Mitochondrial genomes of cichlids
We determined complete or nearly complete mtDNA nucleotide sequences
for six new cichlids from Africa, South America, Madagascar, and
Indo/Sri Lanka (Table 1). The sizes of these genomes ranged from 16,457 to 16,556 bp, including approximately 800 bp in the control region. Tylochromis polylepis alone
appears to have a somewhat longer control region (approximately 1200
bp) although the exact sequence of the region was unable to be
determined because of the long poly-T sequences within the region. We
also analyzed the previously published mitogenomic sequences of four
cichlid species (Table 1). Oreochromis mossambicus (accession no. AY597335) was not included because a congeneric taxon (Oreochromis sp.) sequenced by Mabuchi et al.  had already been sampled.
All 37 genes encoding two rRNAs, 22 tRNAs, and 13 proteins were
identified in these 10 cichlid mitogenomes, basically in the same order
and orientation found for most other vertebrates. Transfer RNA genes
could be folded into secondary structures typical of vertebrate
mitochondrial tRNA . The base composition of cichlid mitogenomes was skewed (data not shown) similarly to those of other vertebrates .
shows the phylogenetic relationships inferred from the Bayesian
analysis among the 52 bony fishes, estimated with two sharks as an
outgroup. The tree topology was identical to that obtained by the
partitioned ML analysis (data not shown). These bony fish taxa included
two sarcopterygians (coelacanth and lungfish), nine basal
actinopterygians (polypterids, acipenseriforms, lepisosteids, and
amiid), and 41 teleosts, including 10 cichlids. The phylogenetic
relationships obtained for non-cichlid taxa (Fig. 1) were largely consistent with those from previous mitogenomic studies [28,43,45], except for a difference in the sister group of holosteans (lepisosteids and amiid).
Although Inoue et al. 
suggested that the "Ancient Fish Clade" unites acipenserids,
lepisosteids, and amiid, our phylogenetic analysis supports the
neopterygian clade (lepisosteids + amiid + teleosts), in agreement with
an analysis of nuclear DNA sequences  and morphological characters .
Relationships between the basal actinopterygians and teleosts were not
stable against changes in taxonomic representations and the genes used
and varied between the two hypotheses (data not shown). We tentatively
assumed the neopterygian relationship for subsequent analyses because
this was consistent in both morphological and molecular (based on
mitochondrial and nuclear sequences) analyses. However, we also
conducted analyses to evaluate how our major conclusions in dating
depend on the two alternative phylogenetic relationships (Table 3).
Table 3. Comparison of divergence time estimates between different time constraints and studies
In terms of the relationships among 20 percomorphs
containing 14 labroids (two labrids, two pomacentrids, and 10
cichlids), we reconfirmed the polyphyly of Labroidei  whereby labrids (designated Labroidei 1 in Fig. 1)
and cichlids + pomacentirids (Labroidei 2) appear in separate lineages
of teleosts. The non-monophyly of the labroid taxa was supported by a
number of nodes with 100% posterior probability and 100% bootstrap
values (Fig. 1).
Figure 1. A Bayesian tree based on mitogenomic DNA sequences.
This is a 50% majority rule consensus tree among 10,000 pooled trees
from two independent Bayesian MCMC runs. The data set comprises aligned
gap-free nucleotide sequences of 10,034-bp length from 54 taxa, which
included 4,887 variable sites and 3,936 parsimony-informative sites.
Partitioned Bayesian analyses were conducted using the GTR + I + Γ
model and with all model parameters variable and unlinked across
partitions. The numerals at internal nodes or branches indicate
Bayesian posterior probabilities (left) and maximum likelihood
bootstrap probability values (right) from 1000 replicates, respectively
(shown as percentage for values above 50%).
