2012 Detecting cryptic speciation in the widespread and morphologically conservative carpet chameleon (Furcifer lateralis) of Madagascar

Detecting cryptic speciation in the widespread and morphologically
conservative carpet chameleon (Furcifer lateralis) of Madagascar
A. M. FLORIO*, C. M. INGRAM,H.A.RAKOTONDRAVONYà,E.E.LOUISJR§
&C.J.RAXWORTHY*
*Richard Gilder Graduate School, Department of Herpetology, American Museum of Natural History, New York, NY, USA
Division of Vertebrate Zoology, American Museum of Natural History, New York, NY, USA
àDe
´partement de Biologie Animale, Universite ´d’Antananarivo, Antananarivo, Madagascar
§Center for Conservation and Research, Henry Doorly Zoo, Omaha, NE, USA
Introduction
Identifying species limits is challenging for recently
evolved groups due to the stochastic nature of gene
sorting (Knowles & Carstens, 2007), difficulties in devel-
oping robust diagnoses (Shaffer & Thomson, 2007) and
branch length heterogeneity (Edwards, 2009). However,
multiple-locus approaches, along with new methods that
utilize morphological characters and niche differences,
can overcome these constraints (Raxworthy et al., 2007;
Rissler & Apodaca, 2007; Hickerson et al., 2010; Glor &
Warren, 2011). This integrated approach has been
promoted to reconcile differences in species delimitation
approaches among disciplines and to strengthen the
validity of species hypotheses (Dayrat, 2005; DeSalle
et al., 2005; Padial et al., 2010). In addition, integrative
taxonomy offers new opportunities for improving the
Correspondence: Antonia M. Florio, Richard Gilder Graduate School,
Department of Herpetology, American Museum of Natural History,
Central Park West at 79th Street, New York, NY, USA.
Tel.: [phone]; fax: [phone]; e-mail: afl[email protected]
ª2012 THE AUTHORS. J. EVOL. BIOL. 25 (2012) 1399–1414
JOURNAL OF EVOLUTIONARY BIOLOGY ª2012 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY 1399
Keywords:
canonical variates analysis;
cryptic species;
ecological niche modelling;
phylogeography;
species delimitation.
Abstract
Species delimitation within recently evolved groups can be challenging
because species may be difficult to distinguish morphologically. Following the
General Lineage Concept, we apply a multiple evidence approach to assess
species limits within the carpet chameleon Furcifer lateralis, which is endemic
to Madagascar and exported in large numbers for the pet trade. Cryptic
speciation within F. lateralis was considered likely because this species (1) has a
vast distribution, (2) occupies exceptionally diverse habitats and (3) exhibits
subtle regional differences in morphology. Phylogenetic trees reconstructed
using nuclear and mitochondrial genes recovered three well-supported clades
corresponding with geography. Morphological results based on canonical
variates analysis show that these clades exhibit subtle differences in head
casque morphology. Ecological niche modelling results found that these
phylogenetic groups also occupy unique environmental space and exhibit
patterns of regional endemism typical of other endemic reptiles. Combined,
our findings provide diverse yet consistent evidence for the existence of three
species. Consequently, we elevate the subspecies F. lateralis major to species
rank and name a new species distributed in northern and western Madagascar.
Initial ecological divergence, associated with speciation of F. lateralis in humid
eastern habitat, fits the Ecographic Constraint model for species diversification
in Madagascar. By contrast, the second speciation event provides some support
for the Riverine Barrier model, with the Mangoky River possibly causing
initial isolation between species. These findings thus support two contrasting
models of speciation within closely related species and demonstrate the utility
of applying a combined-evidence approach for detecting cryptic speciation.
doi: 10.1111/j.1420-9101.2012.02528.x

sensitivity of species recognition, especially in groups that
have undergone recent or cryptic speciation (Padial & De
La Riva, 2009).
