Gene 261 (2000) 3±9

www.elsevier.com/locate/gene

DNA evolution under weak selection

Hidenori Tachida*
Department of Biology, Graduate School of Sciences, Kyushu University, Ropponmatsu, Fukuoka, 810-8560 Japan
Received 24 May 2000; received in revised form 21 August 2000; accepted 25 September 2000
Received by G. Bernardi

Abstract
Some DNA data show patterns of variation not expected under the neutral theory. Here, the independent multicodon (IMC) model, a nearly
neutral mutation model assuming no interaction among codons, was studied when population size changes using computer simulation.
Patterns of variation expected under the model were investigated using statistics for the neutrality tests. The average dispersion index is more
than one when population size changes slowly but it never becomes large. The diversity at linked silent site decreases when the strength of
selection is intermediate and the reduction is larger when population size changes slowly. Tajima's (1989. Genetics 123, 585±595) D is
generally negative. Rejections by the Tajima's test occur more frequently if population size changes quickly but the effect of selection is
confounded with the size change itself in this case. If we apply the test of McDonald and Kreitman (1991. Nature 351, 652±654), the rejection
is always in the direction of excess replacement polymorphisms. The rejection probability decreases as the rate of population size changes
decreases. These results show that the predictions of the IMC model are consistent with the pattern observed in mitochondrial DNA data but
not consistent with some data of nuclear DNA. Interaction among codons or variable selection would be necessary to explain such cases.
q 2000 Elsevier Science B.V. All rights reserved.
Keywords: Nearly neutral model; Neutrality tests; Random genetic drift; Bottleneck effect

1. Introduction
Recent advances in molecular techniques allow us to
obtain DNA sequences fairly easily and now data on varia-
tion of DNA sequences coding for protein genes within and
between species have been accumulating. Some of those
data show patterns of variation not expected under the
neutral theory which states that the main cause of evolu-
tionary change at the molecular level is random ®xation of
selectively neutral or very nearly neutral mutations rather
than Darwinian selection (Kimura, 1968, 1983). First, a
large dispersion index that measures the extent of variation
in substitution rate was observed in mammals (Gillespie,
1989; Ohta, 1995). Secondly, low levels of diversity at silent
sites were observed in regions of low recombination in
Drosophila (Begun and Aquadro, 1992), Lycopersicon
(Stephan and Langley, 1998) and human (Nachman et al.,
1998). Finally, applications of neutrality tests such as those
of Tajima (1989), Fu and Li (1993), HKA (Hudson et al.,
1987) and McDonald and Kreitman (1991) revealed patterns
not compatible with predictions of the strict neutral model
Abbreviations: the IMC model, the independent multicodon model
* Corresponding author. Tel.: 181-92-726-4577; fax: 181-92-726-4644.
E-mail address: htachscb@mbox.nc.kyushu-u.ac.jp (H. Tachida).
(see Moriyama and Powell, 1996). These data suggest some
action of selection on protein coding genes. But we still do
not know what type of selection is acting on these
sequences.
One candidate model involving selection is the nearly
neutral mutation model originally proposed by Ohta
(1973, 1992). In the neutral model, selection coef®cients
of mutants are assumed to have a bimodal distribution.
More concretely, let Ne and s be effective size and selection
coef®cient of a mutant, respectively. In the neutral model,
2Nes is either much larger or smaller than one so that only
very neutral mutations with 2N e s ,, 1 contribute to poly-
morphisms in populations and substitutions among species.
On the other hand, in the nearly neutral model, selection
coef®cients are distributed continuously and some muta-
tions have s with 2Nes of the order of one. Thus, random
genetic drift and selection both affect fates of mutant genes.
In the present paper, I report a simulation study on the
sequence pattern expected under the independent multico-
don (IMC) model (Tachida, 2000) when population size
changes. The model is a variant of the nearly neutral muta-
tion model. With constant population size, it was shown that
the dispersion index was close to one, Tajima's D was
minus and excess replacement polymorphisms compared

