Accumulation of slightly deleterious mutations
in mitochondrial protein-coding genes of
large versus small mammals
Konstantin Popadin*†‡, Leonard V. Polishchuk†§, Leila Mamirova¶, Dmitry Knorre?, and Konstantin Gunbin**
Departments of *Genetics and§General Ecology, Biological Faculty of M.V. Lomonosov Moscow State University, Vorobyevy Gory 1-12, Moscow 119992,
Russia;¶Institute for Information Transmission Problems RAS, Bolshoi Karetny pereulok 19, Moscow 127994, Russia;?A.N. Belozersky Institute of
Physico-Chemical Biology, M.V. Lomonosov Moscow State University, Building A, Moscow 119899, Russia; and **Institute of Cytology
and Genetics SB RAS, Lavrentiev aven. 10, Novosibirsk 630090, Russia
Edited by Tomoko Ohta, National Institute of Genetics, Mishima, Japan, and approved June 19, 2007 (received for review February 13, 2007)
After the effective size of a population, Ne, declines, some slightly
deleterious amino acid replacements which were initially sup-
pressed by purifying selection become effectively neutral and can
reach fixation. Here we investigate this phenomenon for a set of
all 13 mitochondrial protein-coding genes from 110 mammalian
species. By using body mass as a proxy for Ne, we show that large
mammals (i.e., those with low Ne) as compared with small ones (in
our sample these are, on average, 369.5 kg and 275 g, respectively)
have a 43% higher rate of accumulation of nonsynonymous nu-
8–40% higher rate of accumulation of radical amino acid substi-
tutions relative to conservative substitutions, depending on the
type of amino acid classification. These higher rates result in a 6%
greater amino acid dissimilarity between modern species and their
most recent reconstructed ancestors in large versus small mam-
mals. Because nonsynonymous substitutions are likely to be more
harmful than synonymous substitutions, and radical amino acid
substitutions are likely to be more harmful than conservative ones,
our results suggest that large mammals experience less efficient
purifying selection than small mammals. Furthermore, because in
the course of mammalian evolution body size tends to increase
and, consequently, Ne tends to decline, evolution of mammals
toward large body size may involve accumulation of slightly
deleterious mutations in mitochondrial protein-coding genes,
which may contribute to decline or extinction of large mammals.
body mass ? effective population size ? substitution rates ?
purifying selection ? body size-dependent extinction
efficient s such that ?s? ? 1/Neare effectively neutral, in the sense
that their dynamics are affected mostly by random drift (1, 2).
Effectively neutral nucleotide substitutions, both slightly deleteri-
ous and slightly beneficial, play a major role in evolution at the
molecular level (3). Provided that the distribution of selection
effectively neutral must be higher in populations with smaller Ne.
fraction of slightly deleterious mutations can reach fixation.
Purifying selection affects nonsynonymous substitutions much
stronger than synonymous substitutions (4–6). Indeed, when the
ratio of the rate of nonsynonymous (amino acid changing)
substitutions over the rate of synonymous (silent) substitutions,
Ka/Ks, is ?1, it is indicative of purifying selection on nonsynony-
mous substitutions and reflects its strength: the closer the Ka/Ks
to 1, the weaker is the purifying selection (3). Another measure
of the strength of purifying selection is the ratio of the rate of
radical (presumably more harmful) over the rate of conservative
(less harmful) amino acid substitutions, Kr/Kc (7, 8). Further-
more, because species with reduced population sizes experience
atural selection is more efficient in large populations. In a
population of effective size Ne, mutations with selection co-
more deleterious mutations and thus to show a greater amino
acid dissimilarity to its most recent ancestor. One measure of
such dissimilarity is Grantham’s distance (9), which is often used
in the context of protein comparisons (10).
