The anatomy of a 'suture zone' in Amazonian butterflies: a coalescent-based test for vicariant geographic divergence and speciation.

Abstract

Abstract Attempts by biogeographers to understand biotic diversification in the Amazon have often employed contemporary species distribution patterns to support particular theories, such as Pleistocene rainforest refugia, rather than to test among alternative hypotheses. Suture zones, narrow regions where multiple contact zones and hybrid zones between taxa cluster, have been seen as evidence for past expansion of whole biotas that have undergone allopatric divergence in vicariant refuges. We used coalescent analysis of mutilocus sequence data to examine population split times in 22 pairs of geminate taxa in ithomiine and heliconiine butterflies. We test a hypothesis of simultaneous divergence across a suture zone in NE Peru. Our results reveal a scattered time course of diversification in this suture zone, rather than a tight cluster of split times. Additionally, we find rapid diversification within some lineages such as Melinaea contrasting with older divergence within lineages such as the Oleriina (Hyposcada and Oleria). These results strongly reject simple vicariance as a cause of the suture zone. At the same time, observed lineage effects are incompatible with a series of geographically coincident vicariant events which should affect all lineages similarly. Our results suggest that Pleistocene climatic forcing cannot readily explain this Peruvian suture zone. Lineage-specific biological traits, such as characteristic distances of gene flow or varying rates of parapatric divergence, may be of greater importance.

Full text

Page 1

The anatomy of a ‘suture zone’ in Amazonian butterflies:
a coalescent-based test for vicariant geographic
divergence and speciation
KANCHON K. DASMAHAPATRA,* GERARDO LAMAS,† FRASER SIMPSON* and JAMES
MALLET*‡
*Department of Genetics, Evolution and Environment, University College London, 4 Stephenson Way, London NW1 2HE, UK,
†Museo de Historia Natural, Universidad Nacional Mayor de San Marcos, Av. Arenales 1256, Apartado 14-0434, Lima-14,
Peru; ‡Radcliffe Institute for Advanced Study, Byerly Hall, Harvard University, Cambridge, MA 02138, USA
Abstract
Attempts by biogeographers to understand biotic diversification in the Amazon have
often employed contemporary species distribution patterns to support particular
theories, such as Pleistocene rainforest refugia, rather than to test among alternative
hypotheses. Suture zones, narrow regions where multiple contact zones and hybrid
zones between taxa cluster, have been seen as evidence for past expansion of whole
biotas that have undergone allopatric divergence in vicariant refuges. We used coalescent
analysis of mutilocus sequence data to examine population split times in 22 pairs of
geminate taxa in ithomiine and heliconiine butterflies. We test a hypothesis of
simultaneous divergence across a suture zone in NE Peru. Our results reveal a scattered
time course of diversification in this suture zone, rather than a tight cluster of split times.
Additionally, we find rapid diversification within some lineages such as Melinaea
contrasting with older divergence within lineages such as the Oleriina (Hyposcada and
Oleria). These results strongly reject simple vicariance as a cause of the suture zone. At
the same time, observed lineage effects are incompatible with a series of geographically
coincident vicariant events which should affect all lineages similarly. Our results suggest
that Pleistocene climatic forcing cannot readily explain this Peruvian suture zone.
Lineage-specific biological traits, such as characteristic distances of gene flow or varying
rates of parapatric divergence, may be of greater importance.
Keywords: coalescent theory, Heliconius, Ithomiini, phylogeography, Pleistocene refuges, specia-
tion
Received 7 April 2010; revision received 15 July 2010; accepted 19 July 2010
Introduction
The highest terrestrial diversity on the planet is found
in the Amazon basin, yet processes responsible for its
evolution are unclear. There is a widely held view that
most speciation occurs in geographic isolation, or ‘allop-
atry’ (Mayr 1963; Coyne & Orr 2004). However, the
Amazon is a large, continuously forested region with
few obvious geographic barriers. If allopatric speciation
is the predominant mode of speciation, the simplest
prediction would be that the Amazon should have low,
rather than high, species diversity. Many theories have
therefore been put forward to explain spatial patterns
of diversification in the Amazon and other tropical
regions, reviewed in (Haffer 1997, 2008), such as river-
ine barrier hypotheses (Patton & Da Silva 1998; Gascon
et al. 2000), Pleistocene refuge hypotheses (Haffer 1969;
Simpson & Haffer 1978; Haffer & Prance 2001), ecologi-
cal (parapatric) hypotheses (Benson 1982; Endler 1982b),
peripheral speciation (Fjeldsa ˚ 1994), centrifugal specia-
tion (Brown 1957) as well as vicariance hypotheses
based on pre-Pleistocene geographic barriers (Cracraft
& Prum 1988; Patton & Da Silva 1998; Hall & Harvey
Correspondence: Kanchon K. Dasmahapatra, Fax: +44 (0)207
6795052; E-mail: k.dasmahapatra@ucl.ac.uk
? 2010 Blackwell Publishing Ltd
Molecular Ecology (2010) 19, 4283–4301doi: 10.1111/j.1365-294X.2010.04802.x

Page 2

2002). Among these, Pleistocene refuge hypotheses have
received much attention from both supporters and crit-
ics (Endler 1982a; Haffer 1997; Moritz et al. 2000; Co-
linvaux et al. 2001; Carnaval et al. 2009).
The Pleistocene (1 800 000–10 000 years ago) is the
most recent of a number of geological periods with
cyclical climate changes during which there were
repeated glaciations in the temperate zone with a peri-
odicity of ?100 000 years. According to the Pleistocene
refuge hypothesis (Haffer 1969), ice ages in the temper-
ate zone led to periods of aridity in the tropics. For-
merly continuous forest was hypothesized to become
fragmented into refuges separated by expanses of grass-
dominated savannah or desert. The forest biota was
supposed then to have diverged during this period of
confinement. After wetter conditions returned, the for-
est biota expanded from the refuges and came back into
contact with formerly identical conspecifics. If repro-
ductively isolated, these taxa could overlap as separate
species. Alternatively if reproductive barriers were
weak, the divergent taxa either fused or remained in
contact at narrow hybrid zones.
The ability of the refuge hypothesis to give rise to allo-
patric speciation in an area where allopatry seemed at
first sight unlikely, together with apparent correspon-
dence of likely refuge areas with the contemporary
patchwork biodiversity pattern in both temperate (He-
witt 2000) and tropical regions (Haffer 1969; Prance 1973;
Brown 1987a), have contributed to its popularity. A
Pleistocene refuge theory for divergence and diversifica-
tion is widely and generally supported for the temperate
zone. For example, refuges undoubtedly existed in
Southern Europe during repeated glaciations, while Cen-
tral and Northern Europe was scoured by a thick layer of
ice (Hewitt 2000; Stewart et al. 2010). In contrast, applica-
tions of refuge theory to explain patterns of neotropical
biodiversity remain contentious. Pleistocene refuges
have been invoked to explain neotropical species distri-
butions in birds (Haffer 1969), plants (Prance 1973, 1982)
and butterflies (Brown 1977, 1987a). Yet, for a number of
taxa, factors other than isolation in past refuges might
also explain contemporary species distributions (Moritz
et al. 2000). For example, current geographic distribu-
tions of Amazonian birds can be explained via a null
model of random placement of species ranges better
rather than via refuges (Beven et al. 1984), and putative
centres of endemism for Amazonian plants correlate
strongly with collecting density, rather than providing a
strong biogeographic signal (Nelson et al. 1990).
Geological data can be used as evidence for existence
of past refuges, just as they have to establish patterns
and movement of ice in the temperate zone. However,
geological data are harder to come by in the Amazon
than almost anywhere else, and the palaeoecological
implications are disputed. Pollen core data are com-
monly used to reconstruct past vegetation changes, but
palynological records from Amazonia are scarce. Some
studies claim evidence for great expansions of grass-
dominated savannah during the Pleistocene based on
fragmentary palynological and geomorphological data
(Haffer & Prance 2001), but others argue that the forest
overall remained continuous, rather than patchy as pre-
dicted by the refuge hypothesis. At the same time,
changes in Amazonian pollen flora do indeed indicate
changes in forest tree composition during temperature
fluctuations correlated with temperate zone glaciation
(Colinvaux et al. 2000, 2001).
Given that direct evidence from geology is still
unclear, current patterns of distribution of organisms
remain among the most powerful data available to test
refuge theory in the Amazon. One class of biogeograph-
ic evidence that can suggest vicariant divergence (of
which refuge theory is an example) is the existence of
‘suture zones.’ Suture zones are defined as narrow
regions with unusual concentrations of contact zones
and hybrid zones. They have been argued to represent
the meeting places where whole biotas, having emerged
after divergence from a pair of vicariant refuges, have
begun to interact after climate amelioration (Remington
1968). Pleistocene climatic cycles have been implicated
in the formation of a number of possible terrestrial
suture zones, such as the Great Plains suture zone in
North America where numerous avian hybrid zones
cluster (Remington 1968; Moore & Price 1993; Klicka &
Zink 1997), the clustered suture zones in the Alps and
Central Europe (Hewitt 1996, 2000) and the Australian
tropics (Moritz et al. 2009).
If divergence occurred during refuge formation, the
multiple pairs of sister taxa across a suture zone might
be expected to have split at a similar time, correspond-
ing to the timing of the vicariant event itself (Coyne &
Orr 2004). Divergence owing to known or probable sin-
gle vicariant events has been studied mainly in marine
systems, for example in species pairs between the
Atlantic and Gulf coasts of USA (Avise 1994, 2000) and
particularly across the Isthmus of Panama 2.7–3.5 Ma
(Lessios 1979; Vawter et al. 1980; Bermingham & Lessi-
os 1993; Knowlton et al. 1993; Knowlton & Weigt 1998;
Hurt et al. 2009). Distribution patterns and biogeo-
graphic features such as suture zones by themselves
may be insufficient to address biogeographic patterns
with complex underlying evolutionary histories such as
species diversification in the Amazon. Incorporation of
molecular sequence data can help uncover useful tem-
poral information about taxa which shed light on these
complex histories (Elias et al. 2009). At the same time,
phylogeographical analyses based on single genes and
simpleestimates of divergencesuch as sequence
4284 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 3