0378-1119/00/$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.
PII: S 0378-111 9(00)00475-3
4 H. Tachida / Gene 261 (2000) 3±9
to the neutral expectation were detected by the McDonald
and Kreitman (1991) test (Tachida, 2000). However,
because the behavior of the nearly neutral mutation models
depends on population size, we need to explore the pattern
expected under the model when population size changes.
2. Model and simulation
We need to specify two aspects, relationship between
gene and ®tness and population structures, to simulate
protein evolution.
2.1. Gene model
The gene model studied is one of the simplest mimick-
ing a protein coding gene. A gene consists of n codons
each comprising three sites. Each site allows four alleles
and the mutation rate to one of the other three alleles is u/3.
The ®rst and second sites determine the amino acid of the
codon so that there are 16 kinds of amino acids. They are
called replacement sites. The third site is a silent site. The
®tness of an amino acid speci®ed by a codon is determined
by drawing a random number from a normal distribution
with variance s 2 at the beginning of a simulation. The
®tness of a gene is determined multiplicatively by the
®tness of amino acids at respective codons. Because ®tness
is determined multiplicatively, there is no interaction
among codons. Hence, the model is called the independent
multicodon (IMC) model (Tachida, 2000) and one of the
simplest nearly neutral models imitating a protein coding
gene. Another extreme model is the house-of-cards (®xed)
model in which all codons interact (Ohta and Tachida,
1990). Between these extreme models, various intermedi-
ate ones can be conceived (Kauffman, 1993; Ohta, 1997)
but as a ®rst step, we stick to the simplest model. Both
advantageous and deleterious mutations occur but most
mutations become deleterious after population evolves to
high ®tness. We assume that no recombination occurs
within a gene.
2.2. Population structure
A random mating haploid population with discrete
generations (the Wright±Fisher model) is assumed but its
size ¯uctuates stochastically in the simulation. For simpli-
city the size takes only two values N1 and N2 with N 1 $ N2 .
The waiting times for size change are T1 for N1 and T2 for N2
and they are assumed to be distributed geometrically with
means t1 and t2, respectively. In the simulation, exponential
random numbers truncated to integers were used as approx-
imations to geometrical random numbers for generating Ti.
So there are four population parameters, N1, N2, t1 and t2.
In order to compare effects of changing population size, I
de®ne the effective population size based on the average
homozygosity, f, under the in®nite allele model (see
Kimura, 1983) at a site when all sites are neutral. The homo-
zygosity in the equilibrium state is computed to be (M.
Iizuka, H. Tachida and H. Matsuda, personal communica-
tion)
1
f ˆ
…t1 1 t2 †…1 2 A1 A2 †
  …1†
…1 2 A1 †…t1 1 t2 A2 † …1 2 A2 †…t2 1 t1 A1 †
1
1 1 2N1 u 1 1 2N2 u
where
1
Ai ˆ  
t
…1 1 2Ni u† i 1 1
Ni
This formula can be obtained by ®rst noting that the
probability of population size being Ni at a random time is
ti/(t1 1 t2) and then computing the expected homozygosity
applying Eq. (6) of Nei et al. (1975) for each case. As t1 and
t2 get larger compared to N1 and N2, the homozygosity
approaches
     