In this study, we analyze six molecular traits: Ka/Ks, four Kr/Kc
ratios under various types of amino acid classification (based on
volume, charge, polarity, and both polarity and volume), and
Grantham’s distance, all of which characterize, in a variety of
ways, the efficiency of purifying selection. Our hypothesis is that
at least some of these traits would increase with decreasing Ne,
as a response to less efficient purifying selection in small
populations. To test this hypothesis, we take advantage of an
extensive set of complete mammalian mitochondrial genomes
not available in the previous studies (8, 11–15). The choice of
mitochondrial genome is motivated by its having a low effective
population size due to uniparental inheritance, effective hap-
loidy, and the absence of recombination. In addition, the prob-
ability of fixation of a slightly deleterious mutation with a given
selection coefficient is so structured that it is particularly sen-
sitive to changes in Newhen Neis low (16). These features make
the mitochondrial genome particularly suitable to study the rate
of deleterious-mutation accumulation in relation to Ne(17, 18),
although recurrent events of positive selection in mitochondrial
DNA, as may be the case in invertebrates (19) but less likely in
mammals (20), may hamper this dependence.
Because for most mammalian species no direct estimates of Ne
are presently available, we use body mass W as a proxy for Ne,
based on the nearly universal inverse relationship between body
mass and population size (21–24). Thus, this study aims to
determine the rate of accumulation of slightly deleterious mu-
tations (expressed through Ka/Ksand Kr/Kc) and the resulting
amino acid dissimilarity between modern species and their most
recent reconstructed ancestors (measured as Grantham’s dis-
tance) in large versus small-bodied mammals and, by implica-
tion, in small versus large populations.
Average Values of Nonsynonymous to Synonymous Nucleotide Sub-
stitutions, Ka/Ks, and of Radical to Conservative Amino Acid Substi-
tutions, Kr/Kc.All species-specific values of Ka/Ksare well below 1
(typically 0.03–0.06, Table 1). These values indicate that most
and K.P., L.V.P., and L.M. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
†To whom correspondence may be addressed. E-mail: email@example.com or
‡Present address: Institute for Information Transmission Problems RAS, Bolshoi Karetny
pereulok 19, Moscow 127994, Russia.
This article contains supporting information online at www.pnas.org/cgi/content/full/
© 2007 by The National Academy of Sciences of the USA
August 14, 2007 ?
vol. 104 ?
nonsynonymous substitutions (at least 94–97%) produced by
mutation process are subject to purifying selection and thus are
deleterious. The majority of our Kr/Kcvalues are also ?1, and
none is considerably larger than 1 (Table 1), which indicates
stronger purifying selection against radical substitutions.
Comparison of Large Versus Small Mammals. We divided the whole
sample into two groups, 55 species each, with small mammals
having ln W ? 9.04 and large mammals having ln W ? 9.04,
where 9.04 (or ?8,400 g) is the median of logetransformed body
mass W (in grams).
Three substitution rate ratios (Ka/Ks, polarity-based Kr/Kc, and
mammals, polarity–volume-based Kr/Kcis marginally significantly
smaller (P ? 0.07), and charge-based Kr/Kcshows no significant
difference at a 5% significance level (t test, Fig. 1 A–E).
Ordinary Linear Models. In terms of ordinary linear regressions, all
of the rates of nonsynonymous over synonymous, and of radical
a 5% level) positive relationship with body mass (Table 1), albeit
with considerable scatter (r2? 0.16, 0.28, 0.32, 0.06, and 0.07 for
Fig. 2 A–E, respectively). Although such a scatter implies the
influence of multiple factors on the traits under study, the increas-
ing trend with increasing body mass is clearly corroborated.
Data Nonindependence and Mixed-Effects Models.Thelinearmixed-
effects models, which account for nonindependence of the
individual-species data due to the effect of shared ancestry,
differ only slightly from the ordinary linear models in terms of
slopes and intercepts, except for the slope of charge-based Kr/Kc
(Table 1). However, mixed-effects models all show larger P,
indicating that taking species nonindependence into account
makes the relationships of molecular traits on body mass gen-
erally less reliable than implied by ordinary regressions (Table
1). This result is probably because of a lower number of the
effective degrees of freedom left after accounting for noninde-
pendence. A bigger discrepancy in terms of statistical signifi-
cance occurs for charge-based Kr/Kcand for polarity–volume-
based Kr/Kc, which yield significant, at a 5% level, relationships
with body mass based on ordinary models but either nonsignif-
icant (charge-based Kr/Kc) or marginally significant (polarity–
volume-based Kr/Kc) relationships with body mass based on
significant according to either type of models.
