REPORTS
Pollinator-Mediated Selection on Flower
Color Allele Drives Reinforcement
Robin Hopkins* and Mark D. Rausher
Reinforcement is the process by which reduced hybrid fitness generates selection favoring
the evolution of stronger prezygotic reproductive barriers between emerging species. Using
common-garden field experiments, we quantified the strength of reinforcing selection in nature
by demonstrating strong selection favoring an allele conferring increased pigment intensity in
the plant Phlox drummondii in areas of sympatry with the closely related species Phlox cuspidata.
Incomplete hybrid sterility between the two species generates selection for traits that decrease
interspecies hybridization. In contrast, selection on this locus is undetectable in the absence of
P. cuspidata. We demonstrate that reinforcing selection is generated by nonrandom pollinator
movement, in which pollinators move less frequently between intensely pigmented P. drummondii
and P. cuspidata than between lightly pigmented P. drummondii and P. cuspidata.
R
Department of Biology, Box 90338, Duke University, Durham,
NC 27708, USA.
*To whom correspondence should be addressed. E-mail:
robin.hopkins@duke.edu
1090
(F3′5′h) alters the anthocyanin pigment composition of flowers and changes them from blue
to red. At this hue locus, the ancestral “blue”
allele (H) is dominant to the derived “red” allele
(h). Up-regulation of an R2R3-Myb transcription factor increases the amount of pigments
produced, resulting in increased color intensity.
At this intensity locus, the derived “dark” allele
(I) is dominant to the ancestral “light” allele (i).
Western, allopatric P. drummondii populations are
fixed for the i and H alleles, whereas eastern,
sympatric populations are fixed or nearly fixed for
the I and h alleles, and the two recombinant flower colors, light-red (iihh) and dark-blue (I-,H-),
occur only near the boundary between allopatric
and sympatric populations (15).
Patterns of neutral genetic variation across the
range of P. drummondii suggest extensive gene
flow between allopatric and sympatric populations, indicating that natural selection and not
genetic drift is likely responsible for the geographic pattern of flower color variation (15). To
determine whether selection in sympatry is due
primarily to environmental factors acting directly
on flower color variation, rather than to effects of
reinforcement, we performed a common-garden
field experiment designed to detect selection in the
absence of P. cuspidata. We measured average fitness of the four flower-color double-homozygote
genotypes in their natural habitat. Three generations
of crosses were performed to produce seeds of
known flower-color genotype and to randomize the
genetic background of loci unlinked to the two
flower-color loci (19). For clarity, we will refer to the
homozygous color genotypes by their corresponding flower color throughout the remainder of this
paper. A total of 2720 seeds were planted in a randomized block design, with 170 individuals per
genotype per block, at the University of Texas
Stengl research station (Smithville, Texas). This
station is located within the sympatric region of
P. drummondii and P. cuspidata and contains
natural populations of both species (19).
No significant differences in survival or reproductive success among the flower-color genotypes were observed (table S2). We noted that
2 MARCH 2012
VOL 335
SCIENCE
Fig. 1. Fitness components for each flower color
genotype. (A) Survival probability (n = 1909). (B)
Fruit production (n = 1240). (C) Fitness (n =
1909). (D) Relative hybridization rate (n = 931).
Bars indicate 1 SE.
www.sciencemag.org
Downloaded from http://science.sciencemag.org/ on August 8, 2020
einforcement is the evolution of increased
prezygotic reproductive isolation due to
selection favoring decreased hybridization between diverging groups of individuals or
emerging species (1–4). A. R. Wallace first proposed that selection against hybrids might favor
the evolution of novel prezygotic isolating barriers (subsequently termed the Wallace effect) in
1889 (5). Although this idea has been controversial, recent theoretical and empirical work suggests
that reinforcement may often play an important
role in increasing reproductive isolation in nature
(1, 3, 4, 6, 7). However, the magnitude of reinforcing selection in nature is generally unknown,
as are the genes upon which such selection acts.
Theoretical models have demonstrated that direct
environmental selection can be more effective in
influencing trait evolution than reinforcing selection (6, 8–11), but previous investigations of reinforcement have rarely differentiated between
these two types of selection [but see (12, 13)].
Flower color variation in Phlox drummondii
has been hypothesized to be an example of reinforcement (14). The geographic range of this
species partly overlaps with that of a congener,
P. cuspidata, in eastern Texas. Both species have
the light-blue (sometimes called violet or pink)
flower color characteristic of most Phlox species in allopatric areas of their ranges, whereas
P. drummondii has dark-red flowers in regions
sympatric with P. cuspidata (15). In the region of
sympatry, populations of the two species frequently grow in close proximity; produce hybrids
in nature that have high, but not complete, ovule
and pollen sterility; and exhibit some interspecific
gene flow (16, 17). In P. drummondii, the difference between the ancestral light-blue flower
color and the derived dark-red flower color is
caused by mutations in the cis-regulatory regions
of two genes (18). Down-regulation of the gene
coding for the enzyme Flavonoid 3′5′-hydroxylase
survival was slightly lower for two derived genotypes (dark-blue and dark-red), compared with the
ancestral genotype (light-blue), whereas it was
slightly higher for the derived genotype light-red
(Fig. 1A). The number of fruits produced was
slightly higher for all three derived genotypes compared with light-blue (Fig. 1B), but these differences were also not statistically significant (table
S3). There were no detectable differences among
genotypes for number of seeds per fruit (table S4).
Female fitness, the product of survival and fruit
production, was also slightly higher for the derived
genotypes, compared with light-blue genotype
(Fig. 1C), but again none of these differences were
statistically significant (19). Overall, we did not
detect environmental effects acting directly on
flower color favoring the derived allele at either the
hue or the intensity locus in the area of sympatry.
To examine whether reinforcing selection generated by hybridization with P. cuspidata favors
the derived allele at either flower-color locus, we
established blocks consisting of 30 plants of one
of the double-homozygous genotypes (“focal
plants”), as well as 105 light-blue plants of a
stock line. We followed 30 of the light-blue stock
individuals as “reference plants” to control for environmental variation among blocks. In addition,
we planted 115 P. cuspidata plants in each block.
We collected fruits from reference and focal plants
and randomly chose 100 to 150 seeds from each
REPORTS
likely resulted in the subsequent spread of the
dark allele throughout the sympatric region. To
our surprise, there appeared to be no effect of the
red allele at the hue locus on hybridization, even
though this allele is fixed in sympatric populations. Taken together with the lack of difference
in fitness among genotypes in the first experiment,
the difference in hybridization rates between dark
and light plants supports the hypothesis that reinforcing selection is responsible for the fixation
of the dark allele in sympatric populations.
