Revisiting the Insect Mitochondrial Molecular Clock: The
Mid-Aegean Trench Calibration
Anna Papadopoulou,*,1,2Ioannis Anastasiou,3and Alfried P. Vogler1,2
1Department of Entomology, Natural History Museum, London, United Kingdom
2Division of Biology, Imperial College London, Silwood Park Campus, Ascot, United Kingdom
3Department of Biology, University of Athens, Panepistimioupolis, Athens, Greece
*Corresponding author: E-mail: firstname.lastname@example.org.
Associate editor: Barbara Holland
Phylogenetic trees in insects are frequently dated by applying a ‘‘standard’’ mitochondrial DNA (mtDNA) clock estimated
at 2.3% My?1, but despite its wide use reliable calibration points have been lacking. Here, we used a well-established
biogeographic barrier, the mid-Aegean trench separating the western and eastern Aegean archipelago, to estimate
substitution rates in tenebrionid beetles. Cytochrome oxidase I (cox1) for six codistributed genera across 28 islands (444
individuals) on both sides of the mid-Aegean trench revealed 60 independently coalescing entities delimited with a mixed
Yule-coalescent model. One representative per entity was used for phylogenetic analysis of mitochondrial (cox1, 16S rRNA)
and nuclear (Mp20, 28S rRNA) genes. Six nodes marked geographically congruent east–west splits whose separation was
largely contemporaneous and likely to reflect the formation of the mid-Aegean trench at 9–12 Mya. Based on these
‘‘known’’ dates, a divergence rate of 3.54% My?1for the cox1 gene (2.69% when combined with the 16S rRNA gene) was
obtained under the preferred partitioning scheme and substitution model selected using Bayes factors. An extensive survey
suggests that discrepancies in mtDNA substitution rates in the entomological literature can be attributed to the use of
different substitution models, the use of different mitochondrial gene regions, mixing of intraspecific with interspecific
data, and not accounting for variance in coalescent times or postseparation gene flow. Different treatments of these
factors in the literature confound estimates of mtDNA substitution rates in opposing directions and obscure lineage-
specific differences in rates when comparing data from various sources.
Key words: molecular clock calibration, Bayesian relaxed clock, insects, mitochondrial DNA rates, Tenebrionidae,
Evolutionary scenarios frequently rely on estimates of node
agesintime-calibratedphylogenetic trees.Wherefossils are
such estimates. For insect mitochondrial DNA (mtDNA),
the most widely quoted rate of molecular evolution is
Brower’s (1994) calibration based on a set of seven studies
that provided age estimates of lineage splits ranging from
300 to 3,250,000 years ago. Regression on uncorrected pair-
wise distances against inferred number of years since line-
age divergence revealed a surprisingly strong linear
relationship with y 5 2.34 ? 10?6x and r25 0.996, for
a substitution rate of 0.0115 per site per My equal to
2.3% divergence. Although Brower’s (1994) rate appears
to be satisfactory to estimate ages in many groups, its uni-
formity in the seven original studies is surprising given the
mixed use of protein-coding and ribosomal markers, whose
rates are expected to differ greatly. In addition, mutation
rates in recently diverged haplotypes at or near the pop-
ulation level have now been found to be much higher than
phylogenetic rates (Ho et al. 2005, 2007), possibly due to
the delayed effect of purifying selection. This ‘‘time depen-
dency’’ of the clock may affect a period of up to ;1 Ma,
well within the time window from which most of Brower’s
(1994) calibration points were drawn.
Subsequent estimates of insect mtDNA substitution
rates based on biogeographic vicariance, island ages, fossils,
or other independent evidence (supplementary table S1,
Supplementary Material online), reported both decreased
(Sperling et al. 1997; Pruser and Mossakowski 1998;
Andersen et al. 2000) or elevated (Fleischer et al. 1998;
Luchetti et al. 2005; Shapiro et al. 2006) rates, which is gen-
erally attributed to lineage-specific or gene-specific effects
(e.g., a lowerrateof 1.5% divergencein cytochrome oxidase I
[cox1]; Farrell 2001). However, the effects of the method-
ology used for rate or age estimation have not been fully
appreciated. Recently developed ‘‘relaxed-clock’’ methods,
which allow substitution rates to vary among branches
either in an autocorrelated or uncorrelated manner
(Sanderson 2002, 2003; Thorne and Kishino 2002; Yang
accurate calibration of the mtDNA clock and its rate var-
iation among insect lineages. Yet, accurate estimations of
substitution rates require careful choice of the model of
sequence evolution used to correct for multiple hits (Yang
1996; Arbogast et al. 2002), an issue that has received little
attention in the entomological literature for its effect on
© The Author 2010. Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution. All rights reserved. For permissions, please
Mol. Biol. Evol. 27(7):1659–1672. 2010 doi:10.1093/molbev/msq051 Advance Access publication February 18, 20101659
clock estimates. Moreover, the preponderance of biogeo-
requires well documented geological or paleoclimatic
evidence and a demonstration that the assumed barriers
constitute a true obstacle to dispersal for the focal group.
Finally, several factors may confound the correlation be-
tween geological time and lineage divergence, such as an-
cestral polymorphism (Edwards and Beerli 2000) or
postseparation gene flow that delay lineage sorting along
an assumed biogeographic boundary (Carstens and
Improved estimates of the mtDNA clock in insects
therefore require further study of systems that are less af-
fected by frequently encountered sources of error. The
mid-Aegean trench in the eastern Mediterranean is highly
suited for this kind of analysis. This geological formation
originated in the Upper Miocene (12–9 Mya) and led to
the initial split of the united landmass of Aga ¨is (Creutzburg
1963; Dermitzakis and Papanikolaou 1981), separating the
have remained subdivided by the surrounding ocean ever
since, except for the Messinian desiccation of the Mediter-
ranean basin from 5.96 to 5.33 Mya (Krijgsman et al. 1999)
when a land bridge may have existed via a deep rift valley.
The Aegean islands harbor species-rich assemblages of
darkling beetles (Coleoptera: Tenebrionidae) highly suited
for biogeographic clock calibrations. Many species are
flightless with limited dispersal capabilities, in particular,
the ‘‘geophilic’’ lineages associated with ecologically stable
soil types. Genetic variation in a few of these species has
already been shown to be highly structured geographically
and is deeply subdivided along the mid-Aegean trench
(Papadopoulou, Anastasiou, et al. 2009). Codistributed line-
ages of tenebrionids therefore can be used to calibrate rates
a single geological event, thus reducing stochastic effects of
lineage sorting and variation in substitution rates (Edwards
and Beerli 2000; Hickerson et al. 2003). A comprehensive sur-
vey of genetic variation of six flightless, geophilic tenebrionid
genera with wide distribution throughout the Aegean estab-
lished several temporally congruent lineage splits that were
attributable to the formation of the mid-Aegean trench. This
provides an independent calibration of the mtDNA clock in
a major lineage of insects and permits an assessment of the
sensitivity of inferred substitution rates to model selection,
alternative partitioning schemes, the use of relaxed versus
strict clock methods, and gene-specific differences in rates
between protein-coding and rRNA genes. Comparisons with
literature data show that discrepancies in rates found by dif-
ferent studies can be attributed largely to methodological is-
sues, although substitutions rates in the Aegean tenebrionids
may still be higher than most existing estimates for insects.
Material and Methods
Taxon Sampling and DNA Sequencing
A total of 444 specimens from six flightless, geophilic tene-
brionid genera were sampled from 11 to 25 islands each,
and 51–122 individuals were sequenced per genus (table 1,
supplementary table S2, Supplementary Material online). To-
tal genomic DNA was extracted from thorax or leg tissue
using the Promega 96-well plate kit. An 826–829 bp frag-
ment of the 3# end of the cox1 gene was amplified using
primers C1-J-2183 (Jerry) and TL2-N-3014 (Pat) (Simon
et al. 1994) or JerryTen and PatTen (Papadopoulou,
Anastasiou, et al. 2009). A 513–524 bp fragment of the sin-
gle-copy nuclear muscular protein 20 (Mp20) locus, includ-
ing 469 bp of coding region and one intron, was amplified
using the primer pair Mp205# and Mp203# (Pons et al.
2004) for Pimeliinae or Mp20Trib5# and Mp20Trib3# for
Tenebrioninae (Papadopoulou, Anastasiou, et al. 2009).
A 433–437 bp portion of the mitochondrial 16S rRNA gene
(rrnL) was amplified using LR-N-13398 (16Sar) (Simon et al.
1994) and LR-J-12961 (16Sb2) (Cognato and Vogler 2001),
and a 646–655 bp fragment of the nuclear 28S rRNA gene
was amplified using 28SFF and 28SDD (Monaghan et al.