The Relative Efficiency of Purifying Selection. As a measure of the
relative efficiency of purifying selection in large versus small
animals. The logebody mass for a typical small mammal is taken
to be 5.62 (which corresponds to 275 g) and that for a typical
large mammal to be 12.82 (which corresponds to 369,500 g),
Table 1. Averages, standard errors, and parameters of ordinary linear and mixed-effects regressions on loge
body mass along with associated P values for each trait under study
Trait Average (SE)
Ordinary linear models Mixed-effects models
Calculations are based on 110 mammalian species. SE, standard error.
body mass W (in grams). (A) Ka/Ks, P ? 0.001. (B) Polarity-based Kr/Kc, P ? 0.001. (C) Volume-based Kr/Kc, P ? 0.001. (D) Polarity–volume-based Kr/Kc, P ? 0.069.
(E) Charge-based Kr/Kc, P ? 0.20. (F) Grantham distance, P ? 0.001 (t test). Small squares represent mean; rectangles are ?1 SE; and whiskers are 0.95 confidence
Popadin et al. PNAS ?
August 14, 2007 ?
vol. 104 ?
no. 33 ?
these values being the averages of logebody mass over 55 smaller
and 55 larger species in our sample. Ka/Ksis found to be ?43%
higher in large than in small mammals, whereas Kr/Kcis 8–40%
higher, depending on the type of amino acid classification used
(Table 2). These estimates are very similar for ordinary linear
and mixed-effects models (Table 2).
The Amino Acid Dissimilarity Between Modern Species and Their Most
Recent Reconstructed Ancestors in Large Versus Small Mammals.
Because, as shown above, purifying selection acts on most amino
acid changing substitutions versus silent substitutions and is less
efficient in large than in small species, we expect that large
mammals, as compared with small ones, accumulate more
harmful mutations and thus evolve farther away from their most
recent ancestors. The Grantham distance supports this predic-
tion: it is larger in large animals in terms of averages (Fig. 1F),
ordinary linear models (Fig. 2F), and mixed-effects models
(Table 1). Quantitatively, the amino acid dissimilarity between
modern and ancestral sequences is 6% greater in large than in
small mammals (Table 2).
Most of our analyses show that mitochondrial protein-coding
genes of large mammals have a higher rate of accumulation of
nonsynonymous substitutions relative to synonymous substitu-
tions and, among nonsynonymous substitutions, a higher rate of
accumulation of radical substitutions relative to conservative
substitutions. Because nonsynonymous substitutions must be
more harmful than synonymous substitutions, and radical amino
acid substitutions must be more harmful than conservative ones,
this implies that large mammals experience less efficient puri-
fying selection than small mammals. Consequently, large mam-
mals appear to accept more harmful mutations, leading to a
higher level of amino acid dissimilarity between modern species
and their most recent reconstructed ancestors, as observed in
Charge-based Kr/Kcis the only molecular trait that does not
significantly differ in large versus small mammals both in terms
of averages and mixed-effects models (although it does differ on
the basis of ordinary linear models.) To understand this finding,
we note that among four Kr/Kcratios, the charge-based Kr/Kchas
the lowest average (Table 1), which indicates a strong purifying
selection against changes in amino acid charge for the proteins
under study. The selection coefficient s ? 0 is then so large by
absolute value that not only small mammals (with high Ne) but
also large mammals (with low Ne) experience strong purifying
selection, s ? ?1/Ne,large mammals? ?1/Ne,small mammals, and get rid
of such mutations. In contrast, purifying selection against
changes in amino acid polarity and volume is relatively weak
(corresponding Kr/Kc values are the largest, Table 1), and
selection coefficient is intermediate in the sense that it falls in an
effectively neutral zone for large mammals but beyond it for
small mammals, i.e., ?1/Ne,large mammals? s ? ?1/Ne,small mammals.