Manual pollen transfers indicate that there is
no difference between light and dark flowered
plants in fertilization success of P. cuspidata pollen
(14). These data suggest that dark-flowered individuals are not less compatible with P. cuspidata
pollen than light-flowered individuals. It is therefore likely that nonrandom patterns of pollinator
visitation between Phlox species with different
flower colors explain our observed variation in
hybridization. We examined patterns of pollinator
visitation in experimental arrays to test this hypothesis. We constructed three arrays containing
light-blue plants, P. cuspidata plants, and either
light-red, dark-red, or dark-blue focal plants (19).
We observed a total of 181 pollinators making a
total of 2301 transitions between plants.
The primary visitors to both Phlox species in
these arrays, as in natural populations, were Battus
philenor butterflies (108 individuals observed)
and various species of skippers (Lepidoptera,
family Hesperiidae) (73 individuals observed)
(table S6a). Both types of pollinators displayed
Table 1. The percentage of transitions by pollinators across flower types. (A) Pollinator movement
between colors and species within arrays containing light-red P. drummondii. (B) Pollinator movement between colors and species within arrays containing dark-blue P. drummondii. (C) Pollinator
movement between colors and species within arrays containing dark-red P. drummondii. The percentages of transitions between P. cuspidata and P. drummondii plants are in bold.
A Pollinators in arrays with the light-red P. drummondii moved
To (%)
From
Light-red (252)
Light-blue (219)
P. cuspidata (185)
Light-red
Light-blue
P. cuspidata
32
50
34
44
22
31
24
28
35
B Pollinators in arrays with the dark-blue P. drummondii moved
From
Dark-blue (222)
Light-blue (198)
P. cuspidata (274)
Dark-blue
To (%)
Light-blue
P. cuspidata
51
42
7
38
18
32
11
40
61
Dark-red
To (%)
Light-blue
P. cuspidata
37
37
9
50
25
33
13
38
58
C Pollinators in arrays with the dark-red P. drummondii moved
From
Dark-red (255)
Light-blue (335)
P. cuspidata (378)
www.sciencemag.org
SCIENCE
VOL 335
similar movement patterns and visited both Phlox
species extensively (table S6b). In arrays with lightred plants, there is no evidence of pollinator constancy, as measured by the Bateman’s Constancy
Index (19, 20) (table S6c). Pollinators were equally
likely to visit light-blue and light-red plants after
visiting P. cuspidata (Table 1A) (19). This pattern is
consistent with finding no difference in hybridization rates between these two genotypes. In addition,
pollinators were equally likely to visit a P. cuspidata
plant after visiting a light-blue or a light-red plant,
which suggested that pollen wastage through interspecific fertilization does not differ between
these two genotypes (Table 1A) (19).
In contrast, pollinators in arrays containing
either dark-blue or dark-red plants exhibited a
significantly higher species-level Bateman Constancy Index for dark-flowered genotypes compared with light-flowered genotypes (table S6c).
In particular, pollinators were half as likely to
visit dark plants as light-blue plants after visiting
a P. cuspidata—a pattern that explains the reduced hybridization observed in plants with dark
pigmentation (Table 1, B and C) (19). Pollinators were also substantially less likely to visit a
P. cuspidata after visiting a dark-blue or dark-red
plant than after visiting a light-blue plant (Table
1, B and C) (19), which suggested that darkly
pigmented plants waste less pollen on interspecific pollination. Although we did not directly
measure male fitness in our field experiments,
this observation indicates that the dark allele may
significantly increase male fitness, in addition to
female fitness, relative to the light allele. It is
possible that pollinators could be responding to
pleiotropic effects of the intense allele (e.g., nectar volume or concentration), but this seems unlikely given the visual orientation of the primary
pollinator, B. philenor (21).
Our investigations provide no explanation for
why sympatric populations of P. drummondii
have evolved red flowers. Although previously
we used patterns of genetic variation at this locus
and at neutral markers to show that natural selection drove the fixation of the red (h) allele in
the region of sympatry (15), in the current study,
we detected neither fitness differences nor differences in levels of interspecific hybridization between genotypes at the hue locus. One possible
explanation for these contrasting results is that
selection due to environmental factors favors the
red allele in sympatry but that the magnitude of
this selection is too small to detect given the power
of our analysis (as may be evidenced by Fig. 1C).
A second possibility is that environmental selection
operates only intermittently on this locus and was
absent during our experiments. Finally, a third possibility is that a selective agent, such as another type
of pollinator, generated selection in the past on
the hue locus but is no longer present. Although
hitchhiking by the red allele in a selective sweep
involving a closely linked gene is a formal possibility, this seems unlikely because it would have
required that the favored mutation arose on a
rare haplotype carrying the red allele.
2 MARCH 2012
Downloaded from http://science.sciencemag.org/ on August 8, 2020
focal and reference genotype in each block to
genotype and determine whether the paternal
parent was P. drummondii or P. cuspidata (19).
Using this paternity test, we calculated the hybridization rate for focal and reference genotypes
within each block.
Across the four blocks, the hybridization rate
(proportion of seeds sired by P. cuspidata) varied
between 28 and 44% for the light-blue reference
plants, which indicated substantial overall interspecific hybridization. The hybridization rates of
the light-blue and light-red focal plants were similar to those of their respective reference plants
(Fig. 1D and table S5, a and b). In contrast, the
hybridization rates of dark-blue and dark-red
focal plants were more than 50% lower than the
reference plants (Fig. 1D and table S5, a and b).
Thus, we conclude that the dark allele (I) at the
intensity locus significantly decreases hybridization between P. drummondii and P. cuspidata.
Given conservative empirical estimates of hybrid
sterility of ~90% (17) and the average light-blue
reference plant hybridization rate of 0.43, the
reduction in hybridization translates into a selection coefficient of 0.32 favoring the dark allele
(19). The two species of Phlox are commonly
found in intermixed populations with spatial
proximity similar to that in our experiments. In
these populations, the strong reinforcing selection
documented here would increase the frequency
of the dark allele rapidly to fixation. Extensive
gene flow between P. drummondii populations (15),
including those without nearby P. cuspidata, has
1091
REPORTS
1092
Although reinforcement has been studied primarily in animals (3, 7), our work indicates that it
may also be an important contributor to speciation
in plants. If so, this phenomenon may provide a
partial explanation for the tremendous diversity of
floral color, floral morphology, and inflorescence
structure that characterize flowering plants.