The identification of cryptic species is critically impor-
tant for many reasons, including accurate assessment of
biodiversity estimates, facilitating disease and crop plant
pathogen control, and directing conservation efforts
towards vulnerable endemic species (Bickford et al.,
2007). In addition, cryptic speciation may result from
recently evolved sibling species where species limits can
be ambiguous because morphological differences have
not yet accumulated (Knowlton, 1993). These recently
evolved species are good candidates for speciation
research because these groups more closely meet the
assumption that the species’ geographic range has not
changed over time (Losos & Glor, 2003). Further,
research into cryptic species limits can help to both
identify and conserve the processes currently driving
speciation in various groups and regions (Carnaval et al.,
2009).
Situated off the south-eastern coast of Africa, the
island of Madagascar is a model region for cryptic species
identification and studies of speciation (Vences et al.,
2010). Madagascar and Greater India first broke away
from Africa as early as 165 Ma, and Madagascar has
been isolated since separating from Greater India about
88 Ma (Storey et al., 1995). This island continent has a
complicated and poorly understood paleoclimatic his-
tory, but some information is known. Madagascar
experienced a generally dry environment when first
separated from India (Wells, 2003), but the overall
climate became more humid as it drifted northwards
towards the equator. However, the climate again became
drier and cooler during glacial periods, with areas at
lower elevations experiencing more pronounced aridifi-
cation than those at higher elevations (Haffer, 1969).
Presently, the island is composed of striking environ-
mental heterogeneity, with habitat transitions occurring
abruptly. Trade winds and orographic uplift ensure
regular rainfall on the north-east and eastern coast,
and the central mountain chain acts as a barrier causing
a rainfall gradient from the humid north-eastern and
eastern rainforests to the south-western spiny deserts
(Jury, 2003). In addition, there is complex topography
(maximum elevation 2876 m), seasonal rainfall patterns
and drainage systems, all which contribute to the
environmental complexity.
Several hypotheses about the mechanisms driving spe-
ciation in Madagascar have been proposed, and many of
these are related to either the paleoclimate or current
climate of the island. The Watershed hypothesis (Wilme
et al., 2006) proposes that glaciation periods resulting in
severe arid conditions at lower elevations caused species to
move to higher elevations towards more humid areas.
Species distributed in lower elevation watersheds
became trapped in arid pockets, and adapted and diversi-
fied in isolation. The Montane Endemism hypothesis
(Raxworthy & Nussbaum, 1995; Wollenberg et al., 2008)
proposes that some populations of species broadly distrib-
uted during glacial periods became isolated on mountain-
tops during warmer interglacials. The hypothesis of
ecologically mediated speciation (Raxworthy et al., 2007,
2008) proposes that the niches of sister species become
divergent as they adapt to ecotones under disruptive
selection and assortative mating. The Riverine Boundary
hypothesis (Pastorini et al., 2003; Goodman & Ganzhorn,
2004) proposes that the river systems in Madagascar have
restricted gene flow causing isolation between popula-
tions. Lastly, the Ecogeographic Constraint hypothesis
(Yoder & Heckman, 2006) proposes that the abrupt
distinction between the climates, rainfall patterns, and
vegetation of eastern and western Madagascar allows for
initial east–west divergence within widely distributed
species, with subsequent speciation constrained within
eastern and western regions. This hypothesis is an
example of ecologically mediated speciation, where initial
speciation can be either allopatric or parapatric.
Understanding the processes driving speciation in
Madagascar is of interest because an exceptional number
of new endemic species continue to be described (Myers
et al., 2000; Vieites et al., 2009). Some of these new
descriptions have resulted from the splitting of previously
recognized monotypic genera. For example, until
recently (see Miller, 1977) mouse lemurs (Microcebus)
were considered to be a single species (Microcebus muri-
nus), but are now split into at least 15 species using
phylogeographic and morphometric analyses (Rasoloari-
son et al., 2000; Yoder et al., 2000; Louis et al., 2006,
2008). Many other phylogeographic studies have shown
other taxa in Madagascar to represent complexes of
cryptic species (e.g. Raxworthy et al., 2007; Raselima-
nana et al., 2009). In particular, the identification of
recently evolved species, which are often cryptic, offers
important opportunities for improving our understand-
ing of the processes that drive speciation in Madagascar.