difference may also be inadequate (Hurt et al. 2009;
Hickerson et al. 2010). Importantly, even in the case of
the Isthmus of Panama, where a single vicariant event
is undisputed, recent analyses reveal that divergence of
a number of species pairs predates the vicariant event
(Hurt et al. 2009).
We here focus on a well-defined suture zone near
Tarapoto in NE Peru (Fig. 1) between two areas of
endemism in the lowland rainforest: the Rı ´o Mayo⁄
upper Rı ´o Huallaga valley systems on the one hand
(Departments of San Martı ´n in the North and Hua ´nuco
further South), and the lower Rı ´o Huallaga and Rı ´o
Ucayali regions on the other (Departments of San
Martı ´n (Eastern part), Loreto (Western part) in the
North and Ucayali in the South). These two regions are
separated by a low (mostly < 1500 m altitude) eastern-
most extension of the Andes, known as the Cordillera
Escalera⁄Cordillera Azul. Around 40 forest-inhabiting
butterfly species from tribe Ithomiini and subtribe Heli-
coniina (Nymphalidae) show morphologically differen-
tiated geminate pairs of taxa across this suture zone
(Fig. 1). The taxa consist both of subspecies which
hybridize freely, or ‘semi-species,’ which do not (hereaf-
ter we refer to both collectively as ‘subspecies’ for con-
venience). Suture zones of this kind are argued to form
the meeting place for separate biotas recently expanded
from refuges, here from the putative ‘Huallaga’ (Wes-
tern) and ‘Ucayali’ (Eastern) Pleistocene refuges (Brown
et al. 1974; Brown 1979; Lamas 1982; Whinnett et al.
2005b). The Ithomini and Heliconiina contain about 350
and 70 species overall, respectively (Lamas et al. 2004,
2004), each of which consists of up to 30 morphologi-
cally recognizable subspecies distributed across the
Neotropics (Brown 1979; Lamas et al. 2004). Distribu-
tions of heliconiine and ithomiine subspecies have pre-
viously been used to support and indeed to refine
substantially Haffer’s Pleistocene refuge model (Brown
et al. 1974; Brown 1979, 1987b). Characteristically, each
subspecies is a member of a different geographic Mu ¨lle-
rian mimicry ring, varying in step with other similarly
geographically differentiated taxa occurring in the same
areas (Brown 1979).
Ithomiine subspecies pairs across this suture zone
have previously been found to show variable levels of
> 500 m
< 500 m
50 Km
N
6°
Peru
Colombia
Brazil
Ecuador
Pacific
Ocean
7°
76°
75°
25 Km
20
30
20
20
30
40
50
50
40
2030
10
30
40
(b)
(c)
(a)
Río Mayo
Río Huallaga
Río Ucayali
Yurimaguas
Tarapoto
Pucallpa
4
5
6
7
8
9
10
13
12
11
14
15
17
18
19
16
20
2122
23
1
2
3
24
25
26
29
28
27
36
35
34
33
32
31
30
38
40
37
41
42
43
39
Fig. 1 (a) Map of sampling locations across the suture zone. Sampling location numbers refer to the site numbers in Table 2. (b)
Contours of percentage of taxa hybridizing in the suture zone between the Rı ´o Mayo⁄upper Rı ´o Huallaga and the lower Rı ´o Huall-
aga⁄Rı ´o Ucayali regions of endemism. (c) Detail of collections made in the region of more intensive sampling.
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4285
? 2010 Blackwell Publishing Ltd

Page 4

mitochondrial (mtDNA) divergence, suggesting highly
variable split times among different pairs of subspecies
(Whinnett et al. 2005b). If variable nucleotide diver-
gence is typical of the butterflies’ genomes overall, a
single recent forest vicariance event
responsible for divergence. However, several potential
problems prevent drawing this conclusion from the
previous work: (i) Mitochondrial DNA may evolve dif-
ferently from the rest of the genome. For example,
selective sweeps could be important in the nonrecom-
bining mitochondrial genome, especially via indirect
selection from cytoplasmic endosymbionts (Hurst &
Jiggins 2005). (ii) Whinnett et al. (2005b) estimated split
time via a simple mtDNA molecular clock. However,
given that these taxa diverged during the last few mil-
lion years, coalescence times of molecular markers may
be far older than the time when the populations them-
selves split (Hudson & Turelli 2003). The overall DNA
divergence between taxa includes both stochastic coa-
lescence time within ancestral populations before the
split and the time since the populations split (Hudson
& Turelli 2003). The expected time to coalescence
within a population of effective size N is 2N genera-
tions. As many Amazonian ithomiine and Heliconius
species probably have large population sizes, the
degree to which gene coalescence predates population
divergence mightbesubstantial.
methods must therefore be employed to account for
variance in coalescence times.
In this study, we investigated multilocus DNA
sequence divergence across the Tarapoto suture zone to
test a null hypothesis of simultaneous vicariance using
coalescent-based methods. As an alternative, cyclical cli-
matic changes of the Pleistocene could have resulted in
repeated forest expansions and contractions centred on
the same refuges, and a more complex pattern might be
observed. These and other hypotheses are dealt with in
the discussion. We improve on the previous data and
analyses (Whinnett et al. 2005b) as follows:
may not be
Coalescent-based
1 We obtain molecular sequence information from 22
pairs of ithomiine and heliconiine taxa differing mor-
phologically at species or subspecies level across the
suture zone, doubling the number of taxon pairs
studied previously.
2 We double the mitochondrial sequence data per spec-
imen to ?2100 bp
3 We add three nuclear loci: sex-linked Tpi (?800 bp),
as well as autosomal Tektin (?700 bp) and Rpl5
(?650 bp).
4 We obtain sequence data from larger numbers of
individuals per subspecies.
5 We exploit more rigorous coalescent-based analyses
to estimate relative split time, while allowing for
variation in effective ancestral population size (h) and
stochastic coalescence times, instead of using clock-
based simple nucleotide divergence as estimators of
split time
Methods
Ithomiine and heliconiine samples of 22 species, each
with morphologically divergent subspecies across the
suture zone (Table 1), were collected (Fig. 1 and
Table S1). These comprised 17 species from nine ithom-
iine genera and a further five Heliconius species. Collec-
tions were made in 2002, 2004 and 2005. All specimens
were identified by GL and JM. Wings were removed
from specimens and kept in transparent envelopes for
identification, and remaining tissue was preserved in
salt-saturated DMSO and stored at )20 ?C. Both wings
and tissue are held at UCL. Whenever possible, at least
three individuals from each subspecies were sampled
(Table S1) although for a few rarer subspecies, such as
Scada reckia junina and Melinaea satevis tarapotensis, we
obtained only 1–2 specimens. Representative outgroup
taxa were also used (Table 1).
DNA was extracted from one-third of the thorax of
each specimen using the DNeasy Blood and Tissue Kit
(QIAGEN), and DNA extracts stored at )20 ?C. We
amplified ?2130 bp of mtDNA comprising cytochrome
oxidase I (CoI), tRNA-leu and the 5¢-end of cytochrome
oxidase II (CoII), as well as three nuclear loci. The
nuclear loci were Tektin (700–740 bp), ribosomal protein
L5 (Rpl5) (400–900 bp) and triose phosphate isomerase
(Tpi) (690–770 bp) and were chosen as they have been
found to have relatively rapid rates of evolution and
are readily amplifiable in both the Ithomiini and Helico-
nius species (Mallarino et al. 2005; Whinnett et al.
2005a; Dasmahapatra et al. 2007, 2010; Elias et al. 2009;
de-Silva et al. 2010). Details of PCR primers and reac-
tion conditions are provided in Tables S2 and S3. PCR
products were cleaned and cycle sequenced from both
ends using the Big Dye Terminator 3.1 Cycle Sequenc-
ing Kit (Applied Biosystems). Cycle sequenced products
were analysed on a 3730xl Genetic Analyzer (Applied
Biosystems).Chromatograms
edited with ChromasPro 1.41 (Technelysium Pty Ltd).
Sequence alignment was carried out using ClustalX and
checked manually. Both Rpl5 and Tpi spanned several
introns and varied in length owing to indels. When
indel heterozygosity resulted in variable-sized products
from the same individual, indels could often be identi-
fied, and the two alleles ‘deconvoluted’ and haplotyped
using information from the double peak signal reads
beyond the indel after bidirectional sequencing (Flot
et al. 2006; Dasmahapatra et al. 2007). Sequence data
were lodged with Genbank (HM051677–HM052795,
were checkedand
4286 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 5