Among the 10 cichlid taxa that we used, four were from
Africa, two from South America, three from Madagascar, and one from
Indo/Sri Lanka. The tree (Fig. 1)
supports the monophyly of Cichlidae and two other continental groups
from Africa and South America. Four basal taxa from Madagascar and
Indo/Sri Lanka are not monophyletic, and two (Paretroplus from Madagascar and Etroplus from Indo/Sri Lanka) corresponding to Etroplinae sensu Sparks and Smith  form a sister group to all other cichlids. The other two Malagasy taxa (Paratilapia and Ptychochromoides), corresponding to Ptychochrominae sensu Sparks and Smith ,
form a sister group to the African + Neotropical clade. These results
are consistent with previous molecular studies that used a few
mitochondrial or nuclear gene sequences [14-16,48], as well as morphological studies .
However, these previous studies did not fully evaluate the
statistical significance in rejecting alternative hypotheses of cichlid
relationships. We conducted KH and SH tests, as well as a test using
Bayes factor. Based on these tests, alternative hypotheses assuming the
monophyly of Malagasy + Indo/Sri Lankan cichlids (constraint 1), Old
World cichlids (constraint 2), and African + Indo/Sri Lankan cichlids
(constraint 3) are all very unlikely (Table 4).
These results provide statistical support for the paraphyletic
assemblage of the Malagasy + Indo/Sri Lankan taxa to the African +
Table 4. Test of alternative phylogenetic hypotheses for continental cichlid groups
If Cichlidae originated in Cenozoic Africa and migrated into South America, Madagascar, and India via saltwater dispersal [19,49],
Malagasy/Indo Sri Lankan and/or Neotropical taxa would probably be
nested in the African clade, and alternative relationships (e.g., those
corresponding to constraints 2 and 3) would likely appear. However,
these relationships were not found, thus supporting the vicariant
divergence scenario [13,14,18], at least from a topological standpoint.
Timing of cichlid divergences
We conducted divergence time estimation among 54 bony fishes, including 10 cichlids (Fig. 2). Twenty-one time constraints based on extensive fossil evidence for bony fishes (Table 2) were used. Following the advice of Benton and Donoghue 
to set fossil-based time constraints as hard lower boundaries and soft
upper boundaries, we chose older values for upper boundaries. We
estimated the divergence between African + Neotropical cichlids and
Malagasy + Indo/Sri Lankan (ptychochrominae) cichlids to be
approximately 96 MYA (78–115 MYA at 95% credibility). The divergences
of African vs. Neotropical cichlids and Malagasy vs. Indo/Sri Lankan
cichlids within the Etroplinae were estimated to be approximately 89
MYA (72–108 MYA) and 87 MYA (69–106 MYA), respectively.
Figure 2. Divergence times estimated from the partitioned Bayesian analysis.
A posterior distribution of divergence times with 95% credibility
intervals (shaded rectangles) was obtained using mitogenomic DNA
sequences (10,034 sites). Two sharks (Scyliorhinus canicula and Mustelus manazo) were used as an outgroup (not shown). The multidistribute program 
was used to estimate divergence times assuming the tree topology shown
in Fig. 1. Letters indicate nodes at which maximum and/or minimum time
constraints were set (see Table 2 for details of the individual
constraints). Paleogeographical maps at 148 MYA, 120 MYA, 95 MYA, and
85 MYA  are shown. Dark-gray areas on the maps represent those being fragmented within Gondwanaland at those times.
We then compared the estimated divergence times among
cichlids and the probable times of continental fragmentation based on
geological evidence. The divergence time between Malagasy and Indo/Sri
Lankan taxa within Etroplinae (~87 MYA: 69–106 MYA) is very close to
the time of separation between Madagascar and India (85–95 MYA) [50,51].
The divergence time estimated between African and Neotropical clades
(~89 MYA: 72–108 MYA) is also close to the time of separation between
African and South American landmasses (~100 MYA) [50,51].