Multiscreen 96-well plates (Millipore, Billerica, MA) and se-
quenced in both directions using the BigDye technology
and an ABI PRISM 3700 DNA Analyzer (Applied Biosys-
tems). Sequence chromatograms were assembled and edi-
ted using the Sequencher 4.6 software (Gene Codes
Corporation, Ann Harbor, MI). Cox1 and Mp20 sequences
for the genera Dailognatha, Eutagenia, and Zophosis are
from Papadopoulou, Anastasiou, et al. (2009). New sequen-
Database (supplementary tables S2–S3, Supplementary
Species Delimitation, Alignment, and
To minimize the effect of increased mutation rates at the
intraspecific level (Ho et al. 2005, 2007), sequence variation
was divided into within- and between-species groups using
the generalized mixed Yule-coalescent (GMYC) model
(Pons et al. 2006; Fontaneto et al. 2007). This procedure
identifies a threshold value for the shift in branching rate
from coalescent lineage branching to interspecific diversi-
fication onan ultrametric treeand delimits ‘‘independently
evolving’’ mtDNA clusters. The analysis was carried out us-
ing the R package SPLITS (SPecies LImits by Threshold Sta-
tistics) available at http://r-forge.r-project.org/projects/
splits/ with the ‘‘single-threshold’’ option. A clock-
for the full cox1 data sets of each genus, after removal
of identical haplotypes. Tree searches were performed
with Bayesian analysis in MrBayes 3.1.2 (Ronquist and
Huelsenbeck 2003), applying separate models for two
partitions (1st and 2nd codon positions together vs. 3rd
codon position) as selected by Akaike information criterion
(AIC) in MrModeltest 2.2 (Nylander 2004), with two parallel
runs of 5 million generations each and using one cold and
twoincrementallyheatedMarkovchains(k 5 0.1)andsam-
pling every 1,000 steps. Trace plots were visually inspected,
and convergence diagnostics (standard deviation [SD] of
split frequencies, effective sample size), as implemented in
Papadopoulou et al. · doi:10.1093/molbev/msq051
MrBayes and Tracer 1.4.1 (Drummond and Rambaut 2007),
were checked to ensure that the Markov chain had reached
stationarity. After discarding the first 2.5 million generations
as burn-in, trees were summarized using an ‘‘all-compatible’’
consensus. Eachconsensus treewas convertedto ultramet-
ric using penalized likelihood as implemented in r8s 1.7
(Sanderson 2003) with the optimal smoothing parameter
selected by cross-validation of values between 0.01 and
1,000 following the procedure described in the r8s manual.
For the phylogenetic analysis at the species level, a single
exemplar representing each GMYC group in the cox1 anal-
given above. A range of different alignment strategies was
assessed for the length-variable rrnL, 28S, and the intron
region of Mp20. We compared five gap penalty combina-
tions (gap opening penalty: 5-6.66-10-15-20 vs. gap exten-
sion penalty: 6.66) in ClustalW (Thompson et al. 1994) and
three different iterative refinement methods (E-INS-i,
G-INS-i, and L-INS-i) in MAFFT 6.240 (Katoh et al. 2005;
Katoh and Toh 2008). Each resulting alignment was as-
tein-coding regions of cox1 and Mp20 using the
FIG. 1. (a) Relatively dated four-gene tree (combined data set of cox1, rrnL, 28S, and Mp20) of six tenebrionid genera generated using an
uncorrelated lognormal clock in BEAST. Eight east–west nodes (EW1–EW8) are compared for temporal congruence (using a relative scale 0–
100). Node heights correspond to mean values across 9,000 post burn-in trees, whereas gray bars indicate ± 1 SD for the eight focal nodes (see
table 2 for values). (b) Map of the central Aegean region showing the sampled islands and mainland regions on the east and on the west of the
mid-Aegean trench, dashed lines represent the presumed position of the trench.
Revisiting the Insect mtDNA Clock · doi:10.1093/molbev/msq051
Incongruence Length Difference (ILD) test (Farris et al.
1994). Tree lengths were calculated using parsimony
searches in PAUP* 4.0b10 (Swofford 2002) with 1,000 ran-
dom addition sequence replicates and gaps treated as ‘‘5th
state.’’ We separately selected the alignment strategy for
each locus that gave the lowest score for the ratio ILD/length
of the combined-analysis tree (Wheeler and Hayashi 1998)
with respect to the protein-coding regions and then tested
the overall congruence of the resulting data set (partitioned
as cox1—rrnL—28S—Mp20 exon—Mp20 intron) using the
partition homogeneity test in PAUP* with 100 replicates.
Parsimony analysis of the combined data set was per-
formed in PAUP* with 1,000 random addition sequence
data,’’ and nonparametric bootstrap was conducted with
1,000 pseudoreplicates. We used PhyML 3.0 (Guindon
and Gascuel 2003) to perform unpartitioned maximum
likelihood (ML) analysis under a general time reversible
(GTR)þCþI substitution model and calculate bootstrap
support values with 100 replicates. We also carried out par-
titioned ML searches with RAxML 7.0.4 (Stamatakis 2006)
table S4, Supplementary Material online). The GTRMIX
model was employed, so the initial tree searches were
conducted with the GTRCAT approximation but the final
tree topology was evaluated under a separate GTRþCþI
were performed for each partitioning scheme using the
rapid RAxML bootstrapping algorithm (Stamatakis et al.
We tested the null hypothesis that all pairs of east/west
clades are reciprocally monophyletic and sister to each
other, comparing parsimony and RAxML searches under
that topological constraint with those of unconstrained
searches and assessing the significance of the observed differ-
ences with the Shimodaira–Hasegawa test (Shimodaira and
gle GTRþCþI model (as PAUP* does not permit applying
a separate substitution model to each partition) with param-
eters estimated by PhyML, and the significance of the test
was evaluated using RELL sampling with 10,000 replicates.
Relative Node Ages and Clock Calibrations
Node ages and substitution rates were estimated using an
uncorrelated lognormal relaxed clock in BEAST 1.4.8
(Drummond et al. 2006; Drummond and Rambaut
2007). In all analyses, the among-genera relationships
and the monophyly of the eastern and western clades were
constrained according to the results of the topology tests
conducted on the four-gene data set. This was considered
necessary because otherwise, if a small proportion of the
sampled trees from the posterior distribution did not in-
clude all eastern and western clades as reciprocally mono-
phyletic, the 9–12 Mya calibration would have been
partially assigned to incorrect nodes. Relative node ages
(normal prior distribution with a mean of 100 and SD of 1).
Two independent runs of 50 million generations (sampling
using a Yule tree prior and the default options for all other
prior andoperatorsettings. Theconvergence andmixing of
each Markov chain Monte Carlo chain was assessed by in-
spection of the trace plots and the effective sample sizes
using Tracer 1.4.1 (Drummond and Rambaut 2007). Sam-
ples from both independent runs were then pooled after
removing a 10% burn-in using LogCombiner 1.4.8. The
means and standard errors (SEs) of the node heights were
summarized using Tracer, and SDs were calculated by mul-
tiplying the SEs by the square root of the effective sample
size in each case.
Estimates of substitution rates were performed under an
uncorrelated lognormal relaxed clock in BEAST as de-
scribed before,butinsteadoffixing therootnode,anormal
prior distribution was applied on the ages of the selected
calibration points, with a mean of 10.5 My and SD 5 1.5
(0.05 quantile: 8.033, 0.95 quantile: 12.97), to reflect the
geological age of the Aegean trench at 9–12 Mya. We es-
timated rates for the combined cox1 and rrnL data set un-
der an Hasegawa-Kishino-Yano (HKY) model, a GTR
model, a GTRþCþI, and four partitioning schemes (P1:
cox1 vs. rrnL, P2: 3rd codons vs. all other sites, P3: 1st
and 2nd positions vs. 3rd positions vs. rrnL, P4: 1st vs.
2nd vs. 3rd vs. rrnL). We applied a separate substitution
Table 1. Sampled Taxa and Results of the GMYC Model for Species Delimitation.
aNumber of morphologically described species.
bNumber of islands or mainland regions sampled per taxon.
cNumber of cox1 sequences used to apply the GMYC model.
dTotal number of independent entities identified by the GMYC model including singletons (range of entities within 2 logL of the model).
eNumber of entities with more than one individual.
fThe likelihood of the null model.
gThe likelihood of the GMYC model, likelihood ratio test ***P , 0.001, **P , 0.01, *P , 0.05.
hNumber of terminals per genus in the six-genera phylogenetic tree sampled from the east or the west of the mid-Aegean trench.
iGenera that were analyzed in Papadopoulou, Anastasiou, et al. (2009).
Papadopoulou et al. · doi:10.1093/molbev/msq051
model to each partition as selected by the AIC imple-
mented in MrModeltest 2.2 (Nylander 2004) (supplemen-
tary table S5, Supplementary Material online). Moreover,
we used the same priors on the node ages and models se-
lected by the AIC to estimate rates for each of the five gene
regions separately (cox1, rrnL, Mp20 exon, Mp20 intron,
by codon position.
