data set used for the calculation of RIflormech was analyzed again except
that this time the expected proportion of gene flow was set at 0.5. Thus the comparisons of the
observed proportion of dye deposition represent the degree of pollinator-mediated isolation
(RIpollmed) caused by the combined effects of pollinator and floral mechanical isolation. To
obtain an overall view of the strength of pollinator-mediated barriers, data for the two years were
pooled. Equation (1) was used to calculate RIpollmed, in which C and H were, respectively, the
amount of con- and heterospecific dye coverage on the stigmas of each species. ACflormech was

22

calculated as the difference between RIpollmed and RIpollinator, and ACgametic was calculated as
the difference between RIsexual and RIpollmed.
RESULTS
Species Differences in Floral Traits

The flowers of C. allenii and C. villosissimus are morphologically different (Table 1.1).
The labellum (LW) is significantly wider in C. villosissimus flowers than in C. allenii. The ratio
of the width of the stigma to the width of the corolla aperture (SA) is smaller in C. villosissimus
than in C. allenii. The petaloid stamen-labellum distance (PL) is significantly shorter in C. allenii
flowers than in C. villosissimus. In fact, in half of the C. allenii flowers, the petaloid stamens
contacted the labellum (distance = 0). The style (SL) is significantly shorter in C. allenii than in
C. villosissimus. No significant difference was found between species in pollinator handling time
(40.77 ± 10.76 sec in C. allenii, 33.00 ± 11.38 in C. villosissimus, t = 1.15, p = 0.26), suggesting
that the two species produce similar levels of nectar.
Video recordings of pollinator visitation of both species showed that, after a bee landed
on the limb of the labellum, it walked through the space between the petaloid stamen and the
labellum to enter the corolla tube to collect the nectar. The stigma lobes were pushed open as the
bee crawled down the corolla tube and the bee’s dorsal thorax touched the receptive side of the
stigma and the anther. During this process, the bee would deposit the pollen it carried from
previous visits on the stigma and then the pollen from this flower would be deposited on its back.
In flowers of C. allenii, which have a shorter stamen-labellum distance than that of C.
villosissimus, a bee would be forced to lift the tips of the petaloid stamens, and to squeeze
through the tight space, causing its dorsal thorax to brush against the stigma and then the anther.
In flowers of C. villosissimus, the bee also contacts the anther and stigma with its dorsal thorax,

23

but for only a brief time. In both species, after the bee has finished feeding on the nectar, it exits
by crawling backward, at which time more pollen may be deposited on its back. When the bee
exits the flower, the receptive side of the stigma remained closed and self-fertilization would be
avoided. Examples of the pollinator videos on C. allenii and C. villosissimus are available at
http://dl.dropbox.com/u/12153878/20080714allenii.avi and
http://dl.dropbox.com/u/12153878/20080620villosissimus.avi, respectively.
Pollinator Isolation

Pollinators were observed at the arrays for 18 hours over 9 days in 2007 and for 37 hours
over 17 days in 2008. In 2007, a total of 240 bouts consisting of 425 visits were recorded.
Analysis of the first visit by pollinators in each bout indicates a nearly-significant preference for
C. villosissimus (55% of visits) (G = 2.82, df = 1, p = 0.09). When the second flower visited was
C. allenii (n = 48), 40% of the transitions between flowers were conspecific and 60% were
heterospecific (Fig. 1.3A). The low proportion of conspecific transitions resulted in a CI of -0.10
after the insignificant pollinator preference to C. villosissimus was taken into consideration. For
bees that visited C. villosissimus as their second flowers (n = 78), the probabilities of their
previous visits to C. allenii or to C. villosissimus were 36% and 64%, respectively (Fig. 1.3A). A
positive CI of 0.18 suggests a low degree of constancy in bees visiting C. villosissimus. The G
test of independence of whether the species of the second visited flowers depend on the species
of the first visits suggests no significant flower constancy (G = 0.17, df = 1, p = 0.68).
RIpollinator for C. allenii in 2007 was calculated as -0.208 (G = 2.10, df = 1, p = 0.15), while that
of C. villosissimus was 0.282 (G = 6.29, df = 1, p = 0.01). The negative value of RIpollinator for
C. allenii confirmed the field observations that pollinators traveled more frequently from C.
villosissimus to C. allenii than from C. allenii to C. allenii; i.e., heterospecific transitions were

24

more prevalent than conspecific transitions. In contrast, significant isolation was observed in C.
villosissimus, suggesting more heterospecific pollinator transitions to this species in the
pollination array in 2007.
In 2008, 478 bouts consisting of 803 visits were observed. A significant preference for C.
villosissimus was observed, with 69% of the first flower visits to C. villosissimus and 31% to C.
allenii (G = 67.91, df = 1, p < 0.001). Taking pollinator preference for C. villosissimus into
consideration, the proportion of heterospecific pollinator transitions was higher than that of
conspecific transitions in C. allenii (28% conspecific and 72% heterospecific, CI = -0.09, Fig.
1.3B) but lower in C. villosissimus (77% conspecific and 23% heterospecific, CI = 0.20, Fig.
1.3B). The G test of independence indicates no significant flower constancy (G = 0.58, df = 1, p
= 0.45). In 2008, RIpollinator for C. allenii was calculated as -0.449 (G = 19.45, df = 1, p <
0.001), while that of C. villosissimus was 0.532 (G = 43.92, df = 1, p < 0.001). The negative
value of RIpollinator for C. allenii was consistent with the observation in 2007, and again
confirmed the field observations of frequent heterospecific pollinator transitions from C.
villosissimus to C. allenii. The significant level of pollinator isolation in C. villosissimus in 2008
was also consistent with observations in 2007 and suggests a high frequency of conspecific
transitions in the species.
Combining the two years (Fig. 1.3C), RIpollinator for C. allenii was -0.366 (G = 19.49, df
= 1, p < 0.001), while that for C. villosissimus was 0.446 (G = 46.26, df = 1, p < 0.001). In brief,
pollinators preferred C. villosissimus but displayed low constancy, and as a result, pollinator
isolation reduced heterospecific gene flow from C. allenii to C. villosissimus, but not in the
reverse direction.

25

Floral Mechanical Isolation

Floral mechanical isolation was measured for 9 days in 2007 and 22 days in 2008. In
2007, there was no significant difference between species in dye deposition on C. allenii stigmas
(t = 0.90, df = 8, p = 0.40; mean ± 95% CI deposition index of C. allenii dye on C. allenii
stigmas = 3.00 ± 1.39, C. villosissimus dye on C. allenii = 1.89 ± 1.86; Fig. 1.4A). A similar
pattern was observed on C. villosissimus stigmas; the mean ± 95% CI deposition index of C.
villosissimus dye was 3.44 ± 1.28 and that of C. allenii was 2.22 ± 1.37 (Fig. 1.4A), with no
significant difference between species (t = 1.69, df = 8, p = 0.13). Using the proportion of conand heterospecific pollinator transitions from the pollinator isolation experiment in 2007 (C.
allenii: conspecific = 0.40, heterospecific = 0.60; C. villosissimus: conspecific = 0.64,
heterospecific = 0.36) to calculate the expected values of deposition indices, average RIflormech
of C. allenii was 0.443, and that of C. villosissimus was 0.079 in 2007 (Table 1.2; see individual
flower data in Appendix). Although no significant difference was detected when the amounts of
con- and heterospecific dye deposited on stigmas of C. allenii were compared directly with a
paired t-test, a significant RIflormech was found in this species. This is because the paired t-test
only compared the observed amount of dye deposition while RIflormech and the corresponding
repeated G test of goodness of fit also considered the expected dye deposition due to the different
frequencies of pollinator transitions. Despite significant pollinator infidelity in C. allenii, floral
structures favored conspecific pollen deposition on stigmas of C. allenii in 2007.
In 2008, the mean ± 95% CI deposition index of C. allenii dye on C. allenii stigmas was
2.27 ± 0.70, and that of C. villosissimus dye on C. allenii stigmas was 2.73 ± 0.57 (Fig. 1.4B).
There was no significant difference between the amounts of dye from the two species deposited

26

on C. allenii stigmas (t = 1.23, p = 0.23). A significant difference was observed on C.
villosissimus stigmas; more conspecific dye (2.91 ± 0.78) was deposited than heterospecific dye
(1.32 ± 0.62) (Fig. 1.4B; t = 4.23, p < 0.001). Using the proportions of pollinator transitions from
the pollinator isolation experiment in 2008 (C. allenii: conspecific = 0.28, heterospecific = 0.72;
C. villosissimus: conspecific = 0.77, heterospecific = 0.23) to calculate the expected values of
deposition indices, RIflormech of C. allenii was 0.164, and that of C. villosissimus was 0.081
(Table 1.2; see data for individual flower in Appendix). Although a slightly higher proportion of
heterospecific dye was found on the stigmas of C. allenii, floral structural differences between
the two species may have increased the proportion of conspecific dye deposition compared with
the proportion of conspecific pollinator transitions. Despite higher conspecific dye deposition,
reproductive isolation due to floral mechanical isolation per se was not significant for C.
villosissimus. This suggests that the species difference in dye deposition on stigmas of C.
villosissimus was because C. villosissimus flowers were more frequently visited by pollinators
carrying conspecific pollen rather than conspecific pollen being transported more efficiently.
When data for the two years are combined (Fig. 1.4C) and compared with the combined
proportion of con- and heterospecific pollinator transitions in the pollinator isolation experiment
(C. allenii: conspecific = 0.32, heterospecific = 0.68; C. villosissimus: conspecific = 0.72,
heterospecific = 0.28), significant floral mechanical isolation was found for C. allenii (RIflormech
= 0.220) but not for C. villosissimus (RIflormech = 0.095) (Table 1.2; see data for individual
flowers in Appendix).

27

Gametic Isolation

RIgametic was estimated from hand-pollinated fruits which received a 50:50 pollen
mixture. The average proportion (± 95% CI) of hybrids per fruit was 0.13 ± 0.17 (n = 12 fruits)
in C. allenii and 0.32 ± 0.18 (n = 11 fruits) in C. villosissimus. The overall frequency of hybrids
was significantly lower than the expected 50% in both species (Table 1.2). In addition, there was
also significant heterogeneity in hybrid frequency among fruits in C. villosissimus but not in C.
allenii (Table 1.2; see individual fruit data in Appendix). RIgametic was measured as 0.892 in C.
allenii and 0.310 in C. villosissimus.
Sexual Isolation

The cumulative sexual isolation (SI) estimated from the three individual barriers was
0.858 for C. allenii and 0.720 for C. villosissimus (Table 1.3). In contrast, the estimate of sexual
isolation obtained simply from the frequency of hybrid formation in naturally-pollinated flowers
was substantially lower than SI for C. allenii and moderately lower for C. villosissimus. The
average proportion (± 95% CI) of hybrid formation following natural pollination in the array was
0.44 ± 0.22 for C. allenii (n = 15 fruits) and 0.21 ± 0.14 for C. villosissimus (n = 19 fruits). Using
these proportions of hybrid formation as the frequencies of heterospecific gene flow, RIsexual
was 0.121 for C. allenii and 0.581 for C. villosissimus (Table 1.2; see individual fruit data in
Appendix). For C. allenii, the proportion of hybrid seedling formation was not significantly
lower than 0.5; i.e., sexual isolating barriers did not reduce heterospecific gene flow from C.
villosissimus to C. allenii. For C. villosissimus, a significant reduction in the proportion of hybrid
seedling formation was observed, indicating that sexual isolation significantly reduced
heterospecific gene flow from C. allenii to C. villosissimus. The values of RIsexual represent

28

more direct and conservative estimates for the cumulative strength of the three barriers than the
values of SI, which were computed from the estimates of indices of individual barriers. Therefore,
values of RIsexual, which indicated that sexual isolation is significant in C. villosissimus but not
in C. allenii, were used to calculate the contribution of each barrier in each species.
The contribution of each barrier to total sexual isolation (AC) is reported in Table 1.3. In
C. allenii, negative pollinator isolation (ACpollinator = -0.366) and positive floral mechanical
isolation (ACpollinator = 0.337) essentially canceled each other out to give a weak RIpollmed of 0.029. ACgametic of C. allenii was 0.150, calculated as the difference between RIsexual and
RIpollmed. In C. villosissimus, both pollinator and gametic isolation contribute to cumulative
sexual isolation (ACpollinator = 0.446, ACgametic = 0.207), although floral mechanical isolation
did not (ACflormech = -0.072). RIpollmed was 0.374 in C. villosissimus, with pollinator isolation
contributing the most.
DISCUSSION

Based on the proportion of hybrid seedling observed in naturally-pollinated fruits, sexual
isolation was stronger in C. villosissimus (RIsexual = 0.581) than in C. allenii (RIsexual = 0.121),
suggesting asymmetrical isolation between this parapatric species pair. Pollinators preferred C.
villosissimus, but visited both species and exhibited low flower constancy. The strong preference
for C. villosissimus coupled with low constancy means that pollinators regularly moved between
species, but pollinator transitions from C. villosissimus to C. allenii were more common than
those from C. allenii to C. villosissimus. Floral mechanical and gametic barriers significantly
reduced the probability of heterospecific gene flow in C. allenii, but these barriers barely

29

counteracted the effects of frequent heterospecific pollinator transitions (Table 1.3). In C.
villosissimus, pollinator isolation and gametic isolation restricted heterospecific gene flow, while
floral mechanical isolation was not significant. These three barriers together resulted in
significant sexual isolation in C. villosissimus (Table 1.3).
Mechanisms of Sexual Isolation

Pollinator Isolation. The strong preference of pollinators for C. villosissimus reduces gene flow
from C. allenii to C. villosissimus, but not in the reciprocal direction. This is primarily due to
strong pollinator preference for C. villosissimus. It is unclear why C. villosissimus was favored in
the common garden, but here I propose two alternative explanations. First, C. villosissimus may
be truly preferred over C. allenii across their geographic range. Pollinator preference is usually
determined by floral display size and/or reward size (Makino and Sakai, 2007). Pollinator
handling time was not different between species, which implies that the reward size is unlikely to
be the key trait causing pollinator preference. Costus villosissimus produces larger flowers which
may be more visible to pollinators. In addition to flower size, the bright, yellow color of C.
villosissimus flowers (Fig. 1.1B) and the red-striped, cream-colored flowers of C. allenii (Fig.
1.1A) may also display different attractiveness to pollinators. In other systems, it has been
suggested that bumble bees prefer yellow to red (Schemske and Bradshaw, 1999) and solitary
Hylaeus bees prefer white to yellow (Campbell et al., 2010). How flower color may affect
preference of euglossine bees is unclear. Costus allenii and C. villosissimus do not produce an
obvious floral scent (Chen, personal observation), thus differences in scent probably do not
contribute to pollinator preference. Although male Euglossine bees display strong preference for
different floral scents (Zimmermann et al., 2006), all pollinators observed on the two Costus
species in this study were females.

30

A second explanation for the pollinator preference for C. villosissimus is a result of the
difference in flowering phenology of the two species. Euglossine bees are known to have good
memory and display “traplining” foraging behavior (Dressler, 1982). Costus villosissimus
flowers slightly earlier than C. allenii (see Ch. 4), thus bees may be more familiar with C.
villosissimus. Once a bee finds a C. villosissimus inflorescence, it might continue to visit this
species repeatedly. However, the finding that bees exhibit low flower constancy to C.
villosissimus suggests that minor differences in flowering phenology contribute little to
pollinator preference. During the experiment, the inflorescences were randomly relocated on
each observation day, which should have further reduced the opportunity for learning.
Nevertheless, pollinators preferred C. villosissimus but showed low constancy, resulting in
substantial heterospecific gene flow.
Floral Mechanical Isolation. The degree of floral mechanical isolation was also asymmetrical
between C. allenii and C. villosissimus, but in a different direction from that of pollinator
isolation. Significantly less C. villosissimus dye was deposited on C. allenii stigmas than
expected, but this was not the case for C. allenii dye deposited on C. villosissimus stigmas.
Pollinators of C. allenii are more likely to enter the corolla tube under the petaloid stamen
because flowers of C. allenii have a smaller stigma/aperture ratio (SA). In addition, the shorter
stamen-labellum distance (PL) in C. allenii flowers further facilitates contact between the plants’
sexual structures and the pollinators when the bees lift the stamen on their way to the nectaries
(see supplemental video). Both floral traits can affect the efficiency and accuracy of pollen
transport to the stigma.
When a pollinator enters a C. villosissimus flower, it may touch the stigma and anther
briefly without transferring a substantial amount of pollen. If the same bee then visits a C. allenii

31

flower, it would deposit less pollen on the stigma than if it has previously visited C. allenii. Thus,
pollen transport in pollinator transitions from C. allenii to C. allenii may be more efficient than
those from C. villosissimus to C. allenii. When a bee moves from C. allenii to C. villosissimus, it
may carry more pollen from C. allenii due to close contact between the bee’s dorsal thorax and
the anther. In comparison with a flower of C. allenii, there is more space between the labellum
and the petaloid stamen in a flower of C. villosissimus, and therefore less contact between the
bee’s dorsal thorax and the stigma and anther. The bee would only deposit a small proportion of
C. allenii pollen on the stigma of C. villosissimus, presumably a similar amount as if it carries
pollen from a flower of C. villosissimus. The differences in floral structure may explain why
floral mechanical isolation is significant in C. allenii but not in C. villosissimus.
Gametic Isolation. Both species displayed strong gametic isolation, as estimated by the relatively
low frequency of hybrids in experimental pollinations conducted with a 50:50 ratio of pollen
from the two parental species, (RIgametic = 0.892 for C. allenii and 0.310 for C. villosissimus).
The styles of C. allenii were 2.3 mm shorter than those of C. villosissimus. Gametic isolation
caused by differences in pistil length between species has been found in many systems (Williams
and Rouse, 1990; see review in Yost and Kay, 2009). However, whether the gametic isolation
observed between C. allenii and C. villosissimus is because C. allenii have shorter styles, and
presumably shorter pollen tubes, is unknown. To dissect the mechanisms of gametic isolation
between the two species will require in vitro investigations of pollen tube competition.
One important caveat in the interpretation of the strength of gametic isolation regards the
methods used to measure this barrier. Gametic isolation is defined as the reduction in hybrid
formation that is caused by mechanisms operating after mating but before fertilization (Coyne
and Orr, 2004). However, it is often difficult to distinguish gametic isolation from early-acting,

32

intrinsic postzygotic isolation. Specifically, if hybrid zygotes have reduced survival, it is
technically difficult to determine if the cause of reduced hybrid formation is due to prezygotic or
postzygotic barriers (Klips, 1999; Chari and Wilson, 2001). As gametic isolation in this study
was measured by comparing the expected and observed hybrid production from heterospecific
crosses, there is the possibility that some of the isolation observed at this stage is due to intrinsic
postzygotic barriers. However, the seed set of pure interspecific crosses is not significantly less
than that of intraspecific crosses between C. allenii and C. villosissimus (see Ch. 4), suggesting
that gametic isolation is the major post-pollination mechanism of isolation between these species.
Reduced hybrid formation due to gametic isolation, but not intrinsic postzygotic isolation, has
also been shown in Helianthus (Rieseberg et al., 1995), Hibiscus (Klips, 1999), and Silene
(Rahme et al., 2009).
Temporal Variation in Sexual Isolation. Substantial differences were observed between years in
the magnitude of some isolating barriers. In C. villosissimus, pollinator isolation was stronger in
2008 than in 2007 (RIpollinator = 0.534 vs. 0.282), and in C. allenii, floral mechanical isolation
was weaker in 2008 than in 2007 (RIflormech =0.164 vs. 0.443). These differences may be caused
by temporal variation in any number of environmental factors which influence floral display and
thus pollinator visitation (Mothershead and Marquis, 2000; Carroll et al., 2001; Halpern et al.,
2010). Since the pollinators visiting the pollination array were not naïve, differences between
years in the availability of other floral resources may also affect pollinator preferences. It is also
worth noting that the sample size in 2008 was approximately two-fold larger than that in 2007
(478: 240 bouts in the pollinator isolation observation and 22: 9 days in the floral mechanical
isolation experiment). Thus, data from 2008 may be more indicative of the preference, constancy,
and efficiency of the pollinator assemblage.

33

Contribution of Individual Barriers to Sexual Isolation

While pollinator isolation has been shown to be an important barrier in some other
systems (Ramsey et al., 2003; Grant, 1994), the data presented here suggest that pollinatormediated isolation between plant species specializing on the same pollinators may also restrict
heterospecific gene flow. Similar results have been found between diploid and tetraploid
fireweeds (Husband and Sabara, 2003), bee-pollinated Pedicularis species (Yang et al., 2007),
and hummingbird-pollinated Costus species (Kay, 2006). In these three studies, subtle
differences in floral traits caused differences in pollinator preference, constancy, and efficiency,
respectively, and contributed to strong sexual isolation.
For C. villosissimus as the maternal parent of the potential hybridization, the cumulative
strength of sexual isolation computed from pollinator, floral mechanical, and gametic barriers (SI
= 0.720) was slightly greater than that measured from naturally pollinated fruits in the pollination
array (RIsexual = 0.581). The value of SI (SI = 0.858) is much higher than the value of RIsexual
(RIsexual = 0.121) for C. allenii as the maternal parent of the potential hybridization. The
difference between the two estimates of sexual isolation, SI and RIsexual, mainly results from the
disproportional contribution of gametic isolation to RIsexual.
Although strong gametic isolation was observed in C. allenii (RIgametic = 0.892) when
the barrier was measured with fruits produced by hand pollination with a 50:50 pollen mixture,
its contribution to sexual isolation (ACgametic = 0.150) was much smaller than one would expect
from the high RIgametic value. The strength of gametic isolation may depend on the degree of
competition for ovules, and this is likely to vary with the amount of pollen deposited on the

34

stigmas. In this study, flowers were hand pollinated with far more pollen grains than there are
ovules. Thus, if the pollen of one species is competitively superior to the other, it could fertilize
all the ovules, and the gametic isolation measured would be essentially complete. However, as
the pollen load decreases, the opportunity for competitive exclusion is reduced. For example, if a
flower contains 30 ovules and is pollinated with 15 pollen grains from each of two species, the
fruit may produce 15 hybrid and 15 pure species seeds, and no gametic isolation would be
detected. If the same flower is pollinated with 100 pollen grains from each species, all the ovules
could be fertilized by conspecifics. This phenomenon has also been observed in Mimulus in
which stronger gametic isolation was detected when more pollen was deposited on the stigma
(Sobel 2010).
Because the degree of gametic isolation estimated in experimental crosses may vary with
the size of the pollen load, some studies have precisely controlled the amount of pollen applied
(Rieseberg et al., 1995; Ramsey et al., 2003; Rahme et al., 2009). Nevertheless, these studies
were all conducted under greenhouse conditions and did not mimic the composition or amount of
pollen applied. This approach is much more difficult in field experiments, and field studies of
gametic isolation have often used excess pollen (Klips, 1999; Kay, 2006), as was the case here.
The low density of flowers of both Costus species in the present study made it impossible to test
different proportions and densities of pollen.
Sexual Isolation between Parapatric Species

Mechanisms for the evolution of reproductive isolation have been discussed by Albert
and Schluter (2004). Drawing largely on examples and theory based upon animal systems, they
proposed three mechanisms: reproductive isolation can evolve by 1) direct selection on mating
preference, 2) as a byproduct of adaptation to different niches, and 3) through reinforcement.

35

Their findings support the mechanism of direct selection on mating preference in the evolution of
reproductive isolation between sympatric stickleback populations. The proposed mechanisms can
also be applied to plant systems analogously. First, speciation could be driven by sexual isolation
when direct selection acts upon mating preference. Harder and Johnson (2009) proposed three
modes of pollination-mediated diversification, including 1) pairwise coevolution of specific
floral-pollinator interaction, 2) divergent use of the same pollen vector, and 3) pollinator shifts.
In each case, floral traits are the direct targets of selection exerted by pollinators, and adaptation
to pollinators is the major isolating mechanism which contributes to the initial divergence
between populations (Harder and Johnson, 2009; Peakall et al., 2010).
A second mechanism is that sexual isolation is not the direct target of natural selection,
but an incidental consequence of the adaptive evolution of other traits, such as habitat preference.
This mechanism may be more prevalent in allopatric species in which adaptation to different
environmental factors could have pleiotropic effects on sexual isolation. For example, Searcy
and MacNair (1990) suggested that edaphic adaptation to different copper concentrations caused
pollen-pistil incompatibility between copper-sensitive and copper-tolerant populations of
Mimulus guttatus.
In contrast, reinforcement might be responsible for the evolution of increased sexual
isolation between taxa that evolved in allopatry but have come back together in a zone of
secondary contact. In this case, sexual isolation might show geographic variation, with sympatric
populations evolving stronger sexual isolation than allopatric populations (Kay and Schemske,
2008; Yost and Kay, 2009).
For C. allenii and C. villosissimus, sexual isolation was weak in C. allenii (RIsexual =
0.121) and moderate in C. villosissimus (RIsexual = 0.581), thus this barrier alone is not sufficient

36

to prevent heterospecific gene flow. The two species are pollinated by the same pollinators, and
there is no clear evidence of divergent use of the same pollinators or pairwise coevolution
between the plants and their pollinators. Therefore, the direct-selection mechanism is unlikely to
be the case here. The currently parapatric distribution of these species probably reduces selection
for reinforcement, suggesting that the traits involved in sexual isolation between C. allenii and C.
villosissimus are more likely to be a byproduct of local adaptation to their distinct habitats.
Pollinator shifts from bees to hummingbirds have evolved independently multiple times
in Neotropical Costus (Kay et al., 2005). However, pollinator isolation alone does not explain
diversification in the genus because many of the speciation events have occurred within the
clades in which species share the same pollinators (Kay et al., 2005). In some systems, e.g., C.
bracteatus and C. lasius, a pollinator shift has occurred between sister species occupying
different habitats (Kay et al., 2005). In other systems, e.g., the broadly sympatric C.
pulverulentus and C. scaber, species share the same pollinators, yet sexual isolation contributes
significantly to speciation (Kay, 2006).
The results presented here for C. allenii and C. villosissimus are very different from those
observed in C. pulverulentus and C. scaber. In contrast to the parapatric distribution of C. allenii
and C. villosissimus, C. pulverulentus and C. scaber are regularly sympatric, but like C. allenii
and C. villosissimus, they flower at the same time and share the same pollinator. As a result,
heterospecific pollen flow between C. pulverulentus and C. scaber is common, yet hybrids are
rare, and this is due principally to strong sexual isolation that includes both gametic and
mechanical isolation (Kay, 2006). Kay and Schemske (2008) suggested that pollen-pistil
incompatibilities have evolved between these species through reinforcement. However, in
parapatric C. allenii and C. villosissimus, there is little pollinator isolation, and gametic and floral

37

mechanical isolation are not sufficient to isolate the species. Sexual isolation has evolved as a
byproduct of local adaption, and local adaptation to spatially distinct habitats is a key element of
speciation in C. allenii and C. villosissimus. That these two congeneric species pairs display such
striking differences in their isolating barriers suggests that speciation can be achieved through a
diversity of mechanisms.
For allopatric and parapatric speciation, sexual isolation may not be a crucial barrier if
ecogeographic and habitat isolation have eliminated most, if not all, heterospecific gene flow. If
two species are completely isolated by their distinct geographic ranges as a consequence of
genetically-determined adaptive traits, there would be no direct selection for sexual isolation. If
two species are partially sympatric, the contribution of sexual isolation to the total isolation may
still be greatly reduced by ecogeographic isolation. For example, M. cardinalis and M. lewisii are
almost completely isolated by their distinct pollinators, but they are also strongly isolated by
heritable differences in their ecogeographic ranges (Ramsey et al., 2003; Schemske, 2010). A
similar pattern was also observed in phytophagous ladybird beetles (Henosepilachna
vigintioctomaculata and H. pustulosa), where females of both species exhibit high degrees of
assortative mating indicating strong sexual isolation, yet habitat isolation was also substantial
(Matsubayashi and Katakura, 2009). These examples suggest that the relative contribution of
sexual isolation to the total isolation between taxa depends on the degree of ecogeographic
isolation. This dependency is due to the fact that the isolating barriers act in a specific sequence,
and only organisms living in the same geographic range have the opportunity to exchange genes
(Ramsey et al., 2003; Coyne and Orr, 2004).
Different habitats may support different pollinator assemblages (Herrera, 1988), and
different pollinator assemblages may further result in pollinator isolation (e.g., Ramsey et al.,

38

2003). In the present study of C. allenii and C. villosissimus, this is unlikely because most plants
were pollinated by a single species, Euglossa imperialis. The observation that some hybrids are
produced in nature indicates that flowering individuals of the two species may sometimes occur
within the traveling range of individual pollinators. In these rare circumstances, sexual isolation
is an important, though incomplete, barrier to heterospecific gene flow.
On PLR, flowering individuals of the two species are rarely located within 1 km of each
other (see Ch. 2). During the floral mechanical experiments, the stigmas of flowers on nearby
Costus of both species were also inspected. The closest natural plants were approximately 1 km
away from the array, and no dye was ever observed on the flowers of these plants. These
preliminary observations suggest that the habitats of C. allenii and C. villosissimus may be
sufficiently displaced to virtually eliminate pollinators traveling between species. In the
relatively rare cases when flowering individuals of the two species are located within the
traveling range of a single pollinator, sexual isolation may substantially reduce heterospecific
gene flow.
Foraging bees typically visit neighboring plants (see review in Mitchell et al., 2009) and
may groom pollen from each previous visit, thus reducing the likelihood of long distance gene
flow by pollen (Thomson, 1986). Orchid bees also groom pollen from their bodies and move
pollen away from their dorsal thorax (Kimsey, 1984), where pollen can be deposited on a
heterospecific stigma. Thus pollinators may not transfer pollen as far as their travel distance, and
plant species such as C. allenii and C. villosissimus that are adapted to different habits may be
functionally isolated by virtue of their largely non-overlapping spatial distributions.
Few other studies have investigated sexual isolation between parapatric species.
Svensson et al. (2006) studied population divergence and premating isolation in parapatric

39

damselfly (Calopteryx splendens) populations. Courtship success in crosses of males and females
from the same populations was higher than that of males and females from different populations,
indicating sexual isolation between populations. Takami et al. (2007) studied reproductive
isolation between the parapatric ground beetles Carabus yamato and C. albrechti. Reproductive
isolation is strong but incomplete and asymmetric in this system. For C. yamato as the maternal
parent, genital mismatching, a type of mechanical isolation, contributes most to the total
reproductive isolation. For C. albrechti as the maternal parent, male mate choice was the major
barrier. The results from these animal systems are similar to those obtained for C. allenii and C.
villosissimus, suggesting that sexual isolation reduces heterospecific gene flow and contributes to
total reproductive isolation between parapatric species.
Conclusion

In summary, the parapatric Neotropical herb species C. allenii and C. villosissimus are
partially and asymmetrically isolated by sexual isolating barriers. Sexual isolation is probably a
byproduct of local adaptation to distinct habitats, as opposed to the direct result of selection
against hybrid formation, i.e., reinforcement. Further research on the cruising ranges of both
plants and their pollinators, on the correlation between floral traits and pollinator behaviors, and
on the mechanisms of pollen-pistil interactions would shed light on the biology of sexual
isolation in this system. Given the fact that C. allenii and C. villosissimus are unambiguously
distinct species in nature, the modest level of sexual isolation observed suggests that other
barriers must contribute to reproductive isolation. To better understand the relative contribution
of each barrier to speciation requires measurements and comparisons of the strength of all
potential isolating barriers between C. allenii and C. villosissimus. These include isolation that
results from differing ecogeographic ranges and local habitat preferences, sexual isolation, and

40

extrinsic and intrinsic hybrid unfitness. Such a comprehensive study of perennial plants requires
long term experiments involving a variety of approaches. In addition to the results of sexual
isolation presented here, the degree and the mechanisms of habitat isolation and the comparisons
among all isolating barriers will be further investigated to provide a more comprehensive
understanding of speciation in C. allenii and C. villosissimus.

41

Table 1.1. Comparisons of floral traits between C. allenii and C. villosissimus.

mean ± 95% CI (n)
Trait

C. allenii

C. villosissimus

t-value

P

Labellum width (mm)

55.61 ± 2.84 (18)

34.55 ± 2.27 (16)

12.26

< 0.001

Stigma-aperture ratio

0.22 ± 0.01 (15)

0.19 ± 0.01 (15)

3.24

0.003

Petaloid stamen-labellum distance (mm)

1.14 ± 0.48 (34)

5.39 ± 0.68 (65)

10.27

< 0.001

Style length (mm)

55.88 ± 1.28 (16)

58.14 ± 1.60 (14)

2.36

0.025

Pollinator handling time (sec)

40.77 ± 10.76 (24)

33.00 ± 11.38 (24)

1.15

0.260

42

Table 1.2. Reproductive isolation index values and the corresponding results of repeated G tests of goodness of fit on floral
mechanical isolation, gametic isolation, and the cumulative estimate of sexual isolation. *: p < 0.05; **: p < 0.01; ***: p < 0.001.

Isolating Barrier

Year

Species

Floral mechanical isolation

2007

Sexual isolation

2008

27.87 (8) ***

8.47 (1) **

36.35 (9) ***

0.079

11.36 (8)

0.24 (1)

11.60 (9)

C. allenii

0.164

18.64 (21)

0.081

23.71 (21)

C. allenii

0.220

49.71 (30) *

0.095

36.01 (30)

C. allenii

0.892

11.74 (11)

C. villosissimus

2008

0.443

C. villosissimus
Gametic isolation

C. allenii

C. villosissimus
Comb.

GHeterogeneity

C. villosissimus
2008

RI

0.310

C. allenii
C. villosissimus

GPool

15.79 (1) ***
3.03 (1)
22.17 (1) ***
2.78 (1)

GTotal

34.43 (22) *
26.74 (22)
71.89 (31) ***
38.79 (31)

80.68 (1) ***

92.42 (12) ***

38.27 (10) ***

9.53 (1) **

47.80 (11) ***

0.121

66.42 (14) ***

0.01 (1)

66.43 (15) ***

0.581

60.38 (18) ***

43

53.90 (1) ***

114.27 (19) ***

Table 1.3. Components of reproductive isolation and absolute contributions to sexual isolation for the reproductive barriers studied.
Isolating components range from -1 (complete disassortative mating) to 0 (random mating) to 1 (complete assortative mating). For the
barriers of pollinator isolation and floral mechanical isolation, data from 2007 and 2008 were combined to calculate the isolation
indices for each species. SI represents the cumulative sexual isolation index computed from the individual components of RI, and
RIsexual represents the sexual isolation index measured from the proportion of hybrid seedling formation in the naturally-pollinated
fruits from the pollination array. G-tests of goodness of fit were used to assess the significance. Repeated G-tests of goodness of fit
were used to assess the significance of mechanical and gametic isolation and RIsexual. The numbers in bold indicate a p value of less
than 0.05 in these statistical analyses.

Components of RI

AC to sexual isolation

Isolating barriers

C. allenii

Pollinator isolation

–0.366

0.446

–0.366

0.446

Floral mechanical isolation

0.220

0.095

0.337

–0.072

Gametic isolation

0.892

0.310

0.150

0.207

C. villosissimus

0.858

C. villosissimus

Magnitude of RIsexual

Magnitude of SI
Sexual isolation

C. allenii

0.720

0.121

44

0.581

A.