t1 1 t2 1
1
t1 1 t2 1 1 2N1 u t1 1 t2 1 1 2N2 u
as it should. For a given mutation rate, the effective size, Ne,
is de®ned as the size of a population that has constant size
and the same homozygosity as the one with changing size.
The effective size can be computed from Eq. (1). The reason
why the effective size is de®ned in this way is that we can
observe only the current DNA data and have little informa-
tion on past population size. So our strategy is to carry out
simulation with parameter values that lead to similar
nucleotide diversity as the observed values.
2.3. Simulation
Computer simulation was carried out following the model
speci®ed above. After about 5/u generations to avoid effects
of the initial state, the population is split into two with the
same size. These two populations evolve independently
after the split. Their size changes are also independent.
Genes were sampled from two populations with the interval
of 0.05/u generations after the split of the population and
various test statistics for the neutrality were computed. The
statistics examined are, the dispersion index (I), nucleotide
diversity (p ), Tajima's D and statistics for the McDonald
and Kreitman test. In all simulations, N e ˆ 500 and
u ˆ 10 25 so that 2N e u ˆ 0:01. Only cases with N1 ˆ
10N2 and N1 ˆ 40N2 were examined. For N1 ˆ 10N2 ,
effects of the rate of size change were investigated. Intensity
of selection is measured by a parameter a ˆ 2N e s and it is
changed from 0.2 to 50. Population parameters used in the
simulations are listed in Table 1. The number of codons, n,
is 400 so that there are 1200 sites in a gene. The number of
replications was 1000 for Quick, Medium and Slow cases
and 500 for the Large case. For comparison, the result for
the case with constant population size (Tachida, 2000) will
 H. Tachida / Gene 261 (2000) 3±9 5

Table 1
Parameter values for the population structure used in the simulation

Simulation N1 N2 t1 t2 u

Constant 500 500 0.00001

N1 ˆ 10N2
Quick 1000 100 485 100 ±
Medium 1000 100 836 250 ±
Slow 1000 100 3289 2500 ±
(Slow limit) 1000 100 1.25t t ±
N1 ˆ 40N2
Large 4000 100 1428 2500 ±

