Ancient Wings: Animating the evolution of butterfly wing patterns

Abstract

Character optimization methods can be used to reconstruct ancestral states at the internal nodes of phylogenetic trees. However, seldom are these ancestral states visualized collectively. Ancient Wings is a computer program that provides a novel method of visualizing the evolution of several morphological traits simultaneously. It allows users to visualize how the ventral hindwing pattern of 54 butterflies in the genus Bicyclus may have changed over time. By clicking on each of the nodes within the evolutionary tree, the user can see an animation of how wing size, eyespot size, and eyespot position relative the wing margin, have putatively evolved as a collective whole. Ancient Wings may be used as a pedagogical device as well as a research tool for hypothesis-generation in the fields of evolutionary, ecological, and developmental biology.

Full text

BioSystems 71 (2003) 289–295
Ancient Wings: animating the evolution of butterfly wing patterns
Samuel Arbesmana, Leo Enthovenb, Antónia Monteiroa,b,∗
aDepartment of Biological Sciences, University at Buffalo, Buffalo, NY 14260, USA
bInstitute of Biology, Leiden University, P.O. Box 9516, Leiden, RA 2300, The Netherlands
Received 4 November 2002; received in revised form 21 April 2003; accepted 7 May 2003
Abstract
Characteroptimizationmethodscanbeusedtoreconstruct ancestral states at the internalnodes of phylogenetic trees. However,
seldom are these ancestral states visualized collectively. Ancient Wings is a computer program that provides a novel method of
visualizing the evolution of several morphological traits simultaneously. It allows users to visualize how the ventral hindwing
pattern of 54 butterflies in the genus Bicyclus may have changed over time. By clicking on each of the nodes within the
evolutionary tree, the user can see an animation of how wing size, eyespot size, and eyespot position relative the wing margin,
have putatively evolved as a collective whole. Ancient Wings may be used as a pedagogical device as well as a research tool for
hypothesis-generation in the fields of evolutionary, ecological, and developmental biology.
© 2003 Elsevier Ireland Ltd. All rights reserved.
Keywords: Bicyclus; Ancestral reconstruction; Ancestor; Phylogeny; Animation; Butterfly; Wing patterns
1. Introduction
Character state reconstruction on phylogenetic
trees has been used to recreate individual character
states (Felsenstein, 1985; Maddison and Maddison,
1992) and to reconstruct ancestral proteins (Chang
and Donoghue, 2000; Chang et al., 2002), but seldom
is it used to recreate a whole integrated suite of traits
simultaneously. We created a computer program, An-
cient Wings, to illustrate a novel method to visualize
the evolution of suites of morphological traits.
We chose to examine and animate the evolution of
the hindwing patterns of the butterfly genus Bicyclus,
found in sub-Saharan Africa (Condamin, 1973). Out
of a total of 80 Bicyclus species, 54 Bicyclus species
and 6 outgroup species belonging to related genera,
∗Corresponding author. Tel.: +1-716-6452363x135;
fax: +1-716-6452975.
E-mail address: monteiro@buffalo.edu (A. Monteiro).
were used to build a rooted molecular phylogeny for
the genus (Monteiro and Pierce, 2001). This genus has
been studied intensively with regard to wing pattern
plasticity (Brakefield and Reitsma, 1991; Roskam and
Brakefield, 1996), and the genetic and developmen-
tal underpinnings of eyespot formation (French and
Brakefield, 1995; Monteiro et al., 1994, 1997, 2003;
Beldade et al., 2002a,b).
Reconstruction of the putative ancestral wing pat-
terns and animation of these over the phylogenetic tree
provides a visual aid and a hypothesis generating tool
through which the mechanisms and selective agents
regarding wing pattern evolution can be investigated.
