ARTICLE IN PRESS

Theoretical Population Biology ( ) –

Contents lists available at ScienceDirect

Theoretical Population Biology

journal homepage: www.elsevier.com/locate/tpb

Alternatives to the Wright–Fisher model: The robustness of mitochondrial

Eve dating
Krzysztof A. Cyran a , Marek Kimmel b,c,∗
a
Institute of Informatics, Silesian University of Technology, Gliwice, Poland
b
Department of Statistics, Rice University, Houston, TX, United States
c
Systems Engineering Group, Silesian University of Technology, Gliwice, Poland

article info abstract

Article history: Methods of calculating the distributions of the time to coalescence depend on the underlying model
Received 18 March 2009 of population demography. In particular, the models assuming deterministic evolution of population
Available online xxxx size may not be applicable to populations evolving stochastically. Therefore the study of coalescence
models involving stochastic demography is important for applications. One interesting approach which
Keywords: includes stochasticity is the O’Connell limit theory of genealogy in branching processes. Our paper
Coalescence distribution
explores how many generations are needed for the limiting distributions of O’Connell to become adequate
Branching processes genealogy
Wright–Fisher model
approximations of exact distributions. We perform extensive simulations of slightly supercritical
O’Connell model branching processes and compare the results to the O’Connell limits. Coalescent computations under
Computer simulations the Wright–Fisher model are compared with limiting O’Connell results and with full genealogy-based
MtEve dating predictions. These results are used to estimate the age of the so-called mitochondrial Eve, i.e., the
Neanderthal mtDNA root of the mitochondrial polymorphisms of the modern humans based on the DNA from humans and
Neanderthal fossils.
© 2010 Elsevier Inc. All rights reserved.

1. Introduction With the introduction of new sequencing techniques, DNA

The indices of genetic variation, including allele distribution,
heterozygosity or linkage disequilibrium, are affected by popula-
tion history. Therefore a lot of effort has been spent by popula-
tion geneticists to estimate the long-term demographic history of
populations belonging to various species. For this purpose many
statistical tests detecting past population expansion have been
proposed, including King et al. (2000), Bjorklund (2003), Laan et al.
(2005) and Cyran and Myszor (2008). Most of the existing infer-
ence techniques are based on the Wright–Fisher model of genetic
drift, supplemented by models of mutation, recombination, selec-
tion, and demography. This is in agreement with a long-term prac-
tice in genetic research. However, there exists another tradition
of modeling of demographic processes, which uses the branching
process approach. Comparison between these two approaches is
likely to determine how the different underlying assumptions in-
fluence differing estimates of genetic and demographic parame-
ters. This paper attempts to contribute to this task, using the data
on modern human ancestry.
∗ Corresponding author at: Department of Statistics, Rice University, Houston, TX,
United States.
E-mail address: kimmel@rice.edu (M. Kimmel).
0040-5809/$ – see front matter © 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.tpb.2010.06.001
strands taken from many qualitatively different loci of Homo sapi-
ens and even Homo Neanderthalensis fossils have been analyzed.
Examples of these studies include maternally inherited mitochon-
drial DNA (mtDNA) (Briggs et al., 2009; Green et al., 2008; Krings
et al., 2000, 1999, 1997; Ovchinnikov et al., 2000; Serré et al., 2004),
paternally inherited Y chromosomes (Jobling, 2001; Thompson
et al., 2000), X chromosomes (Wooding and Rogers, 2000), autoso-
mal DNA sequences (Green et al., 2010; Burbano et al., 2010; Green
et al., 2006; Noonan et al., 2006; Pennisi, 2007; Yu et al., 2001), or
nuclear short tandem repeats (STRs) (Kimmel et al., 1998). Despite
these efforts, the problem of human population trajectory is still
open, and thus there is a growing interest in the studies of sensitiv-
ity of genetic variation indices to departures from the assumptions
made in different models. In particular, it is interesting how these
departures influence the distributions of the time to coalescence,
and the paper addresses this problem.
For a sample of DNA sequences not undergoing recombination,
it is assumed that all sequences are descendants of an ancestral
chromosome existing some generations ago. The time to this
ancestor is referred to as the time to coalescence of the whole
sample. Analogously, we define the time to the coalescence of
two chromosomes randomly drawn from the sample, as a time to
the most recent common ancestor (MRCA) of these chromosomes.
The coalescent method, which is frequently used to infer the

Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
 ARTICLE IN PRESS
2 K.A. Cyran, M. Kimmel / Theoretical Population Biology (
time to the MRCA, is a powerful tool based on the reverse-
time analysis of the Wright–Fisher model of genetic drift. For
large populations, coalescent models are equivalent to diffusion
process approximations, which depend only on the mean and
on the variance of offspring number distribution. Consequently,
coalescent models are robust for large populations, but during the
population bottlenecks, the diffusion approximation may fail.
How relevant for the model predictions are departures from a
panmictic population (in the case of the Wright–Fisher model) and
from large population size (the case of the coalescent method)?
To answer this question we compare three different models for
calculating the distribution of the time to coalescence of a pair of
chromosomes. These include
• the Wright–Fisher model with discrete generations,
• the coalescent-based model with continuous time scaled by the
size of population variable in time, and
• the less known O’Connell model based on branching processes
(see O’Connell, 1995).
All three models are applied to stochastic population growth
approximated by a slightly supercritical Galton–Watson branching
process. To be able to compare these methodologies reliably,
we designed a computational framework to estimate the two-
chromosome coalescence distribution in any of these models as
well as in a model based on a full record of the population history.
After simulating several thousand genealogies we estimated
parameters with great accuracy.
There might be some doubt whether the use of the time to
coalescence of two chromosomes is an adequate tool to estimate
the age of the MRCA. Therefore, we also considered coalescence in
a sample of n chromosomes randomly chosen from a population.
However, the computational difficulties of the recursion involved
as well as the difficulty of linking the results with the known
genetic indices make the approach troublesome. Analysis of the
aforementioned approach is beyond the scope of this article and
will be discussed in a separate paper.
We consider models suitable for the analysis of data having the
form of numbers of pairwise differences between DNA sequences
in the sample. Although phylogenetic methods, attempting to use
all genetic information contained in a sample to build the geneal-
ogy (e.g. Griffiths and Tavaré, 1995, Griffiths’ Genetree software,
available from http://www.stats.ox.ac.uk/~griff/software.html),
tend to give estimates with a smaller variance than those based on
pairwise differences, they cannot be directly compared with the
O’Connell model playing an important role in our paper.
This model was originally proposed by O’Connell (1995)
for dating the mitochondrial Eve (mtEve) based on a sample
of mtDNA of humans and chimpanzees. The O’Connell’s limit
results are based on the assumption that the population is
growing as a slightly supercritical branching process with progeny
distributions homogeneous in time. Though these are not quite
realistic assumptions, especially that of time homogeneity, the
model is important as an alternative to the Wright–Fisher model,
since it does not assume any particular offspring distribution.
Moreover, asymptotically, for a given expected number of
offspring, the O’Connell model is independent of the shape
of the progeny number distribution, and in particular of its
variance as long as this variance is bounded. This property is
interesting in the light of classical results in which the short-
term inbreeding effective population size is proportional to the
variance of the offspring number distribution, and therefore the
variance influences the shape of the coalescence distribution. The
invariance of the offspring number distribution in the O’Connell
model is theoretically valid in a limit. It remains unknown how
fast, in the terms of number of generations, the coalescence
distributions in a real population converge to O’Connell’s limit.
Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
) –
This could be answered only by time-forward simulation of the
full branching process genealogy and then by comparing the actual
distributions with limiting results. The growing interest in studies
concerning genealogies of branching processes is reflected among
others by the studies of Klebaner and Sagitov (2002) focused on
the geometric distribution of progeny, and by the work of Lambert
(2003) focused on subcritical cases. Still, we consider the O’Connell
model as a standard because of its independence of the offspring
number distribution and our interest in supercritical processes
which can model the long-term growth of the human population
size.
In contrast to the O’Connell model, the Wright–Fisher model is
not limited to any specific growth pattern. Yet except for some
early classical work, such as that of Nagylaki (1990), relatively
little effort has been expended in analyzing its relationship to
other models in terms of sensitivity to departures from the models
assumptions. Addressing this problem, this paper compares
coalescence distributions under a range of Wright–Fisher models
(including those which arose from the time continuous coalescent)
to distributions obtained from the O’Connell model. Finally, using
computer software designed by us for this purpose, the results
of all these models are compared with the actual distributions
obtained from simulations of several thousand full genealogies.
As a real world application of our results, we report estimations
of the time to mitochondrial MRCA of modern humans to show
how sensitive these estimates are to the assumptions made by the
various compared models.
2. Models
The two population genetics models we employ are defined
further on in Sections 2.1 and 2.2. Two approaches will be used
by us to study how sensitive is the distribution of the time to
coalescence to specific model assumptions. The first approach
requires storing the entire simulated genealogy of a population.
Then, by averaging over genealogies, the experimental distribution
of the times to coalescence can be found and then compared to
those obtained in the Wright–Fisher and the coalescent models.
This approach is very general since it is not limited to any
generation-to-generation sampling scheme or assumption about
large population size. In particular it can be applied for an arbitrary
progeny distribution (possibly changing in time) used to model
the evolution of the population as a branching process. However,
with the exception of small population sizes, it requires a large
amount of memory for storing information about each generation,
and therefore it seems practically infeasible for simulations of the
required number of generations.
In the alternative approach, the population history is simulated
and only the population size is recorded as a function of time.
Using this information and approaches such as in Bobrowski and
Kimmel (2004) we compute the coalescence distributions of the
pairs of sequences. The results of such experiments were reported
by Cyran and Kimmel (2004) and Cyran (2007). They were also
used in Cyran and Kimmel (2005) for conservative estimation of
the parameter α (see Section 2.3 for the definition of how this
parameter influences the expansion rate of the population) in a
problem of hypothetical Neanderthal contribution to the modern
human mtDNA gene pool. However, this methodology lacks one
important feature which could be taken into consideration only
in the first approach. Namely, by regarding only the sizes of the
population and not its full genealogy, it is impossible to distinguish
between the length of time of the entire simulation started from
the ancestral individual, and the length of time to the MRCA. The
problem becomes clear if we realize that the founder of the process,
the common ancestor of the population evolved, is rarely the
most recent common ancestor of the extant individuals. Having no
 ARTICLE IN PRESS
K.A. Cyran, M. Kimmel / Theoretical Population Biology ( ) – 3

possibility to distinguish between the two, we assumed in earlier we propagated this result to arbitrarily many generations using the
studies (Cyran and Kimmel, 2004, 2005) that the time between the equation
founder and the MRCA is relatively short compared to the time   
TMRCA

of the whole process. Therefore we treated both as identical, not E K0 = 1 davg

T
having information as to what extent this simplification can be davg
T̂MRCA_y =    =    . (3)
justified.
δE T2c T2c 
δ
T
 
 K0 = 1 E K = 1
But now, with the increase of computational power and mem-
ory capacity, we have returned to the problem of implementing
the first approach indirectly. We designed software capable of
simulating full genealogies of at least 102 generations under an
arbitrarily chosen distribution of offspring number and with pa-
rameters of a branching process identical to those which could
reflect the long-time (about 104 generations) evolution of mod-
ern humans. Subsequently, we checked whether the asymptotic
properties of the O’Connell model hold for a small number of gen-
erations such as 102 . If the asymptotic properties of the slightly
supercritical branching process were already valid for simulations
comprising as few as 102 generations, then this also will be the case
for simulations exceeding 102 generations. Therefore, we use the
O’Connell model as a theoretical standard, which extrapolates the
full genealogy simulation results for arbitrarily many generations.
The comparison of the coalescence distributions in different
models allows us to observe how sensitive the estimate of TMRCA,
the time to the most recent common ancestor, is to specific model
assumptions. We consider a sample of n DNA sequences taken from
a population with the expected duration of a generation (in years)
equal to λ. We denote the average pairwise mutation difference
in such a sample by davg and the mutation rate per nucleotide per
generation by µ. In the infinitely many sites model the genetic
divergence rate between two species, δ , is equal to µ/λ, so we
can estimate the mutation rate using the identity µ̂ = δλ. Then,
denoting the expected time to coalescence of two individuals in a
population by T2c , it follows that
T TMRCA  T  0
The parameters δ and davg included in formula (3) and necessary
for estimating T̂MRCA(y) can be retrieved from genetic diversity
data.
Remark. The Wright–Fisher model implies multinomial sampling
of chromosomes from one generation to the other, which can be
approximated by Poisson sampling. Since the inbreeding effective
population size is proportional to the variance of the number of
offspring, we also consider the binary fission (BF) distribution and
linear fractional (LF) distribution. In this way, we investigate the
effects of the departures from the model’s standard, Poisson (P)
progeny number distribution, in the direction of under-dispersion
(variance less than mean) and over-dispersion (variance greater
than mean).
2.1. Wright–Fisher model for a pair of chromosomes
Our presentation of the models starts from the Wright–Fisher
model applied to the smallest sample exhibiting effects of the
genetic drift, i.e. the sample composed of two chromosomes. The
model assumes a population of haploid individuals, say mtDNA
sequences, which at time t = 0 has the size Nt . Since multinomial
sampling is assumed, two individuals at generation t + 1 are
descendants of the single member of generation t with probability
pt = 1/Nt . Consequently, with probability qt = 1 − pt they
are descendants of two different members. This is reflected in
the following distribution of the time to coalescence, T2c , of two
   randomly drawn chromosomes (e.g. Bobrowski and Kimmel, 2004)
T2c 
E (davg |K0 = 1) = TMRCA µ̂E  K0 = 1 , (1) T −1
Y T −1
Y T −1
Y
TMRCA 
P (T2c = t ) = qk − q k = p T −t −1 qk , (4)
where K0 is the number of these individuals at generation 0 whose k=T −t k=T −t −1 k=T −t