A strict clock was applied for comparison using MrBayes
3.1.2 (Ronquist and Huelsenbeck 2003) under a uniform
prior on branch lengths. All searches were conducted with
5 million generations and two parallel runs using one cold
and twoincrementally heatedMarkov chains(k5 0.1) and
sampling every 1,000 steps, and the first 2.5 million genera-
tions were discarded as burn-in. Mean node heights and 95%
higher posterior limits across all post burn-in trees were cal-
culated using TreeAnnotator 1.4.8, and these numbers were
converted to % divergences My?1assuming that the most
recent node corresponds to separation at a minimum of
9 Mya and the oldest node at a maximum of 12 Mya.
We usedBayesfactors toassesstheMLofdifferentmod-
els and partitioning schemes. The harmonic mean of the
sampled likelihoods was estimated either using the ‘‘sump’’
command inMrBayes or,in thecase oftheBEAST searches,
by Tracer with 1,000 bootstrap replicates. For the interpre-
tation of the Bayes factors, we followed the widely used
cutoff values proposed by Kass and Raftery (1995), when
comparing partition schemes that required equal numbers
of free parameters. When partition schemes differed in total
number of free parameters, we calculated the ratio ln(Bayes
Factor)/Dp (Dp 5 difference in number of total free param-
eters between alternative partition schemes) and used the
recommendations of Pagel and Meade (2004), as applied
by Miller et al. (2009), which suggest at least a 10 lnL increase
in the harmonic mean per additional free parameter before
accepting a more complex model.
MEGA4 (Tamura et al. 2007) was used to calculate av-
erage pairwise uncorrected and Kimura 2-parameter dis-
tances between each east/west pair, and distances were
converted to maximum and minimum divergences per
My, corresponding to 9 and 12 Mya, respectively.
Phylogenetic Analysis of cox1 and mtDNA Cluster
The cox1 Bayesian trees of Dendarus, Pimelia, and Tentyria
(supplementary figs. S1–S3, Supplementary Material online)
and Dailognatha, Eutagenia, and Zophosis (Papadopoulou,
Anastasiou, et al. 2009) revealed strong phylogenetic cluster-
ing and geographical structure. The GMYC model had a sig-
nificantly better fit to the data than the null model of
uniform coalescent branching for all lineages (P , 0.05)
and identified between 2 (Tentyria) and 23 (Dailognatha)
GMYC entities (table 1 and supplementary figs. S1–S3,
Supplementary Material online; Papadopoulou, Anastasiou,
et al. 2009). The number of GMYC groups greatly exceeded
that of Linnaean names, as most of these highly subdivided
lineages are currently described as a single species (table 1).
Each of the GMYC clusters was geographically restricted to
a single island or a group of adjacent islands either on the
eastern or western side of the mid-Aegean trench. A notable
exception was the eastern GMYC cluster in the genus Ten-
tyria found on the volcanic island of Santorini located to the
west of the mid-Aegean trench (supplementary fig. S3, Sup-
plementary Material online). This may be attributed to a re-
of the volcano 3,500 years ago, a pattern that has been sug-
gested for many other taxa (Thornton 2007). In total, 67 en-
tities (55 clusters and 12 singletons) were recognized at the
point of the highest likelihood of the GMYC model. One in-
dividual per cluster and selected singletons were chosen for
sequencing of the rrnL, Mp20, and 28S markers. Confidence
intervals for the number of clusters were calculated (table 1),
but we selected exemplars to match the ML solution, except
in the case of Tentyria, where we used the suboptimal solu-
fig. S3, Supplementary Material online). After failure to se-
ern lineages (fig. 1 and supplementary table S3, Supplemen-
tary Material online).
Alignment and Phylogenetic Analysis
Alignment parameters for each length-variable gene region
were selected to reduce incongruence with the protein-
coding regions (supplementary table S6, Supplementary
Material online). The resulting concatenated matrix was
highly congruent among the five partitions (P 5 0.98; par-
tition homogeneity test). Parsimony and ML analyses sup-
(Dendarus, ((Dailognatha, Tentyria), Zophosis), (Eutagenia,
Pimelia)) was favored by all searches except for parsimony
when gaps were treated as 5th state (supplementary table
S4, Supplementary Material online). Within each genus,
most of the eastern (E) and western (W) lineages were re-
ciprocally monophyletic, with generally high bootstrap
support (fig. 1; EW1–8), except for clades E6 and W6 (Dai-
lognatha) that were unresolved with respect to each other
(supplementary table S7, Supplementary Material online).
Tree scores from searches constrained for monophyly of all
eight east/west pairs and their reciprocal monophyly were
very similar to those from unconstrained analyses (supple-
mentary table S8, Supplementary Material online), and
when scores were slightly lower these differences were
not significant in the Shimodaira–Hasegawa test under
parsimony (all P .. 0.05) nor ML (P . 0.05; supplemen-
tarytableS9,Supplementary Materialonline).These results
provide justification to conduct mtDNA clock calibration
on trees constrained for the monophyly of these nodes.
Testing for Contemporaneous Divergence
The eight east/west pairs were tested for temporal congru-
ence using a relative dating approach (Loader et al. 2007),
by comparing their relative ages under a relaxed-clock
model based on all four genes or only on the mtDNA data
Revisiting the Insect mtDNA Clock · doi:10.1093/molbev/msq051
set. Mean values of node ages were within ± 1 SD of each
other for six east/west pairs (table 2), supporting a hypoth-
esis of contemporaneous divergence. The two remaining
east/west pairs were either marginally (Dailognatha, EW6)
orsignificantly (Zophosis,EW3)morerecent(table 2).Inlight
of the geological history of the region, the six contempora-
neous nodes were attributed to the formation of the mid-
Aegean trench at 9–12 Mya, whereas the EW3 node was as-
sumed to reflect the younger east/west subdivision after the
end of the Messinian desiccation at 5.96 Mya. The EW6 split
may also be associated to the initial formation of the mid-
Aegean trench but affected by postseparation dispersal (e.g.,
to the west of the trench), and the E6/W6 separation was
as calibration points based on the age of the mid-Aegean
Calibration of Substitution Rates Based on the
Mid-Aegean Trench Geological Age
trees calibrated at 9–12 Mya for the six nodes revealed
great differences depending on the substitution model
used for correction. The estimated divergences per My
ranged from 1% when using uncorrected distances and ap-
proximately 1.2% when using a HKY or a GTR model with-
out accounting for rate heterogeneity among sites. When
applying a GTRþCþI model, this divergence estimate in-
creased to 2.23% (strict clock in MrBayes) or 2.39% (relaxed
clock in BEAST) without partitioning and to 2.69% when
using the P3 or P4 partitioning scheme in BEAST (fig. 2).
The differences in estimated rates among partitioning
schemes were consistent between MrBayes and BEAST
analyses (fig. 2) but were in all cases higher by 0.1–0.2%
with the latter. Bayes factor comparisons favored the P3
partitioning scheme in both MrBayes and BEAST analyses
when applying the ln(Bayes Factor)/Dp ? 10 criterion or
the P4 partitioning scheme applying the Kass and Raftery
(1995) criterion of 2lnBF ? 10 (table 3), in either case sup-
porting a rate of 2.7% (relaxed clock; BEAST) or 2.5% (strict
clock; MrBayes). Applying a strict clock in BEAST under the
preferred P3 partitioning scheme gave an estimated rate of
2.6% and Bayes factors comparisons between the strict and
the uncorrelated lognormal relaxed clock favored the re-
laxed clock (lnBF 5 18.32). Our estimates for cox1 on its
own were higher than the average of the two mitochon-
drial genes (3.36% unpartitioned or 3.54% for the preferred
partitioning scheme), whereas they were lower for rrnL
(1.06%). For the nuclear genes, we estimated mean diver-
gence rates of 3.68% My?1for the intron region of Mp20,
0.66% My?1for the exon, and 0.12% My?1for 28S (table 4).
The use of an uncorrelated relaxed-clock approach, as
implemented in BEAST, also permits comparisons of the
clocklikeness in different gene regions (Drummond et al.
2006) and measurement of the degree of rate autocorrela-
tion among lineages (Ho 2009). Our results indicate that
cox1 evolves in a more clock-like manner than rrnL (lower
ucld.mean and coefficient of variation; table 3), which is in
agreement with findings of other studies (Gaunt and Miles
2002). Both mitochondrial regions do not deviate greatly
from the strict clock, whereas all nuclear regions (including
the presumably neutrally evolving intron) showed much
greater variation in rates among branches, as the SD of
branch rates and the coefficient of variation are both greater
than 1 (table 4). In terms of autocorrelation of rates between
neighboring branches, we only found a significantly positive
covariance in the case of the Mp20 exon (table 4).
Applying the ‘‘Standard’’ Insect Mitochondrial
When the mean of the branch rates on the combined
cox1 and rrnL data was fixed at 0.0115, that is, 2.3%
Table 2. Relative Age Estimates (mean age ± 1 SD) for Eight East–
West Nodes, When Fixing the Root Node to 100 Under an
Uncorrelated Lognormal Relaxed Clock in BEAST, using Either the
Four-Gene Data set or Only the mtDNA Data set (combined
cox1 þ rrnL).
aBold letters indicate the six nodes with largely overlapping ranges that were
selected as calibration points.
FIG. 2. Estimated rates of divergence for the combined cox1 þ rrnL
data set based on six calibration points corresponding to the age of
mid-Aegean trench (9–12 Mya) and using either pairwise distances
(uncorrected or Kimura 2-parameter), a strict clock in MrBayes or
a relaxed lognormal uncorrelated clock in BEAST under different
nucleotide substitution models (HKY, GTR, GTR þ C þ I, and four
partitioning schemes P1–P4 as described in the text). Error bars
correspond to the 95% highest posterior density limits (BEAST and
MrBayes analyses) or to the range of values calculated across 6
calibration points (P distances). The dashed line indicates the
standard 2.3% per My rate (Brower 1994).