B.

C.

D.
l

l
AP

LW

ps

ps
PL
S
st
an

pt
ct
SL

pt

sy

Figure 1.1. Flowers of (A) Costus allenii and (B) C. villosissimus with the pollinator Euglossa
imperialis and graphic illustrations of floral structures and measurements of the morphological
traits in a front view (C) and a longitudinal section view (D). Trait measurements: LW: labellum
width; SA = S (stigma width) / A (corolla aperture width); PL: petaloid stamen-labellum distance;
SL: style length. Floral structures: an: anther; ct: corolla tube; l: labellum; p: petal; ps: petaloid
stamen; st: stigma; sy: style. For interpretation of the references to color in this and all other
figures, the reader is referred to the electronic version of this dissertation.

45

↑ Caribbean Sea

PLR

Panama Canal

↓ Pacific Ocean
Figure 1.2. Map of the study site in Central Panama (modified from
http://www.stri.si.edu./images/Mapas/Pipeline_BarroColorado.jpg). Pipeline Road (PLR) is
marked by the red solid line on the right side of the map. The red asterisk indicates the location
of the common garden. The red and yellow shaded areas are the locations of the natural
populations of C. allenii and C. villosissimus, respectively.

46

Species of the 1st visited flower
C. allenii C .villosissimus

Species of the 2nd visited flower
C. villosissimus C. allenii

A.

28

50

26

68

34

112

Species of the 1st visited flower
C. allenii C .villosissimus

Species of the 2nd visited flower
C. villosissimus C. allenii

C.

29

Species of the 1st visited flower
C. allenii C .villosissimus

Species of the 2nd visited flower
C. villosissimus C. allenii

B.

19

45

62

97

162

Figure 1.3. Pollinator transitions between C. allenii and C. villosissimus in the pollination array
in 2007 (A), 2008 (B), and for the two years combined (C).

47

Species of the pollen recipient
C. villosissimus C. allenii

A.

Species of the pollen recipient
C. villosissimus C. allenii

B.

Species of the pollen recipient
C. villosissimus C. allenii

C.

Species of the pollen donor
C. allenii

C .villosissimus

3.00

1.88

2.22

3.44

Species of the pollen donor
C. allenii

C .villosissimus

2.27

2.72

1.32

2.91

Species of the pollen donor
C. allenii

C .villosissimus

2.48

2.48

1.58

3.06

Figure 1.4. Dye deposition scores measured on stigmas of C. allenii and C. villosissimus in the
pollination array in 2007 (A), 2008 (B), and for the two years combined (C).

48

APPENDIX

49

Table A.1. Results of repeated G tests of goodness of fit on floral mechanical isolation, gametic
isolation, and the overall strength of sexual isolation. When an observed value equals zero, ln(0)
was replaces by ln(1e-10) in the calculation of G.

Floral mechanical isolation – estimated in 2007
Observed
Expected
C. allenii Observed
flowers
conspecific
heterospecific conspecific dye
dye index
dye index
ratio
1
0
6
0.3958
2
3
2
0.3958
3
3
6
0.3958
4
2
0
0.3958
5
4
1
0.3958
6
2
1
0.3958
7
2
1
0.3958
8
6
0
0.3958
9
5
0
0.3958
Gheterogeneity = 27.8747
df = 8, p = 0.0004
C.
villosissimus
flowers
1
2
3
4
5
6
7
8
9

Observed
conspecific
dye index
0
3
4
6
4
4
4
2
4

Observed
heterospecific
dye index
0
4
4
3
2
3
0
4
0

Gheterogeneity = 11.3643
df = 8, p = 0.1819

Gpool = 8.4737
df = 1, p = 0.0036
Expected
conspecific dye
ratio
0.6410
0.6410
0.6410
0.6410
0.6410
0.6410
0.6410
0.6410
0.6410
Gpool = 0.2410
df = 1, p = 0.6242

50

Expected
heterospecific
dye ratio
0.6042
0.6042
0.6042
0.6042
0.6042
0.6042
0.6042
0.6042
0.6042

G
6.0469
0.8461
0.1502
3.7070
3.4179
0.8958
0.8958
11.121
9.2676

Gtotal = 36.3484
df = 9, p < 0.0001
Expected
heterospecific
dye ratio
0.3590
0.3590
0.3590
0.3590
0.3590
0.3590
0.3590
0.3590
0.3590

G
0.0000
1.3034
0.6632
0.0260
0.0173
0.1438
3.5575
2.3366
3.5575

Gtotal = 11.6053
df = 9, p = 0.2364

Table A.1. (cont’d)

Floral mechanical isolation – estimated in 2008
Observed
Expected
C. allenii Observed
flowers
conspecific
heterospecific conspecific dye
dye index
dye index
ratio
1
0
4
0.2766
2
2
1
0.2766
3
2
3
0.2766
4
3
3
0.2766
5
6
4
0.2766
6
3
2
0.2766
7
2
1
0.2766
8
4
4
0.2766
9
0
4
0.2766
10
1
2
0.2766
11
4
3
0.2766
12
4
4
0.2766
13
2
2
0.2766
14
3
3
0.2766
15
4
2
0.2766
16
3
2
0.2766
17
0
1
0.2766
18
2
6
0.2766
19
3
3
0.2766
20
2
3
0.2766
21
0
1
0.2766
22
1
2
0.2766
Gheterogeneity = 18.6372
df = 21, p = 0.6084

Gpool = 15.7922
df = 1, p < 0.0001

51

Expected
heterospecific
dye ratio
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234
0.7234

G
2.5903
1.9693
0.3534
1.3361
4.5524
2.2762
1.9693
1.7815
2.5903
0.0465
2.6636
1.7815
0.8908
1.3361
3.9386
2.2762
0.6476
0.0289
0.3534
0.3534
0.6476
0.0465

Gtotal = 34.4295
df = 22, p = 0.0443

Table A.1. (cont’d)

Floral mechanical isolation – estimated in 2008
Observed
Observed
Expected
C.
heterospecific conspecific dye
villosissimus conspecific
flowers
dye index
dye index
ratio
1
0
0
0.7671
2
4
4
0.7671
3
4
2
0.7671
4
2
2
0.7671
5
6
5
0.7671
6
5
2
0.7671
7
3
1
0.7671
8
4
3
0.7671
9
2
0
0.7671
10
6
1
0.7671
11
2
1
0.7671
12
6
1
0.7671
13
2
0
0.7671
14
3
1
0.7671
15
1
1
0.7671
16
3
0
0.7671
17
3
0
0.7671
18
0
3
0.7671
19
2
0
0.7671
20
3
1
0.7671
21
1
1
0.7671
22
2
0
0.7671
Gheterogeneity = 23.7114
df = 21, p = 0.3072

Gpool = 3.0308
df = 1, p = 0.0817

52

Expected
heterospecific
dye ratio
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329
0.2329

G
0.0000
2.6885
0.3117
1.3442
2.5956
0.1043
0.0065
1.3036
1.0604
0.3542
0.1558
0.3542
1.0604
0.0065
0.6721
1.5906
1.5906
8.7435
1.0604
0.0065
0.6721
1.0604

Gtotal = 26.7421
df = 22, p = 0.2212

Table A.1. (cont’d)

Floral mechanical isolation – for the two years combined
Observed
Expected
C. allenii Observed
flowers
conspecific
heterospecific conspecific dye
dye index
dye index
ratio
1
0
6
0.3169
2
3
2
0.3169
3
3
6
0.3169
4
2
0
0.3169
5
4
1
0.3169
6
2
1
0.3169
7
2
1
0.3169
8
6
0
0.3169
9
5
0
0.3169
10
0
4
0.3169
11
2
1
0.3169
12
2
3
0.3169
13
3
3
0.3169
14
6
4
0.3169
15
3
2
0.3169
16
2
1
0.3169
17
4
4
0.3169
18
0
4
0.3169
19
1
2
0.3169
20
4
3
0.3169
21
4
4
0.3169
22
2
2
0.3169
23
3
3
0.3169
2
0.3169
24
4
25
3
2
0.3169
26
0
1
0.3169
27
2
6
0.3169
28
3
3
0.3169
29
2
3
0.3169
30
0
1
0.3169
31
1
2
0.3169
Gheterogeneity = 49.71
df = 30, p = 0.0132

Gpool = 22.1739
df = 1, p < 0.0001

53

Expected
heterospecific
dye ratio
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831
0.6831

G
4.5734
1.6893
0.0111
4.5967
4.9515
1.5398
1.5398
13.790
11.492
3.0489
1.5398
0.1532
0.8639
3.3788
1.6893
1.5398
1.1519
3.0489
0.0037
1.9193
1.1519
0.5759
0.8639
3.0796
1.6893
0.7622
0.1727
0.1532
0.1532
0.7622
0.0037

Gtotal = 71.8888
df = 31, p < 0.0001

Table A.1. (cont’d)

Floral mechanical isolation – for the two years combined
Observed
Observed
Expected
C.
heterospecific conspecific dye
villosissimus conspecific
flowers
dye index
dye index
ratio
1
0
0
0.7232
2
3
4
0.7232
3
4
4
0.7232
4
6
3
0.7232
5
4
2
0.7232
6
4
3
0.7232
7
4
0
0.7232
8
2
4
0.7232
9
4
0
0.7232
10
0
0
0.7232
11
4
4
0.7232
12
4
2
0.7232
13
2
2
0.7232
14
6
5
0.7232
15
5
2
0.7232
16
3
1
0.7232
17
4
3
0.7232
18
2
0
0.7232
19
6
1
0.7232
20
2
1
0.7232
21
6
1
0.7232
22
2
0
0.7232
23
3
1
0.7232
1
0.7232
24
1
25
3
0
0.7232
26
3
0
0.7232
27
0
3
0.7232
28
2
0
0.7232
29
3
1
0.7232
30
1
1
0.7232
31
2
0
0.7232
Gheterogeneity = 36.0139
df = 30, p = 0.2076

Gpool = 2.7804
df = 1, p = 0.0954

54

Expected
heterospecific
dye ratio
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768
0.2768

G
0.0000
2.6597
1.7781
0.1384
0.0923
0.7388
2.5924
3.9341
2.5924
0.0000
1.7781
0.0923
0.8891
1.5755
0.0028
0.0146
0.7388
1.2962
0.7160
0.0461
0.0716
1.2962
0.0146
0.4445
1.9443
1.9443
7.7071
1.2962
0.0146
0.4445
1.2962

Gtotal = 38.7943
df = 31, p = 0.1585

Table A.1. (cont’d)

Gametic isolation – hand-pollinated C. allenii fruits
Fruit No. Observed
Observed
Expected
conspecific
heterospecific conspecific
progeny
progeny
progeny ratio
1
3
0
0.5
2
4
0
0.5
3
5
0
0.5
4
5
1
0.5
5
6
0
0.5
6
7
0
0.5
7
8
0
0.5
8
8
2
0.5
9
9
1
0.5
10
9
2
0.5
11
10
0
0.5
12
10
0
0.5
Gheterogeneity = 11.7399
df = 11, p = 0.3835

Gpool = 80.6791
df = 1, p < 0.0001

Gametic isolation – hand-pollinated C. villosissimus fruits
Fruit No. Observed
Observed
Expected
conspecific
heterospecific conspecific
progeny
progeny
progeny ratio
1
2
1
0.5
2
2
7
0.5
3
2
8
0.5
4
4
4
0.5
5
5
5
0.5
6
6
2
0.5
7
6
3
0.5
8
8
2
0.5
9
9
0
0.5
10
9
1
0.5
11
10
0
0.5
12
2
1
0.5
Gheterogeneity = 38.2709
df = 10, p < 0.0001

Gpool = 9.5339
df = 1, p = 0.0020

55

Expected
heterospecific
progeny ratio
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5

G
4.1589*
5.5452*
6.9315*
2.9110
8.3178*
9.7041*
11.090*
3.8549*
7.3613*
4.8182*
13.863*
13.863*

Gtotal = 92.4190
df = 12, p < 0.0001

Expected
heterospecific
progeny ratio
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5

G
0.3398
2.9419
3.8549
0.0000
0.0000
2.0930
1.0194
3.8549
12.477
7.3613
13.863
0.3398

Gtotal = 47.8048
df = 11, p < 0.0001

Table A.1. (cont’d)

Sexual isolation – naturally-pollinated C. allenii fruits
Fruit No. Observed
Observed
Expected
conspecific
heterospecific conspecific
progeny
progeny
progeny ratio
1
0
3
0.5
2
0
5
0.5
3
0
10
0.5
4
1
2
0.5
5
1
9
0.5
6
2
1
0.5
7
3
0
0.5
8
3
1
0.5
9
3
1
0.5
10
3
2
0.5
11
3
5
0.5
12
5
0
0.5
13
5
1
0.5
14
6
0
0.5
15
6
0
0.5
Gpool = 0.0123
Gheterogeneity = 66.4184
df = 14, p < 0.0001
df = 1, p = 0.9116

56

Expected
G
heterospecific
progeny ratio
0.5
4.1589
0.5
6.9315
0.5
13.863
0.5
0.3398
0.5
7.3613
0.5
0.3398
0.5
4.1589
0.5
1.0465
0.5
1.0465
0.5
0.2014
0.5
0.5053
0.5
6.9315
0.5
2.9110
0.5
8.3178
0.5
8.3178
Gtotal = 66.4308
df = 15, p < 0.0001

Table A.1. (cont’d)

Sexual isolation – naturally-pollinated C. villosissimus fruits
Fruit No. Observed
Observed
Expected
conspecific
heterospecific conspecific
progeny
progeny
progeny ratio
1
1
3
0.5
2
1
5
0.5
3
2
4
0.5
4
3
0
0.5
5
3
0
0.5
6
3
0
0.5
7
3
1
0.5
8
4
1
0.5
9
4
4
0.5
10
6
3
0.5
11
6
3
0.5
12
7
0
0.5
13
8
0
0.5
14
8
0
0.5
15
8
1
0.5
16
9
0
0.5
17
10
0
0.5
18
10
0
0.5
19
10
0
0.5
Gpool = 53.8957
Gheterogeneity = 60.3787
df = 18, p < 0.0001
df = 1, p < 0.0001

57

Expected
G
heterospecific
progeny ratio
0.5
1.0465
0.5
2.9110
0.5
0.6796
0.5
4.1589
0.5
4.1589
0.5
4.1589
0.5
1.0465
0.5
1.9274
0.5
0.0000
0.5
1.0194
0.5
1.0194
0.5
9.7041
0.5
11.090
0.5
11.090
0.5
6.1977
0.5
12.477
0.5
13.863
0.5
13.863
0.5
13.863
Gtotal = 114.2744
df = 19, p < 0.0001

LITERATURE CITED

58

Albert, A. Y. K., and D. Schluter. 2004. Reproductive character displacement of male
stickleback mate preference: reinforcement or direct selection? Evolution 58(5): 1099-1107.
Bolnick, D. I. and T. J. Near. 2005. Tempo of hybrid inviability in centrarchid fishes (Teleostei:
Centrarchidae). Evolution 59(8): 1754-1767.
Boose, D. L. 1997. Sources of variation in floral nectar production rate in Epilobium canum
(Onagraceae): implications for natural selection. Oecologia 110: 493-500.
Campbell, D. R. 2008. Pollinator shifts and the origin and loss of plant species. Annals of the
Missouri Botanical Garden 95(2): 264-274.
Campbell, D. R., M. Bischoff, J. M. Lord, and A. W. Robertson. 2010. Flower color influences
insect visitation in alpine New Zealand. Ecology 91(9): 2638-2649.
Carroll, A. B., S. T. Pallardy, and C. Galen. 2001. Drought stress, plant water status, and floral
trait expression in fireweed, Epilobium angustifolium (Onagraceae). American Journal of
Botany 88(3): 438-446.
Chari, J., and P. Wilson. 2001. Factors limiting hybridization between Penstemon spectabilis and
Penstemon centranthifolius. Canadian Journal of Botany 79: 1439-1448.
Couldridge, V. C. K., and G. J. Alexander. 2002. Color patters and species recognition in four
closely related species of Lake Malawi cichlid. Behavioral Ecology 13(1): 59-64.
Coyne, J. A., and H. A. Orr. 1989. Patterns of speciation in Drosophila. Evolution 43(2): 362-381.
Coyne, J. A., and H. A. Orr. 1997. “Patterns of speciation in Drosophila” revisited. Evolution
51(1): 295-303.
Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer, Sunderland, MA.
Dobzhansky, T. 1937. Genetics and the origin of species. Columbia Univ. Press, New York.
Dopman, E. B., P. S. Robbins, and A. Seaman. 2010. Components of reproductive isolation
between North America pheromone strains of the European corn borer. Evolution 64(4): 881902.
Dressler, R. L. 1982. Biology of the orchid bees (Euglossini). Annual Review of Ecology and
Systematics 13: 373-394.
Engelbrecht, B.M.J., L.S. Comita, R. Condit, T. A. Kursar, M. T. Tyree, B. L. Turner, and S. P.
Hubbell. 2007. Drought sensitivity shapes species distribution patterns in tropical forests.
Nature 447: 80–82.

59

Escobar-Restrepo, J-M., N. Huck, S. Kessler, V. Gagliardini, J. Gheyselinck, W-C. Yang, and U.
Grossniklaus. 2007. The FERONIA receptor-like kinase mediates male-female interactions
during pollen tybe reception. Science 317: 656-660.
Flanagan, R. J., R. J. Mitchell, D. Knutowski, and J. D. Karron. 2009. Interspecific pollinator
movements reduce pollen deposition and seed production in Mimulus ringens (Phrymaceae).
American Journal of Botany 96(4): 809-815.
Fulton, M., and S. A. Hodges. 1999. Floral isolation between Aquilegia Formosa and Aquilegia
pubescens. Philosophical Transactions of the Royal Society of London, B, Biological Science
266: 2247-2252.
Gegear, R. J., and T. M. Laverty. 2005. Flower constancy in bumblebees: a test of the trait
variability hypothesis. Animal Behaviour 69: 939-949.
Grant, V. 1994. Modes and origins of mechanical and ethological isolation in angiosperms.
Proceedings of the National Academy of Sciences, USA 91: 3–10.
Halpern, S. L., L. S. Adler, and M. Wink, 2010. Leaf herbivory and drought stress affect floral
attractive and defensive traits in Nicotiana quadrivalvis. Oecologia 163: 961-971.
Harder, L. D. 1982. Measurement and estimation of functional proboscis length in bumblebees
(Hymenoptera: Apidae). Canadian Journal of Zoology 60: 1073-1079.
Harder, L. D. 1983. Flower handling efficiency of bumble bees: morphological aspects of
probing time. Oecologia 57: 274-280.
Harder, L. D., and S. D. Johnson. 2009. Darwin’s beautiful contrivances: evolutionary and
functional evidence for floral adaptation. New Phytologist 183: 530-545.
Hatfield, T., and D. Schluter. 1999. Ecological speciation in sticklebacks: environmentdependent hybrid fitness. Evolution 53(3): 866-873.
Herrera, C. M. 1988. Variation in mutualisms: the spatio-temporal mosaic of a pollinator
assemblage. Biological Journal of the Linnean Society 35: 95-125.
Heslop-Harrison, J. 1982. Pollen-stigma interaction and cross-incompatibility in the grasses.
Science 215: 1358-1364.
Hurt, C. R., M. Farzin, and P. W. Hedrick. 2005. Examining the process of allopatric speciation
in the endangered Sonoran topminnow using cytonuclear genotype frequencies. Genetics 71:
655-672.
Husband, B. C., and H. A. Sabara. 2003. Reproductive isolation between autotetraploids and
their diploid progenitors in fireweed, Chamerion angustifolium (Onagraceae). New
Phytologist 161: 703-713.

60

Husband, B. C., and D. W. Schemske. 2000. Ecological mechanisms of reproductive isolation
between diploid and tetraploid Chamerion angustifolium. Journal of Ecology 88: 689-701.
Janzen, D. H. 1971. Euglossine bees as long-distance pollinators of tropical plants. Science
171(3967): 203-205.
Johnson, N. A. 2010. Hybrid incompatibility genes: remnants of a genomic battlefield? Trends in
Genetics 26: 317-325.
Jones, K. N. 1997. Analysis of pollinator foraging: tests for non-random behavior. Functional
Ecology 11: 255-259.
Kay, K. M. 2006. Reproductive isolation between two closely related hummingbird-pollinated
Neotropical gingers. Evolution 60(3): 538-552.
Kay, K. M., P. A. Reeves, R. G. Olmstead, and D. W. Schemske. 2005. Rapid speciation and
evolution of hummingbird pollination in Neotropical Costus subgenus Costus (Costaceae):
evidence from nrDNA ITS and ETS sequences. American Journal of Botany 92(11): 18991910.
Kay, K. M., and D. W. Schemske. 2008. Natural selection reinforces speciation in a radiation of
Neotropical rainforest plants. Evolution 62(10): 2628-2642.
Kimsey, L. S. 1984. The behavioural and structural aspects of grooming and related activities in
euglossine bees (Hymenoptera: Apidae). Journal of Zoology 204: 541-550.
Klips, R. A. 1999. Pollen competition as a reproductive isolating mechanisms between two
sympatric Hisbiscus species (Malvaceae). American Journal of Botany 86(2): 269-272.
Levin, D. A. 1978. The origin of isolating mechanisms in flowering plants. Evolutionary Biology
11: 185-317.
Lowry, D. B., J. L. Modliszewski, K. M. Wright, C. A. Wu, and J. H. Willis. 2008. The strength
and genetic basis of reproductive isolating barriers in flowering plants. Philosophical
Transactions of the Royal Society: Biological Science 363: 3009-3021.
Maas, P. J. M. 1972. Costoideae (Zingiberaceae). Flora Neotropica, monograph 8. Hafner, New
York, New York.
Makino, T. T., and S. Sakai. 2007. Experience changes pollinator responses to floral display size:
from size-based to reward-based foraging. Functional Ecology 21: 854-863.
Mallet, J. 2006. What does Drosophila genetics tell us about speciation? Trends in Ecology and
Evolution 21(7): 386-393.

61

Martin, N. H., and J. H. Willis. 2007. Ecological divergence associated with mating system
causes nearly complete reproductive isolation between sympatric Mimulus species. Evolution
61(1): 68-82.
Matsubayashi, K. W., and H. Katakura. 2009. Contribution of multiple isolating barriers to
reproductive isolation between a pair of phytophagous ladybird beetles. Evolution 63(10):
2563-2580.
Mayr, E. 1942. Systematics and the origin of species. Columbia University Press, New York.
Mayr, E. 1947. Ecological factors in speciation. Evolution 1: 263-288.
Mayr, E. 1963. Animal species and evolution. Harvard University Press, Cambridge, MA.
McKinnon, J. S., S. Mori, B. K. Blackman, L. David, D. M. Kingsley, L. Jamieson, J. Chou, and
D. Schluter. 2004. Evidence for ecology’s role in speciation. Nature 429: 249-298.
Mitchell, R. J., R. E. Irwin, R. J. Flanagan, and J. D. Karron. 2009. Ecology and evolution of
plant-pollinator interactions. Annals of Botany 103: 1355-1363.
Mothershead, K., and R. J. Marquis. 2000. Fitness impacts of herbivory through indirect effects
on plant-pollinator interactions in Oenothera macrocarpa. Ecology 81(1): 30-40.
Nagel, L. and D. Schluter. 1998. Body size, natural selection, and speciation in sticklebacks.
Evolution 52(1): 209-218.
Niklas, K. J., and S. L. Buchmann. 1987. The aerodynamics of pollen capture in two sympatric
Ephedra species. Evolution 41(1): 104-123.
Nosil, P. 2007. Divergent host plant adaptation and reproductive isolation between ecotypes of
Timema cristinae walking sticks. American Naturalist 169(2): 151-162.
Nosil, P., T. H. Vines, and D. J. Funk. 2005. Prospective: reproductive isolation caused by
natural selection against immigrants from divergent habitats. Evolution 59(4): 705-719.
Palumbi, S. R. 1998. Species formation and the evolution of gamete recognition loci. In Endless
Forms: Species and Speciation D. J. Howard and S. H. Berlocher (eds.). Oxford University
Press, New York.
Palumbi, S. R., and E. C. Metz. 1991. Strong reproductive isolation between closely related
tropical sea urchins (genus Echinometra). Molecular Biology and Evolution 8(2): 227-239.
Peakall, R., D. Ebert, J. Poldy, R. A. Barrow, W. Francke, C. C. Bower, and F. P. Schiestl. 2010.
Pollinator specificity, floral odour chemistry and the phylogeny of Australian sexually
deceptive Chiloglottis orchids: implications for pollinator-driven speciation. New Phytologist
188: 437-450.

62

Pellegrino, G., F. Bellusci, and A. Musacchio. 2010. Strong post-pollination pre-zygotic isolation
between sympatric food-deceptive Mediterranean orchids. Sexual Plant Reproduction 23(4):
281-289
Presgraves, D. C. 2002. Patterns of postzygotic isolation in Lepidoptera. Evolution 56(6): 11681183.
Price, C. S. C., C. H. Kim, C. J. Groulund, and J. A. Coyne. 2001. Cryptic reproductive isolation
in the Drosophila simulans species complex. Evolution 55(1): 81-92.
Price, C. S. C., C. H. Kim, J. Posluszny, and J. A. Coyne. 2000. Mechanisms of conspecific
sperm precedence in Drosophila. Evolution 54(6): 2028-2037.
Price, T. D., and M. M. Bouvier. 2002. The evolution of F1 postzygotic incompatibilities in birds.
Evolution 56(10): 2083-2089.
Pyke, G. H. 1991. What does it cost a plant to produce floral nectar? Nature 350: 58-59.
Rahme, J., A. Wismer, and S. Karrenberg. 2009. Pollen competition as an asymmetric
reproductive barrier between two closely related Silene species. Journal of Evolutionary
Biology 22: 1937-1943.
Ramsey, J., H. D., Bradshaw, Jr., and D. W. Schemske. 2003. Components of reproductive
isolation between the monkeyflowers Mimulus lewisii and M. cardinalis (Phrymaceae).
Evolution 57(7): 1520-1534.
Rieseberg, L. H., A. M. Desrochers, S. T. Youn. 1995. Interspecific pollen competition as a
reproductive barrier between sympatric species of Helianthus (Asteraceae). American
Journal of Botany 82(4): 515-519.
Rundle, H. D., S. F. Chenoweth, P. Doughty, and M. W. Blows. 2005. Divergent selection and
the evolution of signal traits and mating preferences. PLoS Biology 3(11): 1988-1995.
Schemske, D. W. 1981. Floral convergence and pollinator sharing in two bee-pollinated tropical
herbs. Ecology 62(4): 946-954.
Schemske, D. W. 2010. Adapation and the Origin of Species. American Naturalist 176(S1): S4S25.
Schemske, D. W., and H. D. Bradshaw. 1999. Pollinator preference and the evolution of floral
traits in monkeyflowers (Mimulus). Proceedings of the National Academy of Sciences, USA
96(21): 11910-11915.
Schluter, D. 2000. The ecology of adaptive radiation. Oxford University Press, Oxford, U.K.

63

Searcy, K. B., and M. R. MacNair. 1990. Differential seed production in Mimulus guttatus in
response to increasing concentrations of copper in the pistil by pollen from copper tolerant
and sensitive sources. Evolution 44(6): 1424-1435.
Seehausen, O., J. J. M. van Alphen, and R. Lande. 1999. Color polymorphism and sex ratio
distortion in a cichlid fish as an incipient stage in sympatric speciation by sexual selection.
Ecology Letters 1999(2): 367-378.
Slabbekoorn, H., and T. B. Smith. 2002. Bird song, ecology and speciation. Philosophical
Transactions of the Royal Society of London, B, Biological Science 357: 493-503.
Smith, S. D., S. J. Hall, P. R. Izquierdo, and D. A. Baum. 2008. Comparative pollination biology
of sympatric and allopatric Andean Iochroma (Solanaceae). Annals of the Missouri Botanical
Garden 95(4): 600-617.
Sobel, J. M. 2010. Speciation in the western North American wildflower genus Mimulus. PhD
diss. Michigan State University, East Lansing.
Sobel, J. M., G. F. Chen, L. R. Watt, and D. W. Schemske. 2010. The biology of speciation.
Evolution 64:295–315.
Sokal, R. R., and F. J. Rohlf. 1981. Biometry. 2nd ed. W. H. Freeman and Company, San
Francisco, California, USA.
Sota, T., and K. Kubota. 1998. Genital lock-and-key as a selective agent against hybridization.
Evolution 52(5): 1507-1513.
Specht, C. D., W. J. Kress, D. W. Stevenson, and R. DeSalle. 2001. A molecular phylogeny of
Costaceae (Zingiberales). Molecular Phylogenetics and Evolution 21(3): 333-345.
Svensson, E. I., F. Eroukhmanoff, and M. Friberg. 2006. Effects of natural and sexual selection
on adaptive population divergence and premating isolation in a damselfly. Evolution 60(6):
1242-1253.
Takami, Y., N. Nagata, M. Sasabe, and T. Sota. 2007. Asymmetry in reproductive isolation and
its effect on directional mitochondrial introgression in the parapatric ground beetles Carabus
yamato and C. albrechti. Population Ecology 49: 337-346.
Thomson, J. D. 1986. Pollen transport and deposition by bumble bees in Erythronium: influences
of floral nectar and bee grooming. Journal of Ecology 74(2): 329-341.
Waser, N. M. 1986. Flower constancy: definition, cause, and measurement. American Naturalist
127(5): 593-603.
Wesselingh, R. A., and M. L. Arnold. 2000. Pollinator behaviour and the evolution of Louisiana
iris hybrid zones. Journal of Evolutionary Biology 13: 171-180.

64

Whiteman, H. H., and R. D. Semlitsch. 2005. Asymmetric reproductive isolation among
polymorphic salamanders. Biological Journal of the Linnean Society 86: 265-281.
Wikelski, M., J. Moxley, A. Eaton-Mordas, M. M. Lopez-Uribe, R. Holland, D. Moskowitz, D.
W. Roubik, and R. Kays. 2010. Lange-range movements of Neotropical orchid bees observed
via radio telemetry. PLoS ONE 5(5): e10738.
Williams, E. G., and J. L. Rouse. 1990. Relationships of pollen size, pistil length and pollen tube
growth rates in Rhododendron and their influence on hybridization. Sexual Plant
Reproduction 3: 7-17.
Yang, C-F., R. W. Gituru and Y-K. Guo. 2007. Reproductive isolation of two sympatric
louseworts, Pedicularis rhinanthoides and Pedicularis longiflora (Orobanchaceae): how
does the same pollinator type avoid interspecific pollen transfer? Biological Journal of the
Linnean Society 90: 37-48.
Yost, J. M., and K. M. Kay. The evolution of postpollination reproductive isolation in Costus.
Sexual Plant Reproduction 22: 247-255.
Zimmerman, M., and G. H. Pyke. 1988. Experimental manipulations of Polemonium
foliosissimum: effects of subsequent nectar production, seed production and growth. Journal
of Ecology 76(3): 777-789.
Zimmermann, Y., D. W. Roubik, and T. Eltz. 2006. Species-specific attraction to pheromonal
analogues in orchid bees. Behavioral Ecology and Sociobiology 60(6): 833-843.

65

Chapter 2

Local Adaptation to Different Microhabitats in Neotropical Costus

66

ABSTRACT
Local adaptation to different environmental conditions may drive speciation. When the
spatial distributions of species do not overlap, the probability of hybridization is reduced. If local
adaptation is a primary cause of spatial isolation, such that species are unable to colonize
“foreign” habitats, adaptation can be considered a prezygotic isolating barrier. Furthermore, if
hybrids cannot perform as well as the parental species in either of the parental habitats, local
adaptation may also contribute to extrinsic postzygotic isolation. To understand the relationship
between local adaptation and speciation, I studied the distribution of two recently diverged
Neotropical plant species, Costus allenii and C. villosissimus, in central Panama. Here the two
species display a parapatric distribution that reflects local environmental differences, particularly
in relation to gradients in precipitation across the Isthmus of Panama. I summarized precipitation
data from local weather stations and measured soil moisture in populations of both species. The
locations of individuals of the two species were mapped to determine their spatial distributions,
and these data were used to estimate the strength of habitat isolation. Reciprocal transplant
experiments carried out with seeds and cuttings of mature plants of the two species and their
hybrids were conducted to determine whether the two species are locally adapted to their home
habitats and whether there is evidence of extrinsic postzygotic isolation.
I found that the distinct spatial distributions of these species are strongly associated with
local soil moisture in the dry season--Costus allenii is found in habitats with higher water
availability. Both species performed better in their home habitats than in foreign habitats,
indicating that their parapatric distribution is due to local adaptation. There was evidence of
extrinsic postzygotic isolation only at the seed stage-- F1 hybrid seeds had reduced fitness when
transplanted into C. villosissimus habitats. Therefore, local adaptation contributes strongly to

67

habitat isolation and weakly to extrinsic postzygotic isolation between these two recently
diverged species. Because habitat isolation is the major reproductive barrier in this system, I
conclude that local adaptation is the primary mechanism of speciation between C. allenii and C.
villosissimus.