Table S1). In addition, pre-existing data from Genbank
were also used (Table S1).
Approximate Bayesian computational methods have
been developed to test for simultaneous divergence of
taxa taking into account variance in coalescence times
(Hickerson et al. 2006). However, currently msBayes
(Hickerson et al. 2007) does not allow for variation in
substitution rates between taxa, a feature of our data
set. Instead, we employed the coalescent-based Bayesian
program MCMCcoal 1.2 (Yang 2002; Rannala & Yang
2003) to estimate parameters relevant to neutral coales-
cence for each subspecies pair, followed by the scaling
of parameter estimates to account for this variation in
substitution rates. The program allows for the estima-
tion of split times for a pair of taxa (s = tl, the product
of split time in generations and substitution rate per
generation) as well as of ancestral and current effective
population sizes (h = 4Nl, the product of genetically
effective population size and substitution rate). Both
split time and effective population size parameters are
scaled by substitution rate l, which under the neutral
theory is expected to equal the mutation rate. To correct
for among-lineage variation in substitution rate, we
arranged our taxa into groups of three consisting of a
Table 1 The number of sequences of the four loci used in each of the pairs of subspecies for the 22 species across the suture zone
Species Subspecies pair
mtDNA
TektinRpl5Tpi
Outgroup
taxon
used
ssp1ssp2ssp1 ssp2 ssp1ssp2ssp1 ssp2
Heliconius erato
Heliconius melpomene
Heliconius pardalinus
Heliconius numata*
favorinus
amaryllis
sergestus
tarapotensis +
bicoloratus
ssp. nov.
mothone
tarapotensis
ssp. nov.
ssp. nov.
serdolis
ssp. nov.
mendax
ssp. nov.
quotidiana
ethica
quadrona
ssp. nov.
lamia
rindgei
derasa
gracilis
deceptus +
cf. phasianita
emma
aglaope
dilatus
aurora +
silvana
ucayalensis
phasiana
cydon
hicetus
janarilla
lota
flexibilis
kezia
margarita
batesi
junina
pachiteae
arzalia
pharo
corena
aquinia
aureola
fallax
46
4
4
7
4
4
3
4
5
5
3
6
4
5
4
4
7
3
2
6
10
7
5
2
7
8
3
5
Heliconius demeter ucalayensis
Heliconius ethilla aerotome
Heliconius ethilla aerotome
Heliconius ethilla aerotome
13
4
4
Heliconius demeter
Melinaea marsaeus
Melinaea satevis
Melinaea menophilus
Oleria onega
Oleria gunilla
Hyposcada kena
Hyposcada anchiala
Hyposcada illinissa
Scada zibia
Scada reckia
Brevioleria aelia
Brevioleria arzalia
Napeogenes pharo
Napeogenes sylphis
Ithomia salapia
Pseudoscada florula
Mechanitis mazaeus
7
7
2
13
22
7
5
64
5
5
4
8
4
4
7
3
8
6
8
3
8
358
9
2
11
13
6
3
3
7
5
7
7
4
4
6
4
8
9
12
5
12
7
8
2
13
16
6
4
8
7
5
6
4
6
3
6
4
6
6
10
6
7
Heliconius sara sara
Melinaea ludovica ludovica
Melinaea ludovica ludovica
Melinaea ludovica ludovica
Oleria gunilla serdolis
Oleria onega janarilla
Hyposcada anchiala mendaz
Hyposcada kena flexibilis
Hyposcada anchiala mendax
Scada reckia junina
Scada zibia quotidiana
Brevioleria arzalia arzalia
Brevioleria aelia
Napeogenes larina otaxes
Napeogenes inachia pozziana
Ithomia lagusa peruana
Pseudoscada timna timna
Scada zibia quotidiana†
10
1
22
11
7
3
5
6
6
1
10
3
4
8
6
9
10
20
5
5
9
3
5
5
7
7
4
7
4
6
3
10
6
7
22
27
11
3
5
6
12
1
12
3
7
11
7
4
13
12
5
6
4
9
6
5
1
10
2
9
11
7
7
12
17
10
10
4
4
5
7
1
12
4
2
8
6
7
16
Total number of sequences 390277286300
*Heliconius numata is locally polymorphic (Joron et al. 2001).
†An appropriate outgroup within Mechanitis is not available because of uncertain relationships among Mechanitis species
(Dasmahapatra et al. 2010).
τA
X1
θA
θX1
θX
θX2
τX
tX
X2
Y
Fig. 2 A group of three taxa comprising a subspecies pair, X1
and X2, and an outgroup taxon, Y, used in the MCMCcoal
analysis. Six parameters were estimated from each subspecies
pair: hX1, hX2, hX, hA, sXand sA. For recently diverged pairs of
taxa, the coalescence time, tX, may be considerably longer than
the split time sX.
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4287
? 2010 Blackwell Publishing Ltd