be also shown. More details of the simulation except for

change of population size is described in that paper.
3. Results and discussions
3.1. Dispersion index
Let X(t) be number of substitutions accumulating in t
generations in a lineage after the split of the populations.
The dispersion index, I(t), is de®ned as
Var‰X…t†Š
I…t† ˆ …
E‰X…t†Š
where E[X(t)] and Var[X(t)] denote the mean and variance,
respectively. Under the neutral assumption with a constant
mutation rate, the substitution process is expected to be
Poisson and thus I(t) is one. Therefore, the dispersion
index measures the extent of variability of the substitution
rate compared to the neutral expectation. This index was
estimated by taking a gene sample each from the population
at the 0.1/uth generation before the population split as an
outgroup and two populations after the split. The number of
substitutions was estimated by Jukes and Cantor (1969) 's
method, and their sample mean and variance were computed
for each replication. The ratio of the variance to mean was
Fig. 1. The dispersion index at replacement sites measured at 0.5/u genera-
tion after the population split. For sets (Constant, Quick, Medium, Slow and
Large) of population parameters used in the simulation, see Table 1. The
number of codons is n ˆ 400. a ˆ 2N e s is changed. The same applies to
other ®gures.
Fig. 2. The dispersion index at replacement sites observed at different times
after the split of the population. The parameter set used is Slow.
used to estimate the dispersion index. Because this method
of estimation is known to introduce biases, the correction
suggested by Bulmer (1989) was made.
The dispersion index estimated from gene samples taken
at the 0.5/uth generation after the population split is shown
in Fig. 1. In Constant, Quick and Medium cases, it is close to
one though it increases a little as a increases. When the size
2†
change is slow, the dispersion index becomes larger than
one and up to around two for intermediate a (a < 10).
However, the dispersion index in the Large case is not as
large as that in Slow case even though t2's are the same in
the two cases.
Since the dispersion index is a function of t, the number of
generations after the population split, its time dependency
was also investigated for the Slow case (Fig. 2). The disper-
sion index monotonically increases from one as t increases.
In mammalian and other data used for estimating the disper-
sion index, the number of synonymous substitutions per site
between two sequences compared was on the average less
than one (see Fig. 1 of Ohta, 1995). Thus, the dispersion
index under the present model is expected to be less than
that at the 0.5/uth generation in such cases.
The reason why the dispersion index becomes large in the
nearly neutral mutation model when population size change
is as follows: When population size is large, all populations
have high average ®tness and the substitution rate is very
low. When population size becomes smaller, the average
®tness goes down only in some populations and the substi-
tution rate becomes close to the neutral rate. In these popu-
lations, a burst of adaptive substitutions occur when
population size again becomes large. Since the substitution
rate stays low in other populations that did not experience
®tness decrease when population size was small, differentia-
tion in terms of substitution rate results among populations
and the dispersion index becomes higher. The critical factor
is what proportion of populations experience a decrease of
the average ®tness when population size becomes small. For
Quick and Medium cases, the duration, t2, of the time when
population size is small seems not enough to make the
6 H. Tachida / Gene 261 (2000) 3±9
Fig. 3. Nucleotide diversity at replacement sites.
average ®tness of the population lower. In the Large case, t2
is the same as in the Slow case but t1 seems not long enough
for populations to achieve high average ®tness.
Gillespie (1989) and Ohta (1995) estimated the dispersion
index using protein coding genes of mammals and their
estimates were more than ®ve. Such a large dispersion
index was never observed under the present IMC model.
However, a large dispersion index was observed when
population size was changed under the house-of-cards
model in which all codons were assumed to interact
(Araki and Tachida, 1997). Although more data are neces-
sary to estimate the dispersion index in mammals, the
current data suggest some type of interaction among codons
if we stick to models involving only constant selection. In
Drosophila, the dispersion index is estimated to be 1.6
(Zeng et al., 1998) and it is within the range observed
under the current model.