Consequently, radical substitutions in regard to these traits are
fixed in large species as if conservative (note that for large
precisely at 1; Fig. 1 B and C), but tend to be eliminated in small
species. The slightly deleterious mutations with such an inter-
mediate s will accumulate predominantly in large mammals.
Only slightly deleterious mutations that are fixed in popula-
tions are considered in the previous paragraph. However, in this
paper we are dealing not only with mutations that are fixed in
populations and thus associated with interspecific divergence,
but also with mutations that are not fixed and thus associated
Kr/Kc. (D) Polarity–volume-based Kr/Kc. (E) Charge-based Kr/Kc. (F) Grantham distance. For regression parameters see Table 1.
The ordinary linear regressions of molecular traits on logebody mass W for 110 mammalian species. (A) Ka/Ks. (B) Polarity-based Kr/Kc. (C) Volume-based
Table 2. Ratios of nonsynonymous to synonymous (Ka/Ks) and of radical to conservative (Kr/Kc) substitution
rates, and amino acid dissimilarity between modern species and their most recent reconstructed ancestors
measured in terms of Grantham’s distance for large versus small mammals
The values refer to typical large (369.5 kg, left number) versus small (275 g, right number) mammals in our sample. The values are
calculated from ordinary linear models (OLM) and mixed-effects models (MEM).
www.pnas.org?cgi?doi?10.1073?pnas.0701256104Popadin et al.
with the amount of intraspecific polymorphism, because our
Ka/Ksand other molecular traits represent external-branch esti-
mates to which both components of genetic variation contribute.
Ka/Ksrepresenting internal-branch estimates to which only fixed
mutations would contribute are of little interest here because
body mass on internal nodes is unavailable. The problem with
these two classes of mutations is that polymorphism often
dominates fixed mutations (25–28) but it is only the latter that
Our data do not allow us to distinguish between these classes, so
the question arises whether our Ka/Ksand other molecular traits
reflect the behavior of either, or perhaps both, of them. Clearly,
both classes are closely interrelated because currently fixed
mutations were part of the past polymorphism, and the current
polymorphism may harbor mutations that will become fixed in
the future. Moreover, it seems likely that both classes will change
in the same direction: as long as body mass increases, the
effective population size tends to decrease, which results in an
elevated fixation probability and in enhanced polymorphism due
to less efficient purifying selection against deleterious mutations
in small populations (2). Because less efficient selection drives
both components of genetic variation, they go in the same
direction, so that fixed mutations will most likely show the same
positive trend with increasing body mass as Ka/Ks does. This
reasoning suggests an accumulation of slightly deleterious mu-
tations in large mammals, which might incur a threat to species
Indirectly, we can assess the impact of polymorphic mutations
on Ka/Ksby computing the product S ? s Nefrom our estimates
of Ka/Ks. This assessment can be done through the well known
equation for the probability of fixation of a new mutation (2, 16).
Assuming that silent sites evolve neutrally, the ratio of the
fixation probability for a mutation with selection coefficient s ?
0 to the fixation probability for a neutral mutation with s ? 0 can
be equated with Ka/Ksratio (30)
Ka/Ks? 2S/?1 ? exp? ? 2S??.
By using function uniroot in R package (31), we computed S
from Eq. 1 for every of 110 mammalian species. All of the values
of S turn out to be negative, which once again (see Results) is
indicative of the preponderance of purifying selection. They lie
in a rather narrow range from ?2.97 to ?1.69, with mean S being
?2.38 (SE ? 0.03, n ? 110). Clearly, our estimates of ?S? ? 1
reflect the effect of polymorphic mutations on Ka/Ks, with some
of them being deleterious with ?s? ? 1/Ne. This effect, however,
does not seem to be too strong because the estimates of ?S? are
closer to the lower boundary of the range for ?S?, 1 ? ?S? ? 10,
characteristic for polymorphic mutations (2, 25, 32). This allows
for the effect of fixed mutations on Ka/Ks to appear, and is
another reason why Ka/Ks would follow the behavior of fixed
mutations with regard to body mass.