Page 6

subspecies pair (X1, X2) belonging to species X, together
with a single representative of a closely related out-
group species Y (Fig. 2 and Table 1). Using this model,
MCMCcoal was set to estimate six parameters: scaled
population sizes for the subspecies pairs, hX1, hX2, and
of their ancestors, hX, hA, as well as scaled split times sX
and sA; Fig. 2). In each analysis, the major parameter of
interest was sX, the population split time for the focal
pair of subspecies.
Heredity multipliers (inheritance scalars) of 0.25, 0.75,
1.00 and 1.00 were used in MCMCcoal for mtDNA, Tpi,
Tektin and Rpl5 loci, respectively, as correction factors
for h in mtDNA and sex-linked Tpi, relative to autoso-
mal loci (Yang 2007a). Relative substitution rates (l) of
the four loci were estimated by comparing maximum
likelihood ithomiine and Heliconius tree lengths for each
locus using PAML (Yang 2007b). Tree topologies used
for relative rate estimation were based on (Brower et al.
Table 2 Details of sampled locations shown in Fig. 1
Site no. Site Name LatitudeLongitude
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25a
25b
26
27
28
29
30
31
32
33
34
35
36
37
38
39a
39b
40
41
42
43
Jorge Cha ´vez, San Martı ´n
Puente Serranoyacu, San Martı ´n
Puente Aguas Verdes, San Martı ´n
Km 103.1 Tarapoto-Yurimaguas, San Martı ´n
Km 78 Tarapoto-Yurimaguas, San Martı ´n
Santa Rosa de Davidcillo, San Martı ´n
Convento, San Martı ´n
Km 7.2 Pongo-Barranquita, San Martı ´n
Rı ´o Shucshuyacu, Pongo del Cainarachi, San Martı ´n
Km 43–46 Tarapoto-Yurimaguas, San Martı ´n
Km 22 Tarapoto - Yurimaguas, San Martı ´n
Km 24 Tarapoto-Yurimaguas, San Martı ´n
Km 35 Tarapoto-Yurimaguas, San Martı ´n
Km 21 Tarapoto-Yurimaguas, San Martı ´n
La Antena, San Martı ´n
Bocatoma Rı ´o Shilcayo, San Martı ´n
Tu ´nel⁄Biodiversidad, San Martı ´n
Km 8 Tarapoto-Yurimaguas, San Martı ´n
Urahuasha, San Martı ´n
Shapaja, San Martı ´n
Chumı ´a, San Martı ´n
Rı ´o Pucayaquillo
Chasuta, San Martı ´n
Robashca, PNCAZ, San Martı ´n
Camp 1 Robashca-Quebrada Yanayacu, PNCAZ, Loreto
Camp 2 Quebrada Yanayacu, PNCAZ, Loreto
Laguna del Mundo Perdido, San Martı ´n
Boca del Ushpayacu, Loreto
Rı ´o Pauya⁄ Rı ´o Cushabatay confluence, PNCAZ, Loreto
Quebrada Paco, Loreto
Km 22, Nuevo Lima - La Perla del Ponacillo, San Martı ´n
Km 22, Nuevo Lima - Selva Andina, San Martı ´n
Quebrada Huacanqui, Loreto
Pongo del Rı ´o Pauya, Loreto
Cerro Mira Culo, Loreto
Cachatigre, Rı ´o Biabo, PNCAZ
Can ˜o Negro, Rı ´o Biabo, PNCAZ
Quebrada Tunuya, San Jeronimo, Loreto
Cachiyacu, San Martı ´n
Can ˜o Tushmo, Lago Yarinacocha, Ucayali
El Pandisho, Lago Yarinacocha, Ucayali
IVITA, Ucayali
Bosque von Humboldt, Ucayali
8 km west of Boquero ´n del Padre Abad, Ucayali
Villa Jennifer, Tingo Marı ´a, Hua ´nuco
5?39¢43¢¢S
5?40¢31¢¢S
5?41¢16¢¢S
5?58¢18¢¢S
6?08¢58¢¢S
6?14¢57¢¢S
6?16¢22¢¢S
6?17¢12¢¢S
6?20¢07¢¢S
6?22¢36¢¢S
6?24¢30¢¢S
6?24¢58¢¢S
6?25¢29¢¢S
6?24¢23¢¢S
6?27¢18¢¢S
6?27¢27¢¢S
6?27¢30¢¢S
6?27¢43¢¢S
6?27¢60¢¢S
6?34¢47¢¢S
6?36¢57¢¢S
6?35¢10¢¢S
6?34¢28¢¢S
6?42¢38¢¢S
6?44¢27¢¢S
6?44¢55¢¢S
6?45¢05¢¢S
7? 05¢ 40¢¢S
7?09¢18¢¢ S
7?09¢49¢¢S
7?10¢04¢¢S
7?11¢36¢¢S
7?24¢13¢¢S
7?24¢19¢¢S
7?27¢12¢¢S
7?43¢05¢¢S
7?45¢10¢¢S
7?46¢32¢¢S
6?26¢00¢¢S
8?20¢32¢¢S
8?19¢02¢¢S
8?38¢28¢¢S
8?49¢48¢¢S
9?06¢13¢¢S
9?16¢11¢¢S
77?43¢58¢¢W
77?40¢28¢¢W
77?38¢00¢¢W
76?13¢55¢¢W
76?16¢00¢¢W
76?16¢30¢¢W
76?17¢21¢¢ W
76?13¢54¢¢W
76?15¢15¢¢W
76?17¢01¢¢W
76?19¢30¢¢W
76?18¢24¢¢W
76?15¢02¢¢W
76?18¢07¢¢W
76?17¢54¢¢W
76?20¢58¢¢W
76?17¢15¢’W
76?19¢23¢¢W
76?20¢05¢¢W
76?15¢46¢¢W
76?11¢10¢¢W
76?13¢05¢¢W
76?08¢12¢¢W
76?01¢60¢¢W
75?58¢54¢¢W
75?56¢23¢¢W
75?52¢08¢¢W
75?51¢30¢¢W
75?43¢24¢¢W
75?41¢21¢¢W
76?20¢42¢¢W
76?20¢06¢¢W
75?46¢11¢¢W
75?49¢11¢¢W
75?50¢16¢¢W
76?20¢51¢¢W
76?20¢03¢¢W
74?41¢20¢¢W
76?23¢00¢¢W
74?35¢31¢¢W
74?35¢28¢¢W
74?57¢13¢¢W
75?03¢28¢¢W
75?44¢28¢¢W
76?00¢29¢¢W
4288 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 7

2006) for ithomiines and (Beltra ´n et al. 2007) for Helico-
nius; taxa used for this estimation are shown in Fig. S1.
Relative substitution rates of mtDNA, Tpi, Tektin and
Rpl5 were estimated as 1.00:1.79:0.80:1.23 (ithomiines)
and 1.00:1.02:0.31:0.57 (Heliconius). By setting the mito-
chondrial substitution rates to one, the rate of substitu-
tion at mtDNA is assumed to be the same in the
ithomiine and heliconiine lineages. This allowed the
comparison of parameter estimates among the groups.
For each of the six estimated parameters (Fig. 2),
MCMCcoal requires gamma distributed priors specified
by parameters a and b, which determine the mean (a⁄b)
and variance (a⁄b2) of the prior distributions. In these
analyses, we employed permissive priors for all esti-
mated parameters to minimize any strong influence of
the prior on the posterior distribution. Some subspecies
pairs show few fixed sequence differences, suggesting
recent divergence. Therefore, for both s estimates we set
priors with a < 1, which allow s = 0. The a and b val-
ues for the two s estimates were chosen to reflect the
fact that sX< sA(Fig. 2). Thus, a and b were set to 1.1
and 28 respectively for all four h priors, 0.7 and 3.5 for
sX, and 0.7 and 35 for sA. To avoid bias, identically per-
missive prior distributions were specified for all 22 sub-
species pairs.
MCMCcoal analyses were conducted using a burn-in
of 20 000 iterations, following which 20 000 samples
were taken from the posterior distribution with a sam-
ple interval of 10. ‘Fine-tune parameters’ were opti-
mized for each subspecies pair to maintain acceptance
proportions with the interval of (0.15, 0.7) as suggested
in the manual (Yang 2007a). Output stabilization was
examined to ensure the reliability of parameter esti-
mates. Each analysis was run twice with different ran-
dom seeds to check consistency of parameter estimates
and assess convergence. In all cases, parameter esti-
mates converged on the same values in different runs.
The trees and sequences used to estimate relative
substitution rates l of each locus (Fig. S1) were also
used to test for variation in evolutionary rate among
lineages, or deviation from molecular clock, using
PAUP* 4.0b10 (Swofford 2003). Deviation from a molec-
ular clock was found within both Ithomiini (mtDNA:
2Dln L = 81, P < 10)5; Tektin: 2Dln L = 86, P = 10)6with
d.f. = 32) and Heliconius (mtDNA: 2Dln L = 43, P = 10)4;
Tektin: 2Dln L = 26, P = 0.007; Tpi: 2Dln L = 56, P = 10)7
with d.f. = 11). Thus, it was desirable to correct the esti-
mates of s obtained from MCMCcoal to reflect relative
absolute time among lineages.
The ‘local clock’ model implemented in PAML (Yoder
& Yang 2000; Yang & Yoder 2003) was used to estimate
relative substitution rates for each of the four loci in dif-
ferent branches of the ithomiine and heliconiine trees.
Within the Ithomiini, separate relative substitution rates
were calculated simultaneously for each of the nine
genera using sequence data from 35 species and sub-
species (Fig. S1). Within Heliconius, sequences from 13
species and subspecies were used to calculate separate
substitution rates for each locus for melpomene-wallacei-
hierax and erato-sara groups (Fig. S4). In each lineage
involving a subspecies pair, rates for the four loci were
averaged to provide local rates of substitution which
were then used to scale local estimates of s obtained
with MCMCcoal. Lineage correction factors are shown
in Table 3. The estimated substitution rates varied
rather strongly with a standard deviation of 25%.
Estimates of split time between subspecies pairs, sX,
obtained from MCMCcoal were compared with uncor-
rected average pairwise sequence divergence calculated
in MEGA 4.0 (Tamura et al. 2007) between each sub-
species in a pair. Values of sX from MCMCcoal were
also compared with those from IMa, another coales-
cent-based Bayesiananalysis
Wakeley 2001; Hey & Nielsen 2007). For the purposes
of this investigation, migration rates were set to zero
in IMa. As we regularly found some sites with three
states in these analyses, we used the Hasegawa-Kishi-
no-Yano (HKY) (Hasegawa et al. 1985) substitution
model for all loci rather than the infinite sites model.
(MCMCcoal employs a similar Jukes-Cantor substitu-
tion model which allows multiple changes at a single
site). Inheritance scalars and relative substitution rates
for each locus were set as in MCMCcoal. The trun-
cated uniform priors are specified differently from
those in MCMCcoal. Following trial runs, the upper
bound for the priors on hX1, hX2, hX were set at 15,
while the upper bound for s was set at 10. Metropolis-
coupling was implemented using five chains with a
two step heating scheme (Hey 2009). A burn-in of a
million steps was used, and effective sample size val-
ues were examined to assess convergence, following
which the posterior distribution was sampled for an
hour with a sample interval of 10 (the number of steps
sampled varied between 170 000 and 670 000). Each
data set was analysed twice to ensure consistent
parameter estimates.
To investigate differential patterns of divergence
among lineages further, gene genealogies of example
genera Hyposcada and Melinaea were estimated. Each of
these two genera includes three species showing mor-
phological subspecies differences across the Huallaga
suture zone. After hierarchical likelihood ratio tests
implemented within MODELTEST 3.7 (Posada & Crandall
1998) were used to select the most likely model of
sequence evolution, maximum likelihood trees were
constructed using PAUP* (Swofford 2003) from concate-
nated sequences of all four loci using the GTR+I+G
model.Inaddition, for
program(Nielsen&
each genus,separate
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4289
? 2010 Blackwell Publishing Ltd