References and Notes
1. R. Butlin, Trends Ecol. Evol. 2, 8 (1987).
2. T. Dobzhansky, Genetics and the Origin of Species
(Columbia Univ. Press, New York, 1937).
3. D. J. Howard, in Hybrid Zones and the Evolutionary Process,
R. G. Harrison, Ed. (Oxford Univ. Press, New York, 1993),
pp. 46–69.
4. M. R. Servedio, M. A. F. Noor, Annu. Rev. Ecol. Evol. Syst.
34, 339 (2003).
5. A. R. Wallace, Darwinism: An Exposition of the Theory of
Natural Selection, with Some of Its Applications
(Macmillan, London, 1889).
6. M. Kirkpatrick, V. Ravigné, Am. Nat. 159 (suppl. 3), S22
(2002).
7. D. Ortiz-Barrientos, A. Grealy, P. Nosil, Ann. N. Y. Acad.
Sci. 1168, 156 (2009).
8. M. Kirkpatrick, Proc. R. Soc. London Ser. B 267, 1649
(2000).
9. M. Kirkpatrick, M. R. Servedio, Genetics 151, 865 (1999).
10. M. R. Servedio, Evolution 55, 1909 (2001).
11. M. R. Servedio, Evolution 58, 913 (2004).
12. A. Y. K. Albert, D. Schluter, Evolution 58, 1099 (2004).
13. P. Nosil, B. J. Crespi, C. P. Sandoval, Proc. R. Soc. London
Ser. B 270, 1911 (2003).
14. D. A. Levin, Evolution 39, 1275 (1985).
15. R. Hopkins, D. A. Levin, M. D. Rausher, Evolution 66, 469
(2012).
16. D. A. Levin, Am. J. Bot. 54, 1122 (1967).
17. L. G. Ruane, K. Donohue, Evol. Ecol. 22, 229 (2008).
18. R. Hopkins, M. D. Rausher, Nature 469, 411
(2011).
19. Material and methods are available as supporting
material on Science Online.
20. N. M. Waser, Am. Nat. 127, 593 (1986).
21. M. D. Rausher, Science 200, 1071 (1978).
22. H. D. Bradshaw Jr., D. W. Schemske, Nature 426, 176
(2003).
23. D. R. Campbell, N. M. Waser, E. J. Melendez-Ackerman,
Am. Nat. 149, 295 (1997).
24. M. E. Hoballah et al., Plant Cell 19, 779 (2007).
25. L. Chittka, J. D. Thomson, N. M. Waser, Naturwissenschaften
86, 361 (1999).
26. M. Caisse, J. Antonovics, Heredity 40, 371 (1978).
27. J. Felsenstein, Evolution 35, 124 (1981).
28. L. W. Liou, T. D. Price, Evolution 48, 1451 (1994).
29. M. R. Servedio, Evolution 54, 21 (2000).
Acknowledgments: We thank M. Kirkpatrick, S. Otto,
M. Whitlock, D. Des Marais, and members of the Rausher
and Kirkpatrick laboratory group for advice on this manuscript
and S. Scarpino for statistical consultation. We thank the
University of Texas Stengl Research Station for field
experiment support. This work was supported by NSF grant
0841521 to M.D.R. and a NSF Doctoral Dissertation
Improvement Grant to R.H. and M.D.R. R.H. was supported
by the NSF Graduate Research Fellowship Program. All data
presented here are available in the supporting material.
Supporting Online Material
www.sciencemag.org/cgi/content/full/science.1215198/DC1
Materials and Methods
Figs. S1 and S2
Tables S1 to S9
References
12 October 2011; accepted 12 January 2012
Published online 2 February 2012;
10.1126/science.1215198
Generation of Leaf Shape
Through Early Patterns of Growth
and Tissue Polarity
Erika E. Kuchen,1* Samantha Fox,1* Pierre Barbier de Reuille,2 Richard Kennaway,2
Sandra Bensmihen,1 Jerome Avondo,1 Grant M. Calder,1 Paul Southam,2 Sarah Robinson,1
Andrew Bangham,2† Enrico Coen1†
A major challenge in biology is to understand how buds comprising a few cells can give rise
to complex plant and animal appendages like leaves or limbs. We address this problem through
a combination of time-lapse imaging, clonal analysis, and computational modeling. We arrive
at a model that shows how leaf shape can arise through feedback between early patterns of
oriented growth and tissue deformation. Experimental tests through partial leaf ablation support
this model and allow reevaluation of previous experimental studies. Our model allows a range
of observed leaf shapes to be generated and predicts observed clone patterns in different
species. Thus, our experimentally validated model may underlie the development and evolution
of diverse organ shapes.
T
he shapes of many plant and animal appendages are thought to develop under
the influence of orthogonal organizing
1
John Innes Centre, Norwich Research Park, Norwich, NR4 7UH,
UK. 2School of Computing Sciences, University of East Anglia,
Norwich Research Park, Norwich, NR4 7TJ, UK.
*These authors contributed equally to this work.
†To whom correspondence should be addressed. E-mail:
enrico.coen@jic.ac.uk (E.C.); a.bangham@uea.ac.uk (A.B.)
2 MARCH 2012
VOL 335
SCIENCE
systems (i.e., systems with axes that intersect at
right angles) (1–4). However, it is unclear how
these orthogonal systems lead to changes in tissue shape and how shape changes may themselves
feed back to deform the organizing systems. Consider a square piece of tissue that deforms during
growth (Fig. 1A). The tissue has an initial linear
orthogonal system that organizes the pattern of
morphogenesis (Fig. 1B, arrows). We might en-
www.sciencemag.org
Downloaded from http://science.sciencemag.org/ on August 8, 2020
By measuring reinforcing selection acting on
the dark flower–color allele in P. drummondii under
natural sympatric conditions and by quantifying
selection in the absence of P. cuspidata, we were
able to compare the relative strengths of direct
selection by other environmental factors and
by reinforcing selection on a trait conferring increased premating isolation in a region of sympatry. The absence of detectable fitness differences
among flower color genotypes in the absence of
P. cuspidata indicates that another agent of selection is unlikely to be involved in flower color
divergence in P. drummondii. Although we cannot rule out small, statistically undetectable differences in survival or reproductive success favoring
these genotypes, such differences would be of minor
importance compared with the strong reinforcing
selection acting on the intensity locus.