The divergence time between African + Neotropical cichlids and Malagasy
ptychochrominae cichlids (~96 MYA: 78–115 MYA) appears to be somewhat
more recent than the time generally accepted for the complete
separation of the Indo-Madagascar landmass from Gondwanaland (120–130
MYA) [50,51]. However, some studies 
have postulated an extended connection between India and Antarctica by
approximately 112 MYA, which is within the 95% credibility range for
the African/Neotropical vs. ptychochrominae cichlid divergence. Taken
together, these results are consistent with the vicariant divergence of
continental cichlid groups during Cretaceous times and argue against
their Cenozoic dispersal.
Vences et al. 
calibrated a molecular clock for cichlids that assumed that the
divergence time of the most basal endemic lineages in East African Rift
lakes (e.g., Tanganyika) corresponds to the geological estimate of the
age of the lakes. These estimated divergence times between continental
cichlid clades were all in the Cenozoic (rather than the Mesozoic, as
we demonstrate in Fig. 2)
and supported the hypothesis of long-distance Cenozoic transmarine
dispersal of cichlids. This view of the Cenozoic (or latest Cretaceous)
origin and transmarine dispersal of cichlids has also been supported by
some biogeographers  because it is consistent with cichlid fossil records, which first occur in South America and Africa in the Eocene [20,53]. However, the clock-based dating procedures of Vences et al.  present some problems. The strict molecular clock may not hold for all cichlid lineages ,
and the premise that the oldest endemic cichlid divergence is
synchronized with the formation of the lakes may not be valid. Some
lineages that had diverged outside the lake may have immigrated in
parallel . In addition, there is no definitive, geologically based time estimate for the formation of the lakes.
More recently, Genner et al.  used two mitochondrial (cytochrome b and
16S rRNA) and one nuclear (TMO-4C4) gene fragments to estimate the
divergence times among cichlids. When the cichlid divergence by
Gondwanan vicariance was assumed, the resultant divergence times were
more consistent with those estimated with time constraints from
previous paleontological and molecular studies [2,54-57] than when the Cenozoic cichlid divergence was assumed based on fossil records.
Although we concur on the Gondwanan origin and vicariant divergence of cichlids, Genner et al. 
evaluated this biogeographic hypothesis somewhat indirectly, in that
the fitness of estimated times of cichlid divergences to those obtained
with time constraints from previous studies was qualitatively compared
between alternative assumptions on cichlid biogeography. We evaluated
cichlid divergence times more directly by using longer mitogenomic
sequence data and dozens of non-cichlid taxa, allowing us to set many
time constraints purely from the paleontological data and providing
additional evidence for an ancient cichlid divergence on Gondwanaland,
despite the general paucity of the Mesozoic and Cenozoic
paleontological record on bony fishes.
Gondwana fragmentation as time constraints
In Figure 3, minimum time constraints based on fossil records (see Table 2) are plotted against molecular time estimates of the corresponding divergences (values taken from Fig. 2).
In this figure, minimum age estimates of Gondwanan fragmentations are
also plotted against the corresponding molecular time estimates of
continental cichlid groups. It should be noted here that the latter
data points reflecting Gondwanan fragmentation history (closed
triangles) are plotted well on the line of 1:1 relationship whereas
most of the data points reflecting fossil records (closed circles) are
considerably below the line of the 1:1 relationship. This pattern
suggests that Gondwana fragmentation history that is congruent with the
cichlid phylogeny can be effective time constraints better than most of
the Mesozoic and Cenozoic fossil records used here.
Figure 3. Comparison of paleontological and molecular estimates of divergence times.
Minimum estimates of divergence times deducible from fossil records
(see Table 2) were plotted as closed circles against molecularly
estimated divergence times (mean values for the divergence times shown
in Fig. 2). Closed triangles show plots of the timing of continental
breakups against the molecular time estimates of cichlid divergences
between the corresponding continents (data taken from Fig. 2). The
timings used for complete continental breakups are 112 MYA for (Africa
+ South America) vs. (Madagascar + Indo/Sri Lanka), 100 MYA for Africa
vs. South America, and 85 MYA for Madagascar vs. Indo/Sri Lanka [50-52]. The solid line indicates a 1:1 relationship between paleontological and molecular time estimates.