Page 8

Table 3 Estimates of subspecies split times (sX) and effective population size (h) estimates derived from MCMCcoal. All estimates are scaled by the neutral mutation rate, l, cal-
culated relative to the per site mutation rate of mtDNA. Figures in parentheses are 95% credibility intervals. The lineage correction factors are the multipliers used to correct sX
for among-lineage variation in substitution rate
Species
sX
hX1
hX2
hX
Lineage
correction
Corrected sX
Heliconius demeter
0.003983 (0.002228–0.006391)
0.014496 (0.007755–0.027739)
0.021873 (0.013729–0.035694)
0.041461 (0.023856–0.073746)
1.2
0.004780 (0.002674–0.007669)
Oleria onega
0.003683 (0.002152–0.005274)
0.056224 (0.040405–0.078309)
0.076784 (0.052647–0.118445)
0.041189 (0.021564–0.070165)
1.0
0.003683 (0.002152–0.005274)
Heliconius pardalinus
0.003250 (0.001310–0.006747)
0.005568 (0.001858–0.019430)
0.009163 (0.004195–0.022157)
0.025293 (0.009029–0.073215)
1.1
0.003575 (0.001441–0.007422)
Heliconius erato
0.002856 (0.001512–0.004178)
0.086824 (0.040903–0.187583)
0.090473 (0.041586–0.200876)
0.053195 (0.031076–0.087631)
1.2
0.003427 (0.001814–0.005014)
Napeogenes pharo
0.002310 (0.000913–0.003864)
0.029507 (0.015746–0.073117)
0.027792 (0.013313–0.062165)
0.030085 (0.018087–0.018087)
1.3
0.003003 (0.001187–0.005023)
Scada reckia
0.002648 (0.001062–0.005696)
0.004932 (0.002143–0.012130)
–
0.018092 (0.004333–0.056524)
1.1
0.002913 (0.001168–0.006266)
Hyposcada anchiala
0.002527 (0.001238–0.003805)
0.002486 (0.001179–0.005338)
0.002757 (0.001353–0.005901)
0.002478 (0.000151–0.013348)
1.0
0.002527 (0.001238–0.003805)
Napeogenes sylphis
0.001401 (0.000562–0.002162)
0.009124 (0.004297–0.020502)
0.024823 (0.013378–0.050994)
0.010292 (0.004769–0.019603)
1.3
0.001821 (0.000731–0.002811)
Pseudoscada florula
0.002540 (0.001422–0.003649)
0.020931 (0.009993–0.051900)
0.018260 (0.009491–0.038680)
0.030907 (0.011878–0.065928)
0.7
0.001778 (0.000995–0.002554)
Hyposcada kena
0.001773 (0.000740–0.003246)
0.002135 (0.000915–0.004884)
0.000413 (0.000055–0.001690)
0.023967 (0.009023–0.074433)
1.0
0.001773 (0.000740–0.003246)
Mechanitis mazaeus
0.001163 (0.000700–0.001923)
0.007591 (0.004171–0.014740)
0.022701 (0.012891–0.049628)
0.015580 (0.009927–0.025424)
1.2
0.0001396 (0.000840–0.002308)
Heliconius numata
0.001215 (0.000651–0.001929)
0.079231 (0.029334–0.209320)
0.052320 (0.014840–0.166982)
0.014919 (0.008365–0.026411)
1.1
0.001337 (0.000716–0.002122)
Scada zibia
0.001165 (0.000535–0.001900)
0.033939 (0.011897–0.122042)
0.017957 (0.008800–0.037615)
0.015727 (0.008849–0.026913)
1.1
0.001282 (0.000589–0.002090)
Oleria gunilla
0.001177 (0.000611–0.001980)
0.004542 (0.002086–0.010743)
0.007541 (0.003952–0.015641)
0.004193 (0.001750–0.009067)
1.0
0.001177 (0.000611–0.001980)
Ithomia salapia
0.000753 (0.000393–0.001388)
0.008974 (0.004600–0.021947)
0.003587 (0.001687–0.008238)
0.010504 (0.006053–0.018711)
1.3
0.000979 (0.000511–0.001804)
Heliconius melpomene
0.000425 (0.000030–0.000974)
0.060535 (0.014381–0.178497)
0.018014 (0.006128–0.085410)
0.029296 (0.018691–0.045681)
1.1
0.000468 (0.000033–0.001071)
Hyposcada illinissa
0.000378 (0.000174–0.000703)
0.029416 (0.005346–0.135927)
0.015179 (0.003276–0.099033)
0.003877 (0.002126–0.006908)
1.0
0.000378 (0.000174–0.000703)
Melinaea marsaeus
0.000678 (0.000401–0.001068)
0.017304 (0.010347–0.028736)
0.005319 (0.002843–0.010072)
0.052688 (0.033262–0.089017)
0.5
0.000339 (0.000201–0.000534)
Melinaea satevis
0.000554 (0.000044–0.001833)
0.021756 (0.007108–0.102169)
0.006909 (0.000870–0.081087)
0.022020 (0.013147–0.039588)
0.5
0.000277 (0.000022–0.000917)
Brevoleria arzalia
0.000294 (0.000010–0.000995)
0.028373 (0.004151–0.137638)
0.023330 (0.002771–0.124083)
0.008382 (0.002987–0.017189)
0.8
0.000235 (0.000008–0.000796)
Melinaea menophilus
0.000229 (0.000009–0.000620)
0.030204 (0.004407–0.137586)
0.028671 (0.009503–0.093340)
0.027408 (0.018571–0.040049)
0.5
0.000115 (0.000005–0.000310)
Brevoleria aelia
0.000038 (0.000001–0.000143)
0.033757 (0.003731–0.144522)
0.039463 (0.005265–0.152316)
0.004094 (0.002346–0.006654)
0.8
0.000030 (0.000000–0.000114)
4290 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 9