2. Materials and method
Wing pattern data were collected from digital pho-
tographs of 253 females of 54 species of Bicyclus but-
terflies. Males and females have nearly identical pat-
0303-2647/$ – see front matter © 2003 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/S0303-2647(03)00086-8

290 S. Arbesman et al./ BioSystems 71 (2003) 289–295
Fig. 1. Bicyclus ventral hindwing measurements used in the cal-
culations of ancestral wing patterns.
terns but females are usually lighter in coloration and
the patterns are more contrasting against the back-
ground coloration. The photographs were taken from
specimens collected in Africa (Monteiro and Pierce,
2001), from specimens illustrated in the monograph of
the genus (Condamin, 1973), and from the extensive
collection of Bicyclus found at the Royal Museum for
Central Africa, Tervuren, Belgium. The digital images
were analyzed with Object-Image 2.08 (Vischer et al.,
1994).
Eyespot diameter measurements were collected
by measuring the black central disc parallel to the
wing veins from the seven hindwing eyespots from
five females of each species (for five species we col-
lected data from only four specimens, for two species
we collected data from two specimens, and for two
species from three and one specimen, respectively),
and their average values and standard errors were
calculated (see Fig. 1). The size of an eyespot’s outer
gold ring was extrapolated using a constant ratio (the
gold and black colored portion of an eyespot can also
evolve across the genus Bicyclus but this process was
not modeled here; Condamin, 1973; Monteiro et al.,
1997).
Eyespot positional information was collected using
measurements from a single female from each of the
54 species examined in this study. We took the xand
ycoordinates for the center of each of the seven eye-
spots, as well as the coordinates of two points along
one of the longitudinal wing veins (the one posterior
to the fourth eyespot; Fig. 1). This vein was used to
align the eyespot patterns to the same relative x–yco-
ordinates. The x–ycoordinate at the left end of this
vein was used to anchor all eyespot position coordi-
nates across species. The x–ycoordinate at the right
end of the vein was used to rotate all the position co-
ordinates to the same angle. These calculations were
performed using Perl.
Wing margin positional information was collected
by using measurements from the same single repre-
sentative of each of the 54 species mentioned earlier.
We took the xand ycoordinates for the wing margin
at the intersection with the intervenous fold for each
of the seven eyespots (Fig. 1). The coordinate points
along the reference wing vein were used again to an-
chor and rotate the wing margin coordinates.
We used a phylogenetic tree based on molecular
sequence information as the basis for character re-
construction of internal nodes. This phylogeny was
based on a maximum parsimony reconstruction of
DNA sequence divergence of one nuclear gene (Elon-
gation factor 1α), and two mitochondrial genes (Cy-
tochrome Oxidase I and II; COI and COII, respec-
tively) and is the only molecular phylogeny for the
genus proposed to date (Monteiro and Pierce, 2001).
There was strong bootstrap support all of the tip clades
in the tree, excluding the clade containing B. ena,B.
technatis, and B. vansoni, and the clade of B. igno-
bilus and B. pavonis, where long branch attraction
may have brought these species together (Felsenstein,
1978). A few of the more basal nodes also lacked
strong bootstrap support (see Monteiro and Pierce,
2001).
The eyespot sizes, xand ycoordinates of the eye-
spot centers, and the xand ycoordinates of the wing
margins, were each run separately through COM-
PARE 4.4, a program that can estimate the ancestral
traits of a group of taxa given a phylogenetic tree
(Martins, 2001). The program uses Phylogenetic Gen-
eralized Least Squares (PGLS) with a linear model to
estimate ancestral traits (Martins and Hansen, 1997).
The PGLS method of ancestor reconstruction used the
branch lengths of the phylogenetic tree (the maximum
parsimony estimates of the total number of molecular
changes from all the sequence data combined, from
Fig. 5 in Monteiro and Pierce, 2001), along with the
trait data for the tip nodes, to create weighted aver-
ages for the ancestral nodes. The calculations also

S. Arbesman et al./ BioSystems 71 (2003) 289–295 291
take into account inter-specific variation when this
data is provided.
The ancestral wing pattern calculations for each
of the internal nodes were done using a Brown-
ian motion model of evolution in which phenotypic
changes accumulate in a random fashion at constant
rates. Evolutionary mechanisms that fit a Brownian
motion model include random genetic drift (with or
without mutation), strong stabilizing selection with
randomly changing optima, and directional selection
with random fluctuations in direction (Felsenstein,
1988). This model explicitly assumes that weaker
stabilizing selection on the traits under evaluation has
not occurred. These are obviously strong assumptions
regarding the mode of evolution of these characters.