A candidate cryptic species complex is the carpet
chameleon Furcifer lateralis (Gray, 1845), which is
endemic to Madagascar. The carpet chameleon is a
popular and familiar species in the pet trade and
consequently is exported in high numbers. Less known
is that F. lateralis is also an excellent model organism for
understanding the processes driving speciation. Although
most Malagasy chameleons exhibit considerable regional
endemism, paradoxically, F. lateralis is one of the most
widely distributed endemic reptiles in Madagascar (Rax-
worthy et al., 2003). The species is distributed in nearly
all areas of Madagascar, with the exception of montane
regions above 1780-m in elevation and a small high
precipitation region in the north-east. It is also often
found at high population densities in severely degraded
forests, grasslands, agricultural areas and urban environ-
ments (Karsten et al., 2009; Randrianantoandro et al.,
2009). This chameleon exhibits subtle regional variation
in axillary pit development, white line development
1400 A. M. FLORIO ET AL.
ª2012 THE AUTHORS. J. EVOL. BIOL. 25 (2012) 1399–1414
JOURNAL OF EVOLUTIONARY BIOLOGY ª2012 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY

under the tail, scalation and head casque height (Hille-
nius, 1959; Brygoo, 1971), but for the last 50 years has
been consistently considered a single species. The only
phylogenetic study to date found multiple geographically
structured clades within F. lateralis using the mitochon-
drial gene 16S for 31 individuals, with strongest support
for a southern clade (Boumans et al., 2007).
Although F. lateralis has long been recognized as a single
species, Angel (1921) described Furcifer lambertoni,
collected from Tananarive (=Antananarivo), which he
distinguished from F. lateralis using the following
F. lambertoni features: (1) absence of gular crest, (2) homo-
geneous squamation on the body and limbs, (3) 70 scales
between dorsal ridge and midventral line, and (4) tail
shorter than snout-vent length (SVL). Hillenius (1959)
found that three of these states (2–4) were found within
F. lateralis, decided that the absence of the gular crest was
insignificant and considered F. lambertoni a junior syno-
nym of F. lateralis. This view was later supported by
Brygoo (1971) and Klaver & Bo
¨hme (1997). A geographic
large-sized variant, ‘forme major’ from the arid south-west
region of Madagascar, was described by Brygoo (1971)
and subsequently treated as the subspecies F. lateralis
major by Klaver & Bo
¨hme (1997).
In this study, we use the General Lineage Concept
(GLC), which uses the the term ‘species’ for separately
evolving metapopulation lineages that can be recognized
using diverse secondary recognition criteria (De Queiroz,
2007). We here apply three secondary recognition
criteria to assess species limits within F. lateralis, using
the following approach: (1) identify lineages based
on clades from analysis of multiple molecular loci,
(2) employ canonical variates analysis (CVA) to deter-
mine potential morphological variation associated with
phylogenetic groups and (3) use ecological niche mod-
elling to determine the environmental and geographic
space occupied by groups supported by covariation of
genetic and morphological evidence.
Materials and methods
Focal species and sampling
A total of 111 F. lateralis individuals, collected between
1990 and 2011, were included for phylogenetic analysis.
The close outgroup species Furcifer labordi,Furcifer ousta-
leti and Furcifer verrucosus were included to test the
monophyly of F. lateralis (Raxworthy et al., 2002; Town-
send and Larson, 2002), resulting in a total matrix of 121
terminals. Furcifer campani was used as the far outgroup
taxon to root all phylogenetic trees. In most cases,
chameleons were collected during night surveys during
the rainy season (approximately December through
April) using headlamps to find individuals roosting on
vegetation; a detailed description of collection methods is
provided in Raxworthy & Nussbaum (2006). Date, time
and longitude ⁄latitude of each individual (using GPS,
altimeter or 1 : 100 000 topographic maps) were re-
corded at time of collection. Voucher specimens were
euthanized and fixed in 10% buffered formalin and then
later transferred to 70% ethanol. Liver and ⁄or thigh
muscle was preserved in 95% ethanol or tissue buffer for
DNA extraction. Voucher specimens and tissues are
deposited at the American Museum of Natural History
(AMNH), the University of Michigan Museum of Zoology
(UMMZ), Duke University Department of Biology and
the University of Antananarivo Department of Animal
Biology (UADBA). Abbreviations for field series are RAN
(Ronald A. Nussbaum), RAX (Christopher J. Raxworthy)
and HER (Hery A. Rakotondravony). Localities, sample
numbers, coordinates, and Genbank accession numbers
for all samples are provided in the Table S1.