Mitochondrial genes are highly predisposed to mutation ac-
cumulation. Because mitochondria have a low effective popu-
lation size, molecular evolution of mitochondrial DNA genes
should be associated with a high rate of accumulation of slightly
deleterious mutations as compared with nuclear genes (17). The
absence of recombination can additionally increase the rate of
degradation of mitochondrial DNA due to Muller’s ratchet (33).
A number of comparative-species studies have corroborated
these theoretical expectations (34, 35). It is less clear whether a
higher rate of deterioration of mitochondrial genes can actually
lead to a faster rate of extinction (36), but recent research (37)
lends support to this possibility.
The contrast between small and large modern mammals
examined in this study can be turned to the past and thus viewed
from a long-term evolutionary perspective by taking into ac-
count the general evolutionary trend toward larger body size,
known as Cope’s rule. Although there are some exceptions, this
trend is common among mammals (38, 39); suffice it to mention,
by way of example, that their ancestral forms were small,
shrew-sized creatures (40), whereas modern mammals include
such giants as blue whales and African elephants. We suggest, on
the basis of present findings, that evolution of mammals toward
large body size is accompanied by increasing width of the
selective sieve (manifested by increased Ka/Ksand other molec-
36), which leads to the deterioration of mitochondrially encoded
proteins, and may contribute to decline or extinction of large
species. Two additional lines of evidence are in favor of this
hypothesis. First, many theoretical studies (41–43) conclude that
small populations (that is, those typical of large-bodied species)
can go extinct because of accumulation of slightly deleterious
mutations and resulting mutational meltdown (42, 44). Second,
large animals (45), including mammals (46–48), are more prone
to extinction. However, additional analyses are needed to eval-
uate directly the role of less efficient purifying selection in
elevated extinction risks in large mammals.
Materials and Methods
Source Data and Initial Treatment. The fully sequenced mitochon-
drial genomes from all of the 138 mammalian species available
at early summer 2005 were downloaded from the National
Center for Biotechnology Information database (http://www.
downloaded June 5, 2005). All 13 protein-coding mitochondrial
genes were extracted from each genome and translated into
amino acid sequences. Alignments of the amino acid sequences
for each gene were performed by using Clustal X (49) with
default settings and then reverse transcribed to get nucleotide
alignments. The aligned 13 amino acids and 13 corresponding
nucleotide sequences without stop codons were concatenated
into a single amino acid and a single nucleotide sequence for
each species. The concatenated amino acid sequences were used
to reconstruct a mammalian phylogenetic tree by using PHYML
(50). The concatenated nucleotide sequences were used to
analyze the pattern of nucleotide substitutions.
Ratio of Nonsynonymous to Synonymous Nucleotide Substitutions
(Ka/Ks). The ratio of the rates of nonsynonymous over synony-
mous substitutions (Ka/Ks) was estimated for each branch of the
mammalian phylogenetic tree by using the program codeml from
PAML package (51). Because an independent estimate of Ka/Ks
for each branch of the whole tree would have taken too much
time, the tree was divided into 11 monophyletic (52) subtrees,
each of which was treated separately. 16 species with an uncer-
tain position on the whole tree (i.e., with a low bootstrap value)
were omitted from further analyses. The 122 remaining species
were distributed over 11 groups (subtrees) as follows: Afrotheria
(13 species), Artiodactyla (9 species), Carnivora (12 species),
Cetacea (22 species), Chiroptera (8 species), core insectivores
Lagomorpha (4 species), Metatheria (19 species), Perissodactyla
(5 species), and Primates and Dermoptera (17 species). For each
subtree, two evolutionary models were implemented: model 0
free Ka/Ks estimated separately for each branch (the models’
designation as in PAML). Model 1 was run twice: under
transition/transversion parameter [k in Kimura’s notation (53)]
set at 3 across all subtrees (8), and under transition/transversion
parameter estimated for each subtree individually. Only Ka/Ks
values associated with external branches were used to describe
relationships with species body mass.