Papadopoulou et al. · doi:10.1093/molbev/msq051
divergence My?1(Brower 1994), analyses under five parti-
tioning schemes and using a uncorrelated lognormal re-
laxed clock in BEAST, resulted in estimated mean ages
± 1 SD for the six contemporaneous east/west nodes that
were compatible with the age of the mid-Aegean trench
(table 5). The node EW3 (genus Zophosis) was estimated
to be much more recent (4.78–4.91 Mya), an age compat-
ible with post-Messinian divergence, whereas EW6 ranged
between 7.26 and 7.65 Mya. When the 2.3% standard clock
was applied to the individual mitochondrial genes, ages of
the six contemporaneous east/west nodes were much
higher for cox1 (12.3–17.5 Mya) and much lower for rrnL
(3–4.9 Mya) (table 5). The effect of data partitioning was
investigated, both for the combined cox1 and rrnL data set
(P0–P4) and cox1 on its own. Comparisons of the resulting
ages suggested that a greater number of partitions resulted
in slightly higher estimates (table 5). For example, com-
pared with the unpartitioned data the P4 partitioning
caused an increase in estimated node ages of 0.15–1.7
age corresponding to the most recent node (EW3) and the
highest percentage to the oldest node (EW7).
Comparing Existing Calibrations from the
An extensive literature search starting from publications
citing Brower (1994) found 30 other studies (supplemen-
tary table S1, Supplementary Material online) that esti-
mated substitution rates for insect mtDNA based on
biogeographic, paleoclimatic, fossil, or other independent
evidence. These data were compiled for comparisons, sep-
arately for studies that did not account for rate heteroge-
neity among sites (uncorrected distances or using simple
substitution models) and those that used models
Table 4. Estimated Rates Per Gene Region Based on Six East–West Calibration Points and Using a Lognormal Uncorrelated Relaxed Clock in
BEAST, Mean Values ± 1SD.
0.0168 6 0.0018
0.0177 6 0.0019
0.0054 6 0.0009
0.0120 6 0.0012
0.0133 6 0.0013
0.0184 6 0.0152
0.0033 6 0.0006
0.0006 6 0.0003
0.0017 6 0.0003
0.0169 6 0.0019
0.0178 6 0.0019
0.0049 6 0.0008
0.0119 6 0.0011
0.0131 6 0.0013
0.0496 6 0.2628
0.0024 6 0.0008
0.8273 6 6.2150
0.0012 6 0.0003
0.2571 6 0.0674
0.2973 6 0.0644
0.5106 6 0.1238
0.1863 6 0.0619
0.2602 6 0.0554
2.5717 6 0.5399
1.6353 6 0.2375
3.1057 6 1.0567
1.4434 6 0.2044
0.2609 6 0.0702
0.3031 6 0.068
0.5418 6 0.1448
0.1878 6 0.0632
0.264 6 0.0578
4.4511 6 1.1499
2.3847 6 0.5021
5.4504 6 1.8708
2.0133 6 0.4148
20.0120 6 0.0903
20.0134 6 0.0898
0.0001 6 0.0895
20.0176 6 0.0914
20.0141 6 0.0913
0.0781 6 0.1136
0.2118 6 0.1454
0.0140 6 0.0538
0.2058 6 0.1387
aNumber of substitutions per site divided by tree length.
bMean of branch rates.
cThe SD of the branch rates.
dCoefficient of variation.
eCovariance between parent and child branch rates.
gPreferred partitioning scheme.
Table 3. Bayes Factor Comparisons for Selection of Substitution Model and Partitioning Scheme. a) MrBayes Analyses Under a Strict Clock b)
BEAST Analyses Under an Uncorrelated Lognormal Relaxed Clock.
HKYGTRGTRþCþI P1 P2P3 P4
NOTE.—n/a, not applicable.
aThe harmonic mean of sampled likelihoods as estimated by MrBayes or Tracer.
bNumbers in brackets: total number of free parameters required for each model or partitioning scheme.
cAbove the diagonal: ln(Bayes Factor)/Dp (where Dp: difference in total number of free parameters between two models).
dBelow the diagonal: ln(Bayes Factor).
Revisiting the Insect mtDNA Clock · doi:10.1093/molbev/msq051
incorporating gamma-distributed rate variation. The esti-
mated rates obtained in these studies were plotted against
the average calibration age (fig. 3), and an exponential
curve with three parameters was fitted under the least-
squares criterion (Rate(t) 5 le?ktþ k). Following Ho
stitution rate, which was k 5 1.52 (% divergence per My)
for uncorrected distances and k 5 2.10 for the studies that
had accounted for among-site rate variation. The latter
showed a better fit to the exponential curve (R25 0.97,
fig. 3a) than the uncorrected rates (R25 0.60, fig. 3b).
In both cases, linear regression would result in a much
poorer fit (R25 0.14 for fig. 3a, R25 0.19 for fig. 3b).
The Importance of Model Selection
Accurate estimation of substitution rates and divergence
times relies on the model of sequence evolution used to
correct for multiple hits (Yang 1996; Arbogast et al.
2002). However, Brower’s (1994) widely accepted estimate
was based on uncorrected pairwise distances. The appli-
cation of this‘‘standard rate’’using uncorrected distances
has been common practice in the entomological litera-
ture (e.g., Bernhard et al. 2005; Baker et al. 2008; Canfield
et al. 2008) but may lead to incorrect conclusions, par-
ticularly when estimating older divergences. In the case
of Aegean tenebrionids, if we had applied a 2.3% diver-
gence rate without correcting for multiple hits, estimated
ages of the six east/west nodes would be 3.6–5.4 Mya and
support a scenario of post-Messinian diversification for
Conversely, using the geological age of the mid-Aegean
trench to estimate the substitution rate of the cox1 þ rrnL
data set without using a gamma correction resulted in
rates as low as 1% or 1.2% (fig. 2). However, when ac-
counting for rate heterogeneity among sites, either as-
suming a strict or a relaxed clock, the estimates were
much higher (2.23% with MrBayes, 2.39% with BEAST),
demonstrating that rates are greatly underestimated if
they are derived from uncorrected pairwise distances
(Yang 1996; Arbogast et al. 2002). This was also evident
from the compilation of literature values of studies using
uncorrected versus model-based methods (including
a gamma distribution to account for rate heterogeneity
among sites) to estimate substitution rates, which
showed that the values plateau at 1.5% versus 2.1%,
respectively (fig. 3).
The choice of sequence evolution models requires care-
ful consideration. Recent advances in model-based phylo-
genetics have suggested that it is often preferable to apply
partition-specific rate models instead of using a single
gamma distribution to describe the heterogeneity of
the substitution process across multiple-gene regions
and codon positions (Sullivan and Joyce 2005). Parti-
tioned models have been shown to improve likelihood
scores considerably (Castoe et al. 2004; Brandley et al.
2005), while having a great effect on branch lengths
(Marshall et al. 2006) and divergence time estimates
(Yang and Yoder 2003). In this study, we found the par-
titioning to cause an increase in inferred rates or ages by
up to 12–14% (fig. 2 and table 5), with the deeper nodes
affected proportionally more than the most recent nodes
and a parallel increase in marginal likelihood as assessed
by the harmonic mean estimator (table 4). This is in
agreement with other recent studies showing an effect
of the partitioning scheme on estimated divergence
times (Poux et al. 2008; Torres-Carvajal and de Queiroz
2009), particularly in cases where a single calibration
point is used or when an externally estimated standard
molecular clock is applied (Marshall et al. 2006; Papado-
poulou, Jones, et al. 2009). However, the different criteria
for selecting the preferred partitioning scheme using
Bayes factors (favoring the P3 or the most parameter-
rich partitioning scheme P4) resulted in very similar
substitution rates (2.68% vs. 2.69%), indicating that
overparameterization has little impact on the estimates.
05 1015 2025 3035 40 4550
average calibration age (Mya)
%divergence per My
05 10 15 2025 30 3540 4550
average calibration age (Mya)
%divergence per My
FIG. 3. Rates of mtDNA divergence estimated for different insect
groups by 31 other studies (supplementary table S1, Supplementary
Material online) plotted against the average age of the calibration
points used for the estimation. Estimated rates are based either on
protein-coding genes or a combination of protein-coding and rDNA
or tRNA gene regions. An exponential curve with three parameters
was fitted under the least-squares criterion following Ho et al.
(2005). (a) Studies that did not account for rate heterogeneity
among sites; Rate(t) 5 5.408e?1.794tþ 1.5231, R25 0.60, (b) Studies
that accounted for rate heterogeneity among sites using a gamma
distribution; Rate(t) 5 17.256e?1.157tþ 2.0968, R25 0.97.