The animation program was created using Macro-
media Flash MX, an authoring system with a built-in
programming system language called Actionscript
which is ideal for creating animations and interac-
tive programs for use over the World Wide Web. We
worked on a Power Mac G4 with OS X. Ancient Wings
takes advantage of the graphical and object-oriented
nature of Flash’s system to allow the programmer to
easily manipulate and move on-screen elements. This
was well-suited for allowing the visualization of how
the eyespots have putatively changed size and position
over time. The program runs on Netscape or Internet
Explorer browsers with a Flash Player 6 plug-in.
The tree topology used in the final computer pro-
gram was time calibrated with COI assuming a rate of
2% sequence divergence between pairs of sequences
for every million years of evolutionary separation
(Desalle et al., 1987; Brower, 1994; Monteiro and
Pierce, 2001). Time calibrating the tree with COII
(Monteiro and Pierce, 2001) gave somewhat differ-
ent estimates that diverged considerably for the more
basal nodes (sometimes as much as by 5 million
years). These time estimates should be taken with cau-
tion and were only provided for a very approximate
estimate of the relative age of the different clades.
All phylogenetic and trait data were hard-coded
into the computer program. The algorithm for the
animation of trait evolution followed the same as-
sumption used in calculating the ancestral nodes, i.e.
that all evolutionary change occurred at a constant
rate throughout the tree. Animation from node to node
was accomplished by dividing the estimated trait dif-
ferences between the two nodes by the branch length
between them. Branch length was determined based
on the distance between the node-to-node screen
coordinates based on the clock-constrained COI dis-
tances corrected with a general time reversible model,
a␥distribution of rate heterogeneity across sites, a
proportion of invariable sites and unequal base fre-
quencies (see Monteiro and Pierce, 2001). Note that in
order to display morphological change through time,
the branch lengths displayed by the animation were
not the same as those used to estimate the ancestral
nodes with COMPARE (see Section 3 below). The
program then created a smooth, constant, gradient so
that traits change at a constant rate from node to node.
3. Results and discussion
The Flash animation, Ancient Wings, may be ac-
cessed at http://www.acsu.buffalo.edu/∼monteiro/
ancientwings/. The source code and Flash file for
Ancient Wings is available without restriction at
http://www.acsu.buffalo.edu/∼monteiro/ancientwings/
ancientwings.fla. A guide to the evolutionary tree,
with the names of the tip nodes and the numbering
scheme of the internal nodes as well as an Excel data
file may also be downloaded from the same site (see
Fig. 2).
Ancient Wings can be useful as a teaching device
to allow students to see how putative wing patterns of
a genus of butterflies have changed over evolutionary
history. However, in addition to the educational value
of Ancient Wings, the program also has value for re-
searchers as an hypothesis-generation tool.
Due to the clear visualization of the calculated an-
cestral wing pattern data, a large amount of informa-
tion can be absorbed by the user at a single glance.
This allows the researcher to identify patterns within
the data and formulate hypotheses. These hypotheses
can then be tested both by further rigorous mathemat-
ical analysis of the data or by the design of further
experiments and studies.
One of the assumptions behind the reconstruction of
the ancestral wing patterns is that eyespot morphology
is evolving at a constant rate by random drift, by strong
stabilizing selection with randomly changing optima,
or by directional selection with random fluctuations
in direction (Felsenstein, 1988). We also assumed that
the length of the molecular branch (amount of molec-

292 S. Arbesman et al./ BioSystems 71 (2003) 289–295
Fig. 2. (a) A screenshot from Ancient Wings. (b) Phylogenetic tree supplementary diagram, indicating the numbering and labeling system
of the nodes used in the accompanying Excel data spreadsheet available online at http://www.acsu.buffalo.edu/∼monteiro/ancientwings/.