Many plants have evolved premating reproductive isolation by switching pollinator types
(e.g., from bees to hummingbirds) (22–24). Our
work suggests that increased reproductive isolation
can also be achieved by a single pollinator species
through constancy of individual pollinators. In particular, if pollinators transition between flowers
with similar phenotypes more frequently then between flowers with unlike phenotypes, this will
decrease gene flow between unlike flowers. Constancy is commonly studied in bumble bees but
rarely investigated in butterfly pollinators (20, 25).
That the primary pollinator Battus philenor exhibits this type of constancy is not surprising,
given that females of this species exhibit constancy for leaf shape when searching for oviposition sites (21).
Theoretical models indicate that the likelihood of successful reinforcement is greater when
selection is strong, because this will counteract
gene flow and recombination, which tend to
reduce premating isolation (26–28). Our results
indicate that, at least in some cases, very strong
reinforcing selection may act on a single allele
and lead to increased reproductive isolation.
Theory also indicates that reinforcement is
more easily achieved by a one-allele mechanism
(4, 29), but empirical assessment of this prediction has been difficult because the genetic basis
of reinforcement is understood in few systems
(7). Our current demonstration of reinforcing selection acting on the dark allele indicates that
reinforcement in P. drummondii involves a twoallele reinforcement mechanism. The intensity
locus causes reproductive isolation only if the
dark allele is present in P. drummondii and the
light allele is present in P. cuspidata. Consistent
with theory, we find that strong selection and
high levels of hybrid sterility cause the spread of
the dark allele through sympatric P. drummondii
populations. We suspect all reinforcement mechanisms involving different floral phenotypes to
which pollination vectors must respond will be
two-allele assortative mating mechanisms, because pollinators must be able to discriminate
between the novel phenotype in one species and
the ancestral phenotype in both species.
Pollinator-Mediated Selection on Flower Color Allele Drives Reinforcement
Robin Hopkins and Mark D. Rausher
Science 335 (6072), 1090-1092.
DOI: 10.1126/science.1215198originally published online February 2, 2012
ARTICLE TOOLS
http://science.sciencemag.org/content/335/6072/1090
SUPPLEMENTARY
MATERIALS
http://science.sciencemag.org/content/suppl/2012/02/02/science.1215198.DC1
REFERENCES
This article cites 25 articles, 3 of which you can access for free
http://science.sciencemag.org/content/335/6072/1090#BIBL
PERMISSIONS
http://www.sciencemag.org/help/reprints-and-permissions
Use of this article is subject to the Terms of Service
Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. The title Science is a registered trademark of AAAS.
Copyright © 2012, American Association for the Advancement of Science
Downloaded from http://science.sciencemag.org/ on August 8, 2020
The Constant Pollinator
In ecology, reinforcement is the process by which species prevent hybridization and maintain species boundaries,
but the underlying genetic mechanisms are unclear. Hopkins and Rausher (p. 1090, published online 2 February)
examined reinforcement between two species of a wild flowering plant called Phlox that show incomplete hybrid sterility.
Down-regulation of a flavonoid gene produces red flowers and operates in concert with a color intensity locus to adjust
flower color and tone. A distinct geography of flower color has emerged in which it appears that dark coloration causes
less hybridization between the species because the butterfly pollinators tend to favor light-blue flower color variants. If
pollinators visit flowers with similar phenotypes more frequently than flowers with dissimilar phenotypes, this will decrease
gene flow between the unlike flowers.
www.sciencemag.org/cgi/content/full/science.1215198/DC1
Supporting Online Material for
Pollinator-Mediated Selection on Flower Color Allele Drives
Reinforcement
Robin Hopkins* and Mark D. Rausher
*To whom correspondence should be addressed. E-mail: robin.hopkins@duke.edu
Published 3 February 2012 on Science Express
DOI: 10.1126/science.1215198
This PDF file includes
Materials and Methods
Figs. S1 and S2
Tables S1 to S6
Full References
Other Supporting Online Material for this manuscript includes the following:
(available at www.sciencemag.org/cgi/content/full/science.1215198/DC1)
Table S7. Field Experiment 1 – Fruit-Set and Survival (Excel)
Table S8. Field Experiment 2- Hybridization (Excel)
Table S9. Pollinator Observation (Excel)
Materials and Methods
Growing conditions
In order to induce germination, we soaked all Phlox seeds in 500ppm gibberellic
acid solution for 48 hours, planted them in Metro-Mix 360 (Sun Gro Horticulture,
Bellevue, WA) and stratified them for 7 days at 4°C. Once seeds germinated we grew the
plants in the Duke University Greenhouses (Durham, NC) or the University of Texas at
Austin greenhouses (Austin, TX). We planted experimental seedlings into the field once
the plants started growing true leaves.
Breeding design
We performed a series of crosses to produce seeds of known genotype at the
flower color loci (figure S1). This crossing design randomized the genetic background of
unlinked loci and ensured that experimental seeds of the four double homozygote
genotypes were equally outbred. Seeds from each of the two field experiments were
created from slightly different crossing designs as described below. All genotyping
occurred as described by Hopkins and Rausher (18).
For the first field experiment, in which we tested for ecological selection acting
on flower color alleles, seeds were collected from natural populations in the spring of
2006. P. drummondii seeds were collected from two populations with dark-red flowers
(Poc2104 and 80S1) and two populations with light-blue flowers (Dog95 and 466E7)
(table S1). Individuals from population Dog95 were paired with individuals from
population Poc2104, and individuals from 466E7 were paired with individuals from
80S1. Pairs of individuals were crossed to create F1 individuals and these F1’s were selffertilized to create F2 populations. We created a total of four F2 populations from the
population cross 466E7X80S1 and six F2 populations from the population cross
Dog95XPoc2104. F2 individuals were genotyped at markers in the hue locus (F3’5’h)
and the intensity locus (R2-R3 Myb) in order to identify individuals homozygous at single
nucleotide polymorphisms differentiating the two parents at each locus. F2 populations
created from the same parental source populations were paired for a total of 2 pairs from
F2 populations 466E7X80S1 and 3 pairs from Dog95XPoc2104. Within each pair of F2
populations, individuals with the same homozygous genotype at both flower color loci
were crossed to each other. There were a total of four possible homozygous genotype
combinations within each F2 population pair: light-blue (iiHH), dark-blue (IIHH), lightred (iihh), and dark-red (IIhh). Crosses between paired F2 populations created the seeds
used in the field experiment. This crossing design created five sets of outbred seed
families from two sets of source populations, with known homozygous genotype at the
flower color loci and with randomized genetic background.