Ratio of Radical to Conservative Amino Acid Substitutions (Kr/Kc).The
ratio of the rates of radical over conservative substitutions
Popadin et al. PNAS ?
August 14, 2007 ?
vol. 104 ?
no. 33 ?
(Kr/Kc) was estimated by comparison of the nucleotide sequences
of modern animals with the nucleotide sequences of their most
recent reconstructed ancestors through Zhang’s algorithm (7).
The ancestral nucleotide sequences were reconstructed by using
the method implemented by Yang et al. (54) in PAML. Because
Krand Kcvalues were small enough (?0.3), the Jukes–Cantor
formula was used to correct for multiple hits; that is, our Kr/Kc
ratio is identical to dR/dCratio in Zhang’s notation (7).
The 20 amino acids were classified into groups in four
different ways according to their volume (55), charge, polarity,
and both polarity and volume (7). Amino acid substitutions
within groups (i.e., when ancestral and modern amino acids in
homologous sites belong to the same group) were regarded as
conservative, and those between groups as radical.
Comparison of Evolutionary Models. Our analyses of alternative
evolutionary models (free, branch-specific Ka/Ksvalues versus
constant Ka/Ksacross all branches of a subtree and the transition/
transversion parameter k specific to each subtree versus k fixed
at 3 across all subtrees) provide reason to compute Ka/Ksand
PAML model 1) and subtree-specific estimates of k [see A Check
for Alternative Parameter Estimates in supporting information
(SI) Text for more detail]. The Ka/Ksand Kr/Kcobtained in this
manner are used throughout this paper.
In total, 110 of 122 mammal species were each characterized
by Ka/Ks and Kr/Kc values. Twelve species were deleted from
further analyses because of too few (?20) nonsynonymous
substitutions per 13 protein-coding genes, having occurred since
the most recent ancestor.
Average Grantham Distance. To measure amino acid dissimilarity,
we computed an average physicochemical distance between
modern species and their most recent reconstructed ancestors.
The distance between each ancestral (if substituted) and de-
scendant amino acid was taken from Grantham’s matrix (9), and
averaged over all pairs of substitutions for a given external
Body Mass. The body mass of adult individuals for the majority of
analyzed species was obtained from a database compiled by
species on different continents being reported, body mass was
averaged. A few missing data were taken from other sources. All
of the data are available upon request.
Statistical Treatment. A number of statistical analyses were per-
formed. First, we compared the average values of the investi-
gated molecular traits which belong to small versus large mam-
mals (with body mass below and above the median, respectively).
Then we performed the ordinary linear regressions of molecular
traits on body mass. However, the characters we are dealing with
may not be independent among species due to the effect of
phylogenetic inertia, which might compromise comparative-
species analyses (57, 58). The standard way around this problem
is a method of phylogenetically independent contrasts (59). It is
assumes that increases and decreases in the character of interest
are equally likely, so that the character’s average does not change
over time (60). Changes in body mass may not always agree with
the Brownian motion model, because according to Cope’s rule,
which is justified for mammals (38, 39), body mass tends to
increase over evolutionary time, and thus exhibits a largely
directional change. There are more complex evolutionary mod-
els, which allow for a directional shift in character evolution and
(61). A bigger problem is that molecular traits such as branch-
specific Ka/Ksare already contrasts (because they are based on
a sequence difference between two nodes of the branch),
whereas phenotypic traits such as body mass are of course not
contrasts and possess some phylogenetic inertia. Furthermore, if
we were to try to associate Ka/Kswith contrasts based on body
mass we would find them to be incongruent: the former are due
to differences between the present and the past (present and past
nucleotide sequences), whereas the latter are due to differences
pertinent entirely to the present (that is, to modern species).
These considerations seem to preclude a simultaneous use of
molecular and body-mass traits in an analysis of phylogenetically
To control for the effect of data nonindependence, we carried
out a regression analysis using the linear mixed-effects model,
which does not require an a priori model of character evolution,
nor does it rely on contrasts. Instead, it explicitly takes into
account a hierarchical (nested), and therefore correlated, struc-
ture of comparative-species data (ref. 62 and J. Fox, personal
communication). This method has been widely used in social and
medical sciences (see ref. 63 for review) and recently in com-
parative-species research (64, 65). Here, it was implemented
using function lme in nlme package (66) of R language (31). Our
data are nested into the following levels: species within genera
within families within orders. We used a mixed-effect model
which included both random intercept and random slope.