descendants are presently alive. Assuming that TMRCA(y) = λTMRCA where T denotes the number of generations we consider, and
is the equivalent of TMRCA expressed in years, the moment-based for mathematical consistency we let q−1 = 0 and p−1 = 1.
estimate of TMRCA(y) is The average pairwise mutation difference within a sample, after
scaling by the mutation rate µ, corresponds to the expectation
davg of the coalescence distribution (2), and moreover, the discrete
T̂MRCA(y) =   . (2) nature of generations makes it easy to simulate the demography.
δE T2c

K0 = 1
TMRCA 
Therefore, using Monte Carlo techniques it is possible to estimate
the unconditional coalescence distribution by averaging the
In the Wright–Fisher model-based computations, we obtain the conditional one using a series of Nt realizations required to
expectation E (T2c /T | K0 = 1) by performing computer simulations compute parameters in (4).
of a branching process starting from one individual and calculating
the required ratio of T2c and TMRCA . After simulating several 2.2. Coalescent model
thousand processes we planned to obtain the expectation of
Let us assume a coalescent model with population size Nτ
the ratio. However, only in the model with the record of a full
variable in time and continuous time τ measured backwards.
genealogy, the times T2c and TMRCA were explicitly given and (2)
Suppose also that λ(τ ) = N0 /Nτ and that τ2c is the time to
could be applied directly. In other models, only the time T2c could
coalescence of a pair of chromosomes observed over N0 genera-
be computed and the time TMRCA was not available directly. Instead tions. Then, the tail of the distribution of τ2c is given by
we have at our disposal time T , i.e., duration of the process.  Z τ 
Certainly, time T , being the time to the only individual initiating the
P (τ2c > τ ) = exp − λ(u)du , (5)
branching process, is the time to the common ancestor of the whole 0
evolved population. However, as mentioned, it is rarely also the which is the continuous analog of (4). To ensure existence of a
time to the most recent common ancestor. Nevertheless, we were unique common ancestor, λ(t ) must satisfy
able to estimate the ratio of TMRCA and T in simulations with a fully Z ∞
recorded genealogy, and moreover we confirmed that the limiting λ(u)du = ∞. (6)
properties of coalescence distribution in the O’Connell model are 0
valid for as little as 102 generations for which we could perform For the stochastic Nt , and therefore λ(τ ), the right side of the
the full genealogy simulations. In this way we could relate in the Eq. (5) should be averaged over the process realizations. In the
O’Connell model TMRCA to T and T2c , and applying the limit theorem context of our study it is also worth noticing that the continuous

Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
 ARTICLE IN PRESS
4 K.A. Cyran, M. Kimmel / Theoretical Population Biology (
coalescence model correctly approximates the discrete coalescent
model as long as 1 − 1/Nτ ≈ exp(−1/Nτ ), which certainly is not
true in the early phase of the branching process, when Nt is not
large and undergoes stochastic fluctuations. This fact is reflected in
the differences between experimental distributions of the time to
coalescence in the coalescent model and the O’Connell branching
process model.
2.3. O’Connell model
) –
Moreover, the expectation of the ratio TMRCA /T in (13) is obtained
from the simulations with recorded full genealogies, and x̂
denotes the estimate of parameter x. Therefore, to calculate the
estimated MRCA time T̂MRCA(y) from the genetic variation data, we
need α̂ . However, from the simulation results concordant with
limiting properties of the O’Connell model we can have the ratio
E (TMRCA /T |K0 = 1). Therefore, we can simultaneously estimate
TMRCA(y) and α . From (8), if ZT is substituted as an estimate of its
expected value, it follows that

K0 = 1 σ T̂MRCA(y) exp α̂ − 1
   2
Consider a slightly supercritical time-homogeneous branching T  

ZT = E (15)
process with expected number of offspring E (ξ0 ) = 1 + α/T + TMRCA  2λα̂
o(1/T ) and variance Var(ξ0 ) = σ 2 + O(1/T ). For this model,
and estimates of TMRCA(y) and α are solutions of the system of Eqs.
an asymptotic property of the probability P x (Zt > 0) where
(13) and (15) for given short-term inbreeding effective population
P x denotes probabilities for the process started by x individuals,
size of females ZT , and genetic data davg and δ .
satisfies the O’Connell (1995) formula
2α x 2.4. Simulations
P x (Zt > 0) ∼
 , as T → ∞. (7)
σ 1 − exp −α Tt