bootstrapped neighbour-joining trees for each locus
were constructed using MEGA 4.0 (Tamura et al. 2007) to
investigate any locus-specific effects.
Results
A total of 1282 mitochondrial and nuclear sequences
(390 mtDNA, 277 Tektin, 286 Rpl5, 300 Tpi) comprising
?1.5 Mb total bases from 399 specimens within the 30
species (Tables 1 and S1) were used for the analyses.
The average coverage was seven haplotypes sequenced
per locus for each subspecies (Table 1).
As explained previously, MCMCcoal estimates of
split time for the 22 subspecies pairs, sX, were corrected
for variation among lineages in substitution rate. The
resultant estimates of sX varied more or less continu-
ously from near zero (Brevioleria aelia and Melinaea
menophilus) up to 0.005 (Heliconius demeter and Oleria
onega) with little tendency towards clustering (Fig. 3).
A very similar pattern of scattered split times is found
even for uncorrected sX values, suggesting problems
with correction are not the cause (Table 3). If the suture
zone formed as a result of contact between biotas sepa-
rated by a simple vicariant event, MCMCcoal diver-
gence estimates for the 22 subspecies pairs are expected
to be clustered around a single value. The 95% credibil-
ity intervals (i.e. the interval containing 95% of the pos-
terior probability) of all 22 divergence estimates should
overlap. In this study, 95% credibility intervals for sX
were wide, a consequence of using only four loci; how-
ever, a significant fraction is nonoverlapping (Fig. 3),
strongly indicating that not all pairs of taxa split at the
same time, thereby implicating a problem with a single
split time.
In addition to the assessment based on credibility
intervals, we carried out two ad hoc tests of significance
to gauge statistically whether our estimates could have
resulted from a single vicariance event. In the first ad
hoc test, we repeated the initial MCMCcoal analyses but
constrained sXby specifying a very narrow prior cen-
tred about the same average sX(sX= 0.001695) for all
22 subspecies pairs (by setting priors on sX: a = 20000,
b = 11900000, giving a very low prior variance of
2 · 10)10). Priors for all other parameters were left
unchanged. For each taxon pair, the difference in maxi-
mum likelihood obtained in MCMCcoal runs between
constrained and unconstrained sXwas used to carry out
likelihood ratio tests. As expected, in most (17 of the
22) comparisons the unconstrained calculations showed
a higher likelihood than the constrained calculations,
seven of which were significantly higher likelihoods
after Bonferronicorrection
P < 0.0023; Table S4). Alternatively, summing likeli-
hoods from unconstrained and constrained runs across
all22speciesindicates
d.f. = 22, P ? 10)42) that the same split time does not
apply to all species.
The second ad hoc test was carried out in two stages.
First, using MCMCoal a single sXfor all 22 taxon pairs
was estimated combining sequence data from all species
in a single analysis: effectively calculating a single sXof
0.000922 across the suture zone from 88 loci (four loci
from 22 species). Then, individual analyses for each
taxon pair were repeated with sXconstrained by a nar-
row prior to be centred about the single divergence
value obtained in the first stage (the prior on sXwas
forced to have a median of 0.000922 using a = 20 000,
b = 22 000 000, giving a very low prior variance of
()2Dln L > 9.32, d.f. = 1,
strongly()2Dln L = 263,
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
Heliconius demeter
Oleria onega
Heliconius pardalinus
Heliconius erato
Napeogenes pharo
Scada reckia
Hyposcada anchiala
Napeogenes sylphis
Pseudoscada florula
Hyposcada kena
Mechanitis mazaeus
Heliconius numata
Scada zibia
Oleria gunilla
Ithomia salapia
Heliconius melpomene
Huposcada illinissa
Melinaea marsaeus
Melinaea satevis
Brevioleria arzalia
Melinaea menophilus
Brevioleria aelia
Subspecies split time, τX
Fig. 3 MCMCcoal divergence estimates
(sX, the product of absolute split time
and substitution rate) for the 22 subspe-
cies pairs across the suture zone. Split
times are scaled to allow for lineage-
specific substitution rate variation. For
each species, the median sX is shown
with 95%
credibility
upper and lower dashed lines corre-
spond to the lower credibility interval
for the largest value of sX (Heliconius
demeter) and the higher credibility inter-
val for the smallest sX(Brevioleria aelia).
These highlight the lack of overlap
among estimates of split time among
subspecies pairs.
intervals.The
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4291
? 2010 Blackwell Publishing Ltd

Page 10

4 · 10)11), followed by likelihood ratio testing between
constrained and unconstrained sX as before. No out-
group species were used in this second ad hoc test. After
Bonferroni correction, in three of the 22 comparisons the
unconstrained calculations showed a significantly higher
likelihood than the constrained calculations (Table S4).
However, summing likelihoods from unconstrained and
constrained runs across all 22 species again indicated
strongly ()2Dln L = 85.6, d.f. = 22, P ? 10)8) that a sin-
gle split time cannot be applied across all species pairs.
It should be noted that these ad hoc tests are not
strictly valid as the MCMC procedure is not intended
to generate maximum likelihood estimates but to
explore the region around the likelihood peak in pro-
portion to posterior probability. In the likelihood ratio
tests described, we employed the value showing the
maximum likelihood from the 20 000 iterations sam-
pled: this will rarely be the overall maximum likelihood
value. This explains why not all likelihoods are greater
in unconstrained than constrained analyses (although
unconstrained likelihoods were never significantly less
than constrained likelihoods, Table S4). The visual
inspection of 95% credibility provides no quantified sta-
tistical test but is in theory more appropriate and intui-
tive. Nonetheless, although the tests performed here are
somewhat informal, our ad hoc approach is today used
widely in similar genealogical problems, for example to
test the null hypothesis of zero gene flow among popu-
lations (Hey & Nielsen 2007).
The estimates of sXfor the taxon pairs using the two
different coalescent programs, MCMCcoal and IMa,
were strongly correlated (Fig. 4a, r2= 0.86, P = 10)10, 20
d.f.), demonstrating that the programs recovered similar
information from the data in spite of different prior
specification. Split time sX (from MCMCcoal) is also
correlated with raw net sequence divergence among
subspeciespairs at mtDNA
P = 6 · 10)5, 20 d.f.), and also, more weakly, at Tpi
(Fig. 4c, r2= 0.20, P = 0.04, 20 d.f.). We found no signif-
icant correlations between sXand sequence divergence
at either Tektin or Rpl5 (Fig. 4d, e). The weaker correla-
tion of MCMCcoal sX with sequence divergence than
with IMa sXsuggests that similar additional informa-
tion about split time is recovered by both coalescent-
based analysis programs and that this information is
lacking in raw sequence divergence. There were no
clear differences in the estimated effective population
sizes of pairs of subspecies (hX1and hX2): 95% credibility
intervals overlapped strongly in all subspecies pairs
(Table 3). hX1and hX2estimates were strongly corre-
lated with one another (r2= 0.67, P = 1 · 10)5, 19 d.f.)
showing that estimated effective population sizes of
pairs of subspecies are probably similar to each other
(Table 3).
(Fig. 4b,
r2= 0.56,
A comparison of Hyposcada and Melinaea maximum
likelihood trees (Fig. 5) and neighbour-joining trees at
each locus (Fig. 6) demonstrates the effect of more
recent species and subspecies diversification within the
latter. Within Hyposcada, both conspecifics and subspe-
cies form monophyletic clusters separated by relatively
long branches (Figs 5 and 6). In contrast, within Meli-
naea the branches are shorter, and species (M. marsaeus
and M. satevis) and subspecies paraphyly is widespread
(Figs 5 and 6). These patterns are present across all loci
(Fig. 6). This is probably a result of incomplete lineage
sorting owing to the recency of diversification and lar-
ger ancestral population sizes in Melinaea (hXfor M. sat-
evis = 0.022, M. menophilus = 0.027, M. marsaeus = 0.053)
compared to in Hyposcada (hXfor H. anchiala = 0.0025,
H. illinissa = 0.0039, H. kena = 0.024) (Table 3).
Discussion
Suture zones form the boundaries between centres of
endemism under vicariance theories of diversification,
with centres of endemism interpreted as likely sites for
former Pleistocene refuges in heliconiine and ithomiine
butterflies (Brown et al. 1974; Brown 1979, 1987b). Here,
we use coalescent-based analysis of mutilocus sequence
data from 22 pairs of taxa across an Amazonian butter-
fly suture zone to test a hypothesis of simultaneous
split times. Our analysis clearly reveals a scattered time
course of diversification, rather than a tight cluster of
split times as predicted by a simple vicariant event.
These results suggest that diversification is more com-
plex than that suggested by the simplistic, single-split
vicariance model.
With molecular technology and improved methods of
analysis rapidly advancing, we now have tools to inves-
tigate the effects of coalescence time on overall diver-
gence among pairs of related taxa. The time to
coalescence, tX, of two gene sequences now present in
two descendant species is always longer than the time
back to the split of the two species, sX(Fig. 2), because
coalescence must always occur in the ancestor some
time before species divergence. As the species split time
increases, the discrepancy between the two diminishes
as the fraction of the total time to coalescence within
the ancestral population becomes smaller relative to the
time within the descendants. For recently diverged pop-
ulations with large effective population sizes, as is
likely for many Amazonian butterflies, the discrepancy
can be substantial (Edwards & Beerli 2000; Nichols
2001; Beaumont et al. 2010).
A simple measure of nucleotide divergence will
therefore give an inadequate estimate of split time. In
such cases, coalescent-based analyses such as MCMC-
coal and IMa are useful to estimate split times and
4292 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 11