3.2. Nucleotide diversity
The nucleotide diversity is the average heterozygosity per
site. This quantity was measured for replacement and silent
sites separately. Fig. 3 shows the nucleotide diversity at
replacement sites. As a ˆ 2N e s increases, the nucleotide
diversity decreases. For a , 5, the reduction is largest in
the Large case. As the rate of size change increases, the
reduction decreases but this difference disappears for a . 5.
The reduction of the nucleotide diversity at silent sites
has a different pattern (Fig. 4). As a increases, the nucleo-
tide diversity ®rst decreases and takes the minimum value
at around a ˆ 10. Then it start to increase. Thus, the reduc-
tion is largest when the intensity of selection measured by
a ˆ 2N e s is intermediate. The largest reduction is
observed in the Large case and it is about 50% when
a ˆ 10. The reduction is larger when population size
changes slowly. Reduction of the nucleotide diversity at
silent sites in regions of low recombination has been
observed in Drosophila (Begun and Aquadro, 1992), Lyco-
persicon (Stephan and Langley, 1998) and human X chro-
mosome (Nachman et al., 1998). Strong adaptive selection
(selective sweep, see Kaplan et al., 1989) and negative
selection (background selection, see Charlesworth et al.,
1993) on liked sites have been proposed to explain this
phenomenon. In the former case, ®xations of adaptive
mutations sweep out variation at linked neutral sites. In
the latter case, deleterious mutations reduce the effective
size for linked neutral sites by making chromosomes carry-
ing them less likely to be transmitted to future generations.
In either cases, the region affected by a selected mutation is
larger when recombination rate is low. Weak selection
such as that studied here can also explain the reduction
in regions of low recombination (see also Przeworski et
al., 1999). As shown in Fig. 4, the effect depends on popu-
lation size. This feature distinguishes weak selection from
strong selection whose effects depend only on the mutation
rate, selection coef®cient and recombination rate.
Finally, we brie¯y discuss why the largest reduction is
observed when a is intermediate. In the present model, both
deleterious and advantageous mutations occur. Therefore,
both selective sweep and background selection contribute to
the reduction observed here though relative magnitudes of
contributions from the two causes are not known. In selec-
tive sweep, the amount of reduction increases as the ®xation
rate of advantageous mutations increases or the ®xation time
becomes shorter (Kaplan et al., 1989). Although the ®xation
time becomes shorter as a becomes larger, the ®xation rate
drops rapidly especially for larger a . This would be one
reason why the reduction is maximized at an intermediate
a . In background selection, the amount of reduction
decreases as the selection becomes stronger and the equili-
brium mutant frequency decreases when selection is very
strong (Charlesworth et al., 1993). When selection is very
weak, the distinction between mutant and wild type alleles
becomes vague and mutant lineages persists longer contri-
buting to the population for longer generations. In the limit
of a ˆ 0, the reduction is zero. Thus, again the reduction
would be maximized at an intermediate a . Future theoreti-
cal studies are necessary to quantify the reduction of diver-
sity at linked sites under weak selection.
Fig. 4. Nucleotide diversity at silent sites.
 H. Tachida / Gene 261 (2000) 3±9
Fig. 5. Expected values of Tajima's D at replacement sites. The sample size
is m ˆ 50.
3.3. Tajima' D
Tajima (1989) proposed a test statistics, D, de®ned by
! change but other models of selection also result in negative
k 2 S m =am X
m21
D ˆ p ; am ˆ 1=i …
Var‰k 2 Sm =am Š iˆ1
where k and Sm are the average pairwise difference and the
number of segregating sites, respectively, when m genes are
sampled from a population. The denominator is estimated
from Sm. If variation is neutral, population size is constant
and population is in equilibrium, the expectations of k and
Sm/am are both 2Neu and thus the expectation of the numera-
tor is zero. Therefore, D can be used to test the neutrality of
observed variation.
The expected values of D were estimated by sampling 50
genes and they are shown in Fig. 5. Generally E[D] is nega-
tive and its absolute value increases as a increases. Note
that E[D] is negative even for small a in cases Quick,
Medium, and Large. D is expected to be negative or positive
when population is expanding or diminishing, respectively,
even under the neutrality assumption (Tajima, 1993). In the
Fig. 6. The rejection probability by Tajima's test at the 5% level as func-