The crossing design for the field experiment in which we measured hybridization
had a similar design to the one described above. Individuals from the following
populations were crossed to create F1 individuals: 466E7XPoc2104 and Dog95X80S1
and Dog95XPoc2104. The F1 individuals were self-fertilized to create F2 populations.
F3 experimental seeds were created by crossing F2 individuals homozygous at the flower
color loci from the 466E7XPoc2104 cross to both the Dog95X80S1 F2s, and the
Dog95XPoc2104 F2s.
Light-blue stock lines used as standards in the hybridization experiment were
created by randomly crossing five individuals collected from population 696N1 for two
generations. Offspring from these plants were randomly distributed between blocks. P.
cuspidata seeds that were collected from population 2104circ. Plants were self fertilized
in the greenhouse for two generations and seeds were randomly distributed across blocks.
Planting Procedures
For the first field experiment, which tested for direct selection on flower color,
experimental seeds were planted in a randomized block design across four spatial blocks.
Within each block, an equal number of seeds for each flower color genotype within each
source population were fully randomized. Seeds were allowed to germinate and produce
the first true leaf in the University of Texas at Austin greenhouses (Austin, TX). In early
December 2008, seedlings were planted into four spatially separated blocks at the
University of Texas Stengl research station (Smithville, TX). Within each block,
seedlings were planted 15 cm apart in 8 rows. Seedlings were given supplementary water
every three days for two weeks to enhance transplant survival, after which plants were
left to survive under natural conditions.
The Stengl Field Station (http://www.bfl.utexas.edu/stengl) is located in the
sympatric range of the two Phlox species. This 208 acre preserve is composed of natural
pine forests, and meadows which are managed to encourage the growth of native
herbaceous species. The common garden field experiments were performed in fenced
plots within some of the meadows. Both species of Phlox are found naturally on the field
station property as well as neighboring properties and roadsides. Pollinators have been
observed visiting natural Phlox populations at the station.
Experimental plants overwintered as small vegetative rosettes, grew vegetatively
through February, then began to produce reproductive shoots and flower. They flowered
and set fruit through the end of June at which point they dried-up and died. We surveyed
for overwinter survival at the end of January and continued to monitor survival weekly
for the remainder of the experiment. We counted and removed fruits as they ripened on
each plant and collected a subset of fruits in order to calculate average seed set per fruit.
When almost completely ripe, four to ten fruits were collected on 8 plants of each flower
color genotype from each block for a total of 128 plants, 547 fruits, and 1290 seeds
The second field experiment measuring hybridization rate was planted in a split
plot design with four blocks containing one of the four flower color genotypes as focal
color for the block. Within each block, we planted reference light-blue plants and P.
cuspidata plants as well as focal color plants in a fully randomized design. We planted
individuals 15 cm apart in four rows within each block. All other methods were similar
to those described above. We germinated seeds in the University of Texas at Austin
greenhouses and allowed them to produce the first true leaves prior to transplanting them
in the field in late January, 2010. We provided supplemental water to enhance transplant
survival. Plants were regularly surveyed for survival and fruits were collected to
determine paternity.
Analysis of survival
To determine whether flower color genotypes differed in probability of survival to
flowering, we performed a linear analysis using the procedure Catmod in the SAS v.9.2
statistical package (SAS Institute Inc, Cary, NC) with survival to flower as the dependent
variable. The independent factors were:
(i) block
(ii) L vs. D at the intensity locus
(iii) B vs. R at the hue locus
(iv) source population
Parameters in the model were estimated using maximum likelihood. The
generalized Wald statistic, which is approximately chi-square distributed, is computed to
test hypotheses about linear combinations of parameters. The results are presented in
table S2. Neither the effect of the hue locus nor the effect of the intensity locus nor the
effect of their interaction on survival was significant, indicating no evidence of an effect
of flower color genotype on survival to flowering.
Analysis of fruit production
Fruit production was analyzed using a mixed model Analysis of Variance with the
PROC MIXED procedure in SAS using the method of restricted maximum likelihood.
The fixed-effects in the analysis were intensity genotype, hue genotype, and the
interaction. The random-effects were block and source population (source) and all two
and three-way interactions involving fixed effects. Fruit number was log-transformed
before analysis. Untransformed analysis produced similar results. Neither intensity
genotype, nor hue genotype nor their interaction had a significant effect on fruit number
(table S3a). Significance of random effects was determined by using a likelihood ratio χ2
statistic. First, the complete model was run and second a model was run without one of
the random effects. The difference in the log likelihoods between the two models was
calculated and used to test for the importance of the random effect in the model using a
χ2 distribution with degrees of freedom equal to the difference in the number of
covariance parameters between the two models. This was done for each of the random
effects in succession. Block was the only significant random effect in the model (table
S3b).
Analysis of seeds per fruit
We analyzed number of seeds per fruit on a subset of fruits using a mixed-model
Analysis of Variance as described for the fruit set data above. Neither intensity genotype,
nor hue genotype nor their interaction had a significant effect on number of seeds per
fruit (table S4a). None of the random covariance estimates significantly contributed to
the model fit (table S4b).
Calculation of Fitness
We calculated mean fitness of each genotype by multiplying the probability of
survival by the number of fruits produced. We calculated the variance of mean fitness
from the standard formula for the variance of a product of two random variables:
Var[X· Y] = E[X]2 Var[Y] + E[Y]2 Var[X] + Var[X] Var[Y]
The standard error of fitness is the square root of the variance. The standard errors
overlapped and we therefore judged the mean fitnesses to not be significantly
different.
Genotyping hybrids
We germinated seeds collected from the experimental plants in the common
garden in the Duke University greenhouse. We collected tissue from each individual and
extracted DNA using a modified CTAB extraction protocol (18). An intron length
polymorphism in the gene Anthocyanin synthase (ANS) differentiates the two Phlox
species with P. drummondii having a 10-12 base pair smaller intron than P. cuspidata. A
portion of this intron was amplified using the following two primers:
Forward: GCGGGACATGTCGATTTGGC
Reverse: CCCGTATGGGGAATACATTC
The forward primer was fluorescently labeled with FAM allowing the intron size to be
scored using capillary electrophoresis and fragment analysis on an ABI 3730x1 DNA
Analyzer. Intron size was scored by eye using the program GENE MARKER
(SoftGenetics, 2005, State College, PA). All offspring contained a maternal allele from
P. drummondii and a paternal allele from either P. drummondii or P. cuspidata.
Analysis of hybridization rate
We analyzed variation in hybridization rates between different flower color
genotypes using bootstrapping. The following method accounted for variation in mother
and offspring sample size and for between-block variation in hybridization rate. We first
estimated hybridization rate of each genotype in each block and then tested specific
hypotheses about variation in hybridization rates.