Phylogenetic analysis
DNA was extracted from all tissue samples using the
QIAGEN DNeasy Blood & Tissue kit (Valencia, CA, USA)
following the manufacturer’s instructions. Three mito-
chondrial genes – 16S ribosomal RNA, NADH dehydro-
genase subunit 2 (ND2) and NADH dehydrogenase 4
(ND4) – and two nuclear genes – recombination-activat-
ing gene-1 (RAG1) and matrix remodelling-associated
gene (MXRA5) – were amplified. Polymerase chain
reaction was carried out under locus-specific optimal
annealing temperatures (see Table S2). PCR products
were cleaned using MultiScreen PCRl96 Filter plates
(Millipore, Billerica, MA, USA) and sequenced in both
directions using B
IGIG D YEYE v.3.0 (Applied Biosystems, Foster
City, CA, USA) on an ABI 3730 automated DNA
sequencer. Sequences were edited in
GENEIOUSGENEIOUS v.4.8.3
(Biomatters, Auckland, New Zealand). Multiple sequence
alignments were generated using MUSCLE (Edgar, 2004),
with 1000 iterations and default gap opening cost of )1.
Leading and lagging ends were trimmed to remove any
missing data at the alignment edges.
Phylogenetic analyses were conducted using maxi-
mum parsimony (MP) and maximum likelihood (ML).
MP was carried out with
TNTTNT v1.1 (Goloboff et al., 2008)
and W
INCLADAINCLADA v1.0 (Nixon, 2002) with equal weighing
of all characters, and the heuristic search option set at
500 random addition replicates using the New Technol-
ogy search option. Bootstrap support values were calcu-
lated for MP with 500 random addition replicates under a
full heuristic search with 10 random addition sequences
for each. ML was carried out in RA
XX ML (Stamatakis,
2006) with the RA
XX ML GUIGUI 0.93 (Silvestro & Michalak,
2010) using the ML + thorough bootstrap analysis option
with 10 runs and 500 repetitions. Due to the large
number of individuals included in the analysis and the
low genetic divergence found between individuals, the
GTR + CAT algorithm was used as it allows for a rapid
navigation into a search space in which trees score well
under GTR + Gbut at significantly lower computational
costs and memory consumption (Stamatakis, 2006).
Cryptic species in a Malagasy chameleon 1401
ª2012 THE AUTHORS. J. EVOL. BIOL. 25 (2012) 1399–1414
JOURNAL OF EVOLUTIONARY BIOLOGY ª2012 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY

Gene tree concordance and divergence between
putative species
Gene tree concordance was assessed by analysing each
locus individually using ML. The three mitochondrial
genes were also combined and analysed using both ML
and MP. Haplotypes for nuclear sequences were phased
using an approach employed by
PHASEPHASE v2.1 (Stephens
et al., 2001; Stephens & Sheet, 2005) and implemented
using D
NANA SP v5 (Librado & Rozas, 2009). Runs consisted
of 1000 main iterations with an initial 1000 iterations for
burn-in and a thinning interval of 1.
SPLITSTREESPLITSTREE v4.12.13
(Hudson and Bryant, 2006) was used to identify identical
haplotypes and to reconstruct haplotype median-joining
networks for each nuclear locus. To assess the divergence
between and within potential species, the mean number
of mitochondrial nucleotide substitutions between
groups (D
xy
) and nucleotide diversity (Pi) within groups
(Takahata & Nei, 1985) was assessed also using D
NANA SP
v5, with individuals with ambiguity codes in their
sequences excluded from analysis.