tions of a ˆ 2N e s. The sample size is m ˆ 50.
7
three cases mentioned above, the proportion of time spent in
the expanding phase is longer than those in the diminishing
and constant phases and this is why E[D] is negative when
selection is weak. The rejection probability of the neutrality
at the 5% level was also examined and the result is shown in
Fig. 6. The rejection probability is generally low for a # 2
except for the case Large. But it increases as a increases.
The rejection probability is high when population size
changes quickly and selection is strong. However, the
increase of the rejection probability is not only due to selec-
tion but also confounded by the size change itself as
mentioned above. Thus, it is necessary to separate effects
of selection and size change by, for example, looking at the
statistics at silent sites.
Tajima's D is generally negative in mitochondrial DNA
data and the neutrality is rejected in some cases (Rand et al.,
1994; Nachman et al., 1996). At nuclear loci in Drosophila,
the average values are negative but cases of the rejection are
rare (Moriyama and Powell, 1996). Those data can be
explained by the IMC model by adjusting the rate of size
D (e.g. the SAS-CFF model with B # 2, see Gillespie,
3†
1997).
3.4. McDonald and Kreitman test
The McDonald and Kreitman test (1991) examines the
ratios of replacement to silent differences within and
between species. Under the neutral assumption, expecta-
tions of these two ratios are the same. Thus, we can test
the neutrality by performing a test of independence for a
2 £ 2 table based on two categories, whether differences are
silent or replacement and whether differences are within
species (polymorphic) or between species (®xed).
This test was applied to gene samples from two popula-
tions after the split and the probability of rejection at the 5%
level is shown in Fig. 7. The sample size is 20 for each
populations and the time of sampling was at 0.1/u genera-
Fig. 7. The rejection probability by the McDonald and Kreitman (1991) test
at the 5% level as functions of a ˆ 2N e s. Twenty samples (m ˆ 20) were
taken from each population at 0.1/u generations after the split of the two
populations.
8 H. Tachida / Gene 261 (2000) 3±9
Fig. 8. The distribution of log10 (z ) for Constant, Quick, Medium and Slow
cases with a ˆ 20:0. For comparison, the Constant case with a ˆ 0:0 is
also shown. Twenty samples (m ˆ 20) were taken from each population at
0.1/u generations after the split of the two populations.
tions after the split. The McDonald and Kreitman test can
detect the present type of selection very well if population
size is constant. However, if population size changes, the
rejection probability decreases, especially when it changes
slowly.
In order to see how the rejection occurs, we de®ne a
statistics z . Let N( fr), N( fs), N(pr) and N(ps) be numbers
of ®xed-replacement, ®xed-silent, polymorphic-replace-
ment and polymorphic-synonymous differences, respec-
tively. We de®ne a ratio z as
N… fr†=N… fs†
zˆ
N…pr†=N…ps†
This ratio is expected to be approximately one under the
neutrality. If z , 1, there are excess replacement poly-
morphisms. If z . 1, there are excess replacement ®xations.
The distribution of log10(z ) when the rejection probability is
highest (a ˆ 20:0) is shown in Fig. 8. For comparison, that
in the Constant case with a ˆ 20:0 is also shown. With the
current type of selection, the distribution is shifted toward
negative and wider. Since log10z is negative, the rejection is
always in the direction of excess replacement polymorph-
isms. As the size change becomes slower, the distribution is
shifted toward positive but z rarely becomes more than one
even in the case Slow.
This pattern of excess replacement polymorphisms
appears in mitochondrial data but is rarely observed in
nuclear data. For example, in mitochondrial DNA, log10z
is 20.46 in human-chimpanzee (Nachman et al., 1996),
20.99 in mice (Nachman et al., 1994) and 20.36 (Rand
et al., 1994) and 20.69 (Ballard and Kreitman, 1994) in
Drosophila but it is 0.94 in Adh of Drosophila (McDonald
and Kreitman, 1991). Thus, the IMC model can explain
results of the McDonald and Kreitman test for mitochon-
drial DNA data but can not explain those seen in nuclear
data.
4. Conclusion
In the present paper, the IMC model, assuming no inter-
action among codons, was studied introducing changes of
population size. The IMC model can account for patterns
observed in mitochondrial DNA data but can not explain
those in nuclear data, especially concerning magnitudes of
the dispersion index and the direction of rejection in the
McDonald and Kreitman test. If we introduce interaction
among sites, the dispersion index become large in a certain