68

INTRODUCTION
Ecology has been recognized as a major factor in speciation (Darwin, 1859; Dobzhansky,
1937; Mayr, 1947, 1963; Schluter, 2001; Coyne and Orr, 2004; Sobel et al., 2010). Local
adaptation caused by divergent selection in different environments may lead to reproductive
isolation (Schluter, 2001; Rundle and Nosil, 2005; Nosil et al., 2009; Sobel et al., 2010). When
locally adapted species occupying different habitats are unable to colonize foreign habitats, the
probability of hybridization is reduced (Mayr, 1963; Coyne and Orr, 2004; Nosil et al., 2005).
Local adaptation thus reduces heterospecific gene flow and leads to ecogeographic and/or
microhabitat isolation (reviewed in Schluter, 2001; Coyne and Orr, 2004; Rundle and Nosil,
2005; Hendry et al., 2007; Schemske, 2010; Sobel et al., 2010). If hybrids are formed, divergent
selection against the parental species in the foreign habitats may also reduce survival, growth,
and reproduction of hybrids in either of the parental habitats (reviewed in Coyne and Orr, 2004;
Rundle and Nosil, 2005). The recombination of divergently adaptive traits may result in
unfavorable phenotypes and reduced fitness when hybrids are grown in the parental habitats (e.g.,
Hatfield and Schluter, 1999; Campbell and Waser, 2007; McBride and Singer, 2011). When
hybrids experience reduced fitness in the parental habitats, local adaptation may lead to extrinsic
postzygotic isolation and further reduce heterospecific gene flow (reviewed in Coyne and Orr,
2004; Rundle and Nosil, 2005).
Divergent selection in different habitats and subsequent local adaptation in closely related
species are often reflected in parapatric or allopatric geographic distribution (Kawecki and Ebert
2004; Fine et al., 2005). When two closely related species display nonoverlapping distributions
congruent with environmental differences, it is reasonable to suspect that local adaptation
contributes to spatial isolation. However, differences in species’ geographic ranges may also be

69

influenced by historical factors (Endler, 1982; Coyne and Orr, 2004; Thorpe et al., 2008; Sobel
et al., 2010). To determine the cause of the nonoverlapping geographic distribution, reciprocal
transplant experiments can be conducted to separate ecological from historical factors (Sobel et
al., 2010). Local adaptation to home habitats and maladaptation in foreign habitats can be
demonstrated by artificially moving organisms to environments where they naturally do not
occur (reviewed in Kawecki and Ebert, 2004; Leimu and Fischer, 2008). Organisms usually have
low fitness and fail to establish a viable population when they are transplanted beyond their
current distribution range (e.g., Angert and Schemske 2005). When reduced fitness is observed
in foreign habitats compared to that in home habitats, we conclude that the current distribution is
mainly due to biological constraints, not to history (Schemske, 2000; Sobel et al., 2010).
However, reciprocal transplant experiments are more difficult to conduct with animals than with
plants because animals have greater mobility. Conducting reciprocal transplant experiments with
plants have been commonly used to test adaptation to different environments between closely
related species (e.g., Clausen et al., 1940; Campbell and Waser, 2001; Angert and Schemske,
2005). To examine whether current species distribution is due to local adaptation, I am more
interested in conducting reciprocal transplant experiments with plants.
Just as a reciprocal transplant experiment between closely related species estimates the
degree of prezygotic isolation due to habitat, such an experiment conducted with hybrids
between the parental species can be a measurement of postzygotic isolation. When hybrids have
reduced fitness independent of their environment, intrinsic postzygotic isolation occurs as a
consequence of genetic incompatibilities between the parental genomes. In contrast, when
hybrids have lower fitness than parental species in either of the parental habitats, heterospecific
gene flow is reduced due to this genotype-by-environment interaction, i.e., extrinsic postzygotic

70

isolation (Rundle and Whitlock, 2001; Coyne and Orr, 2004; Kimball et al., 2008). However,
hybrids may also have superior fitness than parental species in the corresponding parental
habitats, especially in early generations (Burke and Arnold, 2001). In this case, postzygotic
isolation does not exist and heterospecific gene flow is not reduced by natural selection against
hybrids. Despite the large number of reciprocal transplant experiments conducted in plant
systems, most have not incorporated hybrids, so estimates of extrinsic postzygotic isolation are
rare (Widmer et al., 2009).
In their review of plant speciation studies, Lowry et al. (2008a) found that most studies
provided evidence of genetically-based habitat isolation, but only a few examined extrinsic
postzygotic isolation. Schemske (2010) reviewed speciation studies of both plant and animal
systems and concluded that habitat isolation makes a significant contribution to speciation while
few studies measured the relative contribution of extrinsic postzygotic isolation. For example,
studies of the narrowly sympatric, closely related species Mimulus cardinalis and M. lewisii
show that plants are adapted to different elevation (Angert and Schemske, 2005; Angert et al.,
2008), and that habitat isolation is the major isolating mechanism between the two species
(Ramsey et al., 2003). However, because of the low likelihood of hybrid formation, hybrid
fitness in parental habitats has not been measured and extrinsic postzygotic isolation has not been
examined in this system. In another study of the role of local adaptation in speciation, Lowry et
al. (2008b) compared inland and coast races of M. guttatus by reciprocal transplanting seedlings
of parental races and F1 hybrids. They demonstrated local adaptation causing strong habitat
isolation but not extrinsic postzygotic isolation between races (Lowry et al., 2008b).
To understand the role of local adaptation in speciation requires studies involving
reciprocal transplant experiments of both parental species and hybrids in parental habitats.

71

Although emerging evidence demonstrates local adaptation in many systems, fitness of parental
species and hybrids in parental habitats was rarely compared through life history. Such
comparisons in recently diverged species are more likely to reflect the ecological and
evolutionary conditions at the time of speciation, before additional adaptive differences
accumulate after two species have been completely speciated (Schemske, 2010).
To understand whether local adaption contributes to microhabitat isolation and extrinsic
postzygotic isolation, I studied two recently diverged species, Costus allenii and C. villosissimus.
Both C. allenii and C. villosissimus are found in Central America and northern South America
but they occupy different habitats (Maas, 1972; Ch.4). The typical habitats of C. allenii are moist
forest understory along ravines, while those of C. villosissimus are more open sites along forest
edges or large gaps (Ch. 3). In this chapter, I examined two environmental factors, precipitation
and soil moisture, in central Panama, where the two species co-occur, and determined the spatial
distribution of the two species in this region. Habitat differences between the two species may
lead to isolation on a large spatial scale, i.e., ecogeographic isolation, as well as on a small
spatial scale, i.e., microhabitat isolation. In a geographic region where both species occur within
a short distance, the low resolution of ecogeographic distribution limits our ability to properly
assess spatial isolation. In such a region, measuring microhabitat isolation is more appropriate to
determine the reduction of gene flow due to habitat differences. To this end, the distribution data
in central Panama were used to calculate the strength of microhabitat isolation.
I also conducted reciprocal transplant experiments to determine whether the current
distribution is a consequence of local adaptation. Parental species and hybrids were reciprocally
transplanted to determine whether there is evidence of extrinsic postzygotic isolation. Because
Costus are perennials and it may take decades to assess the lifetime fecundity of a plant, the

72

reciprocal transplant experiments were conducted with plants of two life stages: seeds and
cuttings of mature plants. This experimental design allows us to compare fitness through life
history of these perennials in a reasonable time frame. Given the observed habitat differences
between the two species, I hypothesized that 1) precipitation is lower in C. villosissimus habitats,
2) the distribution of the two species correlates with soil moisture, 3) the two species are locally
adapted to their home habitats, and 4) the hybrids have reduced fitness in parental habitats.
MATERIALS AND METHODS
Study System
The genus Costus is a clade of perennial, tropical herbs characterized by their spirally
arranged leaves. Costus occupy diverse habitats including tree-fall gaps, forest edges, understory
streamsides, a range of edaphic conditions such as red clay, white sand, and limestone, and are
found from low to mid elevation (Maas, 1972; Schemske, 1983; Kay et al., 2005). Individual
plants usually produce one to several upright stems from their rhizome and terminal
inflorescences, which produce morphologically complex, bee- or hummingbird-pollinated
flowers. The subgenus Costus, which comprises the majority of species in the genus Costus, has
diversified rapidly in the Neotropics during the last three to five million years (Kay et al., 2005).
Costus allenii and C. villosissimus are sister taxa found in Central America and northern
South America (Kay et al., 2005). Costus allenii is found mostly in Panama, Colombia, and
Venezuela, while C. villosissimus has a wider distribution, including Mexico to the northwest,
Ecuador to the south, and Guyana to the east (Maas, 1972; Ch. 4). Although the two species are
found in the same geographic region, they occupy different habitats: C. allenii is located along
ravines in rainforest understory, while C. villosissimus inhabits forest edges. Both species flower
in the early wet season (late May to August; Ch. 4) and are pollinated by euglossine bees (Ch. 1).

73

Fully fertile F1 and F2 hybrids of the two species can be easily made by hand-pollination in the
field or greenhouse, yet hybrids are rarely found in nature.
Study Sites
This study was conducted at a number of sites in central Panama, where natural
populations of both C. allenii and C. villosissimus are found (Maas, 1972; Ch. 4). Water and
light availability differ significantly between microhabitats of the two species (Ch. 3). Across the
Isthmus of Panama, a distance of 80 km, more rain falls on the Atlantic side (mean annual
precipitation = 3234 mm at Cristobal) than on the Pacific side (mean annual precipitation = 1798
mm at Balboa Heights) (Panama Canal Authority Meteorological and Hydrological Service
weather station network). Soberania National Park (SNP), located along the Panama Canal
approximately 25 km north of Panama City, served as the primary study area for the research
described here. SNP encompasses 22,104 hectares, and includes populations of both of the study
species. The spatial distribution of the two species was determined in the vicinity of Pipeline
Road (PLR; Fig. 2.1), which runs through SNP from south to north and parallels the Panama
Canal. Along the southern end of PLR, the road is wide and the canopy is open. As the road
progresses northward, the canopy closes gradually and the road narrows until becoming
completely inaccessible. The gradient of water availability and light availability along PLR spans
the typical habitats of C. allenii and C. villosissimus. I conducted studies of spatial distribution
along 17 km of PLR. Both species were found along this section, with C. allenii predominating
in the northern stretch and C. villosissimus to the south.
For the reciprocal transplant experiments, the “home” site for C. villosissimus (VP) was
located between 1 km and 2 km from the entrance of PLR and the “home” site for C. allenii (AP)
was located between 10 km and 11 km from the entrance (Fig. 2.1). Two additional locations

74

were used in the reciprocal transplant experiments, 1) Gigante Peninsula (AG) in Barro Colorado
Nature Monument was used as representative of a C. allenii habitat, and 2) a site near the
Chagres River in Gamboa (VG), which is representative of the habitat occupied by C.
villosissimus (Fig. 2.1). In summary, two sites per species’ habitat type were included in the
reciprocal transplant experiment: C. allenii sites on PLR (AP) and on Gigante Peninsula (AG),
and C. villosissimus sites on PLR (VP) and in Gamboa (VG) (Fig. 2.1).
Spatial and Seasonal Variation in Water Availability
To compare the differences in precipitation between habitats of C. allenii and C.
villosissimus, I used rainfall data collected by the Panama Canal Authority Meteorological and
Hydrological Service weather station network
(http://striweb.si.edu/esp/physical_monitoring/data/tesp/acp_rain_mon.zip). Monthly
precipitation records from weather stations at Frijolito (9°13’N, 79°43W) and Gamboa (9°06N,
79°42W) were used to represent C. allenii and C. villosissimus habitats, respectively (Fig. 2.1).
From May 1998 to December 2007, the period when precipitation data were available from both
weather stations, monthly precipitation was compared between Frijolito and Gamboa using a
paired t-test. In central Panama, the wet season usually begins in May and ends in December,
while the dry season is from January to April. To examine seasonal differences, I used paired ttests to compare differences in monthly precipitation between the two locations for months in the
wet and dry seasons, respectively. All statistical analyses in this study were done in R, version
2.12.2 (R Development Core Team, 2011).
To examine soil moisture along the distribution of C. allenii and C. villosissimus, I
collected soil samples along PLR and measured gravimetric water content. Two soil cores were
extracted from the top 15 cm of soil from each kilometer point along the road, from the south

75

entrance of PLR (0 km) to the 11 km point, in June 2007 and in March 2008. The samples were
collected at these two time points to represent the wet and dry seasons, respectively. The soil
samples were sealed in a plastic bag and weighed with a digital balance within 12 hours.
Samples were then weighed after being oven dried at 60ºC for 7 days. The gravimetric water
content in the soil sample was calculated as: (wet weight – dry weight) / wet weight. Soil
moisture for each kilometer section was calculated by averaging gravimetric water content of
samples collected at two adjacent kilometer points. The association between gravimetric water
content and the location of the corresponding kilometer section was analyzed using Pearson
correlation for each season to determine the change in soil moisture along PLR, where natural
populations of both species are found.
Spatial Distribution
The spatial distributions of both species were determined by censusing individual plants
in the vicinity of PLR from March 2006 to July 2009. Because only flowering plants can
contribute to gene flow by pollen, I recorded the reproductive status of each plant as “flowering”
or “non-flowering”. Plants that had produced at least one inflorescence during the census period
were classified as “flowering”, and plants without inflorescences were classified as “nonflowering”. According to the species descriptions listed in Maas (1972) and from greenhouse
observation, plants with broad leaves, brownish trichomes, large extrafloral nectaries, rounded
bracts, and small, cream-colored flowers with red stripes were identified as C. allenii, while
plants with narrow leaves, white trichomes, small extrafloral nectaries, leafy bracts, and large,
yellow flowers were identified as C. villosissimus. Flowering hybrids were easily distinguished
from the parental species by their large yellow flowers with red stripes, and non-flowering plants
with narrow leaves but brownish hair or vice versa were identified as putative non-flowering

76

hybrids. The shortest distance between heterospecific flowering individuals was measured to
investigate the maximum likelihood of hybridization. Kolmogorov-Smirnov tests were used to
determine whether the spatial distribution of the two parental species (flowering and nonflowering) were significantly different. To conduct Kolmogorov-Smirnov tests, I calculated the
proportion of plants of each species and stage located within each kilometer section.
Within each kilometer section, I also determined species composition. Species
composition was measured with the proportion of flowering and non-flowering plants which are
identified as C. allenii, representing the abundance of C. allenii in relative to that of C.
villosissimus. To examine the relationship between water availability and the distribution of C.
allenii and C. villosissimus, the correlation between gravimetric water content in each season and
the relative abundance of flowering and non-flowering C. allenii was analyzed. Because
gravimetric water content was only assessed from samples collected between 0 km and 11 km,
only plants found within this section of PLR were included in the analysis.
Reciprocal Transplant Experiments
To compare fitness between the two Costus species and their hybrids in the two types of
parental habitats, reciprocal transplant experiments were conducted with plants of two life stages:
seeds and cuttings of mature plants. The experiment with seeds assesses the differences in seed
germination and early seedling survival after one year of transplanting, while the experiment
with cuttings examines the differences in plant survival, growth, and reproduction after
establishment. This approach was used to increase the likelihood that sample sizes would be
sufficient for statistical analysis across the entire life history. Because the study species are
perennials with a relatively long pre-reproductive period (> 2 years) and low seed germination

77

and seedling survival under natural conditions (Chen, personal observation), it was not feasible
to conduct the reciprocal transplant studies exclusively with seed.
Seed Transplants.
Four categories of seeds were used in the reciprocal transplant experiment at the early life
stage: the two parental species and two F1 hybrids derived from reciprocal crosses between the
parents, with either C. allenii as the female and C. villosissimus as male (F1allenii), or vice versa
(F1villo). These two parental species and two F1 hybrids are thereafter referred as the four
categories of plants. These seeds were produced by hand-pollinating flowers in the natural
populations on PLR in June 2007 and 2008. The flowers were bagged before and after being
hand pollinated to prevent natural pollination. Pollen collected from a flower was applied to the
stigma with a flat toothpick in the mornings to mimic the timing of natural pollinators. Fifty-five
seeds from each of the four categories were transplanted into the C. allenii habitat (AP) and the
C. villosissimus habitat (VP) on PLR in both October 2007 and 2008 for a total of 880 seeds. The
seeds were planted in transects at approximately 50 cm spacing and randomly arranged along the
forest edge in C. villosissimus habitat or along a forested ravine in C. allenii habitat. A plastic
disposable cup (300 ml, 7 cm in diameter and 12 cm in depth) was used as a mini-pot for each
seed. The bottom of the cup was removed in order to allow water to be transported in or out of
the cup naturally. For each transplanted seed, I dug a round hole (approximately 7 cm in
diameter and 8 cm in depth), inserted the bottomless cup, filled the cup with native soil to ground
level, and placed the seed at the center of the cup, approximately 0.5 cm below the surface.
In the seed transplant experiments, two fitness components were measured: seed
germination and seedling survival. The 2007 cohort was monitored on a monthly basis from the
date of transplanting to May 2009, the end of the second dry season experienced by the

78

transplants. For the 2008 cohort, these fitness components were assessed on a monthly basis for
approximately one year. Because the total number of seeds germinating and surviving was low
and the overall pattern remained consistent in the two cohorts, to improve statistical power of
fitness comparisons, the data of the two cohorts were averaged at a monthly basis for the first12
months since transplanting. In the 2007 cohort, the additional data which was collected after one
year of transplanting were used to demonstrate how different categories may respond to the
second dry season in their life history. To compare fitness differences among categories and
between habitats, the proportion of seeds germinating and the proportion of germinated seedlings
surviving in the one-year period was analyzed with logistic regression with the averaged dataset
of the two cohorts. The proportion of seeds germinating and also surviving in the one-year period
was calculated to represent absolute fitness of seed transplants. Absolute fitness was then
compared among categories and between habitats using logistic regression with the averaged
dataset. Within each site, the average absolute fitness of F1allenii and F1villo was compared to the
absolute fitness of the parental species in its home habitat with a G test of goodness of fit.
Cutting Transplants.
Cuttings of C. allenii, C. villosissimus, F1allenii, and F1villo were used in the reciprocal
transplant experiment conducted for estimating the post-seedling phase of the life history. Plants
used to produce the cuttings were derived from hand pollination of flowers of the two parental
species. These seed parents were established from seed collected from natural populations in
Central Panama and were grown to flowering in the greenhouse at Michigan State University.
Clonal cuttings of 16 C. allenii, 36 C. villosissimus, and 5 F1allenii, and 4 F1villo were made in
the MSU greenhouse before being transported to Panama. A healthy stem was cut into several

79

small pieces (9 to 12 cm long) and grown in incubators. Each cutting contained at least one node,
and most leaf tissue was removed from the cutting before being planted. A small section (~ 5
cm2) of healthy, green tissue on the newest leaf was retained to allow growth during cultivation.
Cuttings were placed individually in test tubes (Aquatube #53, Syndicate Sales, Inc., IN) with
distilled water. Incubators were set for 12 hour light (6 am to 6 pm), at temperatures of 30°C
daytime and 24°C nighttime. Cuttings were checked twice a week to remove damaged tissue and
to change water.
After 2 months of growth in the tubes, cuttings that had produced new shoots were sealed
with parafilm and shipped to Panama. These were then planted into 1-L plastic grow bags
containing potting soil (manufactured by Do it Center, Panama), and grown in a greenhouse in
Gamboa. To obtain enough healthy cuttings for all transplant sites, this cloning/shipping process
was repeated three times in 2006 and 2007. Importation permits were obtained from Autoridad
Nacional del Ambiente (SIM/P-3-06, No. SIM/P-3-07), and from Ministerio de Desarrollo
Agropecuario Direccion ( #560569, #560575, #561805, #580448, #580459, #580448, #607295,
#607296, #607297).
Cuttings were randomly assigned to different transplant sites and planted in a random
order along the forest edges at the C. villosissimus sites (VP and VG) and along ravines at the C.
allenii sites (AP and AG). In June 2006, I planted 18 C. allenii, 18 C. villosissimus, 20 F1allenii,
and 26 F1villo cuttings at the AG site and 13 C. allenii, 18 C. villosissimus, 10 F1allenii, and 13
F1villo cuttings at each of the C. villosissimus sites (VP and VG). In June 2007, I planted 13 C.
allenii, 18 C. villosissimus, 10 F1allenii, and 13 F1villo cuttings at the AP site. The number of

80

leaves present on each plant was counted at the time of transplanting. Across all sites and years,
a total of 224 plants were transplanted.
Three fitness components, survival, growth, and reproduction, were measured in the
cutting transplant experiment. The survival and size of each plant was monitored twice a year
until 2010, in March to represent the dry season, and in June to represent the wet season. An
index of plant growth was calculated as the number of leaves present on the stems at a given
census divided by the number of leaves at the time of transplanting. Plants that had maintained
the same number of leaves during a given period would have a growth index of 1, plants that had
increased the number of leaves would have a growth index > 1, and plants that had lost all their
leaves would have a growth index of 0. Plants without aboveground tissue were considered dead,
and growth was calculated only for living plants.
Because plant survival varied among categories and between habitats, some categories
had small sample sizes for the analyses of growth and reproduction. Given that the fitness
patterns are consistent between sites of the same parental habitats, data from the two sites
representing the same parental habitats were combined in all fitness comparisons to improve
statistical power. Plant survival from the beginning of the experiment to June 2010 was
compared among categories and between habitats using logistic regression with initial size, i.e.,
leaf number, as a covariate. To compare the fitness differences among surviving plants, plant
growth from the date of transplanting to June 2010 was also compared among categories and
between habitats using a two-way ANOVA. I compared the average survival and growth of F1
hybrids with that of the parental species in its home habitat with a G test of goodness of fit and
an ANOVA, respectively.

81

To analyze reproduction, the number of plants which produced at least one inflorescence
from 2007 to 2010 and the number of inflorescences produced by flowering plants was compared
among plant categories and between habitats using G tests of goodness of fit. Because only a
small proportion of cutting transplants flowered during the experimental period and some
currently non-flowering plants may produce flowering beyond the experimental period,
reproduction was omitted in the fitness calculation of cutting transplants. Absolute fitness of
cutting transplants was calculated as the product of the first two fitness components, i.e., the
proportion of cuttings surviving and the proportional change of plant size of the surviving plants.
Pest damage to parents and hybrids was measured in the cutting transplant experiment.
The proportion of leaf damage was estimated on new leaves in June and July, 2008 and 2010. In
June, a twist tie was loosely applied to each stem between the second and the third leaf from the
apex. At this time, the third leaf is usually newly-expanded, and can support the weight of the
twist tie. The position of the twist tie does not affect subsequent growth. Pest damage on the new
leaves was estimated approximately 30 days after the plants were marked. Each new leaf
produced in the 30-day period was assigned to a damage level of 0%, 0-25%, 25-50%, 50-75%,
and 75-100%. A damage index was calculated as the average damage level on these new leaves.
Only plants alive at the time of damage measurements were included in the analysis, and
therefore, sample sizes among sites and among species and hybrids differed. Pest damage on new
leaves was arcsine-transformed and then compared among categories and between habitats using
a two-way ANOVA and Tukey’s HSD. For plants which were alive in 2008 and 2010, the
measurements of the two years were averaged to reduce heterogeneity of damage level due to
fluctuation within individual plants. For plants which were alive in 2008 but not 2010, the onetime measurements were included in the analysis as these were the best data available.

82

Total Fitness.
The results of the seed and cutting transplants were combined into an estimate of total
fitness for each category in each habitat. Seed data from the 2007 and 2008 cohorts were
combined to generate one absolute fitness value, the proportion of seeds germinating and also
surviving after one year of transplanting, for each category in each habitat. Absolute fitness of
cuttings transplanted into sites AG and AP were combined for calculating their fitness in C.
allenii habitats, while absolute fitness of cuttings transplanted into site VG and VP were
combined for calculating their fitness in C. villosissimus habitats. As the fitness components
measured in the two transplant experiments are independent of each other, absolute total fitness
was the product of absolute fitness in seed transplants and absolute fitness in cutting transplants.
Relative total fitness was determined by dividing absolute fitness of each category in each habitat
by absolute fitness of the parental species in its home habitat.
Reproductive Isolation
Reproductive isolation due to microhabitat differentiation (RIhabitat) was calculated for
each species using the distribution of the two species along PLR. The distribution was
determined from the number of flowering plants located within each kilometer section because
only flowering plants can contribute to gene flow. Following Sobel and Chen (in prep), RIhabitat
was calculated as

Ai
Vi
RI habitat = 1 − 2∑ (
×
)
i A
total Ai + Vi
for C. allenii and

Ai
V
)
RI habitat = 1 − 2∑ ( i ×
i V
total Ai + Vi

83

for C. villosissimus, with Ai and Vi representing the number of C. allenii and C. villosissimus
individuals found in kilometer section i, respectively, while Atotal and Vtotal representing the
total number of C. allenii and C. villosissimus plants found along PLR, respectively. This
approach assumes that the population sizes are similar for the two species (Sobel and Chen, in
prep). If the isolation index equals zero, the distribution of the two species is completely random
and there is no spatial isolation. If the isolation index equals one, there is no overlap in the
distribution of the two species; thus, they are parapatric or allopatric. In such cases,
heterospecific gene flow would be completely eliminated unless pollinators travel and transfer
heterospecific pollen efficiently across the geographic barrier.
The performance of hybrids in the reciprocal transplant experiments in comparison with
the performance of the transplanted parental species in their home habitats represents the
strength of postzygotic isolation. The differences in fitness between hybrids and parental species
are affected by both genetic and environmental factors, which contribute to intrinsic and extrinsic
postzygotic isolation, respectively. When grown under greenhouse conditions, F1 hybrids
between C. allenii and C. villosissimus germinate, survive, grow, and reproduce as well as the
parental species (Chen, personal observation). In contrast, any differences in fitness between
hybrids and parental species when grown in parental habitats in the field are mostly
environment-dependent. Therefore, reproductive isolation caused by the reduced performance of
hybrids in the reciprocal transplant experiments mostly represents extrinsic postzygotic isolation.
To estimate postzygotic isolation for each species, the fitness of hybrids and parental
species were compared at two life stages. Seed germination and early seedling survival were
used for calculating an index RIseed, while cutting survival and growth was used for calculating

84

an index RIcutting. Because only a few transplants flowered during the experimental period and
the surviving plants may produce inflorescences beyond the experimental period, reproduction
was not included in the calculation of RIcutting. For both RIseed and RIcutting of each species, the
indices were calculated as
⎛ H ⎞
RI = 1 − 2 * ⎜
⎟
⎝C +H ⎠
(Sobel and Chen, in prep). The probability of conspecific gene flow (C) was estimated as the
fitness of one parental species transplanted into its home habitats, and the probability of
heterospecific gene flow (H) was estimated as the average fitness of F1allenii and F1villo
transplanted into the same parental habitats. In the calculation of RIseed, the fitness of each
category was measured as the number of seeds that germinated and survived after one year. In
the calculation of RIcutting, the fitness of each category was measured as the multiplicative
product of the proportion of surviving transplants and the growth of the surviving plants from the
time of transplanting to June 2010.
RESULTS
Spatial and Seasonal Variation in Water Availability

The precipitation data showed that more rain falls in Frijolito (average monthly
precipitation = 200.34 mm) in the north than in Gamboa in the south (183.16 mm; t = 2.60, df =
115, p = 0.01). The precipitation difference between locations was primarily due to a significant
difference in the dry season. In the dry season, from January to April, more rain fell in Frijolito
(92.59 mm) than in Gamboa (69.18 mm) (t = 2.85, df = 44, p = 0.007). In the wet season, from

85

May to December, similar amount of rainfall was recorded in Frijolito (268.62 mm) and Gamboa
(255.41 mm) (t = 2.85, df = 70, p = 0.17) (Fig. 2.2).
Soil moisture increased northward along PLR (Fig. 2.3A). A positive correlation between
gravimetric water content and road kilometer sections was found in both the wet (R = 0.62, p =
0.046) and the dry season (R = 0.66, p = 0.03).
Spatial Distribution

I located 1203 individual plants on PLR, including 254 plants which had flowered at least
once during the four year period and 949 plants which did not flower in these four years. The
flowering plants included 93 C. allenii, 156 C. villosissimus, and 5 putative hybrids, while the
non-flowering plants included 557 C. allenii, 386 C. villosissimus, and 6 putative hybrids. There
was a significant difference in the distribution of flowering C. allenii and C. villosissimus (D =
0.65, p = 0.001), and in the distribution of non-flowering plants (D = 0.65, p = 0.003) (Fig. 2.3).
Among the flowering individuals, 95.5% of C. villosissimus were located toward the south side
of PLR (between 0 km and 5.8 km), while 95.7% of C. allenii were found in the north side of the
road (between 7.8 km and 17 km). For non-flowering plants, 91.2% of C. villosissimus were
located from the south entrance to approximately 5.8 km northwards, while 90.5% of C. allenii
were found beyond the 7.8 km point of PLR. The shortest distance between two heterospecific,
flowering individuals was approximately 75 m, which occurred approximately at the 5.8 km
point of PLR. More than 80% of the flowering individuals are at least 500 m away from a
heterospecific flowering plant. Two flowering and five non-flowering putative hybrids were
located approximately at the 5.8 km point, close to the northern border of C. villosissimus. Two
flowering and one non-flowering putative hybrids were found approximately at the 13.0 km
point, and another flowering putative hybrid was at the 13.7 km point, all of which were located

86

within the range of C. allenii. RIhabitat of C. allenii was calculated as 0.886 and that of C.
villosissimus was 0.932.
In the wet season, there was no significant correlation between gravimetric water content
and species composition (= the proportion C. allenii) in each kilometer section, either for
flowering plants (R = 0.42, p = 0.20) or for non-flowering plants (R = 0.44, p = 0.18). In contrast,
in the dry season, gravimetric water content was positively correlated with species composition
for flowering plants (R = 0.82, p = 0.002) and for non-flowering plants (R = 0.79, p = 0.004).
Reciprocal Transplant Experiments

Seed Transplants.
More transplanted seeds germinated at the C. villosissimus site (VP) than at the C. allenii
site (AP) in the average dataset of the 2007 and 2008 cohorts (p < 0.001; Fig. 2.4A). The
proportion of seeds germinating was significantly different among categories (p < 0.001; Fig.
2.4A). There was no significant interaction between transplant site and plant category (p = 0.44;
Fig. 2.4A). Because the overall germination rate is low at the C. allenii site for all categories (Fig.
2.4A), the timing of germination will only be discussed at the C. villosissimus site (Fig. 2.5B).
For C. allenii, F1allenii, and F1villo, seeds germinated before early January, when the dry season
begins. In contrast, a sharp increase of seed germination in C. villosissimus occurred between
April and May, which coincides with the end of the dry season and the beginning of the wet
season.
The average seedling survival after one year of transplanting was significantly higher at
the C. allenii site than at the C. villosissimus site (p < 0.001; Fig. 2.4B). The seedling survival
across habitats was significantly different among the four categories (p < 0.001, Fig. 2.4B).
There was a significant interaction between transplant site and category (p < 0.001), indicating

87

that plant lineages with high survival in one habitat had lower survival in the other habitat (Fig.
2.4B). For both cohorts, seedlings at the C. allenii site died at a relatively constant rate (Fig.
2.5C). However, heterogeneity in timing of seedling mortality was observed at the C.
villosissimus sites. Most of the seedlings of C. allenii and F1allenii at the C. villosissimus site
died between January and April, while the seedling mortality of F1villo and C. villosissimus was
roughly linear during the one-year period (Fig. 2.5D).
Seed germination and seedling survival was monitored for another seven months after the
one-year period in the 2007 cohort. During these seven months, none of the seeds germinated.
Between October 2008 and January 2009, before the second dry season started, 9 seedlings (2 C.
allenii, 4 F1allenii, 3 F1villo) in C. allenii habitat and 7 seedlings (2 F1villo and 5 C. villosissimus)
in C. villosissimus habitat were dead. During the second dry season (between January and May
2009), 4 seedlings (2 C. allenii, 1 F1allenii, and 1 F1villo) in C. allenii habitat and 1 C.
villosissimus seedling in C. villosissimus habitat were dead. Because the least favorable
categories in each habitat had no survived seedlings at the end of one year, the additional seven
months of observation in seed fitness could not be lower than zero in these categories. Although
the overall pattern of seedling mortality was similar from the end of one year to the end of the
second dry season, the increase of seed mortality in other categories reduced the statistical power
to determine differences among categories and between habitats. Therefore, the measurements of
fitness components were terminated at the end of one year in the 2008 cohort.
The absolute fitness of seed transplants, i.e., the number of transplanted seeds which
germinated and survived for one year, was not different between transplant sites (p = 0.55), nor
among categories (p = 0.21), but there was a significant interaction between transplanting site

88

and category (p < 0.001; Fig. 2.4C). At the end of one year, none of the germinated C. allenii
seeds survived in C. villosissimus habitats, and vice versa. In contrast, few germinated C. allenii
and C. villosissimus seeds survived in their home habitats. Therefore, both species had higher
fitness in home habitats (Fig. 2.4C). F1 hybrid seeds had higher fitness in the habitats of their
maternal parents than in the habitats of their paternal parents (Fig. 2.4C). The average fitness of
F1allenii and F1villo (0.05) was lower than the fitness of C. villosissimus (0.14) in C. villosissimus
habitat (G = 4.58, p = 0.03), but averaged hybrid fitness (0.07) was similar to the fitness of C.
allenii (0.06) in C. allenii habitat (G = 0.07, p = 0.80). RIseed of C. allenii was calculated as
0.067 and that of C. villosissimus was 0.463. The results of the G tests combined with RIseed
values indicate that extrinsic postzygotic isolation was significant in C. villosissimus habitat but
not in C. allenii.
Cutting Transplants.
The overall transplant survival until June 2010 at the C. allenii sites (31.5%) was similar
to that at the C. villosissimus sites (34.3%; p = 0.44; Fig. 2.6A). There was a significant
difference in proportion surviving among categories (C. allenii: 19.3%, C. villosissimus: 18.3%,
F1allenii: 48.0%, and F1villo: 50.8%; p < 0.001; Fig. 2.6A). The survival of both C. allenii and C.
villosissimus was highest in their home habitats, and each species had higher survival than the
other species in their home habitats, as indicated by the significant interaction between transplant
site and category (p < 0.001; Fig. 2.6A). None of the C. allenii transplants survived in C.
villosissimus sites, none of the C. villosissimus transplants survived at one of the C. allenii sites
(AP), and only 10.7% of C. villosissimus survived at the other C. allenii site (AG) (Fig. 2.7). The
survival of F1 hybrids was higher than that of parental species at the two C. villosissimus sites

89

and at one C. allenii site (AG), but lower than C. allenii at the other C. allenii site (AP) (Fig. 2.7).
The mortality rates of C. villosissimus and the F1 hybrids were relatively consistent throughout
the whole experiment and across transplant sites. However, heterogeneity in cutting mortality
was observed in C. allenii transplants at the C. villosissimus sites. These plants had a sharp
decrease in survival between March and June 2007, which was the end of the dry season and
beginning of the following wet season (Fig. 2.7C and 2.7D). In C. allenii habitats, the proportion
of F1 hybrids surviving was similar to that of the C. allenii transplants (G = 0.61, p = 0.43). In C.
villosissimus habitats, the proportion of F1 hybrids surviving was also similar to that of the C.
villosissimus transplants (G = 2.03, p = 0.15).
Of the 83 plants which survived through the experiment (until June 2010), 30.1% had
fewer leaves than at the time of transplanting, 6.0% had the same number of leaves, and 63.9%
had more leaves. Plant growth, measured as the proportional increase in number of leaves, was
significantly lower in C. allenii habitats than in C. villosissimus habitats (p < 0.001, Fig. 2.6B).
There was no significant difference in growth between the two species or between the two F1
hybrids (p = 0.12; Fig. 2.6B). In C. allenii habitats, the growth of F1 hybrids was similar to that
of C. allenii (p = 0.81; Fig. 2.8A and 2.8B). In contrast, in C. villosissimus habitats, the growth
of F1 hybrids was significantly greater than that of C. villosissimus (p = 0.03; Fig. 2.8C and
2.8D).
Combining the two fitness components, survival and growth, the absolute fitness of C.
allenii was zero in C. villosissimus habitats and that of C. villosissimus was nearly zero in C.
allenii habitats (Fig. 2.6C). The hybrids did not have reduced fitness in either of the parental
habitats (Fig. 2.6C). RIcutting of C. allenii was calculated as -0.163 and that of C. villosissimus

90

was -0.689. The negative values of RIcutting indicate that hybrids have higher fitness than the
parental species in the corresponding parental habitats.
None of the plants grown at C. allenii sites reproduced, while plants in each category
reproduced in C. villosissimus sites (6 C. villosissimus, 3 F1allenii, and 2 F1villo at VG; 1 C.
allenii, 4 C. villosissimus, 6 F1allenii, and 8 F1villo at the VP). There was a significant difference
in the number of flowering plants between parental habitats (p < 0.001). There was also a
difference in number of flowering plants among categories (p = 0.002) as there was only one C.
allenii but more plants of the other three categories flowered. For flowering plants, there was a
significant pattern for increasing number of inflorescences produced from C. allenii (1
inflorescence produced by 1 plant), F1allenii (4.33 ± 2.58), to F1villo (7.20 ± 4.06) and C.
villosissimus (9.0 ± 5.37).
The proportion of leaf damage on new leaves (Fig. 2.9) was slightly, but significantly
higher in C. villosissimus sites (0.08 ± 0.04) than in C. allenii sites (0.06 ± 0.03, p = 0.046).
There was also a significant difference among categories (p = 0.04), with more damage on C.
villosissimus (0.10 ± 0.05) than on C. allenii (0.01 ± 0.01, p = 0.04). There was no interaction
between transplant site and category for leaf damage.
Total Fitness.
Combining the results of the two reciprocal transplant experiments, relative fitness of C.
allenii was zero in C. villosissimus habitats, and vice versa (Fig. 2.10). Neither F1allenii and
F1villo in C. allenii habitats had reduced fitness. F1allenii had zero fitness in C. villosissimus
habitats, while F1villo had higher fitness than C. villosissimus in the home habitat (Fig. 2.10).