Morphological analysis
A total of 87 adult males, including Gray’s (1854) adult
syntype of F. lateralis BMNH 1946.8.22.12 (the other
syntype is a juvenile), and 53 adult females were
subjected to morphometric analysis. A list of the exam-
ined specimens is provided in the Table S3. Adults were
defined as exceeding 60 mm SVL, and chameleons sexed
based on the presence of everted hemipenes (males) or
the presence or absence of hemipenal bulges at the tail
base. High-resolution photographs of a lateral view of the
left side of each specimen’s head were obtained using a
Nikon D5000 Digital SLR Camera with a Nikon AF-S DX
18–55 mm lens on graph paper with a ruler visible to
record scale. Ten landmarks (see Fig. S1) were defined
that could be consistently placed across individuals and
provide adequate coverage of form (Zelditch et al., 2004);
these landmarks were placed on the images using the
digital images of the specimen and tpsUtil (Rohlf, 2004)
with tpsDIG (Rohlf, 2001). Specimens with preparation
irregularities that affect landmark placement (i.e. open
mouths or contortions caused by preservation) were
excluded from the study.
The program Morpho-J (Klingenberg, 2011) was used
to apply a Procrustes superimposition allowing standard-
ization of size and landmark configurations, to generate
covariance matrices and to perform CVA. CVA finds
shape values that maximize group means relative to
variation within groups, by assuming that within-group
covariates matrices are identical (Klingenberg, 2010).
Groups for CVA were predefined based on the major
clades recovered from the phylogenetic analyses. At least
one individual from each population included in this
study was sequenced for molecular data, and morpho-
logical specimens lacking molecular data were predefined
based on their population assignment. The male F.
lateralis syntype was not predefined to a group, but
instead included as a separate classification to assess its
relationship to the other groups. Significance for differ-
ences across these groups was determined using permu-
tation tests (50000) with Procrustes and Mahalanobis
distances, using Morpho-J. Both tests were used to assess
significance because P-values can differ due to the
anisotropy (direction dependency) of shape variation
(Klingenberg & Monteiro, 2005). Male and female spec-
imens were analysed independently as F. lateralis exhibits
sexual dimorphism in head morphology (Brygoo, 1971).
The morphology of the specimens was described using
standard morphological terms and methods (see Rax-
worthy & Nussbaum, 2006). The following measure-
ments and scale counts were used for species
identification: head casque height was measured as the
distance on the lateral crest, from where it began turning
vertical to the top of the head casque; head height was
measured as in Hopkins & Tolley (2011 – see Fig. 2) by
measuring the distance from the back of the lower jaw to
the tip of the casque; parietal crest scale counts were
taken by counting the tubercles on the raised parietal
crest.
Distribution mapping and ecological niche models
After deletion of duplicate records, 110 unique localities
were included for the development of the ecological
niche models (ENM) for the F. lateralis complex. As with
the morphometric analysis, localities were partitioned
based on phylogenetic results. Climate data was taken
from the WorldClim database (Hijmans et al., 2005;
http://worldclim.org/), with the 19 bioclimatic variables
used for ENM analyses in M
AXENTV3AXENTV3 .3.31 (Phillips et al.,
2006). All occurrence localities and environmental vari-
ables were resampled to an oblique Mercator projection
at 1 km
2
resolution (Pearson et al., 2007) using A RCRC M APAP
(ESRI, 2011). Default values were used for the maximum
number of iterations (500) and for the convergence
threshold (10
)5
). The minimum training presence [or
lowest predicted value (LPT) of environmental suitabil-
ity] was chosen for each model as the decision threshold.
The ENM was visualized in A
RCRC M APAP by reclassifying the
continuous data to create a binary prediction, and all
values above the LPT were reclassified as suitable
environment.