S. Arbesman et al./ BioSystems 71 (2003) 289–295 293
ular evolution) is correlated with the amount of mor-
phological evolution along the same branch. This last
assumption has been supported in several other stud-
ies where branch lengths from independently derived
molecular and morphological phylogenies were found
to correlate (Omland, 1997). In the final animation of
the program, however, we enforced a molecular clock
and modified the branch lengths in order to portray
evolution of the wing patterns against absolute time.
This clock correction has the effect of slowing down or
speeding up the rates of morphological change along
subsets of the branches.
A recent paper showed that relatively small differ-
ences in ventral eyespots among two recently diverged
species of Lycaenid butterflies played an important
role in mate recognition (Fordyce et al., 2002). This
indicated that small differences in morphology can be
detected by the butterfly’s visual system and be used
in sexual selection or species recognition. In Bicy-
clus, although it has been shown that dorsal eyespot
size may play a role in sexual selection (Breuker and
Brakefield, 2002), no such effect has yet been shown
for the ventral eyespots. It is also not clear what sub-
tle differences in eyespot patterns may play in the in-
teraction with predators. The actual selective forces
driving the evolution of ventral wing patterns in the
genus Bicyclus are largely unknown and thereby the
reconstruction of ancestral traits is only valid as long
as the assumption of constant rates of morphological
evolution is valid.
Recent experimental work on one of the species of
the genus, Bicyclus anynana, has showed that there
are little constraints posed by the developmental sys-
tem in obtaining a rapid response to artificial selection
on eyespot size (Beldade et al., 2002b). In particular,
selecting for opposing eyespot sizes on the same wing
surface was accomplished with relative ease. One may
ask whether if the selection experiments were per-
formed with any of the other species in the genus, with
distinct evolutionary histories the same result would
hold. For instance, Ancient Wings can help us iden-
tify two species in the phylogeny that have a very
different history of eyespot size evolution. In B. any-
nana, for instance, eyespot number 3 is usually smaller
than eyespot 4 on the hindwing. Similar relative sizes
are also found in the two closest relatives to B. any-
nana (B. campus and B. angulosus), and the eyespot
size animation from the most recent common ancestor
(MRCA) to each of these three tip species (at around
7 million years ago) shows little morphological evo-
lution in these two eyespots. In contrast, in the lin-
eage leading to B. sciathis, there is rapid evolution of
the relative sizes of eyespots 3 and 4 from a similarly
aged ancestor. In this ancestor, eyespot 4 has a similar
size to eyespot 3 but becomes much smaller relative to
eyespot 3 in present day B. sciathis populations. Does
this different evolutionary history influence present
day levels of genetic variation and the matrix of ge-
netic correlations for the sizes of these two eyespots as
well as the relative ease that these two eyespots can be
“uncoupled” from each other by present day artificial
or natural selection? Does the developmental system
retain some “memory” of recent events of morpholog-
ical evolution? Will antagonistic selection on the sizes
of eyespots 3 and 4 in B. sciathis be able to retrace the
steps back to the MRCA in a much faster way than
a similar selection on eyespots in B. anynana where
there is no recent history of antagonistic size varia-
tion? Two other species that replicate a recent history
of rapid evolution or stasis (from a 5 million year an-
cestor) involving eyespots 3 and 4, are B. smithi and B.
golo (or B. smithi and B. madetes), respectively. These
groups of paired species can be brought into the lab
and used to test the role of recent evolutionary history
on the speed of response to present day selection.
Another field of inquiry suggested by Ancient
Wings is that of biogeography. Due to the visualiza-
tion of changes in wing types unfolding alongside a
timeline in the program, biogeographic studies can be
examined. For example, the origin of certain taxa with
distinct traits (such as larger wing size or missing eye-
spots) may be correlated with certain points in time.
By conducting an analysis of the climatic changes
occurring at these times, various hypotheses of the
functional significance of wing traits can be examined.