Among the fossil data points, four data points in the
Paleozoic show a fairly good 1:1 relationship, whereas other points
mostly in the Mesozoic are considerably below the line of 1:1
relationship. This might mean that the Mesozoic fossils do not really
represent the oldest fossil for the corresponding lineages whereas this
is not the case for older Paleozoic lineages. This situation is
somewhat reminiscent of the apparent relative paucity of Mesozoic
fossil evidence of tetrapods (mammals and birds) .
Several papers have noticed that molecular time estimations are consistently older than paleontological ones [2,3,5-7,59]. Benton and Ayala 
have pointed out four pervasive biases that make molecular dates too
old: i) too old calibration dates based on previous molecular studies;
ii) undetected fast-evolving genes; iii) ancestral polymorphism that is
maintained through long evolutionary period; and iv) asymmetric
distributions of estimated times, with a constrained younger end but an
unconstrained older end (this is caused because rates of evolution are
constrained to be nonnegative, but the rates are unbounded above zero).
The first factor is not the case for the present study, because we
did not use the calibration dates based on previous molecular studies,
but used only those based on fossil records. The third factor would be
the case when the used genomic regions are under the long-term
balancing selection, but no mitochondrial gene has been reported to be
under such selection. Regarding the second and fourth factors, we
believe that they are also not the case for this study, because we used
mitogenomic sequence data excluding peculiarly rapid evolving region
(e.g., the control region), and because each mitochondrial gene used
here was tested to perform well for dating vertebrate (tetrapod)
divergences . According to Benton and Ayala ,
for reliable dating "careful choice of genes may be a more appropriate
strategy (than the larger data strategy), with a focus on long and
fast-evolving (yet alignable) sequences." Our present study based on
nearly whole mitogenomic sequence data fairly accommodates such
Improved dating of teleostean divergences
We then conducted the divergence time estimation using the Gondwanan
vicariance assumption regarding cichlids as additional time constraints
(Fig. 4). Compared to the results shown in Figure 2
(without the additional time constraints), the means of estimated
divergence times at various nodes are similar or somewhat larger (= 18
million years; see Table 3).
However, the 95% credibility ranges of the estimated times overlap well
between the two results, and the differences in mean values are not
large, compared to potential error ranges in other elements, such as
stochastic errors in molecular evolution and errors in dating fossils.
Figure 4. Divergence
times estimated from the partitioned Bayesian analysis using both
paleontological time constraints (Table 2) and biogeographical
assumptions for the divergences of continental cichlid groups.
The added time constraints on cichlid divergences are as follows: 112
MYA (lower) and 145 MYA (upper) for (Africa + South America) vs.
(Madagascar + Indo/Sri Lanka); 100 MYA (lower) and 120 MYA (upper) for
Africa vs. South America; and 85 MYA (lower) and 95 MYA (upper) for
Madagascar vs. Indo/Sri Lanka [50-52]. See Fig. 2 legend for other details.
The addition of the cichlid constraints appears to
shorten the 95% credibility intervals of the time estimates, especially
for divergences occurring within Acanthomorpha 100–200 MYA. For
example, our Figure 2 and Yamanoue et al. 
estimated the divergence time of torafugu (Tetraodontiformes) and
medaka (Beloniformes) to be approximately 159 (136–183) MYA and 184
(154–221) MYA, respectively. The cichlid constraints considerably
narrowed the 95% credibility interval to 176 (163–191) MYA (Table 3),
and also increased the precision of time estimates for other nodes. The
use of ample molecular data from mitogenomic sequences also helped to
narrow the credibility interval. For example, Kumazawa et al.  used two mitochondrial genes (NADH dehydrogenase subunit 2 and cytochrome b)
and estimated the divergence between torafugu and zebrafish at 284 ± 28
(mean ± standard deviation) MYA, whereas our whole mitogenomic data set
showed the divergence at 288 (268–307) MYA (Table 3).