Model validation was assessed using k-fold partition-
ing of the data with six replicates (k = 6), as imple-
mented in Maxent using the cross-validation option.
With k-fold partitioning, occurrence points are ran-
domly split in k parts, then one part is used as a test
set for assessing model performance and the remaining
sets (k )1) are used to train the model (Fielding &
Bell, 1997). The number of k-fold partitions typically
varies between 3 and 10 depending on the number of
species occurrence records (Hirzel et al., 2006); here,
1402 A. M. FLORIO ET AL.
ª2012 THE AUTHORS. J. EVOL. BIOL. 25 (2012) 1399–1414
JOURNAL OF EVOLUTIONARY BIOLOGY ª2012 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY

k = 6 was chosen so that the data could be evenly
distributed across the partitions. The area under the
receiver operating characteristic curve (AUC; here
reported as a mean AUC calculated among trials) was
also used to evaluate the model as it provides a
measure of accuracy not dependent on a threshold
(Fielding & Bell, 1997).
Results
Phylogenetic analyses
Gene amplification was successful with all 111 ⁄111
F. lateralis individuals sequenced for 368-bp 16S, 622-bp
ND4 and 630-bp RAG1. Additionally, 720-bp ND2 were
sequenced for 110 of 111 F. lateralis individuals, and 635-
bp MXRA were sequenced for 108 of 111 F. lateralis
individuals. Phylogenetic trees from the individual genes
were not well resolved (data not shown). The mitochon-
drial genes 16S and ND2 were inconclusive with respect
to the relationship of the sister species F. labordi to
F. lateralis. ND4 recovered a monophyletic F. lateralis
complex, but this also received low support (40% boot-
strap support), but with some evidence for further genetic
substructuring within the complex. The two nuclear
genes RAG1 and MXRA also provided poor resolution,
although RAG1 provided weak support for the mono-
phyly of the F. lateralis complex, with 54% bootstrap
support. Combined, the three mitochondrial and two
nuclear genes resulted in a 2975 character matrix.
A heuristic search using MP resulted in 403 equally
parsimonious trees (tree length = 3255). The large num-
ber of equally parsimonious trees resulted from minor
incongruence between individuals from geographically
close populations, but deeper tree topology was congru-
ent across all trees. ML recovered a tree congruent with
the MP strict consensus. The ML tree is congruent with
the MP tree and is shown in Fig. 1a. Furcifer lateralis is
supported as a monophyletic group (bootstrap support
values: MP 84%, ML 82%), and there are three obvious
and major clades recovered within the complex that are
well-supported and correspond with geography:
(1) eastern (MP 100%, ML 100%), (2) southern (MP
93%, ML 86%) and (3) north-west (MP 100%, ML
Furcifer campani Furcifer verrucosus
Furcifer labordi
Furcifer oustaleti
**/**
84/82
**/** **/**
**/**
99/**
93/86
**/99
0.02
(a)
North-west
ENM
Eastern
ENM
Southern
ENM
Overlap
between
models
Furcifer lateralis
‘north-west’
Furcifer lateralis
‘eastern’
(b)
Furcifer lateralis
‘southern’
Fig. 1 (a) Phylogenetic relationships within the Furcifer lateralis complex and near outgroups resulting from the analysis of 121 individual
chameleons and 2975 characters (partial 16S,ND2,ND4,RAG1 and MXRA5) reconstructed on the maximum likelihood (ML) tree (congruent
with MP). Three well-supported clades (eastern = blue, southern = yellow and north-west = green) are recovered within the F. lateralis
complex (bootstrap support values for MP ⁄ML, **100%). (b) Ecological niche models for each clade of the F. lateralis complex, projected onto
Madagascar, with collecting localities. MP, maximum parsimony.
Cryptic species in a Malagasy chameleon 1403
ª2012 THE AUTHORS. J. EVOL. BIOL. 25 (2012) 1399–1414
JOURNAL OF EVOLUTIONARY BIOLOGY ª2012 EUROPEAN SOCIETY FOR EVOLUTIONARY BIOLOGY