The topic of functional integration may also be in-
vestigated. We observed how the eyespots change in
size and position relative to each other and relative to
the wing margin, and remarked that despite substan-
tial changes in eyespot size, eyespots always tend to
“dodge” each other by coordinated changes in posi-
tion and/or size of neighbors. This avoids the merging
of adjacent eyespots and produces consistent circular
shapes. Mimicking circular “vertebrate eyes” may be
important for deflecting the attacks of predators and
the subsequent survival of these butterflies (Brakefield,

294 S. Arbesman et al./ BioSystems 71 (2003) 289–295
1996). This functional integration hypothesis awaits
more rigorous testing. Alternatively, patterns of eye-
spot covariation may reflect underlying developmental
compartments of the wing (Monteiro et al., 2003).
In addition, Ancient Wings can be used to analyze
correlations between eyespot position and wing size.
For example, Nijhout (1991, 2001) proposed that the
dimensions of the wing section bordered by veins,
where the eyespots develop, may determine the posi-
tion of the eyespot’s center (the focus) relative to the
margin. If wing cells change shape or size, dependent
morphogenetic processes of focus differentiation such
as the putative mechanism of lateral inhibition of two
diffusable interacting substances secreted by the wing
veins, may change accordingly (Nijhout, 1991). This
hypothesis may be investigated using Ancient Wings
as a preliminary exploratory tool.
A fundamental assumption in the calculation of the
ancestral wing patterns is that eyespot size has evolved
gradually (at a constant rate). This may not have been
the case. For instance, mutations of large phenotypic
effect are known to occur sporadically in lab popula-
tions of one of the species, Bicyclus anynana, either
introducing or removing fully developed eyespots, or
shifting eyespots along the margin in an abrupt fashion
(McMillan et al., 2002; Beldade and Brakefield, 2002;
Monteiro et al., 2003). Presumably, Ancient Wings can
be used to suggest in which lineages a gradual versus
a punctuational type change has occurred on an evo-
lutionary time-scale. Instances where the speed of the
animation of certain eyespots (the rate of phenotypic
change) is accelerated relative to the other eyespots
could indicate that a linear model of ancestor recon-
struction is trying to accommodate what is actually
a much faster, non-gradual, process of evolutionary
change (see Martins, 1994). In other words, by forcing
the phenotypic difference between nodes to be divided
equally by the time available between nodes, a muta-
tion of large phenotypic effect (saltational evolution)
would become represented in the animation as a rapid
gradual change in morphology, standing out against
the slower gradual animation of the other characters.
Ancient Wings, through its novel visualization
method of evolutionary reconstruction, may be used
both in the classroom and in the laboratory, both
encouraging students to learn more about evolution
and stimulating researchers to look at data in new
ways. Its main strength lies in a continuous rendering
of suites of ancestral character states that are dis-
played simultaneously. Hopefully, Ancient Wings,as
a demonstration of this concept, will prompt others to
bring evolutionary data to life.
Acknowledgements
We thank William Piel and two anonymous review-
ers for their comments and constructive criticism.
A.M. was funded while at Leiden by grant RG0058
from the Human Frontier Science Program.
References
Beldade, P., Brakefield, P.M., 2002. The genetics and evo devo of
butterfly wing patterns. Nat. Rev. Genet. 3, 442–452.
Beldade, P., Brakefield, P.M., Long, A.D., 2002a. Contribution of
distal-less to quantitative variation in butterfly eyespots. Nature
415, 315–317.
Beldade, P., Koops, K., Brakefield, P.M., 2002b. Developmental
constraints versus flexibility in morphological evolution. Nature
416, 844–847.
Brakefield, P.M., 1996. Seasonal polyphenism in butterflies and
natural selection. Trends Ecol. Evol. 11, 275–277.
Brakefield, P.M., Reitsma, N., 1991. Phenotypic plasticity, seasonal
climate and the population biology of Bicyclus butterflies
(Satyridae) in Malawi. Ecol. Entomol. 16, 291–303.
Breuker, C.J., Brakefield, P.M., 2002. Female choice depends
on size but not symmetry of dorsal eyespots in the butterfly
Bicyclus anynana. Proc. R. Soc. Lond. B 269, 1233–1239.
Brower, A.V.Z., 1994. Rapid morphological radiation and
convergence among races of the butterfly Heliconius erato
inferred from patterns of mitochondrial DNA evolution. Proc.