Papadopoulou et al. · doi:10.1093/molbev/msq051
Time Dependency of Molecular Rate Estimates:
Phylogenetic Versus Coalescent Dating
Brower’s (1994) estimatewas basedon relativelyrecentcal-
ibration points (300 years–3.25 My) and assumed a linear
relationship of sequence divergence with time over this pe-
riod. This ignores the much higher apparent rate of diver-
gence in recently split lineages (Ho et al. 2005; Ho and
Larson 2006) which is possibly due tothe fact that (slightly)
deleterious mutations persist as transient polymorphisms
in large populations, before they are lost at phylogenetic
scales (Penny 2005). Even if the exact timescale affected
by these higher rates and the exponential distribution de-
scribing this effect has been debated (Emerson 2007; Ho
et al. 2007), the J-shaped curve of inferred rates in the lit-
erature survey fits an exponential function (fig. 3) and is
consistent with an increased intraspecific rate. However,
there is uncertainty with these data because a recent cal-
ibration point is needed when using population-level data
to estimate mutation rates (Ho et al. 2008), which is rarely
available. Consequently, the upturn in this curve is largely
due to a small number of attempts to calibrate an intra-
specific rate for insect taxa (Clarke et al. 2001; Gratton
et al. 2008), which resulted in very high mtDNA divergence
sive studies on mammals (including human populations)
and birds where demographic data are available confirm
this exponential relationship, in broad agreement with
the extrapolations of our figure 3. Yet, studies in insects
continue to apply the standard phylogenetic rate of
2.3% at the population level where intraspecific calibration
points are not available (Pfeiler et al. 2007; Anducho-Reyes
et al. 2008; Avtzis et al. 2008; Leschen et al. 2008; McLean
et al. 2008; Meng et al. 2008) and therefore risk to overes-
timate the inferred evolutionary ages.
Likewise, given these great differences between genea-
logical and phylogenetic rates, it appears critical not to
mix intraspecific and interspecific sequence data when es-
timating lineage ages. Here, we applied an explicit proce-
dure for removing intraspecific variation using the GMYC
model, which separates independently coalescing entities
that are described by different equations in population-
level and phylogenetic data. This mixed modelwas strongly
favored over the null model of a uniform branching pro-
cess, indicating that indeed such coalescent groups exist in
this data set and can be identified by the threshold value
that defines the ML point for the transition in branching
rates. The GMYC model therefore was used here to reduce
the phylogenetic tree to those portions that are expected
to exhibit the long-term substitution rate, thereby elimi-
nating fast-evolving population level variation that might
have resulted in an overall higher rate estimate.
Caveats of Biogeographic Calibrations
The dense sampling across a putative biogeographic barrier
also addresses the problem of a priori species delimitation
and selection of target groups suitable for clock calibra-
tions. Clockcalibrations usingbiogeographicaldata, includ-
ing those used by Brower (1994), are usually assessed based
on population sampling around a ‘‘known’’ boundary,
rather than detected in a broad survey of populations,
and the divergence on either side of a given barrier is au-
tomatically ascribed to the geological events that led to the
formation of the barrier. However, lineage divergence and
barrier formation (and its timing) may be uncorrelated,
even where multiple codistributed taxon pairs are found
to be subdivided across a given barrier (Heads 2005). For
example, the final rise of the Isthmus of Panama at 2.7–
3.5 Mya has been widely employed for molecular clock cal-
ibrations in marine taxa, but their ancestors may have di-
verged well before the closure of the Isthmus (Knowlton
and Weigt 1998; Marko 2002) resulting in overestimated
substitution rates. Here, instead of assuming a priori that
all east/west splits can be attributed to a single biogeo-
graphic event, we sampled populations comprehensively
across a wide geographic area without prior assumption
of a biogeographic boundary and, once a major genetic
break was found consistent with known geological features
(the mid-Aegean trench), only those taxa showing broad
temporal congruence across this border were retained
for the calibration, using a relative dating approach (Loader
et al. 2007).
Table 5. Estimated Mean Ages ± 1 SD for Each of the East–West Nodes When Applying the Standard 2.3% Divergence My?1Rate on the
mtDNA Data set Either Unpartitioned (P0) or Under Four Alternative Partitioning Schemes (P1–P4) or on cox1 and rrnL Separately.
10.9 6 1.37
8.67 6 0.87
4.78 6 0.78
9.67 6 1.65
11.64 6 1.35
7.26 6 0.8
12.22 6 1.96
10.73 6 1.33
14.81 6 2.25
12.32 6 1.46
7.09 6 1.3
13.38 6 2.33
16.41 6 2.03
9.46 6 1.24
17.55 6 3.07
15.57 6 2.16
17 6 2.7
12.9 6 1.61
7.45 6 1.53
13.98 6 2.69
16.88 6 2.13
10.18 6 1.41
17.59 6 3.4
15.53 6 2.35
16.81 6 2.73
13.09 6 1.66
7.46 6 1.55
13.96 6 2.77
16.68 6 2.15
10.17 6 1.43
18.25 6 3.56
15.83 6 2.38
10.97 6 1.47
9.61 6 1.07
4.85 6 0.83
9.87 6 1.72
12.34 6 1.51
7.38 6 0.87
13.68 6 2.21
12.04 6 1.59
11.77 6 1.5
9.47 6 1.01
4.85 6 0.85
9.81 6 1.78
12.38 6 1.5
7.42 6 0.85
12.78 6 2.19
11.52 6 1.52
11.8 6 1.58
9.76 6 1.07
4.92 6 0.87
10.03 6 1.84
12.47 6 1.54
7.65 6 0.91
13.64 6 2.38
11.9 6 1.59
11.79 6 1.58
9.87 6 1.08
4.91 6 0.86
10.01 6 1.82
12.42 6 1.54
7.63 6 0.9
13.98 6 2.41
12.15 6 1.62
4.96 6 1.31
4.34 6 1
2.12 6 0.93
3.03 6 1.25
4.34 6 1.15
3.49 6 0.77
4.15 6 1.41
4.37 6 1.19
bcox1 versus rrnL.
c3rd codons versus all other sites.
d1st and 2nd positions versus 3rd positions versus rrnL.
e1st versus 2nd versus 3rd versus rrnL.
fEach codon position separately.
Revisiting the Insect mtDNA Clock · doi:10.1093/molbev/msq051
Another important consideration when using biogeo-
graphic separations to calibrate molecular clocks is the fact
that gene divergence generally predates population diver-
gence due to ancestral polymorphism (Edwards and Beerli
2000), which may greatly affect the divergence time esti-
mates when looking at recently separated populations.
However, the amount of error will be proportionally small-
er when divergence time is long relatively to the effective
population size of the ancestral population (Edwards and
Beerli 2000; Hurt et al. 2009). In the example of the Isthmus
of Panama, recent coalescent-based reanalyses of multiple
‘‘geminate’’ species data sets found that a large proportion
of the observed variation in genetic distances among taxa
can be explained without invoking scenarios of nonsimul-
taneous divergence if ancestral polymorphism and among-
taxon differences in demographic history are taken into ac-
count (Hickerson et al. 2006; Hurt et al. 2009). Nonetheless,
applying a similar approach on the Aegean tenebrionid
data set was not considered appropriate as this data set
concerns more ancient divergences, whereas each of the
eastern and western clades are composed of multiple in-
dependently coalescing groups confined to sets of adjacent
islands. Lineage sorting therefore might affect the relation-
ships at the level of sets of coalescence groups within each
of the eastern and western clades but is unlikely to go as far
tions of current coalescence-based approaches for time
calibrations that require an estimate of the effective pop-
ulation size to be taken into account (as implemented, e.g.,
in the msBayes software; Hickerson et al. 2007) are not ful-
filled. Moreover, our calibration was performed on all no-
des simultaneously, using an uncorrelated relaxed-clock
method, which permitted us to estimate a mean rate in
the face of stochastic variation in coalescent times and rate
heterogeneity among lineages.
We also intended to minimize the confounding effects
of postseparation gene flow by selecting focal taxa that are
predicted to exhibit very low dispersal propensity given
their morphological traits and their habitat association
(Papadopoulou, Anastasiou, et al. 2009). For example,
the geophilic lineages usedhere coexist onthe same islands
with ‘‘psammophilic’’ tenebrionids confined to ephemeral
sandy habitats (beaches, sand dunes) and winged lineages.
persistence and consequently these taxa show much
weaker geographic structure in the same island system,
(Papadopoulou, Anastasiou, et al. 2009), so they are not
suitable for clock calibrations based on vicariance patterns.
However, despite the careful choice of study taxa, our re-
sults illustrate the difficulty of correlating a geologically
dated event with the timing of the lineage split. Ocean bar-
riers that separated eastern and western lineages may have
broken down during the Messinian drying events, and
therefore, the six older contemporaneous east/west splits
might be attributable to a more recent event along the
same biogeographic boundary. Under this scenario, the
original geological separation that lasted at least 3 My (9
to 5.96 Mya) would have left no signature on the diversi-
fication of these lineages, that is, all lineages would have
acquired wide distribution in the Messinian before again
being subjected to vicariant separation, which seems un-
unexplained which now can be associated with the post-
Messinian split. Attributing the six older divergence events
to the original formation of the trench therefore is consid-
ered the most parsimonious explanation for the observed
Calibrating the Insect Molecular Clock
The final estimate for the mean mtDNA divergence rate in
the Aegean tenebrionids of 2.39% My?1under a relaxed
clock and applying a GTRþCþI or 2.69% under the pre-
ferred partitioning scheme (fig. 2) was remarkably similar
to Brower’s (1994) estimate and might give credibility to
this widely applied rate. However, the current analysis
shows that we should guard ourselves against accepting
this number as a ‘‘universal’’ clock rate. The way these
two figures were arrived at could not be more different,
as they were obtained from very recent (mostly , 1
My) versus older (10.5 ± 1.5 My) calibration ages, uncor-
rected versus corrected substitution rates, and separate
versus simultaneous estimates from multiple calibrations.
versity of data sets from cox1, cox2, rrnS, rrnL, and restric-
tion sites that are not easily comparable. Our estimates for
Aegean tenebrionids are a composite of cox1 and slowly
evolving rrnL genes (for a rate of 3.54% and 1.06%, respec-
tively), and therefore the close similarity to the standard
rate reflects an average of two quite divergent estimates.