91

Discussion
Water Availability and Species’ Distribution

The precipitation data revealed that more rain falls at Frijolito in the north than at
Gamboa in the south. This between-location difference is parallel to the north-south pattern of
more rainfall on the Atlantic side than on the Pacific side across the Isthmus of Panama. The
same pattern is also found along PLR, with an increase in soil moisture as the road progresses
northward. All these observations indicate higher water availability in the northern region.
Because Frijolito and Gamboa represent the preferred habitats of C. allenii and C. villosissimus,
respectively, the between-location difference in precipitation at these localities indicates that C.
allenii habitats receive more rainfall than those of C. villosissimus. Interestingly, the
precipitation difference was found only during the dry season. Moreover, the correlation between
species composition and soil moisture was significant only in the dry season. These findings
indicate that the spatial distributions of C. allenii and C. villosissimus are affected mainly by
water availability in the dry season. Given the finding of higher drought tolerance in C.
villosissimus (Ch. 3), it is suggested that C. villosissimus is adapted to drier habitats, and can thus
survive in southern regions of the Isthmus of Panama that are beyond the range of C. allenii. Due
to the marked differences in water availability, natural populations of C. allenii and C.
villosissimus on PLR display a parapatric distribution, which correlates with the soil moisture
gradient in the area.
Water availability has been shown to be a limiting factor of plant species’ distribution in
many systems (e.g., Whittaker, 1965; Gentry 1988; Duivenvoorden, 1995; Bongers et al., 1999;
Pyke et al., 2001; Engelbretch et al., 2007; Giriraj et al., 2008). Across the Isthmus of Panama,
correlations between drought sensitivity and tree species distribution suggested that soil water

92

availability is a direct determinant of species distribution (Engelbretch et al., 2007). While such
correlations have been found among distantly-related tree species (Engelbretch et al., 2007), the
results presented here demonstrate a correlation between soil moisture and the distribution of
closely related species, and suggest that the underlying cause is divergent adaptation to different
soil moisture. A greenhouse experiment of drought tolerance has shown that C. villosissimus has
higher drought tolerance than C. allenii. This provides an explanation of why C. villosissimus is
found in drier habitats (Ch. 3). A potential mechanism to explain the parapatric distribution of
the two species is that limited water availability in the dry season prevents C. allenii from
invading the habitat of C. villosissimus. This potential mechanism is supported by the results of
high mortality of C. allenii transplanted in C. villosissimus habitats.
Local Adaptation

The reciprocal transplant experiment initiated with seeds showed that the parental species
performed best in their home habitats. Although C. allenii had a high germination rate (58.2%) in
C. villosissimus habitat, none of the seedlings survived (Fig. 2.4). The germination rate of C.
villosissimus in C. villosissimus habitats (30.9%) was lower than that of C. allenii in C.
villosissimus habitat (58.2%), but a few seedlings (44.5% of germinants) survived after one year
of transplanting (Fig. 2.4). The difference in seedling survival between the two parental species
in C. villosissimus habitats is probably a result of the difference between species in germination
timing. Most C. allenii seeds germinated in the late wet season (October to December), while
most of C. villosissimus germinated in the interval between the end of the dry season and the
beginning of the wet season (April to May; Fig. 2.5B). Most C. allenii seedlings died in the early
dry season (January to March), while mortality of C. villosissimus seedlings in the dry season is
not different from that in the wet season (Fig. 2.5D). Therefore, I speculate that the delayed

93

timing of C. villosissimus seed germination is associated with seasonal drought in C.
villosissimus habitats.
Drought-adapted plants usually cope with drought by escape, by avoiding the negative
consequences of dehydration, or by tolerating low tissue water potential (reviewed in Chaves et
al., 2003). A short life cycle or short growing season enables plants to reproduce before
drought—such species are usually categorized as using drought-escape strategies (e.g., McKay et
al., 2003; Wu et al., 2010; Franks, 2011). Delayed seed germination is another mechanism of
escape from drought (Venable and Lowlor; 1980; Volis et al., 2009). Given the differences in the
timing of seed germination and seedling mortality between species, I suggest that seeds of C.
villosissimus germinate later than those of C. allenii to escape the severe drought condition that
is typical of C. villosissimus habitats during the dry season.
Observations of seeds grown under uniform conditions with high water availability in the
greenhouse and in incubators also indicate that seeds of C. allenii germinate earlier than those of
C. villosissimus (Chen, unpublished data). The consistent difference in the timing of germination
between species suggests that germination timing is a heritable trait, and that intrinsic factors, as
opposed to extrinsic environmental cues, may play an important role in germination. Further
investigation into the factors responsible for seed germination may help understand how C.
villosissimus uses seed dormancy to adaptively escape from seasonal drought in its habitat.
In the reciprocal transplant experiment initiated with cuttings, the parental species also
perform best in their home habitats. None of the C. allenii transplants survived in C.
villosissimus habitats, while < 10 % of the C. villosissimus transplants survived in C. allenii
habitats (Fig. 2.6A). The surviving C. villosissimus transplants in C. allenii habitats had fewer
leaves (Fig. 2.6B). Taken together, both parental species had lower fitness when transplanting

94

into foreign habitats (Fig. 2.6C). Interestingly, most mortality of C. allenii in C. villosissimus
habitats occurred in the dry season, while most C. villosissimus mortality in C. allenii habitats
occurred in the wet season (Fig. 2.7). These results suggest a trade-off between being able to
survive in a dry habitat and being able to survive and grow in shade. This is consistent with the
hypothesis proposed by Smith and Huston (1989) that plants which are able to survive in dry
habitats are likely to grow more slowly in shady habitats (also see empirical evidence in BrenesArguedas et al., 2011). Coley et al. (1985) hypothesized that when growing slower in shady
habitats, plants with lower pest resistance may fail to produce new leaves fast enough to
compensate the loss due to pest and thus have lower survival. The observations of higher pest
damage, reduced growth and high mortality of C. villosissimus transplants in C. allenii habitats
support both hypotheses. These observations also suggest that local adaption in C. allenii and C.
villosissimus is a result of natural selection caused by interactions among pest, water, and light
availability in their habitats.
Despite the fact that reproduction was low overall, with no plants flowering in C. allenii
habitat, the proportion flowering of C. villosissimus was higher than that of C. allenii in C.
villosissimus habitats. In fact, the only flowering C. allenii produced one inflorescence in 2009
and was dead in 2010, while flowering C. villosissimus produced an average of nine
inflorescences during the experimental period and more than half of these C. villosissimus were
alive at the end of the experiment. Although reproduction was not included in the calculation of
cutting fitness, the results are also consistent with the prediction of local adaptation in C.
villosissimus.
Both seed and cutting reciprocal transplants indicate local adaptation to different habitats
in C. allenii and C. villosissimus, suggesting that the parapatric distribution of the two species is

95

largely due to local adaptation. As further evidenced, greenhouse measurements revealed that C.
allenii has higher leaf mass per area, a physiological feature typical of shade-adapted species,
and that C. villosissimus has higher drought tolerance (Ch. 3). These results are consistent with
the hypothesis that the preferred habitats of these species, wet, shaded understory habitats for C.
allenii and drier, open gaps and forest edges for C. villosissimus, are a product of local
adaptation.
Local Adaptation and Reproductive Isolation

Reciprocal transplant experiments of the parental species demonstrated that C. allenii and
C. villosissimus are locally adapted to their different home habitats. The parapatric distribution of
the two species leads to prezygotic, habitat isolation. Habitat isolation was calculated by the
distribution of the two species along PLR, and the index values were high in both species
(RIhabitat = 0.886 for C. allenii and RIhabitat = 0.932 for C. villosissimus). Because habitat
isolation acts early in the life history, its relative contribution to total isolation is greater than
later acting, sexual , postmating, and postzygotic isolating barriers (Schemske, 2010). By
calculating the relative contribution of the habitat isolating barrier to total isolation with
comparisons among all barriers (as suggested by Coyne and Orr, 1989, 1997; Ramsey et al.,
2003; see reviewed in Schemske, 2010), habitat isolation is a major contributor to total isolation
between C. allenii and C. villosissimus (Ch. 4). As the two species diverged quite recently (Kay
et al., 2005), local adaptation to different habitats may be the primary isolating mechanism at the
time of speciation between C. allenii and C. villosissimus (Ch. 4).
The reciprocal transplant experiment initiated with seeds showed that the F1 hybrids
performed as well as C. allenii in C. allenii habitats but worse than C. villosissimus in C.
villosissimus habitats. Reproductive isolation measured in these transplants was low in C. allenii

96

(RIseed = 0.067) and moderate in C. villosissimus (RIseed = 0.463). The proportion of hybrid
seeds that germinated and survived depends on transplant sites, suggesting extrinsic rather than
intrinsic postzygotic isolation. Moreover, there was a higher probability of germinating and
surviving when a hybrid seed was transplanted in the habitat of its maternal parent, i.e., F1allenii
in C. allenii habitat and F1villo in C. villosissimus habitat, than in the habitat of its paternal
parent. The fitness differences in hybrids transplanting in different environments may be due to
maternal environmental effects and/or local adaptation of cytoplasmic genes (Kimball et al.,
2008).
The reciprocal transplant experiment initiated with cuttings showed that both F1 hybrids
performed better than the parental species in the corresponding parental habitats. This indicates
that there is no extrinsic postzygotic isolation at the juvenile stage (RIcutting = -0.163 for C.
allenii and RIcutting = -0.689 for C. villosissimus). The trade-off between surviving in drought
and surviving in shade observed in the parental species was not observed in the hybrids, as the
hybrids survive and grow well in both parental habitats. Costus plants may live for decades, yet
the field experiments reported here are short in comparison. Thus, the fitness estimates obtained
in the reciprocal transplant experiments must be viewed with caution. Because the number of
inflorescences produced per flowering plant was significantly higher in C. villosissimus than in
hybrids within the experimental period, one may expect that the hybrids have reduced lifetime
fecundity and reduced fitness comparing to C. villosissimus. Hybrid transplants producing fewer
inflorescences than C. villosissimus beyond the experimental period might result in extrinsic
postzygotic isolation. In addition to differential reproduction, fitness in backcrosses and later
generations of hybrids may also cause extrinsic postzygotic isolation beyond the F1 generation.

97

When F1 hybrids reproduce, they are more likely to backcross to a parental species because
flowering hybrids are rare in nature. For backcross and later generations of hybrids, hybrid
fitness usually decreases and the chance of the foreign alleles carried in the heterozygotic F1
hybrids to be passed through generations is reduced by one-half in each generation (Rhode and
Cruzan, 2005). A more comprehensive assessment of postzygotic isolation in perennials requires
long-term reciprocal transplant experiments through the life history of multiple generations of
hybrids in comparison with parental species.
Despite the finding of extrinsic postzygotic isolation for seeds transplanted into C.
villosissimus habitats, there was no evidence of extrinsic postzygotic isolation for seeds
transplanted into C. allenii habitats or for cuttings transplanted into either parental habitat.
Postzygotic isolation is, therefore, asymmetrical between species and between different life
history stages. When fitness differences between different species are compared at multiple ages
or life stages, variation in fitness differences among ages or life stages is commonly observed
(Aston and Bradshaw, 1961; Gross, 1981; Platekamp, 1990, 1991; Raabová et al., 2007, 2011;
Raabová et al., 2008). One explanation of this variation was proposed by Platekamp (1991) that
younger plants may be affected more by environmental conditions than older, clonally derived
transplants, as the latter expressed greater genetic difference among plant species than the former.
For C. allenii, C. villosissimus and their F1 hybrids, higher overall survival was observed in
cuttings than in seed transplants, indicating that seed transplants are affected more by
environmental conditions. However, the fitness of hybrid seeds was lower than that for C.
villosissimus seeds in C. villosissimus habitat, but survival of hybrid cuttings was not
significantly different from that of cuttings of the parental species in its home habitat. This
difference between plant life stages may be a result of stronger selection against younger plants.

98

When selection acts strongly on seeds and early seedlings, the genotypic differences between
hybrids and the parental species in its home habitats may cause greater fitness differences
between these plants than between older cuttings.
Although the reciprocal transplant experiment initiated with cuttings showed no
postzygotic isolation, the actual likelihood of heterospecific gene flow may be greatly reduced by
other barriers. This is because local adaptation leads to strong habitat isolation and eliminates the
majority of heterospecific gene flow prior to hybridization. When the probability of hybrid
formation is low, the relative contribution of postzygotic isolation is limited (Schemske, 2010).
Although the proportion of hybrid offspring has not been measured in naturally-pollinated fruits
of the two study species, it has been measured in a pollination array where flowering plants of
both species were intermixed artificially. In the pollination array, a lower proportion of hybrid
offspring than that of parental species was observed in the naturally pollinated fruits (Ch. 1),
indicating a reduced likelihood of hybrid formation. The low frequency of hybrids found along
PLR also suggests that heterospecific gene flow is limited although the edges of the natural
distributions are in close proximity to each other.
Reciprocal transplant experiments demonstrated 1) that C. allenii and C. villosissimus are
locally adapted to their home habitats, 2) that local adaptation leads to strong habitat isolation
between the two species, and 3) that local adaptation leads to asymmetrical extrinsic postzygotic
isolation only in seed transplants. Therefore, between the sister species C. allenii and C.
villosissimus, local adaptation contributes mainly to prezygotic but not postzygotic isolating
barriers.
Local adaptation has also been suggested as the primary isolating mechanism in other
plant systems (see reviews in Lowry et al., 2008a, and Schemske, 2010). For example, the

99

closely related and naturally hybridizing species Ipomopsis aggregata and I. tenuituba were
found to be adapted to different light intensities (Wu and Campbell, 2006; Campbell and Waser,
2007). Reciprocal transplant plant experiments conducted with these species demonstrated strong
habitat isolation and weak extrinsic postzygotic isolation (Campbell and Waser, 2001, 2007),
parallel to the findings presented here. Ramsey et al. (2003) found that habitat isolation is the
primary mechanism of reproductive isolation between Mimulus cardinalis and M. lewisii.
Because of almost complete prezygotic isolation, extrinsic postzygotic isolation, which was not
measured in this study, may not be relevant to reducing heterospecific gene flow between the
two species. Reciprocal transplant experiments and greenhouse experiments suggest that the two
species are locally adapted to their different elevation levels (Angert and Schemske, 2005;
Angert et al., 2008).
Local adaptation to drought has been shown to contribute to habitat isolation between
inland and coast races of Mimulus guttatus (Lowry et al., 2008b). Reciprocal transplant
experiments and corresponding greenhouse experiments in this species revealed that selection
caused by seasonal drought reduced fitness of coast transplants in inland habitats while inland
populations with low salt tolerance had low fitness when being transplanted to the coast (Lowry
et al., 2008b). Extrinsic postzygotic isolation is weak in the inland race but negative in the coast
race, (Lowry et al., 2008b), similar to the case of C. allenii and C. villosissimus presented here.
In summary, local adaptation leading to habitat isolation, which is often the primary isolating
barrier, is commonly found in plant systems.
Conclusion and Future Directions

Reciprocal transplant experiments conducted with seeds and cuttings indicate that local
adaptation to different habitats is the cause of the parapatric distribution of C. allenii and C.

100

villosissimus. I conclude that local adaptation contributes to strong habitat isolation and weak
extrinsic postzygotic isolation between these two recently diverged species. As the parapatric
distribution of the two species is correlated with the soil moisture gradient across the Isthmus of
Panama, further greenhouse experiments on how the two species and their hybrids respond to
drought have been conducted (Ch. 3). The reciprocal transplant experiment in this chapter and
the greenhouse experiment in Chapter 3 together suggest that C. villosissimus has higher drought
tolerance. In contrast, C. allenii habitats have lower light availability, thus the higher leaf mass
per area in C. allenii suggest that the leaf physiology of this species is due to local adaptation. By
using the components-of-isolation method to measure the strength of multiple isolating barriers,
the relative contribution of habitat isolation, extrinsic postzygotic isolation, and other isolating
barriers has been compared to determine the primary barriers of reproductive isolation (see Ch.
4). Given that habitat isolation is the primary isolating barrier between C. allenii and C.
villosissimus (Ch. 4) and that local adaptation contributes to habitat isolation, the results
presented here support the idea of local adaptation being the primary mechanism of speciation.
Taken together, this study enhances our understanding of how local adaptation and habitat
isolation contribute to speciation.

101

↑ Caribbean Sea

PLR

Frijolito

AG
Panama Canal

AP

Gigante
Peninsula

VP
VG

Gamboa
↓ Pacific Ocean
Figure 2.1. Map of the study sites in Central Panama (modified from
http://www.stri.si.edu./images/Mapas/Pipeline_BarroColorado.jpg). Pipeline Road (PLR) is
marked by the red solid line on the right side of the map. The precipitation data were collected at
the Frijolito and Gamboa weather stations marked with blue dots. Red dots indicate the location
of the reciprocal transplant sites: C. allenii sites on Pipeline Road (AP) and on the Gigante
Peninsula (AG), and C. villosissimus sites on Pipeline Road (VP) and in Gamboa (VG).

102

monthly rainfall (mm)

350

frijolito

300

gamboa

250
200
150

*

100
50
0
wet

dry

Figure 2.2. Comparisons of monthly precipitation at Frijolito and Gamboa in the wet and dry
seasons. Error bars denote the mean ± 95% CI. *: p < 0.05 in a paired t-test.

103

gravimetric water content

A.

frequency

B.

0.4
wet

0.3

dry

0.2
0.1
0
0

1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17

0.4
C. allenii (n = 93)

0.3

C. villosissimus (n = 156)

0.2
0.1
0
1

frequency

C.

2

3

4

6

7

8

9

10 11 12 13 14 1

16 17

0.4
C. allenii (n = 557)

0.3

C. villosissimus (n = 386)

0.2
0.1
0
1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17

distance along the transect (km)
Figure 2.3. Soil moisture and species’ distribution along Pipeline Road (PLR). (A) Soil moisture
measured as gravimetric water content in each kilometer section in the wet and dry seasons. (B)
Frequency distribution of flowering plants of C. allenii and C. villosissimus in each kilometer
section on PLR. (C) Frequency distribution of non-flowering plants of C. allenii and C.
villosissimus in each kilometer section on PLR.

104

0.7
proportion germinating

A.

0.6
0.5
0.4
0.3
0.2
0.1
0

C.

proportion surviving

0.7

proportion germinating & surviving

B.

C. allenii F1allenii F1villo C. villo C. allenii F1allenii F1villo C. villo

0.6
0.5
0.4
0.3
0.2
0.1
0
0.7

0

0

0

C. allenii F1allenii F1villo C. villo C. allenii F1allenii F1villo C. villo
C allenii site
C villosissimus site

0.6
0.5
0.4
0.3
0.2
0.1

0

0

0

0

C. allenii F1allenii F1villo C. villo C. allenii F1allenii F1villo C. villo
C. allenii site
C. villosissimus site
C. allenii site

C. villosissimus site

Figure 2.4. Fitness components of seed reciprocal transplant experiment for the 2007 and 2008
cohorts combined. Four categories are C. allenii, C. villosissimus (C. villo), F1 hybrids with C.
allenii as the maternal parent (F1allenii), and F1 hybrids with C. villosissimus as the maternal

parent (F1villo). (A) Seed germination; (B) seedling survival; (C) the proportion germinating and
surviving.
105

cumulative proportion germinating

cumulative proportion germinating

B.

0.7

C. allenii
F1allenii
F1villo
C. villosissimus

0.6
0.5
0.4
0.3
0.2
0.1
0

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

0 30 60 90 120 150 180 210 240 270 300 330 360 39
O N D J F M A M J J A S O

0 O30 60D J 12 F M A M J27 30A33S36O
15 18 21 24 J
39
N 90

D.

0.7

cumulative proportion mortality

C.

cumulative proportion mortality

A.

0.6
0.5
0.4
0.3
0.2
0.1
0
0

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

30 60 90 120 150 180 210 240 270 300 330 360 390

O N D J

F M A M J J A S O

30 60 90 120 150 180 210 240 270 300 330 360 390

O N D J

F M A M J J A S O

Figure 2.5. Comparisons of the changes in seed germination and seedling mortality through time between reciprocal transplanting
sites and among C. allenii, F1allenii, F1villo, and C. villosissimus for the 2007 and 2008 cohorts combined. (A) Seed germination in C.
allenii habitats; (B) seed germination in C. villosissimus habitats; (C) seedling mortality in C. allenii habitats; (D) seedling mortality
in C. villosissimus habitats.

106

proportion surviving

A.

0.7
0.6
0.5
0.4
0.3
0.2
0.1

0

0
proportion change in No. of leaves

C.

10

absolute fitness: survival x growth

B.

C. allenii F1C. allenii site C. villo C. alleniiC. villosissimusvillo C. villo
F1allenii F1 site
allenii F1villo

6

8
6
4
2

0

0

C. allenii F1C. allenii site C. villo C. alleniiC. villosissimus villo C. villo
F1allenii F1 site
allenii F1villo
5
4
3
2
1
0

0
C. allenii F1allenii F1villo C. villo C. alleniiC. villosissimusvillo C. villo
F1allenii F1 site
C. allenii site

C. allenii site

C. villosissimus site

Figure 2.6. Fitness components of cutting reciprocal transplant experiment. Data from sites AG
and AP are combined, and data from site VG and VP are combined. (A) Cutting survival; (B)
cutting growth; (C) absolute fitness of cuttings.

107

proportion surviving

0.6
0.4
0.2

0.0
J
’06

proportion surviving

1

0.8

0

C.

B.

1

C. allenii
F1allenii
F1villo
C. villosissimus
1.0
M J
’07

2.0
M J
’08

3.0
M J
’09

4.0
M J
‘10

0.8
0.6
0.4
0.2
0

0.0
J
’06

1.0
M J
’07

2.0
M J
’08

M 3.0
J
’09

M 4.0
J
‘10

0.6
0.4
0.2

0.0
J
’07

D.

1

0.8

0

proportion surviving

proportion surviving

A.

1.0
M J
’08

2.0
M J
’09

M 3.0
J
‘10

M J
1.0
’07

M J
2.0
’08

M J
3.0
’09

M J
4.0
‘10

1
0.8
0.6
0.4
0.2
0

J
0.0
’06

Figure 2.7. Comparisons of cutting survival among C. allenii, F1allenii, F1villo, and C. villosissimus and among transplanting sites AG
(A), AP (B). VG (C), and VP (D). The x axes represent the time of each census. M denotes March, representing a census in the dry
season; J denotes June, representing a census in the wet season.

108

B.

2.5

pro portion size
p ro p o rtio n s ize (# o f leaves)
of

p ro p ortio n size (# o f leaves)

A.

C. allenii
F1allenii
F1villo
C. villosissimus

2

1.5
1

0.5
0

0.0
J
’06

1.0
M J
’07

2.0
M J
’08

M 3.0
J
’09

M 4.0
J
‘10

2.5

2
1.5

1
0.5

0
0.0
J
’07

3.0
M J
‘10

2.0
M J
’08

3.0
M J
’09

4.0
M J
‘10

14

5
4
3
2
1
0
0.0
J

’06

M

1.0
J

’07

M

2.0
J

’08

M

3.0
J

’09

M

4.0
J

‘10

proportion size (# of leaves)

6

2.0
M J
’09

D. 16

proportion s ize
p ro p o rtio n size (# of leaves)
of

C. 7

1.0
M J
’08

12
10
8
6
4
2
0

J0.0
’06

M J1.0
’07

Figure 2.8. Comparisons of cutting growth among C. allenii, F1allenii, F1villo, and C. villosissimus and among transplanting sites AG
(A), AP (B). VG (C), and VP (D). The y axes represent proportion change in the number of leaves from the time of transplanting to a
given census. The x axes represent the time of each census. M denotes March, representing a census in the dry season; J denotes June,
representing a census in the wet season.

109

relative total fitness

5
4
3
2
1
0

0

0

0

C. allenii F1allenii F1villo C. villo C. allenii F1allenii F1villo C. villo

C. allenii site

C. villosissimus site

Figure 2.9. Comparison of relative total fitness among C. allenii, F1allenii, F1villo, and C.
villosissimus and between parental habitats. The product of fitness components measured in seed
and the cutting transplants of each category in each habitat was compared with that of the
parental species transplanted in its home habitats.

110

proportion damage on new leaves

0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0

C. allenii F1allenii F1villo C. villo C. allenii F1allenii F1villo C. villo

C. allenii site

C. villosissimus site

Figure 2.10. Proportion of pest damage on new leaves of C. allenii, F1allenii, F1villo, and C.
villosissimus in the cutting reciprocal transplant experiment. Error bars denote the mean + 95%
CI.

111

LITERATURE CITED

112

Angert, A. L., H. D. Bradshaw Jr., and D. W. Schemske. 2008. Using experimental evolution to
investigate geographic range limits in monkeyflowers. Evolution 62(10): 2660-2675.
Angert, A. L., and D. W. Schemske. 2005. The evolution of species’ distributions: reciprocal
transplants across the elevation ranges of Mimulus cardinalis and M. lewisii. Evolution
59(8): 1671-1684.
Aston, J. L., and A. D. Bradshaw. 1966. Evolution in closely adjacent plant populations II.
Agrostis stolonifera in maritime habitats. Heredity 21: 649-664.
Bongers, F., L. Poorter, R. S. A. R. Van Rompaey, and M. P. E. Parren. 1999. Distribution of
twelve moist forest canopy tree species in Liberia and Côte d’Ivoire: response curve to a
climatic gradient. Journal of Vegetation Science 10: 371-382.
Brenes-Arguedas, T., A. B. Roddy, P. D. Coley, and T. A. Kursar. 2011. Do differences in
understory light contribute to species distributions along a tropical rainfall gradient?
Oecologia 166: 443-456.
Burke, J. M., and M. L. Arnold. 2001. Genetics and the fitness of hybrids. Annual Review of
Genetics 35: 31-52.
Campbell, D. R., and N. M. Waser. 2001. Genotype-by-environment interaction and the fitness
of plant hybrids in the wild. Evolution 55(4): 669-676.
Campbell, D. R., and N. M. Waser. 2007. Evolutionary dynamics of an Ipomopsis hybrid zone:
confronting models with lifetime fitness data. American Naturalist 169(3): 298-310.
Cheves, M. M., J. P. Maroco, and J. S. Pereira. 2003. Understanding plant responses to
drought — from genes to the whole plant. Functional Plant Biology 30: 239-264.
Clausen, J. C., D. D. Keck, and W. M. Hiesey. 1940. Experimental studies on the nature of
species. I. Effect of varied environments on western North American plants. Carnegia
Institute of Washington Publ. no. 520. Washington, DC.
Coley, P. D., J. P. Bryant, and F. S. Chapin. 1985. Resource availability and plant antiherbivore
defense. Science 230(4728): 895-899.
Coyne, J. A., and H. A. Orr. 1989. Patterns of speciation in Drosophila. Evolution 43(2): 362-381.
Coyne, J. A., and H. A. Orr. 1997. “Patterns of speciation in Drosophila” revisited. Evolution
51(1): 295-303.
Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer, Sunderland, MA.
Darwin, C. R. 1859. On the origin of species by means of natural selection, or the preservation
of favoured races in the struggle for life. J. Murray, London.

113

Dobzhansky, T. 1937. Genetics and the origin of species. Columbia Univ. Press, New York.
Duivenvoorden, J. F. 1995. Tree species composition and rain forest-environment relationships
in the middle Caquetá Area, Colombia, NW Amazonia. Vegetatio 120(2): 91-113.
Endler, J. A. 1982. Problems of distinguishing historical from ecological factors in biogeography.
American Zoologist 22(2): 441-452.
Engelbrecht, B.M.J., L.S. Comita, R. Condit, T. A. Kursar, M. T. Tyree, B. L. Turner, and S. P.
Hubbell. 2007. Drought sensitivity shapes species distribution patterns in tropical forests.
Nature 447: 80–82.
Fine, P. V. A., D. C. Daly, G. V. Muñoz, I. Mesones, and K. M. Cameron. 2005. The
contribution of edaphic heterogeneity to the evolution and diversity of Burseraceae trees
in the western Amazon. Evolution 59(7): 1464-1478.
Franks, S. F. 2011. Plasticity and evolution in drought avoidance and escape in the annual plant
Brassica rapa. New Phytologist 190: 249-257.
Gentry A. H. 1988. Changes in plant community diversity and floristic composition on
environmental and geographical gradients. Annals of the Missouri Botanical Garden
75(1): 1-34.
Giriraj, A., M. Irfan-Ullah, B. R. Ramesh, P. V. Karunakaran, A. Jentsch, and M. S. R. Murthy.
2008. Mapping the potential distribution of Rhododendron arboreum Sm. ssp.
nilagiricum (Zenker) Tagg (Ericaceae), an endemic plant using ecological niche
modelling. Current Science 94(12): 1605-1612.
Gross, K. L. 1981. Predications of fate from rosette size in four “biennial” plant species:
Verbascum thapsus, Oenothera biennis, Daucus carota, and Tragopogon dubius.
Oecologia 48: 209-213.
Hatfield, T., and D. Schluter. 1999. Ecological speciation in sticklebacks: environmentdependent hybrid fitness. Evolution 53(3): 866-873.
Hendry, A. P., P. Nosil, and L. H. Rieseberg. 2007. The speed of ecological speciation.
Functional Ecology 21: 455-464.
Kawecki, T. J., and D. Ebert. 2004. Conceptual issues in local adaptation. Ecology Letters 7(12):
1225-1241.
Kay, K. M., P. A. Reeves, R. G. Olmstead, and D. W. Schemske. 2005. Rapid speciation and
evolution of hummingbird pollination in Neotropical Costus subgenus Costus (Costaceae):
evidence from nrDNA ITS and ETS sequences. American Journal of Botany 92(11):
1899-1910.

114

Kimball, S., D. R. Campbell, and C. Lessin. 2008. Differential performance of reciprocal hybrids
in multiple environments. Journal of Ecology 96: 1306-1318.
Leimu, R., and M. Fischer. 2008. A meta-analysis of local adaptation in plants. PLoS One 3(12):
e4010.
Lowry, D. B., J. L. Modliszewski, K. M. Wright, C. A. Wu, and J. H. Willis. 2008a. The strength
and genetic basis of reproductive isolating barriers in flowering plants. Philosophical
Transactions of the Royal Society: Biological Science 363: 3009-3021.
Lowry, D. B., R. C. Rockwood, and J. H. Willis. 2008b. Ecological reproductive isolation of
coast and inland races of Mimulus guttatus. Evolution 62(9): 2196-2214.
Maas, P. J. M. 1972. Costoideae (Zingiberaceae). Flora Neotropica, monograph 8. Hafner, New
York, New York.
Mayr, E. 1947. Ecological factors in speciation. Evolution 1: 263-288.
Mayr, E. 1963. Animal species and evolution. Harvard University Press, Cambridge, MA.
McBride, C. S., and M. C. Singer. 2011. Field studies reveal strong postmating isolation between
ecologically divergent butterfly populations. PLoS Biology 8(10): e1000529.
McKay, J. K., J. H. Richards, and T. Mitchell-Olds. Genetics of drought adaptation in
Arabidopsis thaliana: I. Pleiotropy contributes to genetic correlations among ecological
traits. Molecular Ecology 12: 1137-1151.
Nosil. P., T. H. Vines, and D. J. Funk. 2005. Reproductive isolation caused by natural selection
against immigrants from divergent habitats. Evolution 59(4): 705-719.
Platenkamp, G. A. J. 1990. Phenotypic plasticity and genetic differentiation in the demography
of the grass Anthoxanthum odoratum. Journal of Ecology 78: 772-788.
Platenkamp, G. A. J. 1991. Phenotypic plasticity and population differentiation in seeds and
seedlings of the grass Anthoxanthum odoratum. Oecologia 88(4): 515-520.
Pyke, C. R., R. Condit, S. Aguilar, and S. Lao. 2001. Floristic composition across a climatic
gradient in a Neotropical lowland forest. Journal of Vegetation Science 12(4): 553-566.
R development Core Team. 2011. R: a language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna. http://www.r-project.org.
Raabová, J., M. Fischer, and Z. Münzbergová. 2008. Niche differentiation between diploid and
hexaploid Aster amellus. Oecologia 158: 463-472.

115

Raabová, J., Z. Münzbergová, and M. Fischer. 2007. Ecological rather than geographic or
genetic distance affects local adaptation of the rare perennial herb, Aster amellus.
Biological Conservation 139: 348-357.
Raabová, J., Z. Münzbergová, and M. Fischer. 2011. The role of spatial scale and soil for local
adaptation in Inula hirta. Basic and Applied Ecology 12: 152-160.
Ramsey, J., H. D., Bradshaw, Jr., and D. W. Schemske. 2003. Components of reproductive
isolation between the monkeyflowers Mimulus lewisii and M. cardinalis (Phrymaceae).
Evolution 57(7): 1520-1534.
Rhode, J. M., and M. B. Cruzan. 2005. Contributions of heterosis and epistasis to hybrid fitness.
American Naturalist 166(5): E124-E139.
Rundle, H. D., and P. Nosil. 2005. Ecological Speciation. Ecology Letters 8: 336-352.
Rundle, H. D., and M. C. Whitlock. 2001. A genetic interpretation of ecologically dependent
isolation. Evolution 55(1): 198-201.
Schemske, D. W. 1983. Breeding system and habitat effects on fitness components in three
Neotropical Costus (Zingiberaceae). Evolution 37(3): 523-539.
Schemske, D. W. 2000. Understanding the origin of species. Evolution 54(3): 1069-1073.
Schemske, D. W. 2010. Adapation and the Origin of Species. American Naturalist 176(S1): S4S25.
Schluter, D. 2001. Ecology and the origin of species. Trends in Ecology and Evolution 16(7):
372-380.
Smith, T., and M. Huston. 1989. A theory of the spatial and temporal dynamics of plant
communities. Vegetatio 83(1/2): 49-69.
Sobel, J. M., G. F. Chen, L. R. Watt, and D. W. Schemske. 2010. The biology of speciation.
Evolution 64:295–315.
Thorpe, R. S., Y. Surget-Groba, and H. Johansson. 2008. The relative importance of ecology and
geographic isolation for speciation in anoles. Philosophical Transactions of the Royal
Society: Biological Science 363: 3071-3081.
Venable, D. L., and L. Lawlor. 1980. Delayed germination and dispersal in desert annuals:
escape in space and time. Oecologia 46: 272-282.
Volis, S. 2009. Seed-related traits and their adaptive role in population differentiation in Avena
sterilis along an aridity gradient. Israel Journal of Plant Sciences 57: 79-90.