parameter range as was shown by a study on the in®nite
house-of-cards model (Araki and Tachida, 1997). However,
we need to examine other statistics also and this may be left
to future studies on other nearly neutral mutation models
involving interaction among codons (Ohta, 1997). If all
constant selection models fail to explain the data, then we
need to invoke non-constant selection models such as those
studied by Gillespie (1991) for at least some variation.
Acknowledgements
This work was supported in part by a grant from Program
for Promotion of Basic Research Activities for Innovative
Biosciences (PROBRAIN) and a grant from Uehara
Memorial Foundation to H.T.
References
Araki, H., Tachida, H., 1997. Bottleneck effect on evolutionary rate in the
nearly neutral mutation model. Genetics 147, 907±914.
Ballard, J.W.O., Kreitman, M., 1994. Unraveling selection in the mitochon-
drial genome of Drosophila. Genetics 138, 757±772.
Begun, D.J., Aquadro, C.F., 1992. Levels of naturally occurring DNA
polymorphism correlate with recombination rates in D. melanogaster.
Nature 356, 519±520.
Bulmer, M., 1989. Estimating the variability of substitution rates. Genetics
123, 615±619.
Charlesworth, B., Morgan, M.T., Charlesworth, D., 1993. The effects of
deleterious mutations on neutral molecular variation. Genetics 134,
1289±1303.
Fu, Y.-X., Li, W.-H., 1993. Statistical tests of neutrality of mutations.
Genetics 133, 693±709.
Gillespie, J.H., 1989. Lineage effects and the index of dispersion of mole-
cular evolution. Mol. Biol. Evol. 6, 636±647.
Gillespie, J.H., 1991. The Causes of Molecular Evolution. Oxford Univer-
sity Press, New York.
Gillespie, J.H., 1997. Junk ain't junk does: neutral alleles in a selected
context. Gene 205, 291±299.
Hudson, R.R., Kreitman, M., Aguade, M., 1987. A test of neutral molecular
evolution based on nucleotide data. Genetics 116, 153±159.
Jukes, T.H., Cantor, C.R., 1969. Evolution of protein molecules. In: Munro,
H.N. (Ed.). Mammalian Protein Metabolism, Vol. 3. Academic Press,
New York, NY, pp. 21±132.
Kaplan, N.L., Hudson, R.R., Langley, C.H., 1989. The `hitchhiking effect'
revisited. Genetics 123, 887±899.
Kauffman, S.A., 1993. The Origins of Order. Oxford University Press,
Oxford.
Kimura, M., 1968. Evolutionary rate at the molecular level. Nature 217,
624±626.
 H. Tachida / Gene 261 (2000) 3±9
Kimura, M., 1983. The Neutral Theory of Molecular Evolution. Cambridge
University Press, Cambridge.
McDonald, J.H., Kreitman, M., 1991. Adaptive protein evolution at the Adh
locus in Drosophila. Nature 351, 652±654.
Moriyama, E.N., Powell, J.R., 1996. Intraspeci®c nuclear DNA variation in
Drosophila. Mol. Biol. Evol. 13, 261±277.
Nachman, M.W., Boyer, S.N., Aquadro, C.F., 1994. Nonneutral evolution
at the mitochondrial NADH dehydrogenase subunit 3 gene in mice.
Proc. Natl. Acad. Sci. USA 91, 6364±6368.
Nachman, M.W., Brown, W.M., Stoneking, M., Aquadro, C.F., 1996.
Nonneutral mitochondrial DNA variation in humans and chimpanzees.
Genetics 142, 953±963.
Nachman, M.W., Bauer, V.L., Crowell, S.L., Aquadro, C.F., 1998. DNA
variability and recombination rates at X-linked loci in humans. Genetics
150, 1133±1141.
Nei, M., Maruyama, T., Chakraborty, R., 1975. The bottleneck effect and
genetic variability in populations. Evolution 29, 1±10.
Ohta, T., 1973. Slightly deleterious mutant substitutions in evolution.
Nature 246, 96±98.
Ohta, T., 1992. The nearly neutral theory of molecular evolution. Ann. Rev.
Syst. Ecol. 23, 263±286.
9
Ohta, T., 1995. Synonymous and nonsynonymous substitutions in mamma-
lian genes and the nearly neutral theory. J. Mol. Evol. 40, 56±63.
Ohta, T., 1997. The meaning of near-neutrality at coding and non-coding
regions. Gene 205, 261±267.
Ohta, T., Tachida, H., 1990. Theoretical study of near neutrality I. Hetero-
zygosity and rate of mutant substitution. Genetics 126, 219±229.
Przeworski, M., Charlesworth, B., Wall, J.D., 1999. Genealogies and weak
purifying selection. Mol. Biol. Evol. 16, 246±252.
Rand, D.M., Dorfsman, M., Kan, L.M., 1994. Neutral and nonneutral
evolution of Drosophila mitochondrial DNA. Genetics 138, 741±756.
Stephan, W., Langley, C.H., 1998. DNA polymorphism in Lycopersicon
and crossing-over per physical length. Genetics 150, 1585±1593.
Tachida, H., 2000. Molecular evolution in a multisite nearly neutral muta-
tion model. J. Mol. Evol. 50, 69±81.
Tajima, F., 1989. Statistical method for testing the neutral mutation hypoth-
esis. Genetics 123, 585±595.
Tajima, F., 1993. Statistical analysis of DNA polymorphism. Jpn. J. Genet.
8, 567±595.
Zeng, L.-W., Comeron, J.M., Chen, B., Kreitman, M., 1998. The molecular
clock revisited: the rate of synonymous vs. replacement change in
Drosophila. Genetica 102/103, 369±382.