2

From this it follows (Cyran and Kimmel, 2004) that Simulation mode 1
The first simulation mode implements the full genealogy
σ 2T recording in the branching process model, thus allowing explicit
E (ZT |ZT > 0, Z0 = x) ∼ [exp (α) − 1] , as T → ∞ (8) access to any desired feature of the model. In particular, it is
2α
possible to trace back the genealogy of a pair of individuals and
where symbol ∼ denotes asymptotic equivalence. Let us express
to find their MRCA, and therefore the actual time of coalescence.
the time interval [0, T ] of variable t as a unit interval [0, 1] of
By randomly choosing from simulated population a sample of
variable r = t /T . Then (O’Connell, 1995), as corrected in Kimmel
about 100 individuals and determining coalescence of each pair
and Axelrod (2002), for times T long enough, we have the following
in a sample we obtain a histogram HT 2c |tree of the times to
equation describing the tail of the distribution of DT , the time of
coalescence, conditionally on the simulated tree. This histogram
death of the last common ancestor of two individuals living at T ,
is the experimental approximation of the conditional coalescence
given that we start the population history from x individuals having
distribution P (T2c = t |tree) in the full genealogy model. Having
descendants at T
   obtained the distribution P (T2c = t |tree) we also compute its
DT expected value E (T2c |tree) denoting it as T2c_avg |tree. Additionally,
> r K0 = x

lim P
T →∞ T we trace back the lineages of the whole population to the MRCA,
and therefore we have the time TMRCA |tree as well as the ratios
2qxr h i
= (qr − 1)−x (x − 1)! − F (x − 1, 1 − qr ) , (9) (T2c_avg /TMRCA )|tree and (TMRCA /T )|tree. Finally, by simulating
(x − 1)! many branching processes and then by averaging over trees
where generated, we obtain corresponding unconditional characteristics.
These characteristics include: the histogram HT 2c , the distribution
e−r α − e−α
qr = (10) P (T2c = t ) and its expectation E (T2c ), the distribution P (T2c_avg =
1 − e−α t ) with the expectation E (T2c_avg ), the distribution P (TMRCA = t )
and F : Z+ × (0, 1) → R is defined as with the expected value E (TMRCA ), as well as the histograms and
the expectations over genealogies of the ratios T2c_avg /TMRCA , and
∂n ln(1 − y)
 
F (n, y) = . (11) TMRCA /T . It is important to notice that E (T2c_avg /TMRCA ) obtained
∂ yn y2 in the procedure described above, can be used in this model in
The O’Connell original distribution is continuous, but to compare the Eq. (2) instead of E (T2c /TMRCA ), what yields a smaller variance
it to the discrete empirical distributions described below, we estimator. Such substitution is justified from the genetic point of
consider the discretized version, specified by the tail of the original view by linking the expectation E (T2c_avg ) (scaled by divergence
distribution computed at points r corresponding to integer values rate δ = µ/λ in (1), with the average pairwise mutation difference
of t = rT . For the sake of terminological simplicity, we will still in a sample davg . Note also that the simulations which became
refer to this discrete distribution as the O’Connell distribution. extinct were excluded from computations since problems similar
In the O’Connell model, to those of dating MRCA of modern humans are always solved
   conditionally on non-extinction.
T2c  1 Simulation mode 2
E K0 = 1 = E [(T − DT ) |K0 = 1] , (12)
T  T The second simulation mode stores only the course of
and therefore, the Eq. (3) becomes population size change described by the branching process. This
   mode is used for numerical computation of the distribution in
TMRCA  the Wright–Fisher (7) or the coalescent (8) models, conditional
T̂MRCA(y) = E K0 = 1
T  on Nt . In the discrete Wright–Fisher model, the Eq. (7) can be
directly applied if the history of Nt is known from simulation.
davg
×  , (13) However, in the continuous coalescent model, instead of using (7)
R1
δ 1−2 0 q̂r


q̂r − 1 − ln q̂r dr we apply a Monte Carlo approach by generation of coalescence
(1−q̂r )2 times from the distribution (8) conditional on Nt . This procedure
was repeated 104 times for one simulated branching process.
where
The conditional histogram is used as the approximation of the
e−r α̂ − e−α̂ conditional distribution P (T2c = t | Nt , CM), in which CM
q̂r = . (14) denotes conditioning on the coalescent model. As in the case of
1 − e−α̂
Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
 ARTICLE IN PRESS
K.A. Cyran, M. Kimmel / Theoretical Population Biology ( ) – 5

the full genealogy models, the unconditional with respect to Nt ,

but conditional on the model used, distributions P (T2c = t | CM)
and P (T2c = t | W–F) are obtained by averaging over many
realizations of Nt ; W–F denotes the Wright–Fisher model.

3. Genetic data

To obtain estimates of the time to MRCA we considered the av-

erage pairwise mutation differences davg computed from a com-
plete Neanderthal mitochondrial genome sequence determined by
high-throughput sequencing (Green et al., 2008). We also com-
puted the genetic divergence rate δ based on results reported by
Briggs et al. (2009). We also compared the results based on the re- Fig. 1. Distributions of T2c computed in the full genealogy model.
cent data to the results calculated from a sample of 663 mtDNA
sequences of modern humans and their homolog sequenced from
the Neanderthal fossils (Krings et al., 1999, a single Neanderthal se- a
quence was sequenced at that time). These latter sequences were
taken from the hypervariable control regions I (HVRI) and II (HVRII)
of mtDNA.
As reported by Green et al. (2008), the complete mitochondrial
genome of H. sapiens is composed of 16,568 nucleotides (the
mitochondrial genome of H. neanderthalensis is 3 nucleotides
shorter). The number of average pairwise differences between
species is equal to 210.3 ± 4, and within H. sapiens sample
64.8 ± 26.8. This results in the average mutation difference
between modern humans and Neanderthals of davg ∼ = 0.013, and
the average mutation difference within modern human population
= 0.004.
of davg ∼ b
In the Krings et al. (1999) sample including the two hypervari-
able regions the mutation rate is about 5 times larger. The average
number of pairwise differences is equal to 35.3 ± 2.3. Therefore the
average mutation difference between modern humans and Nean-
derthals is equal to davg ∼= 0.059. Divergence in contemporary hu-
mans results in the average number of pairwise differences equal
to 10.9 ± 5.1, and thus the average mutation difference among
contemporary humans is equal to davg ∼ = 0.018. The average mu-
tation difference among modern humans calculated originally by
O’Connell (1995) was equal to 0.028, but we do not use it because
of O’Connell’s small sample of 19 humans.
Interestingly, the average mutation difference between H.
neanderthalensis and H. sapiens is about 3.25 times greater than the Fig. 2. Distributions in the full genealogy model: general comparison (a),
average mutation difference among contemporary humans based comparison of distributions of T2c in the full genealogy model with binary fission
on the hypervariable regions and based on the complete mtDNA; offspring number distribution and the limit O’Connell model (b).