116

Whittaker, R. H. 1965. Dominance and diversity in land plant communities. Science 147(3655):
250-260.
Widmer, A., C. Lexer, and S. Cozzolino. 2009. Evolution of reproductive isolation in plants.
Heredity 102: 31-38.
Wu, C. A., and D. R. Campbell. 2006. Environmental stressors differentially affect leaf
ecophysiological responses in two Ipomopsis species and their hybrids. Oecologia 148:
202-212.
Wu, C. A., D. B. Lowry, L. I. Nutter, and J. H. Willis. 2010. Natural variation for droughtresponse traits in the Mimulus guttatus species complex. Oecologia 162: 23-33.

117

Chapter 3

Ecological Factors Contributing to Local Adaptation in Two Neotropical Costus

118

ABSTRACT
Local adaptation has been proposed as the primary mechanism of speciation. As
populations adapt to different habitats, prezygotic isolation evolves because locally adapted
individuals are unable to colonize “foreign” habitats. Local adaptation may also contribute to
extrinsic, postzygotic isolation if F1 hybrids experience reduced growth, survival or reproduction
in either of the parental habitats. I compared the habitats of two recently diverged Neotropical
plant species, Costus allenii and C. villosissimus, to identify the principal environmental factors
that are responsible for local adaptation. Although both species are found in Central America and
northern South America, they occupy different habitats and display parapatric distributions in the
region where they co-occur. Costus allenii occurs along ravines in the understory of primary
forests, while C. villosissimus is found along forest edges. Previous reciprocal transplant
experiments conducted with these species in central Panama revealed strong local adaptation,
with each species performing best in their “home” environment. These experiments also
indicated that F1 hybrids had lower fitness at the early seedling stage in the C. villosissimus
habitats but not in the C. allenii habitats, suggesting asymmetrical extrinsic postzygotic isolation.
Light availability was lower and soil moisture was higher in C. allenii habitats. Two
physiological traits, leaf mass per area (LMA) and drought tolerance, were measured in the
parental species and their F1 hybrids to determine if these traits contribute to local adaptation
and to extrinsic postzygotic isolation. LMA was larger in C. allenii than in the hybrids and C.
villosissimus, suggesting that C. allenii produces thicker leaves with lower photosynthetic
capacity and slower growth rate, which are typical features of shade-adapted plants. Higher
drought tolerance was found in C. villosissimus, which coincides with the fact that C.
villosissimus plants are found in drier habitats. The F1 hybrids had C. villosissimus-like LMA,

119

despite the fact that hybrids do not have lower, C. villosissimus-like, fitness in the C. allenii
habitats. The F1 hybrids also had intermediate drought tolerance, which is consistent with the
fact that hybrids had lower fitness than C. villosissimus in the C. villosissimus habitats. Because
the recently diverged C. allenii and C. villosissimus are mainly isolated by their microhabitats,
examining the mechanisms by which these species have adapted to their different habitats
enhances our knowledge of the relationship between adaption and speciation.

120

INTRODUCTION
Since Darwin (1859) proposed that adaptation is the primary mechanism of speciation,
the relationship between adaptation and speciation has been a subject of interest and debate. As
populations adapt to different habitats, locally adapted individuals have higher fitness in their
“home” habitats than in their “foreign” habitats (see review in Leimu and Fischer, 2008). Natural
selection in the foreign habitats may cause the locally adapted populations to be unable to
colonize foreign habitats (Mayr, 1963; Nosil et al., 2005). Local adaptation to divergent selection
may reduce gene flow between populations and cause ecogeographic and/or microhabitat
isolation because the absence of one population in the foreign habitat eliminates the probability
of hybridization between populations (Coyne and Orr, 2004). Locally adapted populations
become species when reproductive isolation is complete (Mayr, 1949). Therefore, adaptation to
local environments can contribute to prezygotic isolation and lead to speciation between closely
related species (reviewed in Schluter, 2001; Coyne and Orr, 2004; Rundle and Nosil, 2005;
Hendry et al., 2007; Schemske, 2010; Sobel et al., 2010). Furthermore, when F1 hybrids
experience reduced fitness in the parental habitats, extrinsic postzygotic isolation may restrict
heterospecific gene flow as a result of local adaptation (reviewed in Coyne and Orr, 2004;
Rundle and Nosil, 2005; McBride and Singer, 2011). As two species are locally adapted to
different habitats, F1 hybrids of the two species may have intermediate traits which may not be
suitable for either of the parental habitats (e.g., Hatfield and Schluter, 1999). As adaptive alleles
of one parental species are contained in a heterozygotic F1 hybrid living in the other parental
habitat, the hybrid may have lower fitness due to genotype-by-environment interactions (e.g.,
Campbell and Waser, 2007). Taken together, local adaptation can contribute to speciation
through ecogeographic, microhabitat, and/or extrinsic postzygotic isolating barriers.

121

Empirical evidence of ecogeographic, microhabitat, and extrinsic postzygotic isolation
has also been provided to support the theory of how adaptation contributes to speciation. In their
review of plant speciation studies in which multiple isolating barriers were examined and
compared, Lowry et al. (2008a) found that more than half of the cases provided evidence of
isolation due to geography and/or immigrant inviability and that a few cases supported the
importance of extrinsic postzygotic isolation. Schemske (2010) reviewed speciation studies of
both plant and animal systems and concluded that habitat isolation makes a significant
contribution to speciation. These speciation studies suggest the importance of ecogeographic,
microhabitat, and extrinsic postzygotic isolating barriers, yet how local adaptation contributes to
these isolating barriers is largely unknown. In order to understand the relationship between
adaptation and speciation, I studied the mechanisms of local adaptation in recently diverged
species by identifying the environmental factors and corresponding traits contributing to
adaptation of closely related species which occupy different habitats. Although differences in
adaptive traits may accumulate after two species have been completely speciated, differences
between recently diverged species are more likely to be the contributing factor at the time of
speciation (Schemske 2010). When heterospecific gene flow is largely reduced by habitat
isolation between recently diverged species, the environmental factors and corresponding traits
contributing to adaptation to different habitats are likely to play major roles in the process of
speciation.
In plants, reciprocal transplant experiments are frequently used to determine if different
species are locally adapted to their habitats (e.g., Clausen et al., 1940; Campbell and Waser,
2001; Angert and Schemske, 2005). Support for local adaptation comes from the finding that the
fitness of each species is higher in its “home” habitat than in the “foreign” habitat. In statistical

122

terms, this is represented by an interaction between source and transplant sites (Leimu and
Fischer, 2008). However, reciprocal transplant experiments alone cannot identify the
environmental factors or traits responsible for local adaptation. To study the mechanisms of local
adaptation and the relationship between adaptation and speciation, it is necessary to identify the
environmental factors that cause selection and identify the traits that have adaptively diverged in
response to these selective agents (Schluter, 2001). Environmental factors which differ between
species’ habitats are potentially the driving forces which select against the immigrants but favor
the local species in a reciprocal transplant experiment. Traits that affect the performance of
plants when exposed to different environmental factors are strong candidates for contributing to
the local adaptation observed (Reeve and Sherman, 1993). Furthermore, fitness trade-offs for
putative adaptive traits are implicated when a species cannot achieve high fitness in both the
home and the foreign habitats, and if hybrids cannot perform as well as the parents in either
habitat (Angert et al., 2008).
A handful of speciation studies have examined the environmental factors and traits
responsible for local adaptation and contributing to speciation. In naturally hybridizing species of
Ipomopsis aggregata and I. tenuituba, habitat isolation contributes most to reproductive isolation
despite the existence of the hybrid zone (Lowry et al., 2008a; Schemske, 2010). Reciprocal
transplant experiments showed that each species had higher fitness in its home habitats than in
the habitats of the other species, suggesting local adaptation (Campbell and Waser, 2007).
Differences in photosynthetic rate in response to different light intensity but not to different
temperature have been identified as an adaptive difference that partially explains the distribution
of the two Ipomopsis species and their hybrids (Wu and Campbell, 2006). The hybrids of the two
species had ecophysiological traits that are intermediate or equivalent to the parental species (Wu

123

and Campbell, 2006) and they performed well in the hybrid zone (Campbell and Waser, 2001;
Campbell and Waser, 2007), resulting in weak extrinsic postzygotic isolation (Lowry et al.,
2008a). Studies in Mimulus cardinalis and M. lewisii provide another example of the relationship
between adaptation and speciation. Mimulus cardinalis and M. lewisii are sister species which
are isolated primarily by their ecogeographic distribution and their distinct pollinators (Ramsey
et al., 2003). The hummingbird-pollinated M. cardinalis is found primarily at low elevations,
while the bee-pollinated M. lewisii is found at high elevations. Reciprocal transplant experiments
demonstrated that the two species are locally adaptive to their home habitats (Angert and
Schemske, 2005). Species’ differences in leaf physiology and flowering phenology under
different temperatures suggest that the reduction in heterospecific gene flow is largely due to
adaptive differentiation between these species (Angert, 2006; Angert et al., 2008). A recent study
on Ainsliaea faurieana and A. apiculata also revealed that adaptation to different light
availability and water currents contributes to speciation between these two riparian and
nonriparian plant species (Mitsui et al., 2011). These studies demonstrated how local adaptation
contributes to speciation of herbaceous plants in temperate regions, but no such studies have
been conducted in the tropical region, where ecological and evolutionary explanations for the
origin of high biodiversity are still a matter of debate (Mittelbach et al., 2007).
To investigate the relationship between adaptation and speciation, recently diverged
species which occupy different habitats can be studied to compare the species’ differences in
habitats and in adaptive traits responsible for local adaptation and speciation. Environmental
differences in understory and forest edge habitats between Costus allenii and C. villosissimus, a
pair of perennial tropical rainforest herb species, present a good opportunity to study the
mechanism of local adaptation and the relationship between adaptation and speciation.

124

Reproductive isolation between these two sister species is largely due to ecogeographic and
microhabitat isolation (Ch. 4). Both species are native to Central America and northern South
America, but they occupy different habitats and display parapatric distributions. Costus allenii is
usually found in wet, shaded habitats along ravines in primary forests, while C. villosissimus is
found in drier, open sites along forest edges. Hybrids are observed at low frequency, although
there is no hybrid zone, suggesting the lack of suitable habitats for hybrids. Reciprocal transplant
experiments have shown that the two species are locally adapted to their home habitats (Ch. 2).
The objectives of this study were to examine possible environmental factors and putative
adaptive traits that may contribute to local adaptation in C. allenii and C. villosissimus. Given the
microhabitat differences observed between species, I hypothesized that spatial differences in the
environmental factors, light availability and soil moisture, and in corresponding plant traits
contribute to local adaptation and speciation of these two Costus species.
Light availability is a potential factor that influences habitat preference and contributes to
local adaptation (Givnish et al., 2004 and papers cited therein; Mitsui et al., 2011), but a few
studies suggested no correlation between light availability and species distribution of adult trees
(reviewed by Théry, 2001). The forest understory is a resource-poor habitat due to its limited
light availability. Shade-adapted plant species typically grow slower, produce leaves with a
lower photosynthetic rate and slower turnover rate, and show a larger investment in herbivore
defense than light-demanding species (Coley et al., 1985). To determine whether light
availability is a determinant of habitat divergence and local adaptation in C. allenii and C.
villosissimus, I compared the difference in photosynthetically active radiation (PAR) between the
habitats of the two species and the adaptive difference in leaf physiology. In this study, a
composite parameter, leaf mass per area (LMA), was measured as an indication of leaf

125

physiology. Slower growth rate and higher herbivore defense investments in shade-adapted
plants result in a higher LMA (Sterck et al., 2006; Lusk and Warton, 2007), which represents
thicker and denser leaves with longer lifespan and lower mass-based photosynthetic capacity
(Wright et al., 2004; Hassiotou et al., 2010). I predicted that C. allenii would have higher LMA
than C. villosissimus because the former is found in habitats with lower light availability.
Precipitation and soil moisture have been long recognized as important factors limiting
the distribution of plant species (Whittaker, 1965; Duivenvoorden, 1995; Pyke et al., 2001;
Engelbrecht et al., 2007). Despite high annual precipitation, seasonal variation in water
availability is a common feature in tropical rainforests, and tropical plants display substantial
differentiation for drought tolerance (Condit et al., 1995; Baltzer et al., 2008; Comita and
Engelbrechi, 2009). Reciprocal transplant experiments found that C. allenii transplants had high
mortality during the dry seasons when grown in C. villosissimus habitats (Ch. 2). To determine if
local differences in water availability contribute to the differences in spatial distributions of C.
allenii and C. villosissimus, I compare differences in soil moisture between the two species’
habitats and examine how the two species respond to drought. Species with high drought
tolerance are predicted to be found in habitats with lower soil moisture.
In order to understand the potential mechanisms of local adaptation, here I compare LMA
and drought tolerance of C. allenii and C. villosissimus. To further understand how natural
selection may act upon hybrids, these two traits were also examined in the F1 hybrids of the two
parental species. Despite that hybrid fitness was similar to fitness of C. allenii transplants in the
C. allenii sites, the F1 hybrid seeds transplanted to the C. villosissimus sites showed lower fitness
than C. villosissimus (Ch. 2). Because no intrinsic postzygotic isolation caused by unfavorable
interactions between divergently adaptive traits has been detected between the two species, the

126

reduction in hybrid fitness in the C. villosissimus habitats was an indication of extrinsic
postzygotic isolation (Ch. 2; Ch. 4). The fitness differences between parental species and hybrids
depend on how natural selection acts on the traits expressed in the hybrids (Burke and Arnold,
2001). The traits which are adaptive in the parental species usually are intermediate or equivalent
to the parental species when expressed in the F1 hybrids (Wu and Campbell, 2006). Hybrids may
have reduced fitness in habitats of one parental species when the adaptive traits are intermediate
or equivalent to the traits of the other parental species. Hybrids may have similar fitness to one
parental species in the corresponding parental habitats when the traits express equally in hybrids
and the parental species. Hybrids may have superior fitness if the interactions between
divergently adaptive traits produce novel, favorable phenotypes (Burke and Arnold, 2001). If
hybrids are intermediate or equivalent to the unfavorable parental species in some traits while
interactions of other traits are favorable, the counteracting effects among traits may result in no
fitness reduction in hybrids. Examining LMA and drought tolerance in C. allenii, C.
villosissimus, and their F1 hybrids enhances our knowledge of how two recently divergent
species adapt to different environmental factors in their habitats, how hybrids respond to these
environmental factors, and how adaptation contributes to reproductive isolation.
MATERIALS AND METHODS
Study System
The genus Costus, commonly known as spiral gingers, is a clade of perennial herbs which
has recently undergone a rapid diversification in the Neotropics (Kay et al., 2005). The species
occupy various habitats such as streamsides, tree-fall gaps, forest edges, are found on limestone,
red clay and white sand soils, and from low to mid elevation (Maas, 1972; Schemske, 1983; Kay
and Schemske, 2008). Individual plants usually consist of one to several upright stems with

127

leaves arranged spirally around the stems. Costus species typically produce terminal
inflorescences bearing morphologically complex flowers which are pollinated by either
hummingbirds or orchid bees (Kay and Schemske, 2003).
The two study species, C. allenii and C. villosissimus, are closely related and occur in the
same geographic region (Maas, 1972), but occupy different habitats. A molecular phylogeny of
Neotropical Costus places these two species as sister taxa (Kay et al., 2005). Both species flower
in the wet season (see Ch. 4), and both are pollinated by euglossine bees (see Ch.1). The two
species can be easily crossed to produce fully fertile F1 and F2 hybrids, yet hybrids between the
two species are rarely found in nature (see Ch. 2).
Study Site
The field component of this study was conducted in Central Panama, the primary center
of distribution of Neotropical Costus (Maas, 1972; Kay et al., 2005). Across the Isthmus of
Panama, a distance of 80 km, more rain falls on the Atlantic side (mean annual precipitation =
3234 mm at Cristobal) than on the Pacific side (mean annual precipitation = 1798 mm at Balboa
Heights) (Panama Canal Authority Meteorological and Hydrological Service weather station
network). The study was conducted in the vicinity of Pipeline Road (Ch. 2), which runs 17 km
from south to north through the primary tropical rainforest in Soberania National Park, parallel to
the Panama Canal. The precipitation gradient observed across the Isthmus of Panama can also be
detected along Pipeline Road: more rain falls to the north than to the south (Ch. 2). Forest
canopy cover also differs, with a more open canopy in the drier, southern region. Natural
populations of both species are found along Pipeline Road and in adjacent forest, with C. allenii
predominating in the northern stretch and C. villosissimus to the south.

128

Light Availability
To determine light availability in natural habitats of C. allenii and C. villosissimus,
photosynthetically available radiation (PAR) was measured at the apex of individual plants of
both species (N = 27 for C. allenii, N = 42 for C. villosissimus). Measurements were taken over a
4-hour period around solar noon to minimize the influence of solar elevation angle. Because the
plants in the field were too widely spaced to measure all individuals in a single 4-hour period, the
measurements of the two species were taken on two overcast days: June 28, 2006 for C. allenii
and July 12, 2006 for C. villosissimus. Because plants of both species generally grow along
ravines or forest edges, light availability for individual plants is higher on the side away from the
forest and lower on the side closer to the forest. Because Costus plants usually contain multiple
stems, the stems within individual plants may receive different amount of light due to the
different directions of stem growth. To measure light availability of each individual in
consideration of this microsite variation among stems, an average of four adjacent measurements
were obtained for each plant. An imaginary line was drawn perpendicular to the ravine or the
forest edge to the base of each plant. In each direction away from the base of the plant, two
points along the imaginary line were marked, one at 30 cm and one at 60 cm. A total of four
measurements of PAR were measured at the height of the tallest stem by a LI-185B photometer
and a quantum sensor LI-190SB (LI-COR Inc., Nebraska, USA) for each plant. These four
measurements covered the range of an average plant sampled in this study. PAR was also
measured in full sun at a nearby large gap or clearing every 30 minutes, and light availability of
each sampling point of each plant was expressed as a percentage of full sun (Haig et al., 2000). A
two-way ANOVA was performed for each species on the individual measurements of light
availability to determine the differences among plant individuals and among sampling points

129

within each plant individual. A t-test was used to compare the average light availability of plant
individuals between species.
Soil Moisture
To determine soil moisture in habitats occupied by C. allenii and C. villosissimus, soil
samples were collected at the base of individual plants of both species (N = 22 for C. allenii, N =
38 for C. villosissimus). In central Panama, the wet season usually lasts about eight months, from
May to December, and the dry season starts around mid-December and ends in late April. To
capture this seasonal variation, samples were collected in both the wet season (July 2006, 2007)
and the dry season (March 2007, 2008). For each plant, a soil core was extracted from the top 15
cm of soil, representing the rooting depth of Costus, and within 30 cm of an individual target
plant. The samples were sealed in a plastic bag and weighed with a digital balance within 12
hours. Samples were then weighed after being oven dried at 60ºC for 7 days. The gravimetric
water content in the soil sample was calculated as: (wet weight – dry weight) / wet weight. The
effects of seasons, species, and their interactive effects on gravimetric water content in soil were
analyzed with ANOVA and Tukey’s HSD.
Leaf Mass per Area
Four categories of plants were used to evaluate leaf mass per area (LMA): the two
parental species, C. allenii and C. villosissimus, and two F1 hybrids derived from reciprocal
crosses between the parents, with either C. allenii as the female and C. villosissimus as male
(F1allenii), or vice versa (F1villo). A total of 36 C. allenii, 35 C. villosissimus, 17 F1allenii, and 25
F1villo were measured. These plants were produced by hand-pollinating C. allenii and C.
villosissimus plants located in the natural populations along Pipeline Road, or from handpollination of greenhouse-raised parental plants germinated from seeds collected from the natural
130

populations on Barro Colorado Island, Panama. All the plants used in this experiment were
germinated in potting soil (High Porosity Professional Mix, Baccto) and raised in the greenhouse
at Michigan State University. The environmental conditions of the greenhouse were set to be
near a maximum of 26°C during the day and a minimum of 15°C during the night, with the
actual temperature being subject to change depending on the conditions outside of the
greenhouse. The plants experienced the natural day length and light levels in Michigan. Despite
their differences in ages at the time of sampling, all plants were well-hydrated, multi-stem adults
of similar sizes.
Sample collection for estimating LMA was done at least 2-3 hours after sunrise and 3-4
hours before sunset, as suggested by Garnier et al. (2001). Samples were taken in June, 2009 on
fully hydrated plants. Four leaf disks per plant were collected using a round squeeze punch
(Fiskars Craft). To determine the area of the leaf disks, a preliminary collection of 39 fresh disks
was sealed in a plastic bag with wet paper towels until the area was measured with a portable
area meter (LI-3000A with a belt conveyer LI-3050A, LI-COR Inc., Nebraska, USA). Because
all disks were obtained using the same squeeze punch, the average area of the fresh disks was
used to represent the area of all the disks. All leaf disks were cut from the third to the fifth leaves
from the apex of a stem. The leaves sampled were young, fully expanded, and free of serious
herbivore or pathogen damage, as recommended by Garnier et al. (2001) and the literature
reviewed therein. The disks were collected between the midridges and the edges of the leaves to
prevent unequal thickness due to leaf architecture. The collected disks were dried at 60ºC for 7
days. The dry weight of the disks was measured with a digital balance. Following Ellsworth and
Reich (1992), LMA of each plant was calculated as the average ratio of dry weight to area (g /

131

m2). The values of LMA were compared among the two parental species and their F1 hybrids
with ANOVA and Tukey’s HSD.
Drought Tolerance
Plants of C. allenii, C. villosissimus, and their F1 hybrids derived from reciprocal crosses
between the parental species were used to evaluate drought tolerance. To produce the seeds for
this experiment, during the wet season in 2006, I hand-pollinated eight C. allenii and seven C.
villosissimus located in natural populations along Pipeline Road. The flowers were bagged
before and after being hand pollinated to prevent natural pollination. Pollen collected from a
flower was applied to the stigma with a flat toothpick. Seeds from 33 C. allenii, 38 F1allenii, 23
F1villo, and 18 C. villosissimus fruits were collected in October 2006. Seeds of each fruit were
immediately sowed in potting soil (High Porosity Professional Mix, Baccto) in a 4-L pot in the
greenhouse at Michigan State University. The environmental conditions of the greenhouse were
set as described above. Once a seed germinated and the first true leaf was fully expanded, the
seedling was transplanted to a 5 cm x 5 cm rose pot. Seedlings were grown in these pots until
May 2007, when for each species and each direction of F1 hybrids, 45 seedlings of similar size
were transplanted to 4-L pots. These plants were randomly arranged and watered daily with
fertilized water (18-9-18 pH Reducer Fertilizer, 100 ppm N, PLANTEX®) until the drought
treatment began on July 2, 2007. The number of leaves and stems per plant was first counted on
June 28 to July 1, before the drought treatment began. At the beginning of the drought treatment,
all plants were watered weekly with progressively reduced volume: 1000 ml on July 9, 500 ml
on July 16, 250 ml on July 23, and finally, 150 ml on July 30. The progressive reduction in water
volume was designed to mimic the beginning of a dry season in the natural habitats of C.
villosissimus, where gradually less precipitation is received, followed by a long dry period. To

132

make sure that all plants received an equal volume of water, each pot was placed on a tray that
retained water. The plants were randomized approximately every other month over the course of
the experiment (approximately 16 months).
To measure drought tolerance, I calculated a drought tolerance index (DTI), which gives
the number of days from the day the drought treatment began (July 2) to the day a plant became
stemless. To estimate DTI, plants were censused for the number of stems biweekly early in the
drought treatment (August to November, 2007) and then weekly until the end of the experiment
(December 2007 to October 2008). It is difficult to determine the exact time of death for plants
and to measure the exact duration a plant can survive in a dry condition. Therefore, the number
of days until a plant becomes stemless was used as an indicator of the date of death. This was
based on the observation in preliminary studies that plants that had become stemless due to
drought never recovered after being rehydrated, while plants that retained stems usually
resprouted after rehydration (Chen unpublished data). To ensure that the absence of stem is a
good indication of plant death, I rehydrated the plants in 15-plant subsets to test their viability.
Because preliminary studies showed that C. allenii has a much lower drought tolerance than C.
villosissimus, the rehydration treatment was first applied when all of the C. allenii plants were
stemless. Fifteen plants per plant group were choose randomly among the experimental plants
and rehydrated to test their viability after drought stress (Fig. 3.1). This rehydration procedure
was repeated when the remaining F1 hybrids and C. villosissimus became stemless (Fig. 3.1).
Only plants which were stemless before being rehydrated were included in the analysis of
DTI. To demonstrate the change in stem numbers during the experimental period, the average
number of stems of each species and each direction of reciprocal crosses of F1 hybrids was
graphed across time. To examine differences in drought tolerance among the two parental

133

species and their F1 hybrids, DTI was compared with ANOVA and Tukey’s HSD. All statistical
analyses in this study were done in R, version 2.12.2 (R Development Core Team, 2011).
RESULTS
Light Availability
Light availability to individual plants, measured as the average percentage PAR of full
sun, varied from 0.3% to 65.7% in C. allenii habitats and from 1.2% to 87.9% in C. villosissimus
habitats. There was a significant variation among plant individuals and among sampling points
for both species (p < 0.001 for each factor in each species). Light availability of individual plants
was significantly lower in the C. allenii plants (mean ± 95% CI = 11.3% ± 5.9) than in the C.
villosissimus plants (26.1% ± 4.9) (t = 3.93, p < 0.001).
Soil Moisture
The ANOVA results for the analysis of gravimetric water content in soil are presented in
Table 3.1. The C. allenii soil samples had significantly higher gravimetric water content (32.2%
± 1.6) than the C. villosissimus samples (23.2% ± 1.4), and samples collected in the wet seasons
had a significantly higher gravimetric water content (31.9% ± 1.2) than those collected in the dry
seasons (21.1% ± 1.6). Furthermore, the difference in soil moisture between species was greater
in the dry seasons than in the wet seasons (Fig. 3.2), as is reflected in the significant season ×
species interaction (Table 3.1).
Leaf Mass per Area
There was significant variation in LMA among C. allenii, C. villosissimus, and their F1
hybrids (F = 13.464, p < 0.001). Costus allenii had significantly higher LMA than F1allenii,
F1villo, or C. villosissimus, but there was no difference among these three latter categories (Fig.
3.3).

134

Drought Tolerance
Plants continued to grow for approximately one month after their last watering,
maintained their sizes for a period, and then started to wilt (Fig. 3.4). Plants suffering from
drought stress first displayed wilted lower leaves. The lower leaves gradually turned yellow and
dried, while the tips of the stems remained green. Eventually all leaves wilted, became brown,
and dropped off the stems. Leafless stems eventually dried and broke from the rhizomes at the
soil surface.
Plants of C. allenii retained most of their stems through early October, and then lost the
stems rapidly in late October and November. Plants of F1allenii and F1villo showed a similar
pattern to each other: they retained most of their stems through November, and then lost their
stems rapidly from December, 2007 to February, 2008. In contrast, plants of C. villosissimus
remained most of their stems through February, and then gradually lost their stem until all plants
were stemless in late September, 2008 (Fig. 3.4).
When all the C. allenii, all the hybrids, and all the C. villosissimus were stemless on
January 14, April 21, and September 22, respectively, the three rehydration treatments were
applied. Across all three rehydration treatments, none of the stemless plants resprouted after
being rehydrated. Under these experimental conditions, all stemless plants were indeed dead.
Among the 58 plants which retained stems before being rehydrated, 5 F1allenii and 5 F1villo did
not recover from the severe drought stress. The hybrid plants that retained stems but did not
recover indicated that these plants were dead before they were stemless. Therefore, a
measurement of DTI may be an overestimate of the number of days the plant can survive under
drought, but it is accurate enough to show the differences in drought tolerance among plant
categories.

135

DTI was recorded for plants which were stemless before being rehydrated, including 45
C. allenii, 30 F1allenii, 31 F1villo, and 16 C. villosissimus. There were significant differences in
DTI among C. allenii, C. villosissimus and their F1 hybrids (F = 179.37, p < 0.001). The DTI of
C. allenii was significantly lower than that of C. villosissimus, and that of F1 hybrids was
intermediate to the parents (Fig. 3.5).
DISCUSSION
The field studies demonstrate that C. allenii occupies habitats with lower light
availability and produces leaves with higher LMA while C. villosissimus occupies habitats with
lower soil moisture and has higher tolerance to drought. Given the findings of strong local
adaptation of C. allenii and C. villosissimus obtained from reciprocal transplant experiments (Ch.
2), the results presented here provide evidence for major differences between species in their
habitats. Moreover, the differences observed in LMA and drought tolerance strongly suggest that
these traits contribute to local adaptation. The drought tolerance of F1 hybrids is intermediate to
the parents, but LMA is similar to that of C. villosissimus. Although hybrids are rarely found in
nature, intermediate drought tolerance in F1 hybrids may be related to the extrinsic postzygotic
isolation observed to act on them in the C. villosissimus habitats. In contrast, the observation that
hybrid LMA is C. villosissimus-like does not reflect on the unreduced hybrid fitness in the C.
allenii habitats.
Adaptation to Low Light Availability in C. allenii Habitats
Average light availability was more than two-fold higher in C. villosissimus habitats than
in C. allenii habitats. Costus villosissimus habitats are usually along forest edges, where canopy
cover is lower than in the primary forest understory typical of C. allenii habitats. Because plants
grow rapidly at the beginning of each wet season (Chen, personal observation), the measurement

136

was conducted for each plant only once in a wet season. Thus, species’ differences in PAR
measurements obtained in the wet season represent the species’ differences in light availability
when plants are actively producing new leaves. One caveat of the comparison of between
species’ habitats is that there may be seasonal variation in light availability, and I only sampled
light in the wet season. Sampling in a dry season may result in a higher PAR measurement
because many trees lose their leaves in the dry seasons and more light may penetrate the canopy
during the dry season (Wirth et al., 2001; Lemos-Filho et al., 2010). The seasonal differences in
light availability may be greater in the understory than along forest edges; nonetheless, the trend
of higher light availability along edges and lower availability in understory would be consistent
despite the seasonal changes (Oshima et al., 1997).
Costus allenii has higher LMA, indicating thicker and/or denser leaves. Higher LMA is
usually correlated with lower photosynthetic capacity, and slower growth. These features are
common in plants growing under low-resource conditions (Poorter et al., 2009), including shady
habitats. Shade-adapted species tend to have tougher leaves per unit dry mass compared with cooccurring sun-adapted species (reviewed in Onoda et al., 2011) and are better protected against
herbivores, with reduced leaf turnover and higher survival under low light conditions (Poorter,
2009). This is consistent with the hypothesis of higher anti-herbivory defense investments in
plants living in habitats with lower resource availability (Coley et al., 1985) and with the
observation of higher LMA leaves experiencing less herbivore damage (Poorter et al., 2009). The
higher LMA found in C. allenii may indicate that it is adapted to the low light availability
characteristic of their natural habitats. Lower LMA in C. villosissimus may prevent the species
from invading C. allenii habitats, as suggested by the low survival of C. villosissimus transplants
in C. allenii habitats observed in the reciprocal transplant experiments (Ch. 2). The underlying

137

mechanisms of local adaption to C. allenii habitats may also involve C. villosissimus having poor
defense against herbivores as there was a higher proportion of pest damage on new leaves of C.
villosissimus than on those of C. allenii (Ch.2).
I hypothesized that C. allenii has higher LMA because it inhabits environments with
lower light availability under which plants generally have lower photosynthetic capacity and
growth rates. However, LMA is not only a heritable trait but also a plastic characteristic which
responds to many environmental factors, including light availability of the growing condition
(Lusk et al., 2008; Poorter et al., 2009; Lusk et al., 2010). Plants within a species increase their
LMA as the light availability of their growing environments increases (Poorter et al., 2009). The
intraspecific correlation between LMA and light availability is contradictory to the interspecific
correlation. The contradiction between intra- and interspecific LMA-light correlation results
from the differences in leaf structure responses to different growing environments (Lusk et al.,
2010). Leaves of shade-adapted species have higher cell wall mass per unit area, which are more
resistant to fracture than those of light-demanding species. However, leaves of plants growing in
shade have less cell content per unit area than those of plants growing in sun (Lusk et al., 2010).
When growing in environments with high light availability, the magnitude of the plastic response
to increasing environmental light is similar between shade-adapted and light-demanding species
(Poorter et al., 2009; Lusk et al., 2010). Light availability in the greenhouse at Michigan State
University around solar noon in June is higher than that in both parental habitats in Panama
(Chen, personal observation). The LMA measurements presented in this study may be higher
than those in the field, but the two species should respond to the field-greenhouse light difference
similarly to each other. Therefore, the differences in LMA among C. allenii, C. villosissimus, and
their hybrids are proper measurements of the inherited differences among these plants.