Finally, Brower’s (1994) estimate refers to a path across
the tree (i.e., % divergence between terminal taxa per time
to-tip axis (expected number of substitutions per site per
time unit). Under certain conditions, these two ways to as-
sess rates can produce different results, particularly if there
is not rate constancy across the tree. Therefore, it would be
preferable to express our rate estimates as substitutions/
site/My rather than % divergence/My, although we main-
tained the latter here (and converted all numbers accord-
ingly) for consistency with the existing literature.
Even if Brower’s (1994) calibration was based on a range
of different genes, as cox1 is widely used in insect phyloge-
netics and in many studies is the only marker available, the
2.3% rate has been often applied to cox1-only data sets.
Furthermore, since Farrell (2001) and Quek et al. (2004)
both reported a rate of 1.5% for cox1, several studies
(Dick et al. 2004; Kandul et al. 2004; Zhang et al. 2005;
Steiner et al. 2006; Bell et al. 2007; Aoki et al. 2008; Canfield
et al. 2008; Kiyoshi 2008; Leschen et al. 2008; Wirta et al.
2008; Kawakita and Kato 2009) employed this lower rate
for cox1 data sets. Here, we found the substitution rate
of cox1 to be more than twice as fast as the estimates
of Farrell (2001) and Quek et al. (2004). Their low estimates
can easily be explained as an artifact of using uncorrected
pairwise distances, whereas their calibration points were
comparatively ancient (up to 20 Mya). Higher rates for
Papadopoulou et al. · doi:10.1093/molbev/msq051
cox1 have been estimated in various insect lineages by Pons
and Vogler (2005) (3.34%; Coleoptera: Cicindelidae),
Shapiro et al. (2006) (3–4%; Orthoptera: Tettigoniidae),
Kiyoshi and Sota (2006) (3.1%; Odonata: Gomphidae),
Percy et al. (2004) (2.35–3.15%; Hemiptera: Psyllidae),
and Nazari and Sperling (2007) (2.3–3.1%; Lepidoptera:
Papilionidae). Nonetheless, cox1 appears to be among
the most slowly evolving protein-coding mitochondrial
genes (Crozier RH and Crozier YC 1993), which is corrob-
orated by recent studies (e.g., Pons and Vogler 2005: Cyto-
chrome b (cob) 4.22% vs. cox1 3.34%). Therefore, our
estimates of the mtDNA clock would likely be higher using
other protein-coding genes. In contrast, the mitochondrial
2001), in agreement with our estimate for rrnL (1.08%)
which is still higher than the rates proposed by Go ´mez-
Zurita et al. (2000) (0.45%, Chrysomelidae), Pons and Vogler
(2005) (0.76%, Cicindelidae) and it falls into the range re-
ported by Percy et al. (2004) (0.95–1.9%, Psyllidae). Hence,
there are consistent, large differences in rates among mito-
chondrial genes and therefore standard rates to date phylog-
enies in the absence of other independent evidence need to
be based on gene-specific calibrations.
Regarding the nuclear genes, the mean rate obtained for
the intron region of Mp20 (3.68%) is close to the numbers
reported for synonymous substitutions in Drosophila (3%
by Rowan and Hunt 1991; 3.08% by Li 1997) and for
a ‘‘numt’’ pseudogene in Cicindela (3.33%; Pons and Vogler
of neutral single-copy nuclear DNA in insects is similar to
the mtDNA rate (Caccone et al. 1988; Sharp and Li 1989),
which is consistent with our results, although our estimate
for the intron rate remains preliminary, as the SD is large
of the substitution process. However, the fact that this rate
is similar to those expected from independent studies pro-
vides confidence in the calibration based on the formation
of the mid-Aegean trench and therefore also supports the
mtDNA rates established here.
Overall, thesubstitution rates found here for theAegean
tenebrionids appear slightly higher than most previous es-
timates for insects, which could be partially due to the par-
ticular mode of diversification of these island lineages. High
speciation rates have been associated with elevated rates of
molecular evolution (Barraclough and Savolainen 2001;
Webster et al. 2003), whereas lineages with small effective
populations sizes are also predicted to have increased sub-
stitution rates due to the greater rate of fixation of nearly
neutral alleles (Ohta 1987; Woolfit 2009). However, the hy-
pothesis that island radiations can speed up the molecular
clock has been tested explicitly in a range of different data
sets and has found no support (Bromham and Woolfit
MtDNA substitution rates are frequently the only source of
information to date historical scenarios. The desire for
a standard rate of change is therefore not surprising but
the uncritical use of these rates to calibrate phylogenetic
trees is fraught with errors. Here, we established the meth-
odological issues that impact rate estimates, with the cor-
rection for rate heterogeneity among sites and the choice
of genes having the greatest effect. However, despite great
differences in reported rate estimates, the discrepancies
can be largely explained by differences in methodology
(fig. 3) without invoking large differences in the actual
mtDNA substitution rate. More data sets of this kind that
reduce stochastic effects by using multiple independent
calibration points are needed to assess the subtle changes
in rates among lineages with greater precision. This will es-
the Aegean tenebrionids are unique to this group or reflect
generally an underestimate of the standard rate for cox1.
at Molecular Biology and Evolution online (http://
We are grateful to Bekir Keskin (Ege University, Izmir) for
help with fieldwork and identification of specimens and F.
Spagopoulou, M. Stalimerou, and S. Terzopoulou (Univer-
sity of Athens) for field assistance. We also thank four
anonymous referees and the Associate Editor for helpful
comments on the manuscript. This work was funded by
a Greek State Scholarship Foundation studentship to
A.P. and Natural Environment Research Council grant
NE/C510908/1 to A.P.V.
Andersen NM, Cheng L, Damgaard J, Sperling FAH. 2000.
Mitochondrial DNA sequence variation and phylogeography of
oceanic insects (Hemiptera: Gerridae: Halobates spp.). Mar Biol.
Anducho-Reyes MA, Cognato AI, Hayes JL, Zuniga G. 2008.
Phylogeography of the bark beetle Dendroctonus mexicanus
Hopkins (Coleoptera: Curculionidae: Scolytinae). Mol Phylogenet
Aoki K, Kato M, Murakami N. 2008. Glacial bottleneck and
postglacial recolonization of a seed parasitic weevil, Curculio
hilgendorfi, inferred from mitochondrial DNA variation. Mol
Arbogast BS, Edwards SV, Wakeley J, Beerli P, Slowinski JB. 2002.
Estimating divergence times from molecular data on phyloge-
netic and population genetic timescales. Annu Rev Ecol Syst.
Avtzis DN, Arthofer W, Stauffer C. 2008. Sympatric occurrence of
diverged mtDNA lineages of Pityogenes chalcographus (Co-
leoptera, Scolytinae) in Europe. Biol J Linn Soc Lond. 94(2):
Baker CH, Graham GC, Scott KD, Cameron SL, Yeates DK,
Merritt DJ. 2008. Distribution and phylogenetic relationships
of Australian glow-worms Arachnocampa (Diptera, Keroplati-
dae). Mol Phylogenet Evol. 48(2):506–514.
Barraclough TG, Savolainen V. 2001. Evolutionary rates and species
diversity in flowering plants. Evolution. 55:677–683.
Revisiting the Insect mtDNA Clock · doi:10.1093/molbev/msq051
Bell KL, Moritz C, Moussalli A, Yeates DK. 2007. Comparative
phylogeography and speciation of dung beetles from the
Australian Wet Tropics rainforest. Mol Ecol. 16(23):4984–4998.
Bernhard D, Fritzsch G, Glockner P, Wurst C. 2005. Molecular
insights into speciation in the Agrilus viridis-complex and the
genus Trachys (Coleoptera: Buprestidae). Eur J Entomol. 102(4):
Brandley MC, Schmitz A, Reeder TW. 2005. Partitioned Bayesian
analyses, partition choice, and the phylogenetic relationships of
scincid lizards. Syst Biol. 54(3):373–390.
Bromham L, Woolfit M. 2004. Explosive radiations and the reliability
of molecular clocks: island endemic radiations as a test case. Syst
Brower AVZ. 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.
Caccone A, Amato GD, Powell JR. 1988. Rates and patterns of
scnDNA and mtDNA divergence within the Drosophila
melanogaster subgroup. Genetics 118(4):671–683.
Canfield MR, Greene E, Moreau CS, Chen N, Pierce NE. 2008.