Natl. Acad. Sci. U.S.A. 91, 6491–6495.
Chang, B.S., Donoghue, M.J., 2000. Recreating ancestral proteins.
TREE 15, 109–114.
Chang, B.S., Jonsson, K., Kazmi, M.A., Donoghue, M.J., Sakmar,
T.P., 2002. Recreating a functional ancestral archosaur visual
pigment. Mol. Biol. Evol. 19, 1483–1489.
Condamin, M., 1973. Monographie du genre Bicyclus
(Lepidoptera, Satyridae). IFAN, Dakar.
Desalle, R., Freedman, T., Prager, E.M., Wilson, A.C., 1987.
Tempo and mode of sequence evolution in mitochondrial DNA
of Hawaiian Drosophila. J. Mol. Evol. 26, 157–164.
Felsenstein, J., 1978. Cases in which parsimony and compatibility
methods will be positively misleading. Syst. Zool. 27, 401–
410.
Felsenstein, J., 1985. Phylogenies and the comparative method.
Am. Nat. 125, 1–15.
Felsenstein, J., 1988. Phylogenies from molecular sequences:
inference and reliability. Annu. Rev. Genet. 22, 521–565.
Fordyce, J.A., Nice, C.C., Forister, M.L., Shapiro, A.M., 2002.
The significance of wing pattern diversity in the Lycaenidae:

S. Arbesman et al./ BioSystems 71 (2003) 289–295 295
mate discrimination by two recently diverged species. J. Evol.
Biol. 15, 871–879.
French, V., Brakefield, P.M., 1995. Eyespot development on
butterfly wings: the focal signal. Dev. Biol. 168, 112–123.
Maddison, W.P., Maddison, D.R., 1992. MacClade. Sinauer,
Sunderland, MA.
Martins, E.P., 1994. Estimating the rate of phenotypic evolution
from comparative data. Am. Nat. 144, 193–209.
Martins, E.P., 2001. Compare: computer programs for the statistical
analysis of comparative data. Distributed by the author via the
WWW at http://compare.bio.indiana.edu/, Bloomington, IN.
Martins, E.P., Hansen, T.F., 1997. Phylogenies and the comparative
method: a general approach to incorporating phylogenetic
information into the analysis of interspecific data. Am. Nat.
149, 646–667.
McMillan, W.O., Monteiro, A., Kapan, D.D., 2002. Development
and evolution on the wing. Trends Ecol. Evol. 17, 125–133.
Monteiro, A., Pierce, N.E., 2001. Phylogeny of Bicyclus
(Lepidoptera: Nymphalidae) inferred from COI, COII and
EF-1a gene sequences. Mol. Phylogen. Evol. 18, 264–281.
Monteiro, A.F., Brakefield, P.M., French, V., 1994. The evolu-
tionary genetics and developmental basis of wing pattern
variation in the butterfly Bicyclus anynana. Evolution 48, 1147–
1157.
Monteiro, A., Brakefield, P.M., French, V., 1997. The genetics
and development of an eyespot pattern in the butterfly Bicyclus
anynana: response to selection for eyespot shape. Genetics 146,
287–294.
Monteiro, A., Prijs, J., Bax, M., Hakkaart, T., Brakefield, P.M.,
2003. Mutants highlight the modular control of butterfly eyespot
patterns. Evol. Dev. 5, 180–187.
Nijhout, H.F., 1991. The Development and Evolution of Butterfly
Wing Patterns. Smithsonian Institution Press, Washington.
Nijhout, H.F., 2001. Elements of butterfly wing patterns. J.
Exp. Zool. 291, 213–225.
Omland, K.E., 1997. Correlated rates of molecular and morpholo-
gical evolution. Evolution 51, 1381–1393.
Roskam, J.C., Brakefield, P.M., 1996. A comparison of tempera-
ture-induced polyphenism in African Bicyclus butterflies
from a savannah-rainforest ecotone. Evolution 50, 2360–
2372.
Vischer, N.O.E., Huls, P.G., Woldringh, C.L., 1994. Object-image:
an interactive image analysis program using structured point
collection. Binary (Bioline) 6.