138

LMA for F1 hybrids produced from both directions of the cross between C. allenii and C.
villosissimus was similar to that of C. villosissimus. This suggests that C. villosissimus has a
dominant allele(s) for LMA and that the gene(s) are unaffected by the cytoplasm. However,
unbalanced sample sizes across the four groups of parents and hybrids may reduce statistical
power in the comparisons of LMA. Although F1 hybrids have low, C. villosissimus-like LMA,
which should be unfavorable in the C. allenii habitats, the fitness of hybrid was not less than that
of the C. allenii transplants in the C. allenii habitats in reciprocal transplant experiments (Ch. 2).
The effects of the unfavorable LMA on hybrid fitness may be compensated by some beneficial
interactions between other divergently adaptive traits. Future examination of other adaptive traits
and interactions among traits may provide a better understanding of how adaptive traits affect
hybrid fitness in the parental habitats.
Adaptation to Low Water Availability in C. villosissimus Habitats
Soil moisture differed substantially between seasons and by species, with a significant
interaction between seasons and species. The rainfall data presented in Chapter 2 also revealed
significant differences between seasons and species’ habitats. The difference in rainfall between
habitats of C. allenii and C. villosissimus was greater in the dry season than in the wet season.
Because the level of soil moisture is a result of the balance between gain (precipitation) and loss
(evaporation and plant transpiration), it is not surprising to find that the soil moisture data has a
similar pattern to the rainfall data. Both rainfall data and soil moisture measurements indicate
higher water availability in C. allenii habitats than in C. villosissimus habitats, higher water
availability in both habitats in wet seasons than in dry seasons, and a greater difference between
species habitats in the dry seasons. This temporal variation becomes important in determining
species distribution when the ability to survive through the dry season in the drier habitats may

139

limit the potential for C. allenii to invade C. villosissimus habitats, as observed in the reciprocal
transplant results (Ch. 2).
The greenhouse drought tolerance experiment showed significant differences between
species and their hybrids in the number of days the plants retained stems under drought stress.
The parental species and their F1 hybrids had the same sample sizes at the beginning of the
drought treatment but the sample sizes reduced unequally in the analysis of DTI because of the
rehydration treatments. Although all C. allenii but fewer F1 hybrids and fewer C. villosissimus
remained in the analysis, highly significant differences in DTI were detected among the two
parental species and their hybrids. The results provide clear evidence of divergence in drought
tolerance between C. allenii and C. villosissimus.
Among the two parental species and their hybrids, C. villosissimus was surprisingly
drought tolerant: the species was able to survive in the greenhouse an average of 381 days
without water (Fig. 3.5). Humidity in the greenhouse was likely higher than in the field in the dry
season, and light intensity in Michigan was not as strong as in central Panama. Therefore, plant
transpiration in the greenhouse drought experiment was potentially much less than that in C.
villosissimus habitats during the dry season. Nevertheless, it is surprising that a tropical
rainforest herb is able to survive >1 year without water.
From assessments of drought sensitivity of 48 tree and shrub species conducted in central
Panama, Engelbrecht et al. (2007) observed a correlation between soil moisture and species
distribution patterns. They found that tree species’ density was correlated with the precipitation
pattern across the isthmus: wet on the north Atlantic side and dry on the south Pacific side. They
also concluded that drought sensitivity is highly correlated to species distribution in the
Panamanian tropical forests. The natural populations of C. allenii and C. villosissimus, although

140

on a relatively local scale (17 km on Pipeline Road), also demonstrate a remarkable difference in
habitat soil moisture (Ch. 2). Soil moisture is identified as an environmental factor responsible
for local adaptation in C. villosissimus habitats, and the higher drought tolerance of C.
villosissimus is a putative adaptive trait which allows C. villosissimus to grow and reproduce in
dry, seasonal environments. Similarly, the low drought tolerance of C. allenii probably prevents
its expansion into C. villosissimus habitats.
The drought tolerance data not only showed a significant difference between species, but
also that the F1 hybrids are intermediate to the two parental species, with no effect of the
direction of the cross. Based upon the broad segregation of drought tolerance observed in an F2
population (Chen unpublished data), the intermediate phenotype of drought tolerance in the F1
hybrids implies that this is a quantitative trait probably controlled by multiple genes with
additive effects. Hybrids that have intermediate drought tolerance may not be able to perform as
well as C. villosissimus plants under drought stress. The intermediate phenotype in the hybrids
may be responsible for the reduced hybrid fitness in the C. villosissimus habitats in the reciprocal
transplant experiment (Ch. 2), suggesting extrinsic postzygotic isolation. Adaptation to drought
may contribute to both pre- and postzygotic isolation in these recently diverged species.
However, drought tolerance may not be the only trait selected by drought stress in C.
villosissimus habitats. Other traits such as late seed germination phenology may also contribute
to the adaptation of C. villosissimus to dry habitats. When seeds of the two species are
germinated in the field and greenhouse, C. villosissimus consistently displays longer seed
dormancy (Ch.2; Chen unpublished data). Seeds of C. allenii begin germination in late October,
approximately one month after seed maturation. In contrast, seeds of C. villosissimus do not
germinate until late April of the following year, when the wet season begins. As a result, C.

141

villosissimus seedlings avoid the dry season in central Panama (January to April) and have lower
mortality (Ch. 2). This delay of germination in C. villosissimus was observed under a range of
light and soil moisture environments, suggesting innate control of germination timing rather than
environmental cues (Ch. 2).
Interaction between Environmental Factors
Soil moisture and light availability were suggested to be important environmental factors
which contribute to local adaptation in C. allenii and C. villosissimus. Although these factors
were measured independently in this study, in natural populations high light availability is
generally associated with low soil moisture, and vice versa. The correlation between soil
moisture and light availability is an integrated consequence of precipitation and canopy coverage.
In the Isthmus of Panama, Brenes-Arguedas et al. (2011) examined the interaction between light
availability and water availability on the performance of seedlings from 24 species, including
trees, shrubs, and lianas. They found that seedlings had better survival and growth under high
light condition with ample water availability but worse performance under high light condition
when water was limiting. Instead of a trade-off that species with lower drought tolerance are
more shade-tolerated, Brenes-Arguedas et al. (2011) found that species with lower drought
tolerance were better able to take advantage of small increases in light and grow faster. They
suggested a trade-off between being tolerant to drought and growing fast in understory where
light availability is low (Brenes-Arguedas et al., 2011). These findings across unrelated species
are different from the findings of closely related C. allenii and C. villosissimus, as the fitness
difference between species transplanted to C. allenii sites lies in the difference in plant survival
instead of growth (Ch. 2). In the system of C. allenii and C. villosissimus, the results of

142

reciprocal transplant experiments and this study suggest a trade-off between surviving in drought
and surviving in shady habitats.
Besides soil moisture and light availability, numerous biotic factors may be important in
determining plant survival. In the tropics, higher pest (herbivore and pathogen) pressure is
suggested to be associated with wetter sites with lower light availability (Brenes-Arguedas et al.,
2009), e.g., the habitats of C. allenii. A general hypothesis of trade-offs between growth and
defense states that a shade-adapted plant will evolve greater resistance to pests because any loss
of photosynthetic area would be costly to replace (Coley and Barone, 1996). Light-adapted
plants, such as C. villosissimus, usually grow fast, as implied by their low LMA, and tolerate leaf
loss without allocating too much energy to pest resistance. When growing in shady habitats, C.
villosissimus may fail to grow fast but is also poorly defended against pests. If C. villosissimus is
less resistant to pests, it is possible that the top-down selection of pests would reduce C.
villosissimus fitness in C. allenii habitats. This hypothesis is supported by the results from the
reciprocal transplant experiments which indicate a lower fitness and a higher level of pest
damage on the new leaves of C. villosissimus than those of C. allenii (Ch. 2).
Local Adaptation and Speciation
The significant differences in drought tolerance and LMA observed between the two
parental species support the hypothesis that these traits cause habitat preference, and therefore
habitat isolation between the two species. The microhabitat differences that largely isolated C.
allenii and C. villosissimus suggest that local adaptation is the driving force of speciation. In this
case, local adaptation directly causes spatial isolation due to different habitat preferences. A
similar pattern has been reported in inland and coastal Mimulus guttatus, in which populations
are adapted to either seasonal drought in inland habitats or to high salinity in coastal habitats

143

(Lowry et al., 2008b). The effect of light availability on habitat differentiation leading to
speciation has been observed between Ainsliaea faurieana and A. apiculata, two naturally
hybridizing riparian and nonriparian plants (Mitsui et al., 2011). Light availability is lower in the
habitats of A. apiculata than those of A. faurieana, and survival rate of A. faurieana under a
greenhouse low light condition was significantly reduced, suggesting maladaptation of A.
faurieana in habitats of A. apiculata (Mitsui et al., 2011). Isolation between Mimulus cardinalis
and M. lewisii is mainly due to their ecogeographic distribution, which is affected by divergent
adaptation to different temperatures at the elevations where the plants occur (Ramsey et al., 2003;
Angert and Schemske, 2005; Angert 2006). All these studies demonstrated the role of local
adaptation as a primary mechanism of speciation. The study of C. allenii and C. villosissimus
agrees with these studies and expands our understanding of the relationship between adaptation
and speciation into the tropics.
In addition to habitat isolation, local adaptation may also contribute to speciation through
extrinsic postzygotic isolation (reviewed in Coyne and Orr, 2004; Rundle and Nosil, 2005;
McBride and Singer, 2011). The hybrids between C. allenii and C. villosissimus have lower
LMA compared to C. allenii and lower drought tolerance compared to C. villosissimus,
suggesting that the hybrids may be less fit than the parents in the corresponding parental habitats,
resulting in extrinsic postzygotic isolation. Reciprocal transplant experiments demonstrated a
reduction in hybrid fitness to inferior seed germination and/or seedling survival but not to
survival or growth in juvenile cuttings in C. villosissimus habitats and no reduction in hybrid
fitness was detected in C. allenii habitats (Ch. 2). Hybrid fitness reduction in C. villosissimus
habitats may be a consequence of the intermediate phenotype of drought tolerance of the hybrids,
suggesting that local adaptation to drought contributes to extrinsic postzygotic isolation. A

144

moderate level of extrinsic postzygotic isolation was found in the early stage of hybrid
establishment in C. villosissimus habitats (Ch. 2; Ch. 4), but there is no evidence of how
established hybrid juveniles may respond to drought and reduce their fitness in C. villosissimus
habitats. In contrast, hybrids had similar fitness to C. allenii in C. allenii habitats, indicating no
extrinsic postzygotic isolation (Ch. 2; Ch. 4). The relationship between the trait of LMA and
hybrid fitness in C. allenii habitats and whether local adaptation to low light environments
contributes to extrinsic postzygotic isolation are still largely unknown. Asymmetrical and
incomplete extrinsic postzygotic isolation between the two species may be reflected by the
occasional finding of natural hybrids located in the parental habitats because there may not be
strong selection against hybrids once their seedlings are established. In C. allenii and C.
villosissimus, most of the heterospecific gene flow is restricted by earlier acting, prezygotic
barriers, mainly ecogeographic and microhabitat isolation, but not postzygotic barriers (Ch. 4). In
fact, there is very little empirical evidence of extrinsic postzygotic isolation in plants (Widmer et
al., 2009)--further investigation of this barrier is needed.
In addition to these direct contributions to speciation, local adaptation to different
habitats may indirectly contribute to speciation through its interaction with other barriers (Sobel
et al., 2010). For example, flowering time may depend on the abiotic conditions of the habitats
and environmentally-mediated differences in flowering time can cause temporal isolation
between populations located in different habitats (Stanton et al., 1997). Geographic separation
caused by local adaptation may have allowed adaptation to different pollinators and resulted in
pollinator isolation between closely related species (Kay and Sargent, 2009). Different edaphic
conditions in different habitats may also cause pollen-pistil incompatibility and lead to gametic
isolation (Searcy and MacNair, 1990). All these studies suggests that local adaption, directly or

145

indirectly, contribute to speciation. However, in the system of C. allenii and C. villosissimus,
flowering time of the two species overlap greatly (Ch. 4), the two species share the same
pollinators (Ch. 1), and gametic isolation does not contribute significantly to total isolation (Ch.
1; Ch. 4). Weak isolation due to phenology, pollinator, and gametic barriers suggests that local
adaptation of C. allenii and C. villosissimus contribute to speciation mainly through habitat
isolation instead of indirectly through interactions with other isolating barriers.
Conclusion
This study shows that high drought tolerance in C. villosissimus is associated with its
drier habitats and high LMA in C. allenii is associated with its shady habitats. Local adaptation
to different habitats causes divergent habitat preferences and thus microhabitat isolation between
the two species. Intermediate drought tolerance in F1 hybrids may cause extrinsic postzygotic
isolation. To confirm whether drought tolerance and LMA are the adaptive traits contributing to
habitat isolation and speciation, studies of the causality between these putative adaptive traits and
natural selection in the parental habitats are necessary. To this end, reciprocal transplant
experiments using F2 populations that display wide segregation for these putative adaptive traits
in the parental habitats may provide insight into whether these traits are currently under selection.
Such experiments may also shed light on whether and how trade-offs between drought tolerance
and shade tolerance contribute to local adaptation and speciation. Preliminary data for a F2
population grown in C. villosissimus habitats show that drought tolerance is significantly
correlated with higher survival through the first dry season after transplanting, suggesting
drought tolerance being an important adaptive trait contributing to habitat isolation. In addition
to the F2 reciprocal transplant experiments, quantitative trait locus mapping on the adaptive traits
may further help us to understand the genetic basis of local adaptation and speciation.

146

Table 3.1. Summary of ANOVA results for soil moisture measured in habitats of Costus allenii
and C. villosissimus in two dry and wet seasons. The numbers in bold indicate a p value of less
than 0.05.
Df

Sum of squares

Mean square

F value

p value

Season

1

6920.0

6920.0

177.35

< 0.001

Species

1

4580.1

4580.1

117.38

< 0.001

Season × Species

1

977.2

977.2

25.04

< 0.001

236

9208.4

39.0

Residuals

147

Drought treatment began (July 2)
45 C. allenii
45 F1allenii
45 F1villo
45 C. villosissimus
all C. allenii were stemless
continue in drought

rehydration

30 C. allenii
30 F1allenii
30 F1villo
30 C. villosissimus

15 C. allenii
15 F1allenii
15 F1villo
15 C. villosissimus

all F1allenii and F1villo were stemless
continue in drought

rehydration

15 C. allenii
15 F1allenii
15 F1villo
15 C. villosissimus

15 C. allenii
15 F1allenii
15 F1villo
15 C. villosissimus

all C. villosissimus were stemless
rehydration
15 C. allenii
15 F1allenii
15 F1villo
15 C. villosissimus
Figure 3.1. Experimental design of the rehydration treatments in the greenhouse drought
tolerance experiment.

148

gravimetric water content (% )

40

a

35

C. allenii
C. villosissimus

b

30

b

25
20

c

15
10
5
0
wet

dry

Figure 3.2. Soil moisture measured in habitats of C. allenii and C. villosissimus in wet and dry
seasons. Error bars denote the mean ± 95% CI. Bars with different letters represent means that
are significantly different at the level of α = 0.05 based on the Tukey’s HSD test.

149

Leaf M ass per Area (g/m2)

60

a

50

b

b

40

b

30
20
10
0
C. allenii

F1allenii

F1villo

C. villosissimus

Figure 3.3. Comparisons among C. allenii, C. villosissimus and their F1 hybrids in Leaf Mass
per Area (LMA). The sample sizes were different among species and hybrids: C. allenii = 36,
F1allenii = 17, F1villo = 25, and C. villosissimus = 35. Error bars represent mean ± 95% CI. Bars
with different letters represent means that are significantly different at the level of α = 0.05 based
on Tukey’s HSD tests.

150

A.
C. allenii

mean number of stems

B.

F1allenii

F1villo

C. villosissimus

12

C. allenii
F1allenii
F1villo
C. villosissimus

10
8
6
4
2
0

0 Jul30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct

Figure 3.4. Changes in plant appearance and size in the drought tolerance experiment. (A)
Pictures of C. allenii, F1allenii, F1villo, and C. villosissimus taken on September 18, 2007. The
four plants were the same size (number of leaves) at the beginning of the experiment but showed
different level of wilting under drought stress. (B) Changes in the number of stems per plant
among C. allenii, C. villosissimus, and the F1 hybrids through the experiment. For each group, n
= 45 before January 14, n = 30 between January 14 and April 22, and n = 15 after April 22.

151

num ber of days until stem less

450

c

400
350
300
250
200
150

b

b

a

100
50
0
C. allenii

F1allenii

F1villo

C. villosissimus

Figure 3.5. Comparisons among C. allenii, C. villosissimus and their F1 hybrids in drought
tolerance (DTI) measured as the number of days until the plant is stemless. Because of the
rehydration treatments, sample sizes were different among species and hybrids: C. allenii = 45,
F1allenii = 35, F1villo = 36, and C. villosissimus = 16. The error bars represent means ± 95% CI.
The bars with different letters represent means that are significantly different at the level of α =
0.05 based on Tukey’s HSD tests.

152

LITERATURE CITED

153

Angert, A. L. 2006. Growth and leaf physiology of monkeyflowers with different altitude ranges.
Oecologia 148: 183-194.
Angert, A. L., H. D. Bradshaw Jr., and D. W. Schemske. 2008. Using experimental evolution to
investigate geographic range limits in monkeyflowers. Evolution 62(10): 2660-2675.
Angert, A. L., and D. W. Schemske. 2005. The evolution of species’ distributions: reciprocal
transplants across the elevation ranges of Mimulus cardinalis and M. lewisii. Evolution
59(8): 1671-1684.
Baltzer, J. L., S. J. Davies, S. Bunyavejchewin, and N. S. M. Noor. 2008. The role of desiccation
tolerance in determining tree species distributions along the Malay-Thai Peninsula.
Functional Ecology 22: 221-231.
Brenes-Arguedas, T., P. D. Coley, and T. A. Kursar. 2009. Pests vs. drought as determinants of
plant distribution along a tropical rainfall gradient. Ecology 90(7): 1751-1761.
Brenes-Arguedas, T., A. B. Roddy, P. D. Coley, and T. A. Kursar. 2011. Do differences in
understory light contribute to species distributions along a tropical rainfall gradient?
Oecologia 166: 443-456.
Burke, J. M., and M. L. Arnold. 2001. Genetics and the fitness of hybrids. Annual Review of
Genetics 35: 31-52.
Campbell, D. R., and N. M. Waser. 2001. Genotype-by-environment interaction and the fitness
of plant hybrids in the wild. Evolution 55(4): 669-676.
Campbell, D. R., and N. M. Waser. 2007. Evolutionary dynamics of an Ipomopsis hybrid zone:
confronting models with lifetime fitness data. American Naturalist 169(3): 298-310.
Clausen, J. C., D. D. Keck, and W. M. Hiesey. 1940. Experimental studies on the nature of
species. I. Effect of varied environments on western North American plants. Carnegia
Institute of Washington Publ. no. 520. Washington, DC.
Coley, P. D., and J. A. Barone. 1996. Herbivory and plant defenses in tropical forests. Annual
Review of Ecology, Evolution, and Systematics 27: 305-335.
Coley, P. D., J. P. Bryant, and F. S. Chapin. 1985. Resource availability and plant antiherbivore
defense. Science 230(4728): 895-899.
Comita L. S., and B. M. J. Engelbrecht. 2009. Seasonal and spatial variation in water availability
drive habitat associations in a tropical forest. Ecology 90(10): 2755-2765.
Condit, R., S. P. Hubbell, and R. B. Foster. 1995. Mortality rates of 205 Neotropical tree and
shrub species and the impact of a severe drought. Ecological Monographs 65(4): 419-439.

154

Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer, Sunderland, MA.
Darwin, C. R. 1859. On the origin of species by means of natural selection, or the preservation
of favoured races in the struggle for life. J. Murray, London.
Duivenvoorden, J. F. 1995. Tree species composition and rain forest-environment relationships
in the middle Caquetá Area, Colombia, NW Amazonia. Vegetatio 120(2): 91-113.
Ellsworth, D. S., and P. B. Reich. 1992. Leaf mass per area, nitrogen content and photosynthetic
carbon gain in Acer saccharum seedlings in contrasting forest light environments.
Functional Ecology 6(4): 432-435.
Engelbrecht, B.M.J., L.S. Comita, R. Condit, T. A. Kursar, M. T. Tyree, B. L. Turner, and S. P.
Hubbell. 2007. Drought sensitivity shapes species distribution patterns in tropical forests.
Nature 447: 80–82.
Garnier, E., B. Shipley, C. Roument, and G. Laurent. 2001. A. standardized protocol for the
determination of specific leaf area and leaf dry matter content. Functional Ecology 15:
688-695.
Givnish, T. J., R. A. Montgomery, and G. Goldstein. 2004. Adaptive radiation of photosynthetic
physiology in the Hawaiian lobeliads: light regimes, static light responses, and wholeplant compensation points. American Journal of Botany 91(2): 228-246.
Haig, A. R., U. Matthes, and D. W. Larson. 2000. Effects of natural habitat fragmentation on the
species richness, diversity and composition of cliff vegetation. Canadian Journal of
Botany 78: 786-797.
Hassiotou, F., M. Renton, M. Ludwig, J. R. Evans, and E. J. Veneklaas. 2010. Photosynthesis at
an extreme end of the leaf trait spectrum: how does it relate to high leaf dry mass per area
and associated structural parameters? Journal of Experimental Botany 61(11): 3015-3028.
Hatfield, T., and D. Schluter. 1999. Ecological speciation in sticklebacks: environmentdependent hybrid fitness. Evolution 53(3): 866-873.
Hendry, A. P., P. Nosil, and L. H. Rieseberg. 2007. The speed of ecological speciation.
Functional Ecology 21: 455-464.
Kay, K. M., P. A. Reeves, R. G. Olmstead, and D. W. Schemske. 2005. Rapid speciation and
evolution of hummingbird pollination in Neotropical Costus subgenus Costus
(Costaceae): evidence from nrDNA ITS and ETS sequences. American Journal of Botany
92(11): 1899-1910.
Kay, K. M., and R. D. Sargent. 2009. The role of animal pollination in plant speciation:
integrating ecology, geography, and genetics. Annual Review of Ecology, Evolution, and
Systematics 40: 637-656.

155

Kay, K. M., and D. W. Schemske. 2003. Pollinator assemblages and visitation rates for 11
species of Neotropical Costus (Costaceae). Biotropica 35(2): 198-207.
Kay, K. M., and D. W. Schemske. 2008. Natural selection reinforces speciation in a radiation of
Neotropical rainforest plants. Evolution 62(10): 2628-2642.
Leimu, R., and M. Fischer. 2008. A meta-analysis of local adaptation in plants. PLoS One 3(12):
e4010.
Lemos-Filho, J. P., C. F. A. Barros, G. P. M. Dantas, L. G. Dias, and R. S. Mendes. 2010. Spatial
and temporal variability of canopy cover and understory light in a Cerrado of Southern
Brazil. Brazilian Journal of Biology 70(1): 19-24.
Lowry, D. B., J. L. Modliszewski, K. M. Wright, C. A. Wu, and J. H. Willis. 2008a. The strength
and genetic basis of reproductive isolating barriers in flowering plants. Philosophical
Transactions of the Royal Society: Biological Science 363: 3009-3021.
Lowry, D. B., R. C. Rockwood, and J. H. Willis. 2008b. Ecological reproductive isolation of
coast and inland races of Mimulus guttatus. Evolution 62(9): 2196-2214.
Lusk, C. H., Y. Onoda, R. Kooyman, and A. Gutiérrex-Girón. 2010. Reconciling species-level vs
plastic responses of evergreen leaf structure to light gradients: shade leaves punch above
their weight. New Phytologist 186: 429-438.
Lusk, C. H., P. B. Reich, R. A. Montgomery, D. A. Ackerly, and J. Cavender-Bares. 2008. Why
are evergreen leaves so contrary about shade? Trends in Ecology and Evolution 23(6):
299-303.
Lusk, C. H., and D. I. Warton. 2007. Global meta-analysis shows that relationships of leaf mass
per area with species shade tolerance depend on leaf habit and ontogeny. New Phytologist
176: 764-774.
Maas, P. J. M. 1972. Costoideae (Zingiberaceae). Flora Neotropica, monograph 8. Hafner, New
York, New York.
Mayr, E. 1949. Speciation and Selection. Proceeding of the American Philosophical Society
93(6): 514-519.
Mayr, E. 1963. Animal species and evolution. Harvard University Press, Cambridge, MA.
McBride, C. S., and M. C. Singer. 2011. Field studies reveal strong postmating isolation between
ecologically divergent butterfly populations. PLoS Biology 8(10): e1000529.
Mitsui, Y., N. Nomura, Y. Isagi, H. Tobe, and H. Setoguchi. 2011. Ecological barriers to gene
flow between riparian and forest species of Ainsliaea (Asteraceae). Evolution 65(2): 335349.

156

Mittelbach, G. G., D. W. Schemske, H. V. Cornell, A. P. Allen, J. M. Brown, M. B. Bush, S. P.
Harrison, A. H. Hurlbert, N. Knowlton, H. A. Lessios, C. M. McCain, A. R. McCune, L.
A. McDade, M. A. McPeek, T. J. Near, T. D. Price, R. E. Ricklefs, K. Roy, D. F. Sax, D.
Schluter, J. M. Sobel, and M. Turelli. 2007. Evolution and the latitudinal diversity
gradient: speciation, extinction and biogeography. Ecology Letters 10 (4): 315–331.
Nosil. P., T. H. Vines, and D. J. Funk. 2005. Reproductive isolation caused by natural selection
against immigrants from divergent habitats. Evolution 59(4): 705-719.
Onoda, Y., M. Westoby, P. B. Adler, A. M. F. Choong, F. J. Clissold, J. H. C. Cornellissen, S.
Diaz, N. J. Dominy, A. Elgart, L. Enrico, P. V. A. Fine, J. J. Howard, A. Jalili, K.
Kitajima, H. Kurokawa, C. McArthur, P. W. Lucas, L. Markesteijn, N. PérezHarguindeguy, L. Poorter, L. Richards, L. S. Santiago, E, E, Sosinski Jr, S. A. Van Bael,
D. I. Warton, I. J. Wright, S. J. Wright, and N. Yamashita. 2011. Global patterns of leaf
mechanical properties. Ecology Letters 14: 301-312.
Oshima, K., Y. Tang, and I. Washitani. 1997. Spatial and seasonal patterns of microsite light
availability in a remnant fragment of deciduous riparian forest and their implication in the
conservation of Arisaema heterophyllum, a threatened plant species. Journal of Plant
Research 110: 321-327.
Poorter, L. 2009. Leaf traits show different relationships with shade tolerance in moist versus dry
tropical forests. New Phytologist 181: 890-900.
Poorter, H., U. Niinemets, L. Poorter, I. J. Wright, and R. Villar. 2009. Causes and consequences
of variation in leaf mass per area (LMA): a meta-analysis. New Phytologist 182: 565-588.
Pyke, C. R., R. Condit, S. Aguilar, and S. Lao. 2001. Floristic composition across a climatic
gradient in a Neotropical lowland forest. Journal of Vegetation Science 12(4): 553-566.
R development Core Team. 2011. R: a language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna. http://www.r-project.org.
Ramsey, J., H. D., Bradshaw, Jr., and D. W. Schemske. 2003. Components of reproductive
isolation between the monkeyflowers Mimulus lewisii and M. cardinalis (Phrymaceae).
Evolution 57(7): 1520-1534.
Reeve, H. K., and P. W. Sherman. 1993. Adaptation and the goals of evolutionary research. The
Quarterly Review of Biology 68(1): 1-32.
Rundle, H. D., and P. Nosil. 2005. Ecological Speciation. Ecology Letters 8: 336-352.
Schemske, D. W. 1983. Breeding system and habitat effects on fitness components in three
Neotropical Costus (Zingiberaceae). Evolution 37(3): 523-539.

157

Schemske, D. W. 2010. Adapation and the Origin of Species. American Naturalist 176(S1): S4S25.
Schluter, D. 2001. Ecology and the origin of species. Trends in Ecology and Evolution 16(7):
372-380.
Searcy, K. B., and M. R. Macnair. 1990. Differential seed production in Mimulus guttatus in
response to increasing concentrations of copper in the pistil by pollen from copper
tolerant and sensitive sources. Evolution 44(6): 1424-1435.
Sobel, J. M., G. F. Chen, L. R. Watt, and D. W. Schemske. 2010. The biology of speciation.
Evolution 64:295–315.
Stanton, M. L., C. Galen, and J. Shore. 1997. Population structure along a steep environmental
gradient: consequences of flowering time and habitat variation in the snow buttercup,
Ranunculus adoneus. Evolution 51(1): 79-94.
Sterck, F. J., L. Poorter, and F. Schieving. 2006. Leaf traits determine the growth-survival tradeoff across rain forest tree species. American Naturalist 167(5): 758-765.
Théry, M. 2001. Forest light and its influence on habitat selection. Plant Ecology 153: 251-261.
Whittaker, R. H. 1965. Dominance and diversity in land plant communities. Science 147(3655):
250-260.
Widmer, A., C. Lexer, and S. Cozzolino. 2009. Evolution of reproductive isolation in plants.
Heredity 102: 31-38.
Wirth R., B. Weber, and R. J. Ryel. 2001. Spatial and temporal variability of canopy structure in
a tropical moist forest. Acta Oecologia 22: 235-244.
Wright, I. J., P. K. Groom, B. B. Lamont, P. Poot, L. D. Prior, P. B. Reich, E-D. Schulze, E. J.
Veneklaas, and M. Westoby. 2004. Leaf trait relationships in Australian plant species.
Functional Plant Biology 31: 551-558.
Wu, C. A., and D. R. Campbell. 2006. Environmental stressors differentially affect leaf
ecophysiological responses in two Ipomopsis species and their hybrids. Oecologia 148:
202-212.

158

Chapter 4

Reproductive Isolation between Two Closely Related Neotropical Herbs

159

ABSTRACT
Comprehensive studies of the mechanisms by which populations diverge to form
different species are required to understand the origins of biodiversity. Speciation research
demands an examination of the full spectrum of isolating barriers and their relative contribution
in gene flow reduction. I investigated and compared nine sequential reproductive isolating
barriers between two closely related, bee-pollinated Neotropical herbs, Costus allenii and C.
villosissimus. Four of the barriers were examined in this chapter: ecogeographic isolation,
temporal isolation, interspecific seed set, and F1 hybrid pollen viability. The other five barriers,
microhabitat isolation, sexual isolation, F1 hybrid seed germination and survival in parental
habitats, F1 juvenile survival in parental habitats, and F1 juvenile growth in parental habitats,
were presented in previous chapters. In both species, ecogeographic and microhabitat barriers
were the most important factors contributing to total isolation. Both of these barriers represent
spatial isolation due to local adaptation to different habitats, but at different spatial scales.
Prezygotic isolation, mainly due to ecogeographic and microhabitat isolation, was found to be
stronger than postzygotic isolation. I also estimated total isolation with barriers of ecogeographic
isolation, the frequency of hybrids naturally found in the suitable habitats for the two species,
and F1 pollen viability. The average total isolation of the two species was 0.988, suggesting that
the two species are highly isolated. Taken together, with the findings of the previous chapters,
the results of this chapter suggest that ecological factors result in local adaptation and contribute
to reproductive isolation. I conclude that local adaptation is the primary mechanism of speciation
between C. allenii and C. villosissimus.

160

INTRODUCTION
As the foundation of biodiversity, speciation is one of the most important topics in the
field of evolutionary biology. How to study speciation largely depends on how species and
species boundaries are defined. As most evolutionists agree that species are reproductively
isolated from each other (biological species concept; Mayr, 1942), analyzing reproductive
isolation helps us understand the mechanism of species formation and maintenance in nature.
When reproductive isolation accumulates, inter-population gene flow decreases, and populations
may diverge to form different species. A number of barriers have been identified as components
of reproductive isolation (Dobzhansky, 1937; Mayr, 1947, 1963; Schluter, 2001; Coyne and Orr,
2004; Nosil et al., 2005), and their relative contributions have been compared in several systems
(e.g., Coyne and Orr, 1989, 1997; Husband and Sabara, 2003; Ramsey et al., 2003; Kay, 2006;
Matsubayashi and Katakura, 2009; Dopman et al., 2010).
Reproductive isolating barriers can be classified according to the timing of their action
throughout the life history (Dobzhansky, 1937; Mayr, 1942; Coyne and Orr, 2004). Prezygotic
isolating barriers include habitat, temporal, and sexual isolation which act before the formation
of hybrids by preventing hybridization. Habitat differentiation caused by local adaptation to
different ecological environmental factors can reduce the probability of hybridization (Mayr,
1963; Coyne and Orr, 2004; Nosil et al., 2005). The genetic-based differences in species’
distribution may lead to ecogeographic at a large scale and microhabitat isolation at a small scale
(reviewed in Schluter, 2001; Coyne and Orr, 2004; Rundle and Nosil, 2005; Hendry et al., 2007;
Schemske, 2010; Sobel et al., 2010). Temporal isolation restricts heterospecific gene flow by
differences in the timing of reproduction (e.g., Cruzan and Arnold, 1994; Martin et al., 2007).
When diverging taxa reproduce at the same location and at the same time, sexual isolation

161

reduces hybridization by having differences in mating preference (e.g., Nagel and Schluter, 1998;
Ramsey et al., 2003) and efficiency (e.g., McCartney and Lessios, 2004; Kay 2006). Postzygotic
barriers act after the formation of hybrids by reducing hybrid fitness. Environment-independent
hybrid unfitness may result in intrinsic postzygotic isolation that occurs as a consequence of
genetic incompatibilities between the parental genomes. Extrinsic postzygotic isolation may
occur when hybrids experience lower fitness dependent upon their environments.
Among all reproductive isolating barriers, intrinsic postzygotic isolation, i.e., the
reduction in fertility and/or viability of hybrids, has traditionally received more attention. This is
in part because it is easier to study genetic incompatibility in the laboratory than to estimate the
importance of multiple ecological isolating barriers in the field (Mallet, 2006). Although strong
intrinsic postzygotic isolation may restrict gene flow between species with large genetic
divergence, it is unclear whether these isolating barriers evolve early or late in speciation
(Schemske, 2000; Coyne and Orr, 2004). In Centrarchid fish, Bolnick and Near (2005)
concluded that hybrid inviability evolved long after reproductive isolation was essentially
complete, suggesting that prezygotic barriers are of primary importance. While studies of
individual barriers show how the focal barriers reduce heterospecific gene flow, comprehensive
comparisons among barriers between closely related species are required to determine which
isolating barrier contributes most at the time of speciation.
To compare the relative importance of isolating barriers acting sequentially throughout
the life history, Coyne and Orr (1989; 1997) developed the components-of-isolation method
(CIM). By using this method, they concluded that prezygotic isolation evolves more rapidly than
postzygotic isolation in Drosophila. Ramsey et al. (2003) generalized the application of the CIM
and compared multiple barriers simultaneously between Mimulus cardinalis and M. lewisii. They

162

suggested that ecogeographic isolation and pollination fidelity are the major isolating
mechanisms. Other studies that employed CIM also indicated that the postzygotic barriers play a
relatively small role in speciation (e.g., Husband and Sabara, 2004; Kay, 2006; Matsubayashi
and Katakura, 2009; Dopman et al., 2010). In his review of seven plant speciation studies
(including the case presented here), Schemske (2010) suggested that prezygotic barriers often
play the major role in speciation while intrinsic postzygotic isolation typically evolves after
speciation is complete. Unfortunately, comparisons of the relative contribution among multiple
isolating barriers have only been conducted in a few species pairs. Here I used CIM in two
recently diverged species to identify the primary isolating barriers and to determine the primary
mechanism of speciation.
Although first articulated by Mayr (1947), the relative importance of ecological isolating
barriers has recently been revisited by evolutionists (reviewed by Schluter, 2001; Nosil et al.,
2005; Lowry et al., 2008; Schemske, 2010; Sobel et al., 2010). By observing recently diverged
species, ecological factors contribute to speciation directly and/or indirectly (Sobel et al., 2010).
The direct contribution results from ecogeographic (e.g., Nagel and Schluter, 1998), microhabitat
(e.g., Matsubayashi and Katakura, 2009), temporal (e.g., Dopman et al., 2010), sexual (e.g.,
Mendelson, 2003), or extrinsic postzygotic isolation (e.g., Hatfield and Schluter, 1999). In
addition, ecological factors promote divergence indirectly through pleiotropy or linkage
disequilibrium with other isolating barriers, for instance, environment-dependent postmating
prezygotic isolation (Searcy and Macnair, 1990) and Dobzhansky-Muller incompatibilities (e.g.,
Fishman and Willis, 2001). Ecological factors generating divergent selection and resulting in
local adaptation in allopatric populations were suggested to be one of the common mechanisms
of speciation.