Exploring phenotypic plasticity and biogeography in emerald
moths: a phylogeny of the genus Nemoria (Lepidoptera:
Geometridae). Mol Phylogenet Evol. 49(2):477–487.
Carstens BC, Knowles LL. 2007. Shifting distributions and speciation:
species divergence during rapid climate change. Mol Ecol. 16(3):
Castoe TA, Doan TM, Parkinson CL. 2004. Data partitions and
complex models in Bayesian analysis: the phylogeny of
Gymnophthalmid lizards. Syst Biol. 53(3):448–469.
Clarke TE, Levin DB, Kavanaugh DH, Reimchen TE. 2001. Rapid
evolution in the Nebria gregaria group (Coleoptera: Carabidae)
and the paleogeography of the Queen Charlotte Islands.
Cognato AI, Vogler AP. 2001. Exploring data interaction and
nucleotide alignment in a multiple gene analysis of Ips
(Coleoptera: Scolytinae). Syst Biol. 50:758–780.
Creutzburg N. 1963. Paleogeographic evolution of Crete from
Miocene till our days. Cretan Ann. 15–16:336–342.
Crozier RH, Crozier YC. 1993. The mitochondrial genome of the
honeybee Apis mellifera: complete sequence and genome
organization. Genetics 133(1):91–117.
Dermitzakis DM, Papanikolaou DJ. 1981. Paleogeography and
geodynamics of the Aegean region during the Neogene. Annales
Ge ´ologique des Pays Helle ´nique. 30:245–289.
Dick CW, Roubik DW, Gruber KF, Bermingham E. 2004. Long-
distance gene flow and cross-Andean dispersal of lowland
rainforest bees (Apidae: Euglossini) revealed by comparative
Drummond AJ, Ho SYW, Phillips MJ, Rambaut A. 2006. Relaxed
phylogenetics and dating with confidence. PLoS Biol. 4(5):
Drummond AJ, Rambaut A. 2007. BEAST: Bayesian evolutionary
analysis by sampling trees. BMC Evol Biol. 7:214
Edwards SV, Beerli P. 2000. Perspective: gene divergence, population
divergence, and the variance in coalescence time in phylogeo-
graphic studies. Evolution 54(6):1839–1854.
Emerson BC. 2007. Alarm bells for the molecular clock? No support
for Ho et al.’s model of time-dependent molecular rate
estimates. Syst Biol. 56(2):337–345.
Farrell BD. 2001. Evolutionary assembly of the milkweed fauna:
cytochrome oxidase I and the age of Tetraopes beetles. Mol
Phylogenet Evol. 18(3):467–478.
Farris JS, Ka ¨llersjo ¨ M, Kluge AG, Bult C. 1994. Testing significance of
incongruence. Cladistics 10:315–319.
Mol Ecol. 13(12):
Fleischer RC, McIntosh CE, Tarr CL. 1998. Evolution on a volcanic
conveyor belt: using phylogeographic reconstructions and K-
Ar-based ages of the Hawaiian Islands to estimate molecular
evolutionary rates. Mol Ecol. 7(4):533–545.
Fontaneto D, Herniou EA, Boschetti C, Caprioli M, Melone G,
Ricci C, Barraclough TG. 2007. Independently evolving species in
asexual bdelloid rotifers. PLoS Biol. 5:914–921.
Gaunt MW, Miles MA. 2002. An insect molecular clock dates the
origin of the insects and accords with palaeontological and
biogeographic landmarks. Mol Biol Evol. 19(5):748–761.
Go ´mez-Zurita J, Juan C, Petitpierre E. 2000. The evolutionary history
of the genus Timarcha (Coleoptera, Chrysomelidae) inferred
from mitochondrial COII gene and partial 16S rDNA sequences.
Mol Phylogenet Evol. 14:304–317.
Gratton P, Konopinski MK, Sbordoni V. 2008. Pleistocene
evolutionary history of the Clouded Apollo (Parnassius
mnemosyne): genetic signatures of climate cycles and a ‘time-
dependent’ mitochondrial substitution rate. Mol Ecol. 17(19):
Guindon S, Gascuel O. 2003. A simple, fast, and accurate algorithm
to estimate large phylogenies by maximum likelihood. Syst Biol.
Heads M. 2005. Dating nodes on molecular phylogenies: a critique of
molecular biogeography. Cladistics 21(1):62–78.
Hickerson MJ, Gilchrist MA, Takebayashi N. 2003. Calibrating
a molecular clock from phylogeographic data: moments and
likelihood estimators. Evolution 57(10):2216–2225.
Hickerson MJ, Stahl E, Takebayashi N. 2007. msBayes: pipeline for
testing comparative phylogeographic histories using hierarchical
approximate Bayesian computation. BMC Bioinformatics. 8:268.
Hickerson MJ, Stahl EA, Lessios HA. 2006. Test for simultaneous
divergence using approximate Bayesian computation. Evolution
Ho SYW. 2009. An examination of phylogenetic models of
substitution rate variation among lineages. Biol Lett. 5(3):421–424
Ho SYW, Larson G. 2006. Molecular clocks: when times are
a-changin’. Trends Genet. 22(2):79–83.
Ho SYW, Phillips MJ, Cooper A, Drummond AJ. 2005. Time
dependency of molecular rate estimates and systematic over-
estimation of recent divergence times. Mol Biol Evol. 22(7):
Ho SYW, Saarma U, Barnett R, Haile J, Shapiro B. 2008. The effect of
inappropriate calibration: three case studies in molecular
ecology. PLoS One. 3(2):e1615.
Ho SYW, Shapiro B, Phillips MJ, Cooper A, Drummond AJ. 2007.
Evidence for time dependency of molecular rate estimates. Syst
Hurt C, Anker A, Knowlton N. 2009. A multilocus test of
simultaneous divergence across the Isthmus of Panama using
snapping shrimp in the genus Alpheus. Evolution 63(2):514–530.
Kandul NP, Lukhtanov VA, Dantchenko AV, Coleman JWS,
Sekercioglu CH, Haig D, Pierce NE. 2004. Phylogeny of
Agrodiaetus Hubner 1822 (Lepidoptera: Lycaenidae) inferred
from mtDNA sequences of COI and COII and nuclear sequences
of EF1-alpha: karyotype diversification and species radiation. Syst
Kass RE, Raftery AE. 1995. Bayes factors. J Am Stat Assoc. 90(430):
Katoh K, Kuma K, Toh H, Miyata T. 2005. MAFFT version 5:
improvement in accuracy of multiple sequence alignment.
Nucleic Acids Res. 33:511–518.
Katoh K, Toh H. 2008. Recent developments in the MAFFT multiple
sequence alignment program. Brief Bioinform. 9(4):286–298.
Kawakita A, Kato M. 2009. Repeated independent evolution of
obligate pollination mutualism in the Phyllantheae-Epicephala
association. Proc R Soc Lond B Biol Sci. 276(1656):417–426.
Papadopoulou et al. · doi:10.1093/molbev/msq051
Kiyoshi T. 2008. Differentiation of golden-ringed dragonfly Ano-
togaster sieboldii (Selys, 1854) (Cordulegastridae: Odonata) in
the insular East Asia revealed by the mitochondrial gene
genealogy with taxonomic implications. J Zool Syst Evol Res.
Kiyoshi T, Sota T. 2006. Differentiation of the dragonfly genus
Davidius (Odonata: Gomphidae) in Japan inferred from
mitochondrial and nuclear gene genealogies. Zool Sci. 23(1):1–8.
Knowlton N, Weigt LA. 1998. New dates and new rates for
divergence across the Isthmus of Panama. Proc R Soc Lond B Biol
Krijgsman W, Hilgen FJ, Raffi I, Sierro FJ, Wilson DS. 1999.
Chronology, causes and progression of the Messinian salinity
crisis. Nature 400(6745):652–655.
Leschen RAB, Buckley TR, Harman HM, Shulmeister J. 2008.
Determining the origin and age of the Westland beech
(Nothofagus) gap, New Zealand, using fungus beetle genetics.
Mol Ecol. 17(5):1256–1276.
Li WH. 1997. Molecular evolution. Sunderland (MA): Sinauer.
Loader SP, Pisani D, Cotton JA, Gower DJ, Day JJ, Wilkinson M. 2007.
Relative time scales reveal multiple origins of parallel disjunct
distributions of African caecilian amphibians. Biol Lett. 3:
Luchetti A, Marini M, Mantovani B. 2005. Mitochondrial evolutionary
rate and speciation in termites: data on European Reticulitermes
taxa (Isoptera, Rhinotermitidae). Insectes Soc. 52(3):218–221.
Marko PB. 2002. Fossil calibration of molecular clocks and the
divergence times of geminate species pairs separated by the
Isthmus of Panama. Mol Biol Evol. 19(11):2005–2021.
Marshall DC, Simon C, Buckley TR. 2006. Accurate branch length
estimation in partitioned Bayesian analyses requires accommo-
dation of among-partition rate variation and attention to
branch length priors. Syst Biol. 55(6):993–1003.
McLean AJ, Schmidt DJ, Hughes JM. 2008. Do lowland habitats
represent barriers to dispersal for a rainforest mayfly, Bungona
narilla, inSouth-East Queensland?