Estimating hybridization rates. Using bootstrapping we calculated standard errors around
our estimates of hybridization rate for each genotype. This involved sampling with
replacement mothers from within each genotype and block. Once mothers were sampled
we sampled offspring from that mother with replacement. We then calculated the
hybridization rate for that genotype to create a single bootstrap replicate estimate. We
repeated this 10,000 times and determined the standard error by calculating the standard
deviation of the bootstrap replicate estimates. We performed this for all 8 genotypes (4
focal colors and 4 reference) (table S5a.). In order to control for between-block variation,
we standardized the focal plant hybridization rates by dividing by the hybridization rate
of the reference plants in that block. Specifically, we divided the vector of 10,000
bootstrap replicate estimates of the focal color by the vector of 10,000 bootstrap replicate
estimates of the corresponding reference for the block. This gave us 10,000 bootstrap
replicate estimates of the standardized hybridization of each flower color genotype. We
calculated a new standard error for this estimate of relative hybridization (table S5a).
Hypothesis testing. We used bootstrap re-sampling to test specific hypotheses about the
effects of the flower color loci on hybridization rate. Specifically, we tested whether
genotype at the intensity locus affected hybridization, if genotype at the hue locus
affected hybridization, and if there was an interaction between genotypes at the hue and
intensity locus. We first calculated the observed value for each of these comparisons. In
order to control for variation between blocks we again standardized the hybridization
rates of the focal plants by dividing by that of the reference plants.
The first comparison of interest is the effect of the intensity locus:
Main Effect of Intensity (MI) = ½ [(DB + DR) – (LB + LR)]
(1)
(DB=relative hybridization of dark-blue, DR=relative hybridization of dark-red,
LB=relative hybridization of light-blue, LR=relative hybridization of light-red.)
The second comparison estimates the effect of the hue locus:
Main Effect of Hue (MH) = ½ [(DB + LB) – (DR + LR)]
(2)
And the final comparison estimates the effect of the interaction between loci.
Interaction Effect (IE) = ½ [(DB + LR) – (DR + LB)]
(3)
For each of these comparisons we calculated the observed value from the field collected
data. We then performed re-sampling with replacement from pooled data to determine the
distribution under the null hypothesis that there is no effect of either locus nor an
interaction. In order to do this we first created a “focal pool” with all the focal mother
plants from all of the four blocks and a “reference pool” with all the reference mothers
from all of the four blocks. From these pools we drew our re-sampled dataset. This
involved 3 steps:
1. Sample random mothers (with replacement) for each focal color from the
“focal pool”. The sample size was the same as the observed sample size for
each focal genotype.
2. Sample random mothers (with replacement) for the standards from the
“standard pool”. Again, the sample sizes for each standard set was the same
as observed in each block.
3. For each mother in each group of focal and standard samples we re-sampled
the offspring with replacement. Each sampled offspring maintained its
identity as a hybrid or not. Re-sampled offspring sample sizes were the same
as observed for each mother.
Once we had a re-sampled dataset for each of the 4 focal and 4 reference genotypes, we
calculated a sample relative hybridization rate for each focal color. Sample comparisons
(equations 1-3 above) were calculated with the sample relative hybridization rates.
We re-sampled the pooled dataset 10,000 times. Each time we calculated the above three
comparisons to create the null distribution for each comparison. Finally we determined
where the observed value for each comparison fell in the null distribution and calculated
a P-value as the proportion of bootstrap values that were greater than or equal to the
observed value (table S5b). Because we are testing the null hypothesis against the a
priori alternative hypothesis that the derived allele has a lower hybridization rate, the test
is one-tailed.
Selection. We calculated the magnitude of selection on allelic variation at the intensity
locus from the hybridization rates in the dark-red and the dark-blue focal color plots.
Previous work has shown that hybrids are approximately 90% sterile (16, 17), therefore
fitness was estimated as (1-0.9hkk), where hkk equals the hybridization rate for genotype
kk. Average hybridization of light-blue reference plants across both dark intensity plots
equaled 0.43, and average hybridization of focal plants in both dark plots equaled 0.21.
The selection coefficient was calculated as (1-0.9 hII)/(1-0.9 hii) -1 = (1 – 0.9x0.21)/(1 –
0.9 x 0.43) - 1 = 0.32.
Pollinator Array Experiments
Array Design
The pollinator observation arrays consisted of three plant types each: P. cuspidata
individuals were alternated with P. drummondii individuals and the P. drummondii
individuals, alternated light-blue flower color genotype with one focal color (a derived
flower color genotype). Each array had 20 P. drummondii individuals (10 of each of two
flower color genotypes), and 35 P. cuspidata plants (figure S.2). In total we had three
arrays, one included light-red individuals (iihh) to quantify the effect of the hue locus on
pollinator movement, one included dark-blue individuals (IIHH) to quantify the effect of
the intensity locus on pollinator movement, and the third included dark-red individuals
(IIhh) to determine if there is an interaction between loci effecting pollinator movement
between Phlox species. Each day the total number of open flowers for each P.
drummondii genotype were manipulated to be equal within each array. There were
approximately the same number of P. cuspidata flowers as P. drummondii flowers (Mean
ratio of cuspidata/drummondii=0.988, SD=0.285). We observed movement of 181
pollinators making a total of 2301 transitions between plants, and used these data to
calculate the transition rate between species for each flower color genotype.
Observations were performed over 10 days during May 2011, between 10am and 3pm on
sunny or mostly-sunny days. This time of year corresponded to when natural Phlox
populations experienced high concentrations of blooms. Although butterflies were seen
foraging earlier then 10am and later then 3pm, most activity tended to be concentrated in
the middle of the day.
Analysis of Pollinator Movements
The number of transitions between different genotypes in the pollinator arrays are
given in table S6a. Two comparisons were performed: (1) proportion of visits to lightblue vs. focal color (light-red, dark-red, or dark-blue) following a visit to P. cuspidata;
and (2) proportion of visits from light-blue vs. light-red, dark-red, or dark-blue to P
cuspidata.