163

Although many studies have demonstrated how ecological factors contribute to individual
isolating barriers, the relative contribution of these barriers to total isolation is not always clear.
In studies of which relative contributions of multiple barriers were compared by using CIM,
ecogeographic isolation is often neglected (Sobel et al., 2010). One major impediment to
incorporating ecogeographic isolation in estimates of reproductive isolation comes from the
difficulties in distinguishing ecogeographic isolation from effective geographic isolation (Sobel
et al., 2010). While both ecological and historical factors may determine species’ distribution and
result in effective geographic isolation, measuring ecogeographic isolation decouples the
ecological factors from the historical ones and provides a potential geographic isolation due to
local adaptation (Sobel et al., 2010). As ecogeographic isolation acts as the first potential barrier
in the life history (Ramsey et al., 2003), calculating contributions of other barriers without
considering the effects of ecogeographic isolation could be misleading. In a review of 19 plant
species pairs in which multiple barriers were measured (Lowry et al., 2008), only 2 studies
(Ramsey et al., 2003; Kay, 2006) estimated effective geographic isolation; in both cases, its
contribution was substantial. In a species pair examined in one of these two studies, a strong
degree of ecogeographic isolation was proved in a reciprocal transplant experiment (Angert and
Schemske, 2005). Having ecogeographic isolation in the comparisons of multiple isolating
barriers is urged to properly estimate the role of ecology in reproductive isolation.
Although a few studies compared relative contribution of multiple isolating barriers and
showed the importance of ecology in speciation, whether these studies demonstrated general
patterns of speciation requires more investigation. Comparisons of multiple isolating barriers in
recently diverged taxa allow us to determine the importance of different barriers at the time of
speciation (Schemske, 2010). Such comparisons including barriers which are consequences of

164

local adaptation to ecological factors may lead to determination of whether ecology is the
primary mechanism of speciation (Schemske, 2010; Sobel et al., 2010). To conduct
comprehensive comparisons of isolating barriers, ecogeographic isolation must be included
(Schemske, 2010; Sobel et al., 2010). Further examinations of the identification and the genetic
basis of adaptive traits that are responsible for reproductive isolation rely on studies of these
comprehensive comparisons (Schemske, 2010; Sobel et al., 2010). To this end, here I studied the
relative contribution of multiple isolating barriers, from ecogeographic isolation to intrinsic and
extrinsic postzygotic isolation, between two closely related plant species. The main questions are:
1) What are the primary isolating barriers? 2) Are prezygotic or postzygotic isolating barriers
more important? 3) How much of the total isolation is due to ecological factors? To answer these
questions, I studied two closely related Neotropical herbs, Costus allenii and C. villosissimus.
Costus allenii is found along moist ravines in dense forests, while C. villosissimus lives in more
open, drier areas. These sister taxa (Kay et al., 2005) have similar flowering times, utilize the
same pollinators, and are interfertile. Therefore, I hypothesized that ecogeographic and
microhabitat isolation, which act prezygotically and are driven by ecological factors, are
important components of reproductive isolation between these species.
MATERIALS AND METHODS
Study System
Costus allenii and C. villosissimus are perennial herbs found in Neotropical rainforests
(Maas, 1972). The two species are segregated by their habitats: C. allenii occupies along ravines
in rainforest understory while C. villosissimus is found along the forest edges. Reciprocal
transplant experiments and greenhouse assessment of leaf physiology and drought tolerance
revealed that the C. allenii is adapted to low light availability and C. villosissimus is adapted to

165

low water availability in their home habitats (Ch. 2; Ch. 3). The two species are morphologically
distinct (Maas, 1972). Costus allenii produces broad leaves and has brownish trichomes on stems
and leaves, while C. villosissimus produces narrow leaves and the trichomes on stems and leaves
are usually white. The inflorescences of C. allenii are comprised of flowers with rounded bracts
and large extrafloral nectaries, while those of C. villosissimus are comprised of flowers with
leafy bracts and small extrafloral nectaries. Costus allenii has cream-colored flowers with a small
red-striped labellum, and C. villosissimus has yellow flowers with a large labellum. In spite of
their floral morphological differences, both species are pollinated by euglossine bees (Ch. 1).
The two species are easily crossed to generate fully fertile F1 and F2 hybrids, but only 2% of the
natural populations were found to be hybrids in the study region (Ch. 2). Phylogenetic analyses
of the internal and external transcribed spacer (ITS and ETS) regions of nuclear ribosomal DNA
suggest that C. allenii and C. villosissimus are sister taxa (Kay et al., 2005). As the subgenus
Costus, which comprises the majority of species in the genus Costus, has undergone a rapid
diversification and given rise to more than 50 species in the Neotropics in just three to five
million years (Kay et al., 2005), the divergence between this pair of sister taxa is very recent.
Study Sites
The field components of this study were conducted in a number of sites across the
Isthmus of Panama, where both natural populations of C. allenii and C. villosissimus are found
(Maas, 1972). A steep rainfall gradient has been observed in this region, from the wet Atlantic
side (mean annual precipitation = 3234 mm at Cristobal) to the dry Pacific side (mean annual
precipitation = 1798 mm at Balboa Heights) (Panama Canal Authority Meteorological and
Hydrological Service weather station network).

166

Specific study sites included Pipeline Road (PLR), Gamboa, and Gigante Peninsula. PLR
runs 17 km from south to north through the primary forest of Soberania National Park, parallel to
the Panama Canal. The precipitation gradient observed across the Isthmus of Panama is also
reflected in the soil moisture gradient along PLR (Ch.2). The southern stretch of PLR is drier
with more open canopy, and, as the road progresses northward, both soil moisture and canopy
cover increase. The gradient of soil moisture and light availability along PLR represent typical
habitats of C. allenii and C. villosissimus, which are found in the northern and southern stretch of
the road, respectively. Natural populations of both species were used to measure microhabitat,
phenology, sexual isolation, and postzygotic isolation. To measure extrinsic postzygotic isolation,
reciprocal transplant experiments were conducted with two sites representing habitats of each
parental species: Gigante Peninsula and the northern stretch of PLR for C. allenii, and Gamboa
and the southern stretch of PLR for C. villosissimus (Ch. 2).
Studied Barriers
Nine reproductive isolating barriers were included in sequential order of their life-history
stages in the estimate of total isolation. Ecogeographic, microhabitat, temporal and sexual
isolation are prezygotic barriers. The studied postzygotic barriers include interspecific seed set,
F1 hybrid seed germination and early seedling survival in parental habitats, F1 juvenile survival
in parental habitats, F1 juvenile growth in parental habitat, and F1 hybrid pollen viability.
Detailed examination of microhabitat isolation, sexual isolation, F1 hybrid seed and juvenile
performance in parental habitats were reported in Chapter 1 (sexual isolation) and Chapter 2
(microhabitat and F1 hybrids in parental habitats). Therefore, only summaries of the methods,
findings, and interpretations of these barriers are included here.

167

The strength (RI) of each individual isolating barrier was measured independently.
Because reproductive isolation may evolve asymmetrically between species, all barriers were
estimated separately for C. allenii and C. villosissimus, as suggested by Kay (2006) and Martin
and Willis (2007). While several studies (e.g., Coyne and Orr, 1989, 1997; Ramsey et al., 2003)
used CIM to measure and to compared insolating barriers, there was no consistency in
calculating the strength of isolation (Martin and Willis, 2007; Sobel and Chen, in prep.). Martin
and Willis (2007) employed calculations that are comparable between both pre- and postzygotic
barriers. Sobel and Chen (in prep.) further adjusted these calculations and proposed a set of
equations which are comparable across barriers with a linear relationship with the frequency of
heterospecific gene flow. The general format of the equations proposed by Sobel and Chen (in
prep.) is

⎛ H ⎞
RI = 1 − 2 * ⎜
⎟
⎝C + H ⎠

(1)

where H represents the proportion of heterospecific gene flow and C represents the proportion of
conspecific gene flow. This equation generates RI indices which have a biologically and
mathematically meaningful range of -1 (complete heterospecific mating or heterosis) to 0 (no
isolation) to 1 (complete isolation). The index values stand for the relative under-representation
of heterospecific gene flow and the relative over-representation of conspecific gene flow in
respect to expectations under random mating or equal fitness in parental species and hybrids.
Therefore, all RI indices in this study were calculated using the procedures of Sobel and Chen (in
prep.).
Ecogeographic Isolation

I determined the geographic distribution of C. allenii and C. villosissimus using
ecological niche modeling (Phillips et al., 2006) with georeferenced locality data of herbarium
168

specimens at the Missouri Botanical Garden. Following Sobel (2010), a predicted map of
suitable environmental conditions from spatial data on abiotic variables and species occurrence
was built for each species using ArcGIS (version 9.2, ESRI, Redlands, CA). The map was
restricted to the area of Latin America between 24°N and 8°S, which are the northern and
southern distributional limits, respectively, of the two species (Maas, 1972; www.Tropicos.org,
Missouri Botanical Garden). Twelve climatic variables, a mix of both temperature and
precipitation variables, were selected from the dataset WORLDCLIM (www.worldclim.org)
(Table 4.1). These variables, consisting of a grid of 1 km2 resolution pixels, were used to build a
ecological niche model for each species in the mapped area using Maxent software (Maxent
version 3.3.3e; http://cs.princeton.edu/~sharire/maxent) (Phillips et al., 2006; Sobel, 2010).
Specimens of 31 C. allenii and 72 C. villosissimus were used to establish the Maxent
models. For each species, georeferenced locality data for 75% of the specimens were randomly
selected and used for training the model while the remaining 25% were used for testing. This
training/testing process was run for 500 iterations. Using a threshold of 10 percentile training
presence, a value of 1 was assigned to every pixel of suitable habitat and 0 was assigned to pixels
of unsuitable habitats for each species. The predicted maps of suitable habitats of the two species
were then superimposed. The number of pixels predicted as suitable only for C. allenii was used
as the unshared area for C. allenii, the number of pixels suitable only for C. villosissimus was
used as the unshared area for C. villosissimus, and the number of pixels predicted as suitable for
both species was used as the shared area. Ecogeographic isolation index (RIecogeo) was
computed as:

RI ecogeo = 1 −

shared area
shared area + unshared area

169

(2)

for each species (Sobel and Chen, in prep).
The relative contribution of each climatic variable to the predicted distribution for each
species was determined by Maxent software (Maxent version 3.3.3e;
http://cs.princeton.edu/~sharire/maxent). For the three variables that contributed most to the
Maxent model of each species, their differences were compared between species to determine
whether these variables also contributed to the differences of the distribution between the two
species. Values of these variables at the location of each specimen were compared between
species with t-test.
Microhabitat Isolation

The estimate of microhabitat isolation was presented in Chapter 2, and here I summarize
the approach. In central Panama, where the two species co-occur, microhabitat isolation was
measured from species distribution data collected along a 17 km transect on PLR. The spatial
distributions of both species were determined by censusing individual flowering plants of the
two species in the vicinity of PLR from March 2006 to July 2009. The numbers of flowering C.
allenii and C. villosissimus individuals located within each kilometer section of PLR were
counted and the distributions of the two species were compared using a Kolmogorov-Smirnov
test.
Following Sobel and Chen (in prep), RIhabitat was calculated as
Sp i
Sp 2 i
1
RI habitat = 1 − 2∑ (
×
)
i Sp
1total Sp1i + Sp 2 i

(3)

for each species, with Sp1i and Sp2i representing the number of individuals of the focal species
and the other species found in kilometer section i, respectively, while Sp1total and Sp2total are,

170

respectively, the total number of plants of the focal species and the other species found along
PLR. See detailed methods of measuring microhabitat isolation in Chapter 2.
Temporal Isolation

Each Costus terminal inflorescence typically produces one flower per day which opens at
dawn and wilts by mid-afternoon. This one-flower-per-day pattern was used to estimate the
flowering span of individual plants. I marked the bract which had the current day’s flower and
recorded the date. The plants were examined approximately two weeks later to count bracts
between the current flower and the marked bract. The flowering rate of each inflorescence was
thus estimated. This measurement was repeated every two weeks for the whole flowering season.
The number of bracts below the first marked flower and above the last examined flower was
counted to determine the start and end dates of flowering for the inflorescence by multiplying the
number of flowers by the flowering rate. Flowering periods of 10 C. allenii and 19 C.
villosissimus inflorescences from natural populations on PLR were estimated in 2007, and those
of 26 C. allenii and 56 C. villosissimus inflorescences were estimated in 2008. For each species,
the data were plotted as the proportion of flowers that opened on each day across the whole
flowering season. The Julian dates representing the midpoint of individual inflorescence
flowering periods were compared between species. This comparison in flowering phenology
between species was conducted yearly with a Wilcoxon rank sum test.
The temporal isolation index (RItemporal) of each species was computed yearly as
⎛

Sp1 i

∑ i ⎜ Sp1
⎜
⎝

×

Sp 2 i
Sp1 i + Sp 2 i

⎞
⎟
⎟
⎠

total
Sp 2 total (Sp1 total + Sp 2 total )
RI temporal = 1 −
⎛ Sp
⎞
⎛ Sp
⎞
Sp i
Sp2 i
1
1
1
∑ i ⎜ Sp1 i × Sp1 + Sp2 ⎟ ∑ i ⎜ Sp1 i × Sp1 + Sp2 ⎟
⎜
⎟
⎜
⎟
total
i
i⎠
total
i
i⎠
⎝
⎝
+
Sp total (Sp total + Sp 2 total ) Sp2 total (Sp total + Sp2 total )
1
1
1
2×

171

(4)

for each species (Sobel and Chen, in prep). In this equation, Sp1i and Sp2i represent the number
of flowers that opened on day i for the focal species and the other species, respectively, while
Sp1total and SP2total represent the total number of flowers of the focal species and the other
species produced throughout each season, respectively. This approach was originally developed
by Martin and Willis (2007) in their studies of reproductive isolation between Mimulus guttatus
and M. nasutus. While their equations represent reproductive isolation as 1 – (proportion of
heterospecific gene flow / proportion of conspecific gene flow), Sobel and Chen (in prep.)
modified the equations to fit a unified calculation of reproductive isolation, i.e., equation (1) in
this chapter. This equation gives an accurate estimate of temporal isolation, which considered
both the duration of the flowering period and the frequency distribution of the flowers during this
period for each species. RItemporal of 2007 and 2008 were averaged to generate a single estimate
of the strength of temporal isolation for each species.
Sexual Isolation

The estimate of sexual isolation was presented in Chapter 1. In summary, sexual isolation
was measured by the proportion of hybrid formation in naturally-pollinated fruits in a pollination
array in June and July 2008. This estimate includes isolation due to pollinator isolation, floral
mechanical isolation, and gametic isolation (Ch. 1). The pollination array was located on PLR
close to the edges of the natural distributions of the two species (approximately 6.9 km north of
the road entrance near Gamboa). The pollination array consisted of multiple plants of each
species and a total of two flowers per species per day. The flowers were naturally pollinated.
Mature fruits were collected in late September, and the seeds were sown in potting soil (High
Porosity Professional Mix, Baccto) shortly thereafter in the greenhouse at Michigan State

172

University. Eighty-one seedlings from 15 C. allenii fruits and 131 seedlings from 19 C.
villosissimus fruits were genotyped with species-specific AFLP markers to determine the
proportion of hybrid seedlings in each fruit. The strength of sexual isolation (RIsexual) was
calculated with Equation (1) (Sobel and Chen, in prep.), in which the proportion of hetero- (H)
and conspecific gene flow (C) were represented by the proportion of hybrid and parental
seedings. See detailed descriptions of the pollination array, greenhouse setting, and the methods
of analysis of hybrid seedling formation in Chapter 1.
Postzygotic Isolation

Five sequential barriers of postzygotic isolation were measured in this study: interspecific
seed set, F1 hybrid seed germination and survival in parental habitats, F1 hybrid juvenile
survival in parental habitats, F1 hybrid juvenile growth in parental habitats, and F1 hybrid pollen
viability. Hybrid inviability was measured in the first four barriers, while hybrid sterility was
measured as the reduction in pollen viability. In reciprocal transplant experiments with both
parents and hybrids, seed germination and survival, juvenile survival, and juvenile growth of
hybrids were compared to those of parental species growing in the home habitats. These fitness
components are thereafter referred to as transplant performance in parental habitats. Reduction in
hybrid fecundity (number of inflorescences, flowers, or seeds) was not included as a barrier here.
Assessing plant fecundity was not applicable in the time frame of this study because Costus
plants are perennials which take at least two years to mature and flower once a year for decades.
For the same reason, hybrid unfitness was not measured in backcrosses or later generations,
either. Among the five studied postzygotic barriers, seed set and pollen viability represent
intrinsic barriers while transplant performance in parental habitats represent extrinsic barriers.
Interspecific Seed Set

173

I compared seed set of heterospecific crosses with that of conspecific crosses. Four
categories of crosses were made: the two intraspecific crosses, C. allenii pollinated with C.
allenii and C. villosissimus pollinated with C. villosissimus, and two reciprocal interspecific
crosses between the parents, with either C. allenii as the female and C. villosissimus as male, or
vice versa. The progeny produced by these two intraspecific crosses and two interspecific crosses
are thereafter referred to as “the four categories” or as C. allenii, C. villosissimus, F1allenii, and
F1villo, respectively. The crosses were made using plants from natural populations along PLR.
The parental plants included 9 C. allenii and 7 C. villosissimus in 2006, 5 C. allenii and 6 C.
villosissimus in 2007, and 7 C. allenii and 3 C. villosissimus in 2008. A total of 59 C. allenii
fruits, 66 F1allenii fruits, 45 F1villo fruits, and 50 C. villosissimus fruits were produced across 3
years by hand pollination. The flowers were bagged before and after being hand pollinated to
prevent natural pollination. Pollen collected from a flower was applied to the stigma with a flat
toothpick.
Mature fruits were collected in October and the seeds in each fruit were counted. For each
parental plant, the number of seeds per fruit was averaged separately for intraspecific cross and
for interspecific cross. These averages are referred to as the seed sets of a given cross for each
parental plant. The seed sets were compared among years, between species of the maternal
parent, and between species of the paternal parent using a multi-factorial ANOVA (seed set = µ
+ year + maternal species + paternal species + maternal species*paternal species + year*maternal
species*paternal species + ε) and Tukey’s HSD. In the ANOVA model, the interactive term of
maternal species and paternal species represented interactions between parental species’
genomes (Kay, 2006).

174

Because there was no interaction between year and the interactive term of maternal and
paternal species (see results below), I combined data of three years to estimate the strength of
postzygotic isolation at the seed production stage. Equation (1) was used to calculate the strength
of isolation (RIseedset) due to differences in seed set between parental and hybrid fruits (Sobel
and Chen, in prep.). For C. allenii, the proportion of hetero- (H) and conspecific gene flow (C) in
the calculation of RIseedset were represented by the mean seed set of F1allenii and C. allenii fruits,
respectively. For C. villosissimus, H and C were represented by the mean seed set of F1villo and
C. villosissimus fruits, respectively.
For sexual isolation and isolation at the seed set stage, interspecific seed set and the
proportion of hybrid offspring were compared to intraspecific seed set and the proportion of
parental offspring, i.e., comparisons between F1allenii and C. allenii or between F1villo and C.
villosissimus). These comparisons estimate the differences between inter- and intraspecific
crosses within the same maternal plants.
F1 Hybrid Seed Germination and Early Seedling Survival in Parental Habitats
The estimate of F1 hybrid seed germination and early seedling survival in parental
habitats was described in Chapter 2-- a summary of the approach is presented here. A reciprocal
transplant experiment with seeds of C. allenii, C. villosissimus, F1allenii, and F1villo was
conducted in C. allenii (AP) and C. villosissimus habitats (VP) on PLR in both 2007 and 2008.
Fifty-five seeds per category per site were transplanted each year. Because the proportion of
germinated seeds surviving was low and the patterns were similar between years, data of the two
years were combined to increase statistical power. The average number of F1 hybrid seeds that
germinated and survived (after one year) was compared to that of the parental species

175

transplanted into its home habitat using a G-test of goodness of fit. Equation (1) was used to
calculate the strength of isolation (RIseed) due to differences in germination and early survival
between parental and hybrid seeds (Sobel and Chen, in prep.). The frequency of conspecific gene
flow (C) was estimated as the proportion of seeds of parental species that germinated and
survived after one year in its home habitats, and the frequency of heterospecific gene flow (H)
was estimated as the average proportion of hybrid seeds that germinated and survived after one
year in the same parental habitat. See detailed description of the reciprocal transplant experiment
and the calculation of RIseed in Chapter 2.
F1 Hybrid Juvenile Survival in Parental Habitats
The estimate of F1 hybrid survival for the juvenile stage was presented in Chapter 2, and
is summarized here. A reciprocal transplant experiment initiated with cuttings of mature plants of
the four categories was conducted in C. allenii (AP and AG) and C. villosissimus habitats (VP
and VG) from 2006 to 2010. These cuttings were at juvenile stages when they were transplanted
into parental habitats. Because survival was low in each category at each site, data from the two
sites representing the same parental habitats were combined to increase statistical power.
Survival from the time of transplanting to June 2010 was calculated for each category in each
habitat. Within each parental habitat, the average number of F1allenii and F1villo transplants
survived was compared to that of the parental transplants in its home habitat using a G-test of
goodness of fit. Equation (1) was used to calculate the strength of isolation (RIjuvsur) due to
differences in juvenile survival between parents and hybrids (Sobel and Chen, in prep.). The
frequency of conspecific gene flow (C) was estimated as the proportion of parental species
surviving in its home habitat, and the frequency of heterospecific gene flow (H) was estimated as

176

the average proportion of F1allenii and F1villo transplants surviving in the same parental habitat.
See detailed description of the reciprocal transplant experiment and the estimate of transplant
survival in Chapter 2.
F1 Hybrid Juvenile Growth in Parental Habitats
The estimate of F1 hybrid growth for the juvenile stage was presented in Chapter 2, and
is summarized here. In the reciprocal transplant experiment of juveniles, growth of the surviving
transplants was measured for each category in each habitat. Plant growth was calculated as the
proportional change in the number of leaves from the time of transplanting to June 2010 for the
surviving plants. Because survival was low in each category at each site, data from the two sites
representing the same parental habitats were combined to increase statistical power. Within each
parental habitat, the average growth of F1allenii and F1villo transplants was compared to that of
the parental transplants in its home habitat using an ANOVA. Equation (1) was used to calculate
the strength of isolation (RIjuvgrow) due to differences in juvenile growth between parents and
hybrids (Sobel and Chen, in prep.). The frequency of conspecific gene flow (C) was estimated as
the growth of parental transplants in its home habitat, and the frequency of heterospecific gene
flow (H) was estimated as the average growth of F1allenii and F1villo transplants in the same
parental habitat. See detailed description of the reciprocal transplant experiment and the estimate
of transplant growth in Chapter 2.
F1 Hybrid Pollen Viability
To compare male fertility between hybrids and parental species, I examined pollen tube
germination, a common measure of pollen viability, for the four categories of plants. The plants
were grown from seed in the greenhouse at Michigan State University. See descriptions of

177

greenhouse setting in Chapter 3. From May to September 2009, 8 C. allenii, 12 F1allenii, 17
F1villo, and 13 C. villosissimus plants flowered in the greenhouse. Fresh pollen was collected
from three to five flowers of each plant in the early morning. Pollen of each flower was then
immediately placed in a microcentrifuge tube with 200 µl of pollen tube growth medium (15%
sucrose, 0.03% calcium nitrate, and 0.02% boric acid in distilled water, modified from the
formula used in Schemske and Fenster, 1983). After incubation at room temperature in the
growth medium for at least 2 hours, approximately 80µl of the pollen/growth medium mix was
placed on a microscope slide with a cover slip. For each flower, 200 pollen grains were counted
and the proportion of germinating pollen was recorded. Pollen was scored as viable when a
pollen tube had elongated outside of the circular pollen grain and when the tail was longer than
the diameter of the grain itself.
The proportion of viable pollen was averaged among flowers within a plant. Averages
were compared among the four categories using an ANOVA. Equation (1) was used to calculate
the strength of isolation (RIpollen) due to differences in pollen viability between parental and
hybrid flowers (Sobel and Chen, in prep.). The frequency of conspecific gene flow (C) was
estimated as the average proportion of viable pollen in flowers of the parental species, and the
frequency of heterospecific gene flow (H) was estimated as the mean of the average for both
F1allenii and F1villo flowers.
The fitness components of F1allenii and F1villo were averaged and then compared to those
of a parental species for the last four postzygotic barriers, F1 seed germination and survival in
parental habitats, F1 juvenile survival in parental habitats, F1 juvenile growth in parental habitats,
and F1 hybrid pollen viability. Because progenies of both directions of reciprocal crosses

178

potentially occur in the parental habitats and mate with the parental species, these comparisons
represent the differences between parental species and hybrids regardless the direction of
reciprocal crosses.
Hybrid Frequency in Suitable Habitats for Both Species

The frequency of flowering hybrids in the suitable habitats for both species predicted in
the ecological niche models was examined in Chapter 2. In summary, plants with large yellow
flowers with red stripes were identified as hybrids. These hybrids also had phenotypes of leafy
bract and extrafloral nectary that were intermediates of the phenotypes of the two parental
species. These intermediate phenotypes observed in natural hybrids were consistent with those
observed in greenhouse-bred F1 hybrids. I located all flowering plants, including parental species
and hybrids, in the vicinity of PLR from 2006 to 2009. I estimated the frequency of hybrids as
the number of flowering hybrids divided by the sum of the number of flowering parental species
and hybrids in this region.
Total Reproductive Isolation

Total reproductive isolation was estimated using a multiplicative function of the
individual isolating barriers for each species as suggested by Sobel and Chen (in prep.). The
product of the proportion of hetero- (H) and conspecific gene flow (C) in each isolating barrier
was calculated to represent the overall proportion of hetero- and conspecific gene flow through
sequential stages in the life history. Total reproductive isolation (T), which also varies from -1 to
0 to 1, is:
T = 1−

2 × ∏ Hi

(5)

∏ Ci + ∏ H i

179

(Sobel and Chen, in prep.), where the subscripted i denotes the order of an individual barrier: 1
for ecogeographic isolation, 2 for microhabitat isolation, 3 for temporal isolation, 4 for sexual
isolation, 5 for seed set, 6 for seed fitness in parental habitats, 7 for juvenile survival in parental
habitats, 8 for juvenile growth in parental habitats, and 9 for pollen viability.
For isolating barriers affecting co-occurrence of the two species, i.e., ecogeographic
isolation, microhabitat isolation, and temporal isolation, the proportion of hetero- (H) and
conspecific gene flow (C) was not directly assessed in the methods described above. To include
these barriers in the calculation of total isolation, H and C of these barriers were calculated from
their estimated RI values:

H=

1 − RI
2

(6)

and
C = 1− H

(7).

To dissect the individual contribution of each isolating barrier to total isolation (T), the
absolute contribution (AC) of each barrier was calculated as

ACi =

2 × ∏ H i −1

−

2 × ∏ Hi

∏ Ci −1 + ∏ H i −1 ∏ Ci + ∏ H i

(8)

(Sobel and Chen, in prep.), where the subscripted i denotes the order of an individual barrier as
described above. Because ecogeographic isolation was the first-acting barrier measured in this
study, ACecogeo was set to be the same as RIecogeo as suggested by Ramsey et al. (2003).
The relative contribution (RC) of an isolating barrier presents the relative influence of the
given barrier to total isolation. For barrier i, the relative contribution is:
RCi =

ACi
T

(9)

180

(Ramsey et al., 2003; Sobel and Chen, in prep). A greater RC value indicates a greater
contribution to total isolation, and the barrier with the greatest RC was identified as the primary
isolating barrier. When total isolation is almost complete, a RC value is almost identical to the
AC value of a given barrier. When total isolation is not complete, a RC value represents the
proportion of total isolation that is due to a given barrier.
I also estimated reproductive isolation from hybrid frequency in suitable habitats for both
species, substituting hybrid frequency for the multiplicative effects of microhabitat isolation,
temporal isolation, sexual isolation, interspecific seed set, F1 seed germination and survival, and
F1 juvenile survival and growth. This was based on the assumption that all hybrids found in
nature were F1s. The values of H, C, and RI for ecogeographic isolation and F1 pollen viability
were averaged between the two species. Estimates of total reproductive isolation (T) and the
corresponding AC and RC were computed with the averaged ecogeographic isolation, hybrid
frequency in suitable habitats for both species, and F1 pollen viability.
RESULTS
Ecogeographic Isolation

The distribution of C. villosissimus is different from that of C. allenii. The herbarium
specimens of C. allenii indicated that this species is found in Costa Rica, Panama, and Columbia
while those of C. villosissimus indicated that this species is found in Costa Rica, Panama,
Columbia, Mexico, Ecuador, Nicaragua, and Venezuela. Ecological niche models of the two
species predicted that the suitable habitats of C. allenii and C. villosissimus included 68,596 and
1,136,490 pixels, respectively, which were slightly larger than the distribution of the herbarium
specimens (Fig. 4.1). The shared area which was suitable for both species included 58,827 pixels
(Fig. 4.1). Thus, RIecogeo was 0.142 for C. allenii and 0.948 for C. villosissimus.

181

Climatic variables contributing to the niche models were similar in the two species. The
three variables contributing the most to the model for the distribution of C. allenii were mean
temperature diurnal range (BIO2, 42%), precipitation of warmest quarter (BIO18, 13%), and
precipitation of the wettest quarter (BIO16, 12%). The three climatic variables contributing the
most to the model of C. villosissimus were mean temperature diurnal range (BIO2, 34%),
precipitation of driest quarter (BIO17, 17%), and precipitation of the warmest quarter (BIO18,
16%). The mean temperature diurnal range (BIO2) of C. villosissimus (mean ± 95% CI = 84.69 ±
3.39) was significantly higher than that of C. allenii (77.65 ± 3.69; t = 2.83, p = 0.006). The
precipitation of the wettest quarter (BIO16) of C. allenii (1300.03 ± 106.92) was significantly
higher than that of C. villosissimus (1052.19 ± 73.07; t = 3.86, p < 0.001). There was no
differences in precipitation of the driest quarter (BIO17) between C. allenii (194.73 ± 97.53) and
C. villosissimus (125.92 ± 29.45; t = 1.37, p = 0.18). There was no difference in precipitation of
the warmest quarter (BIO18) between C. allenii (514.38 ± 87.84) and C. villosissimus (489.09 ±
54.45; t = 0.49, p = 0.62).
Microhabitat Isolation

As described in the results of Chapter 2, there was a significant difference in the
distribution of flowering C. allenii and C. villosissimus (D = 0.65, p = 0.001). I located 93
flowering C. allenii and 156 flowering C. villosissimus in the vicinity of PLR. Most C. allenii
plants were found on the northern stretch of the road, where soil moisture is relatively high and
the canopy is closed. Most C. villosissimus plants were found on the southern stretch of the road
where soil moisture is lower and the canopy is more open. RIhabitat of C. allenii was 0.886 and
that of C. villosissimus was 0.932 (Ch. 2).

182

Temporal Isolation

Both C. allenii and C. villosissimus flower in the early wet season. In 2007, the mean
estimated flowering period for C. allenii was from June 20 to July 27, while it for C.
villosissimus was from June 22 to July 19. The Julian dates of the midpoint of the flowering
periods showed no difference between species, indicating that the two species essentially
flowered at the same time (W = 106.5, p = 0.61). RItemporal was 0.028 for C. allenii and 0.019
for C. villosissimus (Fig. 4.2A). In 2008, the mean estimated flowering period for C. allenii was
from June 28 to August 6, while it for C. villosissimus was from June 30 to August 1. As was the
case in 2007, the Julian dates of the midpoint of the 2008 flowering periods were not different
(W = 803.5, p = 0.4547, Fig. 4.2B). In 2008, RItemporal was 0.061 for C. allenii and 0.035 for C.
villosissimus. The average RItemporal of 2007 and 2008 was 0.045 for C. allenii and 0.027 for C.
villosissimus.
Sexual Isolation

Sexual isolation was observed in C. villosissimus, but not in C. allenii. As described in
Chapter 1, the average proportion (± 95% CI) of hybrid formation following natural pollination
in the array was 0.44 ± 0.22 for C. allenii (n = 15 fruits), which was not significantly lower than
the null expectation of 0.50. Thus, sexual isolating barriers did not reduce heterospecific gene
flow from C. villosissimus to C. allenii. In contrast, there was a significant reduction in the
proportion of hybrid seedling formation observed for C. villosissimus, with an average
proportion of hybrid formation in the array of 0.21 ± 0.14 (n = 19 fruits). RIsexual was estimated
as 0.121 for C. allenii and 0.581 for C. villosissimus.

183

Postzygotic Isolation

Interspecific Seed Set
A total of 9,571 seeds from the experimental crosses were collected and counted from
2006 to 2008. Seed sets differed among years (p < 0.001): more seeds were produced per fruit in
2007 (58.44 ± 9.40) than in 2008 (43.10 ± 8.35) and 2006 (33.63 ± 6.06). Seed sets were higher
in fruits produced by C. villosissimus plants (58.22 ± 7.76) than in fruits produced by C. allenii
plants (32.40 ± 3.71; p < 0.001). There was no significant difference in seed set between fruits
with different paternal parents (p = 0.21), and no significant interaction between year and
category (p = 0.11).
For all fruit categories taken together, i.e., intra-and interspecific crosses, seed set
differed significantly among categories (p = 0.01; Fig. 4.3), indicating incompatibility between
species. For C. villosissimus as the maternal parent, seed set was lower in interspecific crosses
(51.67 ± 10.85) than in intraspecific crosses (64.76 ± 11.32; p = 0.03). For C. allenii as the
maternal parent, seed set of interspecific crosses (30.78 ± 4.94) was similar to seed set of
intraspecific crosses (34.01 ± 5.90). Therefore, there was significant isolation in C. villosissimus
(RIseedset = 0.112) but an insignificant isolation of C. allenii (RIseedset = 0.050).
F1 Hybrid Seed Germination and Early Seedling Survival in Parental Habitats
As described in Chapter 2, reduction in hybrid seed germination and survival was found
in C. villosissimus habitats. The average number of F1allenii and F1villo seeds (5.5) that
germinated and also survived was lower than the number of C. villosissimus seeds (15) that
germinated and also survived in C. villosissimus habitat (G = 4.58, p = 0.03). However, the
averaged germination and survival of hybrid seeds (7) was similar to that of C. allenii (8) in C.
allenii habitat (G = 0.07, p = 0.80). RIseed of C. allenii was calculated as 0.067 and that of C.
184

villosissimus was 0.463. The results of the G tests combined with RIseed values indicated that
postzygotic isolation at the seed stage in F1 hybrids was significant in C. villosissimus habitat but
not in C. allenii (Ch.2).
F1 Hybrid Juvenile Survival in Parental Habitats
As described in Chapter 2, there was no reduction in hybrid juvenile survival in parental
habitats. The average proportion of F1 hybrids surviving was similar to the proportion of the C.
allenii transplants surviving in C. allenii habitats (G = 0.61, p = 0.43), and was also similar to
that of the C. villosissimus transplants in C. villosissimus habitats (G = 2.03, p = 0.15; Ch. 2).
RIjuvsur of C. allenii was -0.129 and that of C. villosissimus was -0.240.
F1 Hybrid Juvenile Growth in Parental Habitats
As described in Chapter 2, there was no reduction in hybrid juvenile growth in parental
habitats. In C. allenii habitats, the growth of the surviving F1 hybrids was similar to that of
surviving C. allenii (p = 0.81). In contrast, the growth of surviving F1 hybrids was significantly
greater than that of C. villosissimus in C. villosissimus habitats (p = 0.03). Therefore, RIjuvgrow
of C. allenii was -0.014 and that of C. villosissimus was -0.493.
F1 Hybrid Pollen Viability
The average proportion of viable pollen was 0.35 ± 0.10 for C. allenii, 0.42 ± 0.10 for
F1allenii, 0.38 ± 0.09 for F1villo, and 0.47 ± 0.09 for C. villosissimus (Fig. 4.4). There was no
significant difference among categories (p = 0.39). Isolation was not significant for either C.
allenii (RIpollen = -0.069) or C. villosissimus (RIpollen = 0.074).