Meng XF, Shi M, Chen XX. 2008. Population genetic structure of
Chilo suppressalis (Walker) (Lepidoptera: Crambidae): strong
subdivision in China inferred from microsatellite markers and
mtDNA gene sequences. Mol Ecol. 17(12):2880–2897.
Miller KB, Bergsten J, Whiting MF. Forthcoming. 2009. Phylogeny
and classification of the tribe Hydaticini (Coleoptera: Dytisci-
dae): partition choice for Bayesian analysis with multiple nuclear
and mitochondrial protein-coding genes. Zool Scr. 38(6):591–615
Monaghan MT, Inward DJG, Hunt T, Vogler AP. 2007. A molecular
phylogenetic analysis of the Scarabaeinae (dung beetles). Mol
Phylogenet Evol. 45:674–692.
Nazari V, Sperling FAH. 2007. Mitochondrial DNA divergence and
phylogeography in western Palaearctic Parnassiinae (Lepidop-
tera: Papilionidae): how many species are there? Insect Syst Evol.
Nylander JAA. 2004. MrModeltest v2. Program distributed by the
author. Evolutionary Biology Centre, Uppsala University.
Ohta T. 1987. Very slightly deleterious mutations and the molecular
clock. J Mol Evol. 26(1–2):1–6.
Pagel M, Meade A. 2004. A phylogenetic mixture model for
detecting pattern-heterogeneity in gene sequence or character-
state data. Syst Biol. 53(4):571–581.
Papadopoulou A, Anastasiou I, Keskin B, Vogler AP. 2009.
Comparative phylogeography of tenebrionid beetles in the
Aegean archipelago: the effect of dispersal ability and habitat
preference. Mol Ecol. 18:2503–2517.
Papadopoulou A, Jones AG, Hammond PM, Vogler AP. 2009. DNA
taxonomy and phylogeography of beetles of the Falkland Islands
(Islas Malvinas). Mol Phylogenet Evol. 53(3):935–947.
Penny D. 2005. Evolutionary biology—relativity for molecular clocks.
Percy DM, Page RDM, Cronk QCB. 2004. Plant-insect interactions:
double-dating associated insect and plant lineages reveals
asynchronous radiations. Syst Biol. 53(1):120–127.
Pfeiler E, Erez T, Hurtado LA, Markow TA. 2007. Genetic
differentiation and demographic history in Drosophila pachea
from the Sonoran Desert. Hereditas 144(2):63–74.
Pons J, Barraclough TG, Gomez-Zurita J, Cardoso A, Duran DP,
Hazell S, Kamoun S, Sumlin WD, Vogler AP. 2006. Sequence-
based species delimitation for the DNA taxonomy of unde-
scribed insects. Syst Biol. 55:595–609.
Pons J, Barraclough TG, Theodorides K, Cardoso A, Vogler AP. 2004.
Using exon and intron sequences of the gene Mp20 to resolve
basal relationships in Cicindela (Coleoptera: Cicindelidae). Syst
Pons J, Vogler AP. 2005. Complex pattern of coalescence and fast
evolution of a mitochondrial rRNA pseudogene in a recent
radiation of tiger beetles. Mol Biol Evol. 22:991–1000.
Poux C, Madsen O, Glos J, de Jong WW, Vences M. 2008. Molecular
phylogeny and divergence times of Malagasy tenrecs: influence
of data partitioning and taxon sampling on dating analyses. BMC
Evol Biol. 8:102
Pruser F, Mossakowski D. 1998. Low substitution rates in
mitochondrial DNA in Mediterranean carabid beetles. Insect
Mol Biol. 7(2):121–128.
Quek SP, Davies SJ, Itino T, Pierce NE. 2004. Codiversification in an
ant-plant mutualism: stem texture and the evolution of host use
in Crematogaster (Formicidae: Myrmicinae) inhabitants of
Macaranga (Euphorbiaceae). Evolution 58(3):554–570.
Ronquist F, Huelsenbeck JP. 2003. MrBayes 3: Bayesian phylogenetic
inference under mixedmodels.
Rowan RG, Hunt JA. 1991. Rates of DNA change and phylogeny
from the DNA-sequences of the alcohol-dehydrogenase for 5
closely related species of Hawaiian Drosophila. Mol Biol Evol.
Sanderson MJ. 2002. Estimating absolute rates of molecular
evolution and divergence times: a penalized likelihood approach.
Mol Biol Evol 19:101–109.
Sanderson MJ. 2003. r8s: inferring absolute rates of molecular
evolution, divergence times in the absence of a molecular clock.
Shapiro LH, Strazanac JS, Roderick GK. 2006. Molecular phylogeny of
Banza (Orthoptera: Tettigoniidae), the endemic katydids of the
Hawaiian Archipelago. Mol Phylogenet Evol. 41(1):53–63.
Sharp PM, Li WH. 1989. On the rate of DNA-sequence evolution in
Drosophila. J Mol Evol. 28(5):398–402.
Shimodaira H, Hasegawa M. 1999. Multiple comparisons of log-
likelihoods with applications to phylogenetic inference. Mol Biol
Simon C, Frati F, Beckenbach A, Crespi B, Liu H, Flook P. 1994.
Evolution, weighting, and phylogenetic utility of mitochondrial
gene sequences and a compilation of conserved polymerase
chain reaction primers. Ann Entomol Soc Am. 87:651–701.
Sperling FAH, Spence JR, Andersen NM. 1997. Mitochondrial DNA,
allozymes, morphology, and hybrid compatibility in Limnoporus
water striders (Heteroptera: Gerridae): do they all track species
phylogenies? Ann Entomol Soc Am. 90(4):401–415.
Stamatakis A. 2006. RAxML-VI-HPC: maximum likelihood-based
phylogenetic analyses with thousands of taxa and mixed models.
Stamatakis A, Hoover P, Rougemont J. 2008. A rapid bootstrap
algorithm for the RAxML web servers. Syst Biol. 57(5):758–771.
Steiner FM, Schlick-Steiner BC, Konrad H, Linksvayer TA, Quek SP,
Christian E, Stauffer C, Buschinger A. 2006. Phylogeny and
Revisiting the Insect mtDNA Clock · doi:10.1093/molbev/msq051
evolutionary history of queen polymorphic Myrmecina ants Download full-text
(Hymenoptera: Formicidae). Eur J Entomol. 103(3):619–626.
Sullivan J, Joyce P. 2005. Model selection in phylogenetics. Ann Rev
Ecology Evol Syst. 36:445–466.
Swofford DL. 2002. PAUP*: phylogenetic analysis using parsimony.
Version 4.0b. Sunderland (MA): Sinauer Associates.
Tamura K, Dudley J, Nei M, Kumar S. 2007. MEGA4: molecular
evolutionary genetics analysis (MEGA) software version 4.0. Mol
Biol Evol. 24(8):1596–1599.
Thompson JD, Higgins DG, Gibson TJ. 1994. CLUSTAL W: improving
the sensitivity of progressive multiple sequence alignment
through sequence weighting, position specific gap penalties
and weight matrix choice. Nucleic Acids Res. 22:4673–4680.
Thorne JL, Kishino H. 2002. Divergence time and evolutionary rate
estimation with multilocus data. Syst Biol. 51(5):689–702.
Thornton I. 2007. Island colonization: the origin and development of
Island communities. New York: Cambridge University Press.
Torres-Carvajal O, de Queiroz K. 2009. Phylogeny of hoplocercine
lizards (Squamata: Iguania) with estimates of relative divergence
times. Mol Phylogenet Evol. 50(1):31–43.
Trewick SA, Wallis GP. 2001. Bridging the ‘‘beech-gap’’: New Zealand
invertebrate phylogeography implicates Pleistocene glaciation
and Pliocene isolation. Evolution 55(11):2170–2180.
Webster AJ, Payne RJH, Pagel M. 2003. Molecular phylogenies link
rates of evolution and speciation. Science 301(5632):478–478.
Wheeler WC, Hayashi CY. 1998. The phylogeny of the extant
chelicerate orders. Cladistics 14:173–192.
Wirta H, Orsini L, Hanski I. 2008. An old adaptive radiation of forest
dung beetles inMadagascar.
Woolfit M. 2009. Effective population size and the rate and pattern
of nucleotide substitutions. Biol Lett. 5(3):417–420.
Yang Z. 2004. A heuristic rate smoothing procedure for maximum
likelihood estimation of species divergence times. Acta Zool
Yang ZH. 1996. Maximum-likelihood models for combined analyses
of multiple sequence data. J Mol Evol. 42(5):587–596.
Yang ZH, Yoder AD. 2003. Comparison of likelihood and Bayesian
methods for estimating divergence times using multiple gene
loci and calibration points, with application to a radiation of
cute-looking mouse lemur species. Syst Biol. 52(5):705–716.
Zhang AB, Kubota K, Takami Y, Kim JL, Kim JK, Sota T. 2005. Species
status and phylogeography of two closely related Coptolabrus
species (Coleoptera: Carabidae) in South Korea inferred from
mitochondrial and nuclear gene sequences. Mol Ecol. 14(12):
Papadopoulou et al. · doi:10.1093/molbev/msq051