(1) Proportion of visits to light-blue vs. focal plants following a visit to P. cuspidata.
We first tested for heterogeneity between Battus philenor and skippers in the
proportion of visits to light-blue and to focal plants using a G-test (table S6b). There was
no significant heterogeneity for either the light-red or dark-blue arrays, so for these arrays
data for Battus and skippers were combined for analysis. There was significant
heterogeneity for the dark-red array, likely due to the small number of skippers observed
visiting dark-red from P. cuspidata; therefore, Battus and skippers were analyzed
separately.
To test whether the proportion of visits to light-blue vs. focal color differed
following a visit to P. cuspidata we used a G-test to ask whether the numbers of visits
from P. cuspidata differed from equality using maximum likelihood. We compared the
expected number of visits to focal color and light-blue from P cuspidata to the predicted
number of visits given equal visitation to both colors.
In the light‐red arrays, pollinators were equally likely to visit light‐blue and
light‐red after visiting a P. cuspidata plant (χ21 = 0.207, n.s.). In the dark‐blue arrays,
pollinators were significantly less likely to visit a dark‐blue than a light‐blue after
visiting a P. cuspidata plant (χ21 = 46.22, P<0.001). In the dark‐red arrays, both
Battus and skippers were significantly less likely to visit a dark‐red than a light‐blue
after visiting a P. cuspidata plant (χ21 = 30.3, P<0.001 and χ21 = 29.84, P<0.001,
respectively).
(2) Proportion of visits from light-blue vs. focal plant to P cuspidata.
We first tested for heterogeneity between Battus philenor and skippers using a 3way G-test (table S.6c). In this analysis, the three factors were (i) pollinator (Battus vs.
skippers), (ii) initial visit to light-blue vs. focal color (light-red, dark-red, or dark-blue),
and (iii) Number of visits to P. drummondii (light-blue plus focal color) vs. to P.
cuspidata. Significant heterogeneity is indicated by the 3-way interaction term. There
was no significant heterogeneity for any of the three arrays, so data for Battus and
skippers were combined for analysis.
To test whether the proportion of visits from light-blue vs. focal color to P
cuspidata differed, we used G-tests on the combined pollinator data. Visits to light-blue
and focal color were pooled because they represent transitions within P. drummondii and
we are interested in comparing the proportions of intra-specific visits to interspecific
visits. The two factors in this test were thus (i) number of visits to light-blue and focal
color vs. number of visits to P. cuspidata, and (ii) initial visit to light-blue vs. focal color.
In the light-red arrays, pollinators were equally likely to move from light-blue vs.
light-red to P. cuspidata (χ21 = 0.7, n.s.). By contrast, in the dark‐blue and dark‐red
arrays, pollinators were significantly less likely to move from dark‐blue or dark‐red
to P. cuspidata than to move from light‐blue (χ21 = 48, P<0.001 and χ21 = 17.4,
P<0.001, respectively).
Bateman’s Constancy Index. In order to quantify constancy within each plot we
calculated Bateman’s Constancy Index (BCI). This index varies from ‐1, indicating
movement only between unlike plants, to 1, indicating movement only between like
plants, with zero indicating random movement of pollinators. Within each array
type we calculated two BCI’s one for the focal color and P. cuspidata and one for the
light‐blue plants and P. cuspidata. Transition tables were constructed as illustrated
below and the BCI was calculated using the standard formula:
FROM
TO
P. drummondii P. cuspidata
P. drummondii
A
B
P. cuspidata
C
D
BCI = [(AD)½ – (BC)½]/[(AD)½ + (BC)½]
Fig. S1.
Crossing design for field experiments. Field collected parents with light-blue and darkred genotypes were crossed to create F1’s which were self fertilized to create F2 families.
Experimental seeds were created by crossing F2 individuals with the same homozygous
genotype between families.
Fig. S2.
Schematic of the dark-red pollinator observation array. The light-red and dark-blue
arrays were a similar design.
Supporting Tables
Table S1.
Seed source populations, locations, and flower color
Population
Species
Color
Dog95
P. drummondii
light-blue
466_E7
P. drummondii
light-blue
Poc2104
P. drummondii
dark-red
80S1
P. drummondii
dark-red
696N1
P. drummondii
light-blue
2104circ
P. cuspidata
light-blue
Location
W -97.3423
N 30.2926
W -97.7291
N 29.5049
W -97.0871
N 30.0813
W -97.6613
N 29.6403
W -97.2902
N 30.3236
W -97.0878
N 30.0707
Table S2.
Maximum likelihood analysis of variance for survival differences among flower color
genotypes.
Source
DF
Chi Square
intercept
intensity
hue
source
block
intensity*hue
intensity*source
intensity*block
hue*source
hue*block
intensity*hue*source
intensity*hue*block
intensity*source*block
hue*source*block
intensity*hue*source*block
1
1
1
1
3
1
1
3
1
3
1
3
3
3
3
493.17
0
0
304.1
518.26
0.01
0.01
9.95
0
1.74
0
2.06
8.1
2.16
1.24
(Wald score)
P
<.0001
0.9657
0.9509
<.0001
<.0001
0.9396
0.9404
0.019
0.9646
0.6286
0.9585
0.56
0.0441
0.5389
0.7434
Table S3.
Mixed model analysis of variance for fruit-set. a. No significant fixed effects of
intensity, hue or their interaction. b. Covariance of random effects reveal a significant
block effect.
a.
Fixed Effect Variance
Parameter
DF
F
P
Hue
Intensity
Hue X Intensity
1
1
1
0.08
0.03
0.08
0.8194
0.8944
0.8294
Random Effect
Covariance Parameter
Estimate
Chi Square
DF
Source
Block
Source X Block
Hue X Source
Hue X Block
Intensity X Source
Intensity X Block
Hue X Intensity X Source
Hue X Intensity X Block
Residual
0
0.7439
0
0
0
0.006614
0.002892
0.02256
0.004686
1.2721
0
9.8
0
0
0
0
0
2.2
0.2
1
1
1
1
1
1
1
1
1
full model log likelihood
3852.3
b.
P
1
0.0017*
1
1
1
1
1
0.138
0.6547
Table S4.
Mixed model analysis of variance for number of seeds-per-fruit. a. No significant fixed
effects of intensity, hue, or their interaction. b. No significant random effects.
a.