the corresponding ratios differ by less than 1%. This suggests

among other that the number of mutations in the hypervariable
regions is small enough to allow ignoring reverse mutations
occurring in both lineages from the time of their MRCA. We
assume this time Td to be about 511,000 years ago, based on Briggs
et al. (2009) results, which are based on analysis including the
information about the dating of the Neanderthal fossils. Combining
Briggs et al. (2009) and Green et al. (2008) data and applying the
infinitely many sites model, we calculate the rate of divergence
for the complete mtDNA as δ = davg /Td(MN) ∼ = 0.013/511,000 =
2.5 × 10−8 mutations per nucleotide per year.
4. Results
For consistency of comparison, the results will be presented in
discrete time measured backwards in generation units. Results of
the models that traditionally use differently measured time have
been scaled accordingly. Note that despite this, the distributions
are depicted for visual convenience as continuous curves.
We confirmed that the O’Connell limit model can approximate
the models with full genealogy even if the number of generations
is as small as 102 . The models with full genealogy yield
visually indistinguishable distributions P (T2c = t ), P (T2c (avg) =
t ), P (TMRCA = t ) and P (T2c (avg) /TMRCA = x) for the same mean
number of progeny regardless of the offspring number distribution.
In particular, in the agreement with the O’Connell model, these
distributions are independent of the variance of the progeny count.
This fact is illustrated in Fig. 1 for P (T2c = t ). The comparison
of the shapes of the coalescence distributions obtained in different
stochastic models with the coalescence distribution conditional on
the expected size of the branching process P (T2c = t |E (Nt )) for a
Poisson offspring number distribution is given in Fig. 2(a).
More importantly, the distributions P (T2c = t ) obtained for any
offspring distribution are also visually identical to the O’Connell
limiting distribution for as few as 102 generations with O’Connell
parameter α = 10 (see Fig. 2(b) for the BF distribution). Specifi-
cally, O’Connell (1995) argued that any value between 10 and 14
is not contradicted by demographic evidence and has little effect
on estimates. We choose α = 10. Visual inspection of Fig. 2(b)
and a comparison of expectations presented in Table 1 prove that
the limiting O’Connell distribution P (T2c = t |OC) almost perfectly
approximates the distributions of T2c obtained from the full geneal-
ogy model for 102 generations. Therefore we can map the expec-
tation of TMRCA available directly from the full genealogy model
on a time scale of the O’Connell process, and can obtain the expec-
tation of the ratio TMRCA /T required in (3). Because of the asymp-
totic character, the result becomes valid for an arbitrary number of

Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
 ARTICLE IN PRESS
6 K.A. Cyran, M. Kimmel / Theoretical Population Biology ( ) –

Table 1 Table 2
Expectations of the ratio T2c /T ± SD in the O’Connell and the full genealogy models. Comparison of the expectations of T2c /T computed in the Wright–Fisher and the
coalescent models for different progeny number distributions.
Model E (T2c /T )
Progeny number distribution E (T2c /T |WF) E (T2c /T |CM)
O’Connell 0.8054 ± 0.1591
Full genealogy with BF progeny 0.8097 ± 0.1585 BF 0.7497 0.7585
Full genealogy with P progeny 0.8008 ± 0.1645 P 0.8005 0.8078
Full genealogy with LF progeny 0.8002 ± 0.1662 LF 0.8454 0.8550

a Table 3
Expectations of different ratios of the coalescence times and their standard
deviations computed in the full genealogy model for various distributions of the
number of progeny.
Parameter Binary fission Poisson Linear fractional

E (T2c /T ) 0.8097 ± 0.1585 0.8008 ± 0.1645 0.8002 ± 0.1662

E (T2c_avg /T ) 0.8097 ± 0.1057 0.8009 ± 0.1124 0.8001 ± 0.1150
E (TMRCA /T ) 0.9094 ± 0.0950 0.9032 ± 0.1011 0.9017 ± 0.1040
E (T2c_avg /TMRCA ) 0.9068 ± 0.0482 0.9027 ± 0.0532 0.9035 ± 0.0535

Table 4
Expected values of the time to the MRCA of modern humans computed in the
O’Connell model, the branching process full genealogy models, the Wright–Fisher
models, and the coalescent models.
b Model T̂MRCA(y) (years × 10−3 )

O’Connell limit 176

Full genealogy, binary fission 174
Full genealogy, Poisson 174
Full genealogy, linear fractional 174
Wright–Fisher, binary fission 189
Wright–Fisher, Poisson 178
Wright–Fisher, linear fractional 168
Coalescent, binary fission 187
Coalescent, Poisson 175
Coalescent, linear fractional 165