185

Hybrid Frequency in Suitable Habitats for Both Species

Five flowering hybrids were found in the vicinity of PLR (Ch. 2). Given that a total of
249 flowering plants were observed in this region, the hybrid frequency was estimated as 2%.
Total Reproductive Isolation

Total isolation was higher in C. villosissimus (T = 0.999; Table 4.2) than in C. allenii (T =
0.924). The high value of total isolation in C .villosissimus indicates that it was almost
completely isolated from C. allenii. In contrast, there remains a reasonably high probability of
gene flow from C. villosissimus to C. allenii. For each species, prezygotic barriers contributed
much more than postzygotic barriers (Fig. 4.5A). Specifically, ecogeographic and microhabitat
isolation are the two strongest barriers in both species (Table 4.2), contributing 15.37% and
83.46%, respectively, to the total isolation in C. allenii, and 94.85% and 5.02%, respectively to
the total isolation in C. villosissimus (Table 4.2; Fig. 4.5A). Because ecogeographic isolation and
microhabitat isolation represented spatial isolation due to local adaptation to different habitats,
the strong isolation due to these barriers indicates that habitat differentiation was is the most
important isolating mechanism between C. allenii and C. villosissimus.
The observed frequency of flowering hybrids in the natural populations of the two species
(2%) was similar to the expected frequency calculated multiplicatively with the estimates of
individual barriers from microhabitat isolation to F1 hybrid juvenile growth in parental habitats
(2.55%). When the barriers from microhabitat isolation to F1 juvenile growth were replaced with
hybrid frequency in suitable habitats for both species, the average total isolation was estimated to
be 0.988. In this estimate, ecogeographic isolation and hybrid frequency in suitable habitats for
both species contributed 55.2% and 44.8% to total isolation (Fig. 4.5B). Therefore, the two
species are highly isolated, mainly due to their ecogeographic differences.

186

DISCUSSION
Ecogeographic Isolation

Local adaptation to different environments often causes differences in species’
distributions and thus leads to spatial isolation (Mayr, 1947). Most speciation studies have been
conducted in regions of sympatry, without considering the effects of geographic isolation. By
using ecological niche modeling, ecogeographic isolation can be distinguished from effective
geographic isolation caused by historical factors, and can be estimated and incorporated into
reproductive isolating measures (Sobel et al., 2010). I found that ecogeographic isolation for C.
villosissimus (RIecogeo = 0.948) was considerably stronger than that for C. allenii (RIecogeo =
0.142). Because suitable habitats for C. villosissimus are more widespread than those of C.
allenii, most suitable habitats for C. villosissimus are not suitable for C. allenii. Therefore, the
likelihood of C. villosissimus being in proximity to C. allenii is low, as reflected in the high
RIecogeo value. In contrast, most suitable habitats for C. allenii are also suitable for C.
villosissimus, suggesting a higher likelihood of C. allenii being in proximity to C. villosissimus.
From results of this experiment, I conclude that ecogeographic isolation is strong and
asymmetrical between C. villosissimus and C. allenii.
Mean diurnal temperature was the best predictor of the ecogeographic distribution of both
species, and was significantly different between the two species. Large diurnal temperature range
is associated with low humidity and high daily solar radiation (Bristow and Campbell, 1984). A
larger diurnal temperature range associating with a larger diurnal humidity range and higher
solar radiation were found along forest edges in comparison to adjacent forests (Chen et al.,
1993). Therefore, diurnal temperature range may be a determinant of the distribution of C. allenii

187

and C. villosissimus because these species are found in moist forest and along forest edges,
respectively.
In addition to diurnal temperature range, precipitation in the warmest and wettest quarter
contributes to the niche model of C. allenii, while precipitation in the driest and warmest quarter
plays an important role in predicting the distribution of C. villosissimus. Precipitation differs
significantly in the wettest quarter, but not in the driest nor in the warmest quarters between the
locations of herbarium specimens of C. allenii and those of C. villosissimus. However, this is not
consistent with the comparisons with precipitation and soil moisture directly measured in the
habitats of C. allenii and C. villosissimus (Ch. 2). Specifically, C. allenii habitats have
significantly higher precipitation and higher soil moisture than C. villosissimus habitats in the dry
season (Ch. 2; Ch. 3). Along PLR, dry season soil moisture is highly correlated with the
distribution of the two species (Ch. 2). Although the resolution of the variables available in
dataset WORLDCLIM was 1 km2, I suspect that the measurements of these variables were not as
accurate as the direct measurements conducted in the natural populations of the two species.
Nevertheless, ecogeographic isolation is associated with different water and light availability in
the habitats of C. allenii and C. villosissimus. Therefore, I suggest that adaptation to different
water and light availabilities causes habitat differentiation and leads to ecogeographic isolation
between C. allenii and C. villosissimus.
There are two potential caveats of using ecological niche modeling to assess
ecogeographic isolation. One big limitation is the availability of environmental variables.
Although the WORLDCLIM database consists of 19 climatic variables of temperature and
precipitation, other environmental variables that may be important, such as edaphic composition
and biotic factors, are not available. Edaphic composition has been suggested as one major

188

determinant of distribution of Neotropical tree species (Fine et al., 2005). Plant distribution may
also be affected by the distribution of their pollinators, seed dispersers, and symbiotic
microorganisms. The lack of edaphic layers and biotic layers may reduce the accuracy of the
prediction of species distribution.
Another potential problem is the level of resolution of environmental variables. I used a
resolution of 1 km2 to examine the distribution of the two study species, yet environmental
variables may vary within a grid of 1 km2. As a consequence, species may be spatially
differentiated over a small spatial scale and show little ecogeographic isolation. Compared to a
finer resolution, a grid of 1 km2 cells in the model may overestimate the range of suitable
habitats for each species alone and for both species. Therefore, the strength of ecogeographic
isolation may be underestimated due to the limitation of the resolution of environmental
variables. Detailed studies of microhabitat isolation conducted at a finer scale are needed to
improve the predictions of ecological niche modeling. Nevertheless, for C. allenii and C.
villosissimus, the resolution of 1 km2 may be sufficient because their shared pollinators,
euglossine bees, are able to fly long distances (Janzen, 1971; Dressler, 1982; Wikelski et al.,
2010). The distance of pollen dispersal, in which the frequency of hybrid formation is quantified
in natural populations, is required to determine the proper resolution of environmental variables
to establish ecological niche models.
Microhabitat Isolation

Costus allenii and C. villosissimus display a parapatric distribution along PLR. The high
values of RIhabitat (RIhabitat = 0.886 for C. allenii; RIhabitat = 0.932 for C. villosissimus)
indicate that microhabitat isolation is strong. The reciprocal transplant experiments conducted on
these species provided evidence of local adaptation (Ch. 2). Therefore, the high values of

189

RIhabitat are caused by local adaptation to different environmental factors in the natural habitats
of the two species. As soil moisture and light availability were significantly different between
parental habitats, these environmental factors are suggested to be responsible for local adaptation
(Ch. 3). In addition, two traits which are associated with light and water availability were
different between species. Leaf mass per area, a composite physiological trait, was larger in C.
allenii, suggesting that C. allenii is adapted to low light environment. Higher drought tolerance
was found in C. villosissimus, suggesting that C. villosissimus is adapted to low soil moisture (Ch.
3). Therefore, leaf mass per area and drought tolerance are putative adaptive traits that contribute
to the evolution microhabitat isolation. Further studies of natural selection experiments and QTL
mapping will be required to determine the effects of these traits on reproductive isolation and to
investigate their genetic basis.
Temporal Isolation

Both C. allenii and C. villosissimus flower in the early wet season, mostly in June and
July. In comparison with the flowering phenology in 2008, the length of the flowering period
seems to be shorter and the frequency distribution does not present as a normal distribution in
2007. This may be due to the smaller sample size examined in 2007. By averaging the results of
the two years, flowering phenology was not different between C. allenii and C. villosissimus. In
addition, reproductive isolation due to this temporal barrier was not significant.
Temporal isolation was estimated by incorporating the number of flowers that opened on
each day for each species (Sp1i and Sp2i) and the total number of flowers of each
species(Sp1total and Sp2total), as suggested by Sobel and Chen (in prep.). This approach was
originally developed by Martin and Willis (2007), who found strong temporal isolation between

190

Mimulus guttatus and M. nasutus (Martin and Willis, 2007). In contrast, there was no evidence
of temporal isolation between C. allenii and C. villosissimus. Of note, flowering phenology of
the two species was only measured along PLR. It is likely that different environmental factors in
other localities may change flowering phenology and result in variation in the estimate of
temporal isolation (Sobel, 2010). However, given the insignificant isolation in the highly
contrasting environments of C. allenii and C. villosissimus along PLR, I suspect that temporal
isolation between the two species is low across their geographic ranges.
Sexual Isolation

While both C. allenii and C. villosissimus are pollinated by euglossine bees, there is
substantial sexual isolation in C. villosissimus (RIsexual = 0.581) but not in C. allenii (RIsexual =
0.121; Ch1). In the pollination array, pollinators preferred C. villosissimus but displayed low
floral constancy (Ch. 1). Significant sexual isolation in C. villosissimus was largely due to
pollinator and gametic isolation but not floral mechanical isolation (Ch. 1). The strong
preference for C. villosissimus increased the likelihood of intraspecific crosses among C.
villosissimus flowers. Despite that floral mechanical and gametic isolation decreased
heterospecific gene flow, sexual isolation was not significant in C. allenii (Ch. 1).
I estimated sexual isolation by the proportion of hybrid formation. This measurement not
only represented the effects of pollinators, floral mechanical structures, and pollen-pistil
interactions on hybrid formation, it also included the potential differences between parents and
hybrids in seed set, seed germination and seedling survival. It is difficult to measure sexual
isolation independent of these early-acting intrinsic postzygotic barriers. While intrinsic
postzygotic isolation was weak between C. allenii and C. villosissimus (see results above), sexual
isolation measured in Chapter 1 was largely due to prezygotic but not postzygotic barriers.

191

Interspecific Seed Set

Reduction in Interspecific seed set can be viewed as an intrinsic postmating, mostly
postzygotic, barrier which acts during hybrid formation (e.g., Kay, 2006) and during the
development of hybrid embryo (e.g., Runions and Owens, 1999; Yasumoto and Yahara, 2008).
Pollen-pistil interaction may reduce the frequency of heterospecific pollen fertilizing the ovules
(e.g., Kay 2006; Escobar-Restrepo et al., 2007), while genetic incompatibility causes hybrid seed
abortion (e.g., Runions and Owens, 1999). Among the four categories of seeds, it is clear that
seed production is higher in the fruits that had C. villosissimus as the maternal parent than those
that had C. allenii as the maternal plant. This is presumably due to species-level differences in
the number of ovules per fruit. For fruits produced with C. allenii as the maternal parent, there
was no difference in seed set between fruits of intra- and interspecific crosses, and there was no
evidence of intrinsic postzygotic isolation during the formation of the hybrids or the
development of hybrid embryo. For fruits produced with C. villosissimus as the maternal parent,
C. villosissimus fruits produced more seeds than F1villo fruits, resulting in significant, but weak
isolation. The results clearly indicate the possibility of making viable F1 hybrid seeds. Isolation
observed in interspecific seed set was absent or weak, allowing me to use the proportion of
hybrid progeny for estimating gametic isolation (Ch. 1). Gametic isolation, which acts prior to
fertilization but after mating, was estimated with the proportion of hybrid progeny in fruits
produced by hand pollination with a 50 interspecific: 50 intraspecific pollen mixture (Ch. 1).
Theoretically, both gametic and early stage intrinsic postzygotic isolation is included in this
estimate. Given that intrinsic postzygotic isolation in seed set is absent or weak, isolation
observed in the estimate of gametic isolation is mainly due to gametic isolation per se. While

192

gametic isolation was included as a component of sexual isolation, using proportion of hybrid
offspring to estimate sexual isolation was also appropriate.
In Costus pulverulentus and C. scaber, Kay (2006) demonstrated severe reduction in seed
set of interspecific crosses. For fruits produced with C. pulverulentus as the maternal parent, the
reduction in hybrid seed set is because the pollen tubes of C. scaber do not grow long enough to
fertilize the ovules of C. pulverulentus. For hybrid fruits with C. scaber as maternal parent,
pollen of C. pulverulentus fail to adhere and germinate on the stigma of C. scaber. In both
species, the reduced hybrid seed formation is a consequence of gametic isolation. However, the
degree of isolation varies across populations: sympatric heterospecific populations are more
isolated from each other than allopatric heterospecific populations (Kay and Schemske, 2008).
The stronger gametic isolation in sympatry suggests that direct selection on pollen-pistil
interactions reinforces speciation in these species. This is very different from the findings in C.
allenii and C. villosissimus since there is no sympatric population of C. allenii and C.
villosissimus for such reinforcement to take place.
F1 Hybrid Transplant Performance

Reciprocal transplant experiments with parents and hybrids demonstrated that F1 hybrids
had reduced fitness at the seed stage in C. villosissimus habitats, but not at other stages or in C.
allenii habitats (Ch. 2). Therefore, extrinsic postzygotic isolation was only significant for C.
villosissimus at the seed stage. Although the importance of local adaption in ecogeographic and
microhabitat isolation was recognized in this study system, local adaptation did not contribute to
extrinsic postzygotic isolation. Unfortunately, plant fecundity and hybrid performance in later
generations were not included in this study. The lack of these potentially important barriers may
result in an underestimation of extrinsic postzygotic isolation.

193

F1 Hybrid Pollen Viability

Hybrid male sterility is an intrinsic postzygotic barrier which has been commonly
examined in speciation studies of animals (e.g., Coyne and Orr, 1989, 1997; Price and Bouvier,
2002). Hybrid pollen inviability, the equivalent measure in plants, is also frequently examined
(e.g.. Fishman and Willis, 2001; Moyle, et al., 2004; Scopece et al., 2008). In a guild of fooddeceptive orchids with weak sexual isolation, significant male sterility reduced hybrid fitness and
caused intrinsic postzygotic isolation (Scopece et al., 2008). However, in C. allenii and C.
villosissimus, there was no evidence of hybrid male sterility. In this study, the average proportion
of viable pollen was only 41%, which was lower than expected. This reduction of pollen viability
across categories may be due to the unnatural humidity conditions in the greenhouse, where the
sampled plants were grown. Hybrid male sterility was also irrelevant to reproductive isolation
between C. pulverulentus and C. scaber (Kay, 2006). This is potentially because C.
pulverulentus and C. scaber are recently diverged and have not had time to accumulate genetic
incompatibility that causes intrinsic postzygotic isolation (Kay and Schemske, 2008).
Total Isolation

The estimates of total isolation demonstrate that C. allenii is highly, but not completely
isolated from C. villosissimus (T = 0.924), which is in turn almost completely isolated from C.
allenii (T = 0.999). Total isolation was also estimated by replacing the barriers from microhabitat
isolation to F1 juvenile growth in parental habitats with observed hybrid frequency in suitable
habitats for both species. In this estimate, the average total isolation (T = 0.988) improved, but
was still not complete. As C. allenii and C. villosissimus are good species, some barriers might
have been underestimated, which lead to the estimate of incomplete total isolation. One barrier
which might have been underestimated is sexual isolation, which was measured as the proportion

194

of hybrid offspring produced by naturally pollinated flowers in a pollination array (Ch. 1).
During this process, the amount of pollen deposited on the stigma may be lower than natural
pollen loads. The source of pollen was limited as there were only two flowers per species in the
pollination array and as one of them had its pollen stained with florescent dye and the staining
process caused pollen loss. If few pollen grains were deposited on stigmas, the pollen
germination rate could be reduced (Schemske and Fenster, 1983), and the proportion of
heterospecific fertilization might be underestimated. After seeds were collected, it was difficult
to distinguish the hybrids from the parental seeds and seedlings by their morphology. Instead,
hybrid seedlings were identified with genetic markers. It was assumed that the proportion of
hybrid offspring remained consistent from seed formation, seed viability, seed germination, to
early seedling survival (Ch. 1). However, it is certainly possible that hybrids and parental seeds
differ in these fitness components, which were not measured in this study, and cause isolation.
A second possible explanation of the estimated incomplete total isolation in C. allenii
may be that F1 hybrid transplant performance did not properly represent hybrid fitness in
parental habitats. Because none of the transplants in C. allenii habitats flowered during the
experimental period (Ch. 2), the likelihood of hybrid reproduction was not included as a
postzygotic barrier. In addition, hybrid fitness was studied only in F1s. Previous studies have
shown that F1 hybrids may have better fitness than parental habitats but fitness of hybrids may
be reduced in later generations (Rhode and Cruzan, 2005). Including hybrid reproduction and
hybrid fitness in later generations in the calculation of total isolation may result in a higher value
of total isolation observed in C. allenii, yet measuring these barriers requires a much longer
period of time, which was beyond the scope of this research.

195

Relative Importance of Isolating Mechanisms

Reproductive isolation between C. allenii and C. villosissimus is mainly due to the
difference between their spatial distributions. Ecogeographic isolation was the primary isolating
barrier in C. villosissimus, and this barrier also contribute substantially to total isolation in C.
allenii. While heterospecific gene flow was not completely eliminated by ecogeographic
isolation, microhabitat isolation reduced most of the remaining heterospecific gene flow (Fig.
4.5A). Microhabitat isolation between C. allenii and C. villosissimus was shown to be a direct
consequence of local adaptation based on the finding of zero fitness of parental species
transplanted into foreign habitats (Ch. 2). As local adaptation contributes 100% to ecogeographic
and microhabitat isolation and these two barriers together contribute > 99% to total isolation, I
conclude that local adaptation is the primary mechanism of speciation in C. allenii and C.
villosissimus.
Furthermore, both ecogeographic and microhabitat isolation are estimates of spatial
isolation, which are caused by habitat differences between species. The habitats of C. allenii and
C. villosissimus differ in water and light availability (Ch. 3). Plant physiological response (leaf
mass per area) to low light availability in C. allenii habitats and high drought tolerance to cope
with low water availability in C. villosissimus habitats are putative adaptive traits in these species
(Ch. 3). Given that ecogeographic and microhabitat isolation are the major contributor to total
isolation, reproductive isolation between C. allenii and C. villosissimus is almost all due to
prezygotic barriers. The large contribution of prezygotic isolation found in this study agrees with
the statement that prezygotic barriers play a major role in speciation and that postzygotic
isolation evolves mainly after speciation is complete (Schemske, 2010).

196

Speciation in Costus

Reproductive isolation has been examined in two species pairs of Costus, and in both
cases, prezygotic barriers are more important than postzygotic barriers. Costus pulverulentus and
C. scaber (Kay, 2006) are hummingbird-pollinated species which are partially isolated by its
habitats. As small patches of habitats of C. pulverulentus and C. scaber are often intermingled
with each other, isolating barriers, other than habitat isolation, play important roles in speciation
of these two species. Costus pulverulentus and C. scaber are mostly isolated by the combined
effects of their spatial distribution, floral mechanical isolation, and gametic isolation. This is
different from the case of C. allenii and C. villosissimus, of which habitat isolation contributes
the most to total isolation. While extrinsic postzygotic isolation does not contribute to total
isolation between C. allenii and C. villosissimus, this barrier was not evaluated in C.
pulverulentus and C. scaber. Although both studies recognized that ecological factors contribute
to speciation, the mechanisms were different between species pairs. Natural selection against
parental species in foreign habitats results in a parapatric distribution of C. allenii and C.
villosissimus (Ch. 2), while reinforcement through natural selection of hybrid pollen-pistil
incompatibility strengthen reproductive isolation in sympatric populations of C. pulverulentus
and C. scaber (Kay and Schemske, 2008). Studies of these two species pairs suggest that
ecological factors contribute to reproductive isolation, while the details and adaptive mechanisms
differ between species pairs.
Conclusion

Overall, I found that strong ecogeographic and microhabitat isolation are the major
isolating barriers between C. allenii and C. villosissimus. With the findings of the previous
chapters, I demonstrated how ecological factors contribute to adaptive divergence and ultimately

197

lead to reproductive isolation. Here I conclude that local adaptation is the primary mechanism of
speciation in this system. Future studies of taxa with varying genetic distance are needed to
compare the relative importance of pre- and postzygotic barriers and to determine whether local
adaptation is the general mechanism of reproductive isolation at the time of speciation.

198

Table 4.1. Climatic variables used in ecological niche modeling. All variables were from the
publicly available dataset WORLDCLIM (www.worldclim.org). A quarter is a three-month
period.

Code
BIO1
BIO2
BIO4
BIO5
BIO6
BIO7
BIO12
BIO15
BIO16
BIO17
BIO18
BIO19

Description
Annual mean temperature (°C)
Mean diurnal range (mean of monthly (maximum temp – minimum temp))
Temperature seasonality (standard deviation * 100)
Maximum temperature of the warmest month
Minimum temperature of the coldest month
Temperature annual range (BIO5 – BIO6)
Annual precipitation
Precipitation seasonality (coefficient of variation)
Precipitation of wettest quarter
Precipitation of driest quarter
Precipitation of warmest quarter
Precipitation of coldest quarter

199

Table 4.2. Estimates of the studied reproductive isolating barriers between C. allenii and C. villosissimus. The proportion of hetero(H) and conspecific gene flow (C) causing the strength (RI) of individual barriers and the absolute contribution (AC) to total isolation
(T) are listed for each species. *: H and C of the isolating barrier were calculated from RI values.

Isolating barriers
Ecogeographic isolation*
Microhabitat isolation*
Temporal isolation*
Sexual isolation
Interspecific seed set
F1 seed in parental habitats
F1 juvenile survival in parental habitats
F1 juvenile growth in parental habitats
F1 pollen viability
Total

C. allenii
H
C
0.43
0.57
0.06
0.94
0.48
0.52
0.44
0.56
0.48
0.52
0.47
0.53
0.46
0.54
0.51
0.49
0.53
0.47
0.00017 0.00441

RI
AC
0.142
0.142
0.886
0.771
0.045
0.007
0.121
0.017
0.050
0.006
0.067
0.007
-0.129
-0.014
-0.014
-0.002
-0.069
-0.009
T = 0.924

200

C. villosissimus
H
C
RI
AC
0.03
0.97
0.948
0.948
0.03
0.97
0.932
0.050
0.49
0.51
0.027
<0.001
0.21
0.79
0.581
0.001
0.44
0.56
0.112
<0.001
0.27
0.73
0.463
<0.001
0.62
0.38
-0.240
-0.000
0.75
0.25
-0.493
-0.000
0.46
0.54
0.074
<0.001
0.000002 0.00804
T = 0.999

unsuitable
suitable for C. allenii
suitable for C. villosissimus
suitable for both species

Figure 4.1. Overlay map showing ecogeographic isolation between C. allenii and C. villosissimus. The ecological niche models show
that 9,769 pixels are suitable for C. allenii alone, 1,077,663 pixels were suitable for C. villosissimus alone, and 58,827 pixels are
suitable for both species. The sizes of the shared and unshared suitable habitats results in RIecogeo = 0.142 for C. allenii and RIecogeo
= 0.948 for C. villosissimus.

201

proportion of total flowers

A.

0.03

0.02

0.01

0

proportion of total flowers

135

B.

C. allenii
C. villosissimus

2007

145

155

165

175

185

195

205

215

225

235

245

155

165

175

185

195

205

215

225

235

245

0.03

2008
0.02
0.01
0
135

145

Julian Dates

Figure 4.2. Flowering phenology of C. allenii and C. villosissimus in 2007 (A) and 2008 (B).
There was no significant difference between the flowering phenology in neither year (p > 0.05).
RItemporal = 0.028 for C. allenii and 0.019 for C. villosissimus in 2007; RItemporal = 0.061 for C.

allenii and 0.035 for C. villosissimus in 2008. The average RItemporal of 2007 and 2008 was
0.045 for C. allenii and 0.027 for C. villosissimus.

202

no. of seeds per fruit

100
90
80
70
60
50
40
30
20
10
0

c
b
a

C. allenii

a

F1allenii

F1villo

C. villosissimus

Figure 4.3. Comparisons of seed set among C. allenii, F1allenii, F1villo, and C. villosissimus.
Error bars represent mean ± 95% CI. Bars with different letters represent means that are
significantly different at the level of α = 0.05 based on Tukey’s HSD tests.

203

p ro p o rtio n p o llen g ermin atin g

0.6
0.5
0.4
0.3
0.2
0.1
0
AA
C. allenii

AV
F1allenii

VA
F1villo

VV
C. villosissimus

Figure 4.4. Comparisons of proportion of pollen germinating, representing male fertility, among
C. allenii, F1allenii, F1villo, and C. villosissimus. Error bars represent mean ± 95% CI. There was
no significant difference among categories (p > 0.05).

204

0.4
0.2
0

F1 pollen viability -

F1 juvenile growth-

F1 juvenile surv-

F1 seed fitness -

Intersp. seed set -

Postzygotic
Sexual -

Temporal -

Microhabitat -

-0.2

Prezygotic

0.6
0.4
0.2
0
-0.2

F1 pollen viability -

0.6

0.8

Hybrid frequency -

C. allenii
C. villosissimus
average of the two species

0.8

1

Ecogeographic -

1

Relative contribution to total reproductive isolation

B.

Ecogeographic -

Relative contribution to total reproductive isolation

A.

Figure 4.5. Relative contribution of isolating barriers to total isolation in C. allenii and C. villosissimus. (A) Relative contribution of
each barrier. (B) Relative contribution of ecogeographic isolation, hybrid frequency in suitable habitats for both species, and F1
hybrid pollen viability to total isolation, based on species averages.

205

LITERATURE CITED

206

Angert, A. L., and D. W. Schemske. 2005. The evolution of species’ distributions: reciprocal
transplants across the elevation ranges of Mimulus cardinalis and M. lewisii. Evolution
59(8): 1671-1684.
Bolnick, D. I. and T. J. Near. 2005. Tempo of hybrid inviability in centrarchid fishes (Teleostei:
Centrarchidae). Evolution 59(8): 1754-1767.
Bristow, K. J., and G. S., Campbell. 1984. On the relationship between incoming solar radiation
and daily maximum and minimum temperature. Agricultural and Forest Meteorology 31:
159-166.
Chen, J., J. F. Franklin, and T. A. Spies. 1993. Contrasting microclimates among clearcut, edge,
and interior of old-growth Douglas-fir forest. Agricultural and Forest Meteorology 63:
219-237.
Coyne, J. A., and H. A. Orr. 1989. Patterns of speciation in Drosophila. Evolution 43(2): 362-381.
Coyne, J. A., and H. A. Orr. 1997. “Patterns of speciation in Drosophila” revisited. Evolution
51(1): 295-303.
Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer, Sunderland, MA.
Cruzan, M. B., and M. L. Arnold. 1994. Assortative mating and natural selection in an Iris
hybrid zone. Evolution 48(6): 1946-1958.
Dobzhansky, T. 1937. Genetics and the origin of species. Columbia Univ. Press, New York.
Dopman, E. B., P. S. Robbins, and A. Seaman. 2010. Components of reproductive isolation
between North America pheromone strains of the European corn borer. Evolution 64(4):
881-902.
Dressler, R. L. 1982. Biology of the orchid bees (Euglossini). Annual Review of Ecology and
Systematics 13: 373-394.
Escobar-Restrepo, J-M., N. Huck, S. Kessler, V. Gagliardini, J. Gheyselinck, W-C. Yang, and U.
Grossniklaus. 2007. The FERONIA receptor-like kinase mediates male-female
interactions during pollen tube reception. Science 317: 656-660.
Fine, P. V. A., D. C. Daly, G. V. Muñoz, I. Mesones, and K. M. Cameron. 2005. The
contribution of edaphic heterogeneity to the evolution and diversity of Burseraceae trees
in the western Amazon. Evolution 59(7): 1464-1478.
Fishman, L., and J. H. Willis. 2001. Evidence for Dobzhansky-Muller incompatibilities
contributing to the sterility of hybrids between Mimulus guttatus and M. nasutus.
Evolution 55:1932-1942.

207

Hatfield, T., and D. Schluter. 1999. Ecological speciation in sticklebacks: environmentdependent hybrid fitness. Evolution 53(3): 866-873.
Hendry, A. P., P. Nosil, and L. H. Rieseberg. 2007. The speed of ecological speciation.
Functional Ecology 21: 455-464.
Husband, B. C., and H. A. Sabara. 2003. Reproductive isolation between autotetraploids and
their diploid progenitors in fireweed, Chamerion angustifolium (Onagraceae). New
Phytologist 161: 703-713.
Janzen, D. H. 1971. Euglossine bees as long-distance pollinators of tropical plants. Science
171(3967): 203-205.
Kay, K. M. 2006. Reproductive isolation between two closely related hummingbird-pollinated
Neotropical gingers. Evolution 60(3): 538-552.
Kay, K. M., P. A. Reeves, R. G. Olmstead, and D. W. Schemske. 2005. Rapid speciation and
evolution of hummingbird pollination in Neotropical Costus subgenus Costus (Costaceae):
evidence from nrDNA ITS and ETS sequences. American Journal of Botany 92(11):
1899-1910.
Kay, K. M., and D. W. Schemske. 2008. Natural selection reinforces speciation in a radiation of
Neotropical rainforest plants. Evolution 62(10): 2628-2642.
Lowry, D. B., J. L. Modliszewski, K. M. Wright, C. A. Wu, and J. H. Willis. 2008. The strength
and genetic basis of reproductive isolating barriers in flowering plants. Philosophical
Transactions of the Royal Society: Biological Science 363: 3009-3021.
Maas, P. J. M. 1972. Costoideae (Zingiberaceae). Flora Neotropica, monograph 8. Hafner, New
York, New York.
Mallet, J. 2006. What does Drosophila genetics tell us about speciation? Trends in Ecology and
Evolution 21(7): 386-393.
Martin, N. H., A. C. Bouck, and M. A. Arnold. 2007. The genetic architecture of reproductive
isolation in Louisiana irises: flowering phenology. Genetics 175: 1803-1812.
Martin, N. H., and J. H. Willis. 2007. Ecological divergence associated with mating system
causes nearly complete reproductive isolation between sympatric Mimulus species.
Evolution 61(1): 68-82.
Matsubayashi, K. W., and H. Katakura. 2009. Contribution of multiple isolating barriers to
reproductive isolation between a pair of phytophagous ladybird beetles. Evolution 63(10):
2563-2580.
Mayr, E. 1942. Systematics and the origin of species. Columbia University Press, New York.

208

Mayr, E. 1947. Ecological factors in speciation. Evolution 1: 263-288.
Mayr, E. 1963. Animal species and evolution. Harvard University Press, Cambridge, MA.
McCartney, M. A., and H. A. Lessios. 2004. Adaptive evolution of sperm bindin tracks eggs
incompatibility in Neotropical Sea Urchins of the genus Echinometra. Molecular Biology
and Evolution 21(4): 732-745.
Mendelson, T. C. 2003. Sexual isolation evolves faster than hybrid inviability in a diverse and
sexually dimorphic genus of fish (Percidae: Etheostoma). Evolution 57(2): 317-327.
Moyle, L. C., M. S. Olson, and P. Tiffin. 2004. Patterns of reproductive isolation in three
angiosperm genera. Evolution 58(6): 1195-1208.
Nagel, L. and D. Schluter. 1998. Body size, natural selection, and speciation in sticklebacks.
Evolution 52(1): 209-218.
Nosil. P., T. H. Vines, and D. J. Funk. 2005. Reproductive isolation caused by natural selection
against immigrants from divergent habitats. Evolution 59(4): 705-719.
Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximum entropy modeling of species
geographic distributions. Ecological Modeling 190: 231-259.
Price, T. D., and M. M. Bouvier. 2002. The evolution of F1 postzygotic incompatibilities in birds.
Evolution 56(10): 2083-2089.
Ramsey, J., H. D., Bradshaw, Jr., and D. W. Schemske. 2003. Components of reproductive
isolation between the monkeyflowers Mimulus lewisii and M. cardinalis (Phrymaceae).
Evolution 57(7): 1520-1534.
Rhode, J. M., and M. B. Cruzan. 2005. Contributions of heterosis and epistasis to hybrid fitness.
American Naturalist 166(5): E124-E139.
Rundle, H. D., and P. Nosil. 2005. Ecological Speciation. Ecology Letters 8: 336-352.
Runions, C. J., and J. N. Owens. 1999. Sexual reproduction of interior spruce (Pinaceae). II.
Fertilization to early embryo formation. International Journal of Plant Sciences 160(4):
641-652.
Schemske, D. W. 2000. Understanding the origin of species. Evolution 54(3): 1069-1073.
Schemske, D. W. 2010. Adapation and the Origin of Species. American Naturalist 176(S1): S4S25.
Schemske, D. W., and C. Fenster. 1983. Pollen-grain interactions in a Neotropical Costus: effects
of clump size and competition. Pp. 405-410 In Pollen: Biology and Implications for Plant

209

Breeding by D. L. Mulcahy and E. Ottaviano (eds.). Elsevier Science Publishing
Company, Inc., New York.
Schluter, D. 2001. Ecology and the origin of species. Trends in Ecology and Evolution 16(7):
372-380.
Scopece, G., A. Widmer, and S. Cozzolino. 2008. Evolution of postzygotic reproductive
isolation in a guild of deceptive orchids. American Naturalist 171: 315-326.
Searcy, K. B., and M. R. Macnair. 1990. Differential seed production in Mimulus guttatus in
response to increasing concentrations of copper in the pistil by pollen from copper
tolerant and sensitive sources. Evolution 44(6): 1424-1435.
Sobel, J. M. 2010. Speciation in the western North American wildflower genus Mimulus. PhD
diss. Michigan State University, East Lansing.
Sobel, J. M., G. F. Chen, L. R. Watt, and D. W. Schemske. 2010. The biology of speciation.
Evolution 64:295–315.
Wikelski, M., J. Moxley, A. Eaton-Mordas, M. M. Lopez-Uribe, R. Holland, D. Moskowitz, D.
W. Roubik, and R. Kays. 2010. Lange-range movements of Neotropical orchid bees
observed via radio telemetry. PLoS ONE 5(5): e10738.
Yasumoto, A. A., and T. Yahara. 2008. Reproductive isolation on interspecific backcross of F1
pollen to parental species, Hemerocallis fulva and H. citrina (Hemerocallidaceae).
Journal of Plant Research 121: 287-291.

210