effective population sizes of ancestral and descendant
populations using mutlilocus data sets, while account-
ing for coalescence in the ancestral population. We
report a strong correlation (r2= 0.56) between sX and
mitochondrial sequence divergence. The correlation of
sXwith sequence divergence at Tpi is weaker, and cor-
relations were not significant with sequence divergence
at Rpl5 and Tektin. This decline in correlation may
partly reflect the larger relative effective population
sizes of the nuclear genes (n.b., mtDNA, Tpi, Rpl5 and
Tektin ‘heredity multipliers’ of population size are
expected to be 0.25, 0.75, 1, 1, see Methods), although it
is also likely to be due to smaller sequence read lengths
at each nuclear gene and increased stochasticity owing
to indel and noncoding DNA variation among lineages.
Similar discrepancies between coalescent-based diver-
gence estimates and sequence divergence were reported
by Hurt et al. (2009) in their analysis of divergences
between species pairs across the Isthmus of Panama.
While we report a strong correlation between sXand
mitochondrial sequence divergence, there is substantial
unexplained variance (Fig. 4b). For example, mtDNA
divergence within H. demeter is ?6%, which would
suggest a split time of 6–7 Myr from applying the
r2 = 0.56
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
Raw mtDNA sequence divergence
r2 = 0.20
0
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
Raw Tpi sequence divergence
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
Raw Tektin sequence divergence
0
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
Raw Rpl5 sequence divergence
Subspecies split time (MCMCcoal), τX
Subspecies split time (MCMCcoal), τX
00.0010.002 0.003 0.0040.005 0.0060.0070 0.0010.0020.0030.004 0.0050.0060.007
Subspecies split time (MCMCcoal), τX
Subspecies split time (MCMCcoal), τX
00.001 0.002 0.003 0.0040.005 0.0060.0070 0.0010.0020.0030.0040.005 0.0060.007
r2 = 0.01
r2 = 0.06
0
100 000
200 000
300 000
400 000
500 000
600 000
00.0010.002 0.0030.0040.005 0.0060.007
Subspecies split time in years, T (IMa)
r2 = 0.86
Subspecies split time, τX (MCMCcoal)
(a)
(b) (c)
(d) (e)
Fig. 4 Relationship between sXper nucleotide estimated using MCMCcoal and (a) split time in years estimated using IMa (assuming
a substitution rate of 2.3% per million years at mtDNA), (b) raw percentage sequence divergence between subspecies at mtDNA, (c)
Tpi, (d) Tektin, (e) Rpl5.
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4293
? 2010 Blackwell Publishing Ltd

Page 12

traditionally accepted mtDNA substitution rate of 2.3%
per branch per million years for arthropods (Brower
1994). Our analyses indicate that in this case, the split
time is in fact less than half of this (Fig. 4b, Table 3).
The opposite effect is seen in H. erato, where the net
mtDNA sequence divergence between subspecies is
only 0.05%, yet split times from both coalescent pro-
grams are considerably higher (Fig. 4B, Table 3), sug-
gesting that the subspecies are not of as recent origin as
that expected based from sequence divergence alone.
Mean coalescence time in the ancestor is 2Ne genera-
tions, and these apparent inconsistencies are presum-
ably owing to variation in ancestral and current
population sizes as well as stochastic variation in coa-
lescence time in ancestral populations: population size
estimates, ancestral (hX) and particularly current popu-
lation sizes (hX2and hX1), in H. erato are higher than in
the other species (Table 3). Such discrepancies demon-
strate the problem with simple application of sequence
divergence from single loci for dating split times and
the importance of approaches that simultaneously esti-
mate split times as well as effective population sizes.
Although we correct for variation in overall substitu-
tion rate among lineages within both Heliconius and It-
homiini, we have assumed that the mean rate of
substitution at mtDNA is the same in both groups to
make comparisons between Heliconius and ithomiine
subspecies pairs. There are no recent fossil calibrations
for nymphalid butterflies (Wahlberg et al. 2009), and sat-
uration of mtDNA changes among nymphalid groups
makes it difficult to assess relative substitution rates for
ithomiines and heliconiines. Even if average substitution
rates are not equal in the two groups, it is unlikely that
the conclusions reached here will be altered. In the It-
homiini, some taxa have high sX(Oleria onega and Scada
reckia), some show intermediate values (Pseudoscada flor-
ula and Oleria gunilla), and others have virtually no
divergence at all (Brevioleria aelia and Melinaea menophi-
lus) (Fig. 3). Similarly, within Heliconius, estimates of sX
vary from high (H. demeter and H. pardalinus) to near
zero (H. melpomene). The same pattern of variable split
times across the suture zone is found within each group.
There are many potential pitfalls with a coalescent-
based approach. For example, all substitutions are
02-512 H. anchiala interrupta
02-2105 H. anchiala interrupta
04-305 H. anchiala interrupta
02-1644 H. anchiala mendax
04-407 H. anchiala mendax
02-1645 H. anchiala mendax
02-1667 H. kena spp. nov.
02-1670 H. kena spp. nov.
02-1591 H. kena flexibilis
05-872 H. kena flexibilis
05-927 H. kena flexibilis
04-403 H. illinissa ssp. nov..
04-380 H. illinissa ssp. nov
04-381 H. illinissa ssp. nov.
05-806 H. illinissa margarita
02-1438 H. illinissa margarita
02-190 H. illinissa margarita
04-470 Oleria gunilla
100
100
100
100
100
100
79
86
98
02-382 M. marsaeus mothone
02-383 M. marsaeus mothone
02-377 M. marsaeus mothone
02-2119 M. marsaeus rileyi
02-1923 M. marsaeus phasiana
02-1924 M. marsaeus phasiana
02-1941 M. marsaeus phasiana
04-408 M. satevis tarapotensis
02-1254 M. satevis cydon
04-138 M. satevis cydon
04-214 M. satevis cydon
04-441 Scada reckia
02-688 M. isocomma simulator
02-689 M. isocomma simulator
02-2054 M. ludovica ludovica
02-542 M. ludovica ludovica
02-415 M. menophilus ssp. nov.
02-980 M. menophilus ssp. nov.
02-981 M. menophilus ssp. nov.
02-1262 M. menophilus hicetus
02-1421 M. menophilus hicetus
04-253 M. menophilus hicetus
100
86
100
GTR divergence
0.05
(a)
(b)
Fig. 5 Maximum likelihood trees based on concatenated mitochondrial and nuclear sequences of (a) Hyposcada and (b) Melinaea spe-
cies found on either side of the suture zone. Nodes with 50% or greater bootstrap support are labelled.
4294 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 13

Fig. 6 Neighbour-joining trees of mitochondrial CoI and CoII, Tektin, Rpl5 and Tpi for the ithomiine genera Melinaea and Hyposcada. Scale bar indicates K2P distances. Nodes with
50% or greater bootstrap support are labelled.
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4295
? 2010 Blackwell Publishing Ltd