The expected values and the standard deviations of various

Fig. 3. Comparison of the distributions of T2c computed in the Wright–Fisher, the
coalescent and the O’Connell models assuming the binary fission (a), and Poisson
(b) progeny number distribution.
generations exceeding 102 . Therefore, even if it is not possible to
directly estimate E (TMRCA /T ) in a full genealogy model, we may in-
directly combine the results with the limiting O’Connell properties
and obtain E (TMRCA /T ) for the number of generations of the order
104 corresponding to the time elapsed from the death of the mtEve.
In our work we have also characterized the relationship of
the Wright–Fisher discrete model to the continuous coalescent
model in stochastic population histories approximated by a slightly
supercritical branching process. Both models considered deviate
from the O’Connell model for offspring distributions other than
Poisson (see Fig. 3(a) for the BF offspring number distribution).
Since the continuous coalescent model is equivalent to the
diffusion process limit, which in turn depends on the variance of
progeny number, this result can be explained by the variances of
binary fission and linear fractional distributions deviating from
the variance of Poisson distribution in opposite directions. There
is one more interesting fact which can be observed. Namely, for
times t close to T (corresponding to the beginning of branching
process) the continuous approximation assumed in coalescent
theory loses validity and the distribution differs more and more
from the Wright–Fisher distribution. This is finally reflected in the
atom of probability at t = T required for probabilities to sum to
one (see Fig. 3(b)). However, despite this visually striking feature,
the expectations of T2c |WF and T2c |CM remain very similar (see
Table 2).
ratios of coalescence times available directly only in the full
genealogy model are collected in Table 3.
Using the O’Connell model with the TMRCA moment computed
based on the full genealogy model, we can estimate the time to
the mtEve. The estimates of this time, assuming δ = 2.5 × 10−8
and davg = 0.003, for different population histories, are given in
Table 4.
Comparison of the numbers in the Table 4 with the expectation
based on the hypervariable regions equal to 163 × 103 years
with the 95% confidence interval [111 × 103 , 260 × 103 ]
as reported by Krings et al. (1999) shows that all stochastic
model predictions based on the recent complete mtDNA data fall
into the phylogenetically obtained interval, although particular
coalescence time distributions vary among the models considered.
Table 5 presents the data required to compute the expected value
of TMRCA and the 95% confidence interval in the full genealogy
model with the use of Eq. (2). As can be seen, we obtain the
expected value of TMRCA = 174 × 103 years with the 95% confidence
interval [96 × 103 , 449 × 103 ]. The confidence interval is computed
in a conservative way; i.e., to compute the lower bound of TMRCA
we used the lower bound of davg and the upper bound of δ and
T2c /TMRCA , whereas for the upper bound of TMRCA we used the upper
bound of davg and the lower bound of δ , and T2c /TMRCA , respectively.
To compute the lower bound of δ we used the lower bound of
the average mutation difference between modern humans and
Neanderthals dHN and the date of the MRCA of the two species of
389 × 103 years ago, while for the upper bound of δ we used the
upper bound of dHN and the date of the MRCA of modern humans
and Neanderthals 641 × 103 years ago (Briggs et al., 2009).

Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
 ARTICLE IN PRESS
K.A. Cyran, M. Kimmel / Theoretical Population Biology (
Table 5
Expected value and the 95% confidence interval of the estimates.
Parameter Lower confidence bound
davg 0.003
dHN 0.0122
TMRCA HN(y) 389 × 103
δ 2.04 × 10−8
T2c /TMRCA 0.6
TMRCA(y) 96 × 103
5. Discussion
Until the last decade, estimation of the divergence rate
in pre-modern and modern humans could rely only on hu-
man–chimpanzee divergence data. Methods used were based on
phylogenetic trees constructed either by maximum likelihood or
maximum parsimony and rooted using the chimpanzee as an out-
group. However, due to the relatively long time to this divergence,
all estimates of this time were very inaccurate, ranging from 4 to
9 million years. Consequently, the estimated divergence rate and
time to the MRCA of modern humans could not be accurate, with
expected values ranging from 200,000 years ago (Wilson and Cann,
1992; Vigilant et al., 1991) to 300,000 years ago (Hasegawa and
Horai, 1991). Additionally, many possible patterns of human pop-
ulation growth were assumed. The simplified exponential models
were often used, but also the logistic growth of human popula-
tion proved to be not inconsistent with the mtDNA variation data
(Polanski et al., 1998).
The majority of these estimates were in agreement with the
out-of-Africa scenario and in contradiction to the multiregional
theory of origin of modern humans, supported by some paleon-
tologists (Thorne and Wolpoff, 1992). These researchers claimed
that the time to the MRCA should be placed about a million years
ago or even earlier. It should be emphasized that the genetic data
did not necessarily contradict the multiregional theory, as shown
by O’Connell (1995). He inferred, using the branching process
model that the genetic diversity of modern humans was consistent
with estimates of the mtEve existing between 700 thousand and
1.5 million years ago. These estimates depended on an inaccurate
inference of the human–chimpanzee divergence time and on the
methods of the inference. To validate his conclusions, O’Connell
(1995) also indicated the weak points of the outgroup methods
when the genetic distance between the outgroup and the sample
was large. Since the application of different methods to the same
genetic data gave results differing by almost one order of magni-
tude, the multiregional hypothesis could not be rejected solely be-
cause it was in contradiction with the majority of genetic models,
while there were still models supporting it.
The situation changed after 1997 (Krings et al., 1997), when,
for the first time, the mtDNA from Neanderthals dated to live
until about 40,000 years ago (Schmitz et al., 2002) was sequenced.
However, fewer than 400 base pairs were sequenced; hence, any
estimates based on this data were not very reliable. The next
successful sequencings of Neanderthal mtDNA in 1999 (Krings
et al., 1999, 2000; Ovchinnikov et al., 2000; Krings et al., 2000)
confirmed the accuracy of the first experiment and radically
changed the estimates of the time to the most recent common
female ancestor of modern humans. Since it seems from the genetic
data (Krings et al., 1999; Green et al., 2008; Briggs et al., 2009)
that Neanderthals did not contribute mtDNA to the lineages of
presently living modern humans, the time of the mtEve should
be placed after the H. sapiens – H. neanderthalensis divergence.
Even if later studies (Serré et al., 2004; Cyran and Kimmel, 2005)
indicated that interbreeding between two human forms could not
be excluded, and moreover that there is an evidence of a small-
scale gene flow (Green et al., 2010), it remains true that the root of
Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
) – 7
Expected value Upper confidence bound
0.0039 0.0055
0.0127 0.0131
511 × 103 641 × 103
2.5 × 10−8 3.14 × 10−8
0.9 1
174 × 103 449 × 103
mtDNA of all currently living humans should be placed after that
of humans and Neanderthals. A slight admixture of at most 25%
(Serré et al., 2004) or 15% (Cyran and Kimmel, 2005) disappeared
as a result of the genetic drift. Therefore, even if the results of
the Neanderthal Genome Project suggest possible interbreeding
between the Neanderthals and the archaic Europeans yielding
about 3% admixture of the nuclear DNA (Plagnol and Wall, 2006;
Pennisi, 2006; Green et al., 2010), treating the Neanderthals as a
mtDNA outgroup is justified.
In the paper we compared the distributions of the time to
coalescence of a pair of chromosomes obtained by conceptually
different methods. In particular, we proved that a branching
process evolving for as little as 102 generations is approximated by
the O’Connell model with an accuracy of less than 2%. Moreover,
the result holds for three offspring number distributions, and due
to the asymptotic character of the O’Connell results, it also remains
true for the evolution of branching processes with an arbitrarily
large number of generations. Having this result, we were able
to obtain the estimate of the expected value of the ratio of the
coalescence times of two individuals and that of all individuals in
the population for generation 104 , even if it is infeasible to apply a
full genealogy model in this case.
Finally, we applied our approach to estimate the age of the
root of the mtDNA polymorphism of modern humans based on
the genetic material belonging to contemporary humans and
Neanderthal fossils, as reported by Krings et al. (1999), Green et al.
(2008) and Briggs et al. (2009). For all stochastic trajectories we
analyzed, the resulting time falls into a 95% confidence interval
of the estimate based on phylogenetic trees (Krings et al., 1999).
Moreover, the result depends mainly (in fact linearly) on the
assumed time to the MRCA of Neanderthals and modern humans
rather than the method that was applied. Therefore, we conclude
that the stochastic models based on branching processes provide
similar estimates to those obtained using phylogenetic analysis,
each supporting the other.
Hence, our results indicate that the estimates of the time
to coalescence in the Wright–Fisher and the coalescent models
with random population size are quite robust in terms of their
insensitivity to the model assumptions. They deviate by less
than 8% (see Table 4) from the O’Connell model predictions,
and the asymptotic O’Connell prediction differs from the actual
value computed in the full genealogy model by only 1.6%. Such
small differences are likely to be negligible compared to the
large range of confidence intervals obtained not only in pairwise
difference-based methods considered in the paper, but also in
the phylogenetic studies (Krings et al., 1999; Green et al., 2008;
Briggs et al., 2009). The greatest uncertainty of the mathematical
expectations is caused by the scaling factors such as the divergence
rate between species (the rate of the molecular clock) and not by
deviations from the particular assumption of the method used.
This validates both the Wright–Fisher and the coalescent models
with stochastic population sizes also for reproduction schemes not
following assumptions of these models. In particular, this provides
support to the results about inferring population trajectory from
the genetic diversity data, as reviewed in Wooding and Rogers
(2002); results which implicitly relied on the Wright–Fisher model
assumption but which remain valid for a much larger spectrum of
possible demographies.
 ARTICLE IN PRESS
8 K.A. Cyran, M. Kimmel / Theoretical Population Biology (
Acknowledgments
The authors would like to thank the anonymous reviewers for
their suggestions and comments. KC was supported by a Grant
from the Polish Ministry of Science and Higher Education No.
NN519 31 9035 from funds for supporting science in 2008–2010.
MK was supported by a CPRIT grant number RP101089 and by a
grant from the Polish Ministry of Science and Higher Education No.
N N519 579938. Assistance and collegiality of Dr. Jan Hewitt of Rice