Fixed Effect Variance
Parameter
Hue
Intensity
Hue X Intensity
DF
F
P
1
1
1
0.94
0.46
5.32
0.5106
0.6191
0.2605
Estimate
Chi Square
DF
0
0.00335
0
0
0.002264
0
0.007688
0
0
1.2721
0
0.1
0
0
0.1
0
0.7
0
0
1
1
1
1
1
1
1
1
1
b.
Random Effect
Covariance Parameter
Source
Block
Source X Block
Hue X Source
Hue X Block
Intensity X Source
Intensity X Block
Hue X Intensity X Source
Hue X Intensity X Block
Residual
full model log likelihood
59.5
P
1
0.751
1
1
0.751
1
0.402
1
1
Table S5.
Hybridization data for each genotype (focal and reference). (a) Bootstrap standard errors
indicated in parentheses. The two light genotypes show a relative hybridization rate not
significantly different from 1 indicating these genotypes behaved similar to the light-blue
reference plants. The two dark genotypes had half the hybridization rate of the reference
light-blue plants. (LB=light-blue, LR=light-red, DB=dark-blue, DR=dark-red). (b)
Results from hypotheses tests indicating an effect of the intensity locus but no effect of
the hue locus or a genetic interaction. We re-sampling with replacement from pooled
data in order to calculated a null distribution against which we tested the observed data .
a.
Block
color
Hybrids
LB
28
LR
48
DB
21
DR
21
Focal
NonHybridization
hybrids
rate
0.301
65
(0.069)
0.421
66
(0.104)
0.207
80
(0.061)
0.212
78
(0.063)
b.
Effect
P
Intensity
0.0477
Hue
0.5118
Interaction
0.4927
Hybrids
46
31
50
69
LB reference
NonHybridization
hybrids
rate
0.275
121
(0.072)
0.373
52
(0.121)
0.427
67
(0.106)
0.439
88
(0.083)
Relative
hybridization
rate
1.09 (0.472)
1.13 (0.824)
0.486 (0.220)
0.482 (0.184)
Table S6.
Analysis of pollinator transitions. a. Number of pollinator transitions in the three array
types. b. G-test for heterogenetity between pollinator types in proportions of transitions
both to and from P. cuspidata. c. Bateman’s Constancy Index within each of the three
pollinator arrays. BCI was calculated for transitions within and between P. cuspidata and
each P. drummondii focal color and light-blue in each array. Array is indicated by
column title.
a.
Pollinator
Array
Focal
Color
From Light-Blue
To
Battus
philenor
Skippers
Total
LR
DB
DR
LR
DB
DR
LR
DB
DR
lightblue
43
29
73
6
7
9
49
36
82
focal
color
90
57
120
18
25
4
108
82
124
Cusp
30
16
90
32
64
39
62
80
129
From focal color
lightblue
92
63
122
20
21
5
112
84
127
b
To P.
cuspidata
From P.
cuspidata
Array Type
LR
DB
DR
LR
DB
DR
G-value
0.01
1.61
3.9
0.02
0
0.01
DF
1
1
1
1
1
1
c.
focal color
light-blue
LR
0.162
0.0829
DB
-0.035
0.752
DR
-0.024
0.622
P
ns
ns
0.05
ns
ns
ns
focal
color
75
101
95
5
12
0
80
113
95
Cusp
28
7
29
32
18
4
60
25
33
From P. cuspidata
lightblue
27
15
87
31
73
39
58
88
126
focal
color
30
6
29
33
14
5
63
20
34
Cusp
16
7
52
48
159
166
64
166
218
References and Notes
1. R. Butlin, Trends Ecol. Evol. 2, 8 (1987).
2. T. Dobzhansky, Genetics and the Origin of Species (Columbia Univ. Press, New York,
1937).
3. D. J. Howard, in Hybrid Zones and the Evolutionary Process, R. G. Harrison, Ed.
(Oxford Univ. Press, New York, 1993), pp. 46-69.
4. M. R. Servedio, M. A. F. Noor, Annu. Rev. Ecol. Evol. Syst. 34, 339 (2003).
5. A. R. Wallace, Darwinism: An exposition of the theory of natural selection, with some
of its applications (Macmillan, London, 1889).
6. M. Kirkpatrick, V. Ravigne, Am. Nat. 159, S22 (2002).
7. D. Ortiz-Barrientos, A. Grealy, P. Nosil, Year in Evolutionary Biology 2009 1168, 156
(2009).
8. M. Kirkpatrick, Proc. R. Soc. Lond. Ser. B Biol. Sci. 267, 1649 (2000).
9. M. Kirkpatrick, M. R. Servedio, Genetics 151, 865 (1999).
10. M. R. Servedio, Evolution 55, 1909 (2001).
11. M. R. Servedio, Evolution 58, 913 (2004).
12. A. Y. K. Albert, D. Schluter, Evolution 58, 1099 (2004).
13. P. Nosil, B. J. Crespi, C. P. Sandoval, Proc. R. Soc. Lond. Ser. B Biol. Sci. 270, 1911
(2003).
14. D. A. Levin, Evolution 39, 1275 (1985).
15. R. Hopkins, D. A. Levin, M. D. Rausher, Evolution 10.1111/j.15585646.2011.01452.x (2011).
16. D. A. Levin, Am. J. Bot. 54, 1122 (1967).
17. L. G. Ruane, K. Donohue, Evol. Ecol. 22, 229 (2008).
18. R. Hopkins, M. D. Rausher, Nature 469, 411 (2011).
19. Material and methods are available as supporting material on Science Online.
20. N. M. Waser, Am. Nat. 127, 593 (1986).
21. M. D. Rausher, Science 200, 1071 (1978).
22. H. D. Bradshaw, D. W. Schemske, Nature 426, 176 (2003).
23. D. R. Campbell, N. M. Waser, E. J. MelendezAckerman, Am. Nat. 149, 295 (1997).
24. M. E. Hoballah et al., Plant Cell 19, 779 (2007).
25. L. Chittka, J. D. Thomson, N. M. Waser, Naturwissenschaften 86, 361 (1999).
26. M. Caisse, J. Antonovics, Heredity 40, 371 (1978).
27. J. Felsenstein, Evolution 35, 124 (1981).
28. L. W. Liou, T. D. Price, Evolution 48, 1451 (1994).
29. M. R. Servedio, Evolution 54, 21 (2000).