Page 14

assumed neutral. This will not be strictly true for cod-
ing regions such as those we have selected. However,
given that the majority of divergence is synonymous
and that we have corrected for relative substitution
rates among loci (see Methods), we have minimized the
problem. A further simplifying assumption is the
absence of gene flow between subspecies since separa-
tion. Narrow hybrid zones between subspecies of many
of our species are known to occur naturally in this
region (Mallet 1993; Mallet & Lamas 2002) and have
been well studied in H. erato, H. melpomene and H. nu-
mata (Mallet 1989; Mallet et al. 1990; Joron et al. 2001);
in these zones, hybrid genotypes show unimodal distri-
butions as expected when rates of hybridization in the
zone are high (Jiggins & Mallet 2000). In contrast, con-
tact zones between the highly divergent Oleria onega
subspecies used in this study (de Silva et al. in prep.)
and Mechanitis mazaeus (Dasmahapatra et al. 2010) show
bimodal distributions of genotypes, implying low rates
of hybridization and gene flow within the zone. The fre-
quency of hybridization
decreases with increasing genetic divergence and fol-
lows an approximately log-linear distribution in Helico-
nius (Mallet et al. 2007), and the parapatric taxa studied
here are scattered across this spectrum. However,
unlike sympatric species, many of the subspecies stud-
ied here are largely isolated by distance, with the possi-
bility of limited gene exchange only within the narrow
contact zone: it is likely that gene flow has little effect
on divergence of each subspecies as a whole. Hybrid
zones usually ensure a strong barrier to gene flow, as
evidenced by strong frequency differences or fixed dif-
ferences maintained across the zone, delaying neutral
allele diffusion by many thousands of generations
(Barton & Hewitt 1983, 1985). We have not taken the
potential for long-range gene flow among subspecies
into account, so that our estimates of split time could
be somewhat underestimated. If compatibility does fol-
low an exponential decline, it is possible that fractional
underestimation would be greater for more recently
diverged subspecies, and the appropriate correction
could have the effect of reducing the variance in esti-
mated split times across the suture zone. In general
however, because split time is defined as the time since
gene flow was completely abolished, there will inevita-
bly be conflation of gene flow since split time and
recency of actual split time.
We have successfully compared relative split times sX
without a reliable estimate of absolute substitution rate,
l. An estimate of the absolute time of diversification
would require calibration, for example using fossils, but
these are not available for such recent events in butter-
flies. Average base divergence rate estimates for mito-
chondrial CoI and CoII vary from 2.3% per million
amongsympatrictaxa
years estimated in arthropods (Brower 1994) to 0.78–
1.02% per million years in swallowtail butterflies
(Zakharov et al. 2004). These rates suggest that diver-
gence across the suture zone began 1500–200 000 or
4000–500 000 years ago, depending on which calibration
is used. As substitution rates within CoI and CoII are
known to vary widely in insects (Zakharov et al. 2004),
these estimates are hardly reliable. Nonetheless, it is
clear that most taxa across this suture zone are recent,
probably diverging in the Pleistocene and Holocene.
The highly variable split times lead us to rule out a
single vicariant event as the cause of the suture zone.
Of course, our null hypothesis is simplistic, so several
alternative hypotheses can be examined. If the Amazon
basin went through glaciation-associated arid periods
during the Pleistocene, climatic oscillations could have
resulted in repeated forest contraction and expansion
centred on the same forest refuges. Variable split times
clearly do not rule out the possibility that different
pairs of subspecies were formed during each forest con-
traction⁄expansion episode. However, by rejecting a
simple ‘climatic forcing’ argument, we are left with a
model of ‘climatic assistance’ that generates nonreplica-
ble patterns across multiple species under identical
environmental cycles. A simple, parsimonious null
model becomes hard to reject: that the pattern of diver-
gences we observe is similar to that resulting from a
simple stochastic model of speciation, independent of
climatic oscillations, such as a constant low probability
of species⁄subspecies divergence and extinction per unit
time, perhaps approximately Poisson-distributed (cf.
Fig. 3). A ‘climatic assistance’ explanation involving
multiple split times but geographically coincident vicar-
iant events would be expected to affect all lineages sim-
ilarly: variable split times across the suture zone should
be found within all lineages. The same would be true
for simple stochastic null models of divergence. How-
ever, in the butterfly data presented here, there appear
to be strong lineage effects that are unexpected under
either hypothesis.
Within the genus Melinaea, all the taxa studied are of
veryrecentdivergence(Fig. 3).
fully sympatric species (M. menophilus, M. marsaeus,
M. satevis and M. isocomma), three of which show sub-
species differentiation across the zone, are themselves
virtually indistinguishable based on the loci sequenced
here: they have polyphyletic genealogies at each of
these loci (Figs 5 and 6). The species themselves are
extremely young (perhaps <100 000 years old), so that
their subspeciesmust be
?10 000 years or less). These results are in stark con-
trast to those from subtribe Oleriina (Hyposcada and Ole-
ria) which show considerably greater genetic divergence
(Figs 5 and 6) and longer split times among pairs of
Furthermore, four
even younger(perhaps
4296 K. K. DASMAHAPATRA ET AL.
? 2010 Blackwell Publishing Ltd

Page 15

subspecies (Fig. 3), greater even than Melinaea show
among species (Figs 5 and 6). Thus, both the random
speciation model and the climatic assistance model
seem inadequate. Even if climatic assistance ‘explains’
divergence of subspecies and species, lineage-specific
biological factors seem to have greater importance.
While ecological factors such as larval host plant
adaptation (Willmott & Freitas 2006) and changes in
pheromone signalling (Schulz et al. 2004; Estrada et al.
2010) may affect diversification in the Ithomiini and He-
liconiina, there is increasing evidence that changes in
wing colour patterns are important. Mimetic shifts cou-
pled with wing colour–mediated mate preference are
known to be important reproductive isolation factors
between several recently diverged Heliconius sister-spe-
cies as well as colour pattern races within the same spe-
cies (Jiggins et al. 2001; Chamberlain et al. 2009). In the
Ithomini, the diversity of colour patterns found varies
from genera to genera. Within Ithomia, wing pattern
changes are associated with diversification above the
species level (Jiggins et al. 2006). By contrast, in the
mimetically more homogenous Oleriina, local geo-
graphic isolation is thought to be more important in
diversification (de-Silva et al. 2010), demonstrating how
biological factors vary among lineages.
We should emphasize that we here use pairs of mor-
phologically distinct taxa only to test the null hypothe-
sis of simultaneous divergence. More than 40 other
species in these same butterfly groups, which are often
as abundant and widespread as the divergent taxa, are
also distributed across the suture zone and yet show no
obvious morphological or subspecies-level divergence.
Under the vicariant model, these nondivergent taxa
would have had to have gone through the single or
multiple range splits without diverging at all.
Generalizing to other taxa, we might expect that for-
est-inhabiting birds would also diversify in rainforest
refuges with the butterflies, given a perhaps naı ¨ve
application of the Haffer and Brown models (correla-
tions of centres of endemism among birds, butterflies
and other taxa have often been claimed). Yet although
there are endemic bird taxa in the Eastern Andes, there
are virtually no ‘Huallaga’ vs. ‘Ucayali’ geminate
subspecies or species pairs similar to those in the but-
terflies studied here (Haffer 1987): most lowland birds
in this region have gone through the hypothesized
cycles of geographic isolation without diverging at all.
In contrast, poison frogs of the genus Dendrobates and
Epipedobates, which themselves form mimicry rings (Sy-
mula et al. 2001), show very narrow patterns of ende-
mism and genetic differentiation in the same region of
Peru, with multiple different forms on opposite banks
of the Rı ´o Mayo, as well as up and down the Mayo val-
ley (Symula et al. 2001; Roberts et al. 2006, 2007). These
patterns of endemism are much narrower than those
shown by ithomiine and heliconiine butterflies. Similar,
highly local patterns of differentiation in Epipedobates
are seen in Ecuador (Graham et al. 2004).
If the suture zone near Tarapoto, Peru was not pro-
duced as a result of simple contact between two
recently re-expanded forest refuge biotas, what caused
the well-defined pattern of biotic contacts (Fig. 1)? An
alternative explanation for the existence of this suture
zone is required. Hybrid zones and clines are poten-
tially mobile, and their position is maintained by a bal-
ance between selection and dispersal (Barton & Hewitt
1985; Barton & Hewitt 1989). In neotropical butterflies,
there is empirical evidence that hybrid and contact
zones can move rapidly (Blum 2002; Dasmahapatra
et al. 2002). Differential selection for or against particu-
lar genotypes across an ecotone will cause a moving
cline to settle on boundaries between environments, or
in regions of low population density which act as sinks
for migration (Barton 1979; Barton et al. 1985; Mallet
2010, 1993). Therefore, the suture zone could be the
result of multiple contact and hybrid zones moving inde-
pendently of one another and becoming trapped on a
common ecological hiatus. In some cases, the transitions
are between geminate mid-elevation-adapted and low-
land forms (e.g. Hyposcada kena, Napeogenes pharo and
Ithomia salapia); in others, the transitions are among
geminate lowland rainforest taxa (e.g. Oleria onega, He-
liconius erato, H. melpomene and H. pardalinus). The
Tarapoto suture zone corresponds approximately with
the eastern base (rather than the ridge top) of the Cor-
dillera Escalera range (Fig. 1) east of which lies the vir-
tually unbroken Amazon basin. The eastern slopes of
this ridge attract high levels of orogenic rainfall (Pongo
del Cainarachi, site 8 on Fig. 1, has 3637 mm rain per
year) compared with further out into the Amazon basin
(e.g. Yurimaguas, 2279 mm), and with the rain shadow
Southwest of the Cordillera Escalera [e.g. Tarapoto,
1004 mm, see Fig. 1 for locations (Mallet 1993)]. Given
that butterflies fly little during wet weather, reducing
their potential rates of increase, the wetter eastern
slopes could therefore act as a population sink which
traps clines and hybrid zones.
Thus, we see a range of levels of local geographic dif-
ferentiation and diversification for forest taxa, from
weak (birds, some butterflies), to strong (the geminate
taxa of butterflies studied here), to very strong (dendro-
batid frogs). There are clear lineage-specific effects on
this broader phylogenetic scale, just as found on a finer
scale for genetic divergence and split time among the
geminate subspecies of butterflies. The lineage-specific
rates of divergence argue that climate forcing or cli-
matic assistance vicariance models provide a very
incomplete explanationfor diversification.Climate-
A COALESCENT TEST FOR SIMULTANEOUS DIVERGENCE 4297
? 2010 Blackwell Publishing Ltd