University in the editing of the final version of the manuscript is
hereby gratefully acknowledged.
References
Bjorklund, M., 2003. Test for a population expansion after a drastic reduction in
population size using DNA sequence data. Heredity 91, 481–486.
Bobrowski, A., Kimmel, M., 2004. Asymptotic behavior of joint distributions of
characteristics of a pair of randomly chosen individuals in discrete-time
Fisher–Wright models with mutations and drift. Theor. Popul. Biol. 66, 355–367.
Briggs, A.W., Good, J.M., Green, R.E., Krause, J., Maricic, T., Stenzel, U.,
Lalueza-Fox, C., Rudan, P., Brajkovi, D., Kucan, Z., Gui, I., Schmitz, R.,
Doronichev, V.B., Golovanova, L.V., de la Rasilla, M., Fortea, J., Rosas, A.,
Pääbo, S., 2009. Targeted retrieval and analysis of five Neanderthal mtDNA
genomes. Science 325, 318–321.
Burbano, H.A., Hodges, E., Green, R.E., Briggs, A.W., Krause, J., Meyer, M., Good, J.M.,
Maricic, T., Johnson, Ph.L.F., Xuan, Z., Rooks, M., Bhattacharjee, A., Brizuela, L.,
Albert, F.W., de la Rasilla, M., Fortea, J., Rosas, A., Lachmann, M., Hannon, G.J.,
Pääbo, S., 2010. Targeted investigation of the Neanderthal genome by array-
based sequence capture. Science 328, 723–725.
Cyran, K.A., 2007. Simulating branching processed in the problem of Mitochondrial
Eve dating based on coalescent distributions. Int. J. Math. Comput. Simul. 1,
268–274.
Cyran, K.A., Kimmel, M., 2004. Robustness of the dating of the most recent common
female ancestor of modern humans. In: Proc. Tenth National Conference on
Application of Mathematics in Biology and Medicine, Swiȩty Krzyż, Poland.
pp. 19–24.
Cyran, K.A., Kimmel, M., 2005. Interactions of Neanderthals and modern humans:
what can be inferred from mitochondrial DNA? Math. Biosci. Eng. 2, 487–498.
Cyran, K.A., Myszor, D., 2008. New artificial neural network based test for the
detection of past population expansion using microsatellite loci. Int. J. Appl.
Math. Inform. 2, 1–9.
Green, R.E., Krause, J., Briggs, A.W., Maricic, T., Stenzel, U., Kircher, M.,
Patterson, N., Li, H., Zhai, W., Fritz, M.H.-Y., Hansen, N.F., Durand, E.Y.,
Malaspinas, A.-S., Jensen, J.D., Marques-Bonet, T., Alkan, C., Prüfer, K.,
Meyer, M., Burbano, H.A., Good, J.M., Schultz, R., Aximu-Petri, A., Butthof, A.,
Höber, B., Höffner, B., Siegemund, M., Weihmann, A., Nusbaum, Ch., Lander, E.S.,
Russ, C., Novod, N., Affourtit, J., Egholm, M., Verna, Ch., Rudan, P.,
Brajkovic, D., Kucan, Ž., Gušic, I., Doronichev, V.B., Golovanova, L.V.,
Lalueza-Fox, C., de la Rasilla, M., Fortea, J., Rosas, A., Schmitz, R.W.,
Johnson, Ph.L.F., Eichler, E.E., Falush, D., Birney, E., Mullikin, J.C.,
Slatkin, M., Nielsen, R., Kelso, J., Lachmann, M., Reich, D., Pääbo, S., 2010.
A draft sequence of the Neanderthal genome. Science 328, 710–721.
Green, R.E., Krause, J., Ptak, S.E., Briggs, A.W., Ronan, M.T., Simons, J.F., Du, L., Egholm,
M., Rothberg, J.M., Paunovic, M., Pääbo, S., 2006. Analysis of one million base
pairs of Neanderthal DNA. Nature 444, 330–336.
Green, R.E., Malaspinas, A.-S., Krause, J., Briggs, A.W., Johnson, Ph.L.F.,
Uhler, C., Meyer, M., Good, J.M., Maricic, T., Stenzel, U., Pruefer, K., Siebauer, M.,
Burbano, H.A., Ronan, M., Rothberg, J.M., Egholm, M., Rudan, P., Brajkovic, D.,
Kucan, Z., Gusic, I., Wikstrom, M., Laakkonen, L., Kelso, J., Slatkin, M., Pääbo, S.,
2008. A complete Neanderthal mitochondrial genome sequence determined by
high-throughput sequencing. Cell 134, 416–426.
Griffiths, R.C., Tavaré, S., 1995. Unrooted genealogical tree probabilities in the
infinitely-many-sites model. Math. Biosci. 127, 77–98.
Please cite this article in press as: Cyran, K.A., Kimmel, M., Alternatives to the Wright–Fisher model: The robustness of mitochondrial Eve dating. Theoretical Population
Biology (2010), doi:10.1016/j.tpb.2010.06.001
) –
Hasegawa, M., Horai, S., 1991. Time of the deepest root for polymorphism in human
mitochondrial DNA. J. Mol. Evol. 32, 37–42.
Jobling, M., 2001. In the name of the father: surnames and genetics. Trends Genet.
17, 353–357.
Kimmel, M., Axelrod, D.E., 2002. Branching Processes in Biology. Springer-Verlag,
New York.
Kimmel, M., Chakraborty, R., King, J., Bamshad, M., Watkins, W., Jorde, L., 1998.
Signatures of population expansion in microsatellite repeat data. Genetics 148,
1921–1930.
King, J.P., Kimmel, M., Chakraborty, R., 2000. A power analysis of microstallite-based
statistics for inferring past population growth. Mol. Biol. Evol. 17, 1859–1868.
Klebaner, F.C., Sagitov, S., 2002. The age of a Galton–Watson population with a
geometric offspring distribution. J. Appl. Probab. 39, 816–828.
Krings, M., Capelli, C., Tschentscher, F., Geisert, H., Meyer, S., von Haeseler,
A., Grossschmidt, K., Possnert, G., Paunovic, M., Pääbo, S., 2000. A view of
Neanderthal genetic diversity. Nat. Genet. 26, 144–146.
Krings, M., Geisert, H., Schmitz, R., Krainitzki, H., Pääbo, S., 1999. DNA sequence of
the mitochondrial hypervariable region II from the Neanderthal type specimen.
Proc. Natl. Acad. Sci. USA 96, 5581–5585.
Krings, M., Stone, A., Schmitz, R., Krainitzki, H., Stoneking, M., Pääbo, S., 1997.
Neanderthal DNA sequences and the origin of modern humans. Cell 90,
19–30.
Laan, M., Wiebe, V., Khusnutdinova, E., Remm, M., Pääbo, S., 2005. X -chromosome
as a marker for population history: linkage disequilibrium and haplotype study
in Euroasians populations. Eur. J. Hum. Genet. 13, 452–462.
Lambert, A., 2003. Coalescence times for the branching process. Adv. Appl. Probab.
35, 1071–1098.
Nagylaki, T., 1990. Models and approximations for random genetic drift. Theor.
Popul. Biol. 37, 192–212.
Noonan, J.P., Coop, G., Kudaravalli, S., Smith, D., Krause, J., Alessi, J., Chen, F.,
Platt, D., Pääbo, S., Pritchard, J.K., Rubin, E.M., 2006. Sequencing and analysis
of Neanderthal genomic DNA. Science 314, 1113–1118.
O’Connell, N., 1995. The genealogy of branching processes and the age of our most
recent common ancestor. Adv. Appl. Probab. 27, 418–442.
Ovchinnikov, I., Götherström, A., Romanova, G., Kharitonov, V., Lidén, K.,
Goodwin, W., 2000. Molecular analysis of Neanderthal DNA from the northern
Caucasus. Nature 404, 490–493.
Pennisi, E., 2007. No sex please, we’re Neanderthals. Science 318, 967.
Pennisi, E., 2006. The dawn of the stone age genomics. Science 314, 1068–1071.
Plagnol, V., Wall, J.D., 2006. Possible ancestral structure in human populations. PLoS
Genet. 2, 972–979.
Polanski, A., Kimmel, M., Chakraborty, R., 1998. Application of time-dependent
coalescence process for inferring the history of population size changes from
DNA sequence data. Proc. Natl. Acad. Sci. USA 95, 5456–5461.
Schmitz, R., Bonani, G., Smith, F.H., 2002. New research at the Neanderthal type site
in the Neander Valley of Germany. J. Hum. Evol. 42, A32.
Serré, D., Langaney, A., Chech, M., Teschler-Nicola, M., Paunovic, M., Mennecier, P.,
Hofreiter, M., Possnert, G., Pääbo, S., 2004. No evidence of Neanderthal mtDNA
contribution to early modern humans. PLOS Biol. 2, 313–317.
Thompson, R., Pritchard, J., Shen, P., Oefner, P., Feldman, M., 2000. Recent common
ancestry of human Y chromosomes: evidence from DNA sequence data. Proc.
Natl. Acad. Sci. USA 97, 7360–7365.
Thorne, A., Wolpoff, M.H., 1992. The multiregional evolution of humans. Scientific
American 266, 76–83.
Vigilant, L., Stoneking, M., Harpending, H., Hawkes, K., Wilson, A.C., 1991. African
populations and the evolution of human mitochondrial DNA. Science 253,
1503–1507.
Wilson, A.C., Cann, R.L., 1992. The recent African genesis of humans. Scientific
American 266, 68–73.
Wooding, S., Rogers, A., 2000. A pleistocence population X -plosion? Hum. Biol. 72,
693–695.
Wooding, S., Rogers, A., 2002. The matrix coalescence and an application to human
single-nucleotide polymorphisms. Genetics 161, 1641–1650.
Yu, N., Zhao, Z., Fu, Y., Sambuughin, N., Ramsay, M., Jenkins, T., Leskinen, E.,
Patthy, L., Jorde, L., Kuromori, T., Li, W., 2001. Global patterns of human DNA
sequence variation in a 10-kb region on chromosome 1. Mol. Biol. Evol. 18,
214–222.