Check for Saturation Effect. An additional check for saturation
effect was performed. After omitting species which tipped the
relatively long branches of ?1 nucleotide substitution per codon,
and examining the remaining data set (n ? 89), the pattern of
the relationships between molecular traits and body mass remains
largely unchanged (see A Check for Saturation Effect in SI Text).
We thank Alexey Kondrashov for comments and many important
suggestions, and Vladimir Aleshin, Shamil Sunyaev, Egor Bazykin,
Alexey Ghilarov, Fydor Kondrashov, Dmitrii Filatov, and Nikolai Muge
for helpful discussions. Special thanks go to Jianzhi Zhang for discussion
of Kr/Kc estimation, Tomoko Ohta for discussion of the effectively
and ideas. This research was supported by the Presidium RAS Program
‘‘Origin and Evolution of Biosphere,’’ by Russian Foundation for Basic
Research Grants 07-04-00521 and 07-04-00343, and by the Russian
Academy of Sciences ‘‘Molecular and Cellular Biology’’ Program. L. M.
was partially supported by Howard Hughes Medical Institute Grant
55005610 and INTAS Grant 05-1000008-8028.
1. Wright S (1931) Genetics 16:97–159.
2. Kimura M (1983) The Neutral Theory of Molecular Evolution (Cambridge Univ
Press, Cambridge, U.K.).
3. Ohta T (1992) Annu Rev Ecol Syst 23:263–286.
4. Mamirova L, Popadin K, Gelfand MS (2007) BMC Evol Biol 7:17.
5. Paland S, Lynch M (2006) Science 311:990–992.
6. Bachtrog D (2005) Genome Res 15:1393–1401.
7. Zhang J (2000) J Mol Evol 50:56–68.
8. Eyre-Walker A, Keightley PD, Smith NGS, Gaffney D (2002) Mol Biol Evol
9. Grantham R (1974) Science 185:862–864.
10. Yampolsky LY, Kondrashov FA, Kondrashov AS (2006) Hum Mol Genet
11. Li W-H, Tanimura M, Sharp PM (1987) J Mol Evol 25:330–342.
12. Ohta T (1993) Proc Natl Acad Sci USA 90:4548–4551.
13. Ohta T (1995) J Mol Evol 40:56–63.
14. Weinreich DM (2001) J Mol Evol 52:40–50.
15. Woolfit M, Bromham L (2005) Proc R Soc London Ser B 272:2277–2282.
16. Kimura M (1962) Genetics 47:713–719.
17. Cortopassi GA (2002) Free Radical Biol Med 33:605–610.
18. Ballard JWO, Whitlock MC (2004) Mol Ecol 13:729–744.
19. Bazin E, Gle ´min S, Galtier N (2006) Science 312:570–572.
20. Mulligan CJ, Kitchen A, Miyamoto MM (2006) Science 314:1390.
21. Ghilarov MS (1944) Rep Acad Sci USSR 43:283–285.
22. Damuth J (1981) Nature 290:699–700.
23. Damuth J (1987) Biol J Linn Soc 31:193–246.
www.pnas.org?cgi?doi?10.1073?pnas.0701256104Popadin et al.
24. Peters RH (1983) The Ecological Implications of Body Size (Cambridge Univ Download full-text
Press, Cambridge, U.K.).
25. Nachman MW (1998) Genetica 102/103:61–69.
26. Nielsen R, Weinreich DM (1999) Genetics 153:497–506.
27. Hasegawa M, Cao Y, Yang Z (1998) Mol Biol Evol 15:1499–1505.
28. Gerber AS, Loggins R, Kumar S, Dowling TE (2001) Annu Rev Genet
29. Petrov DA (2001) Trends Genet 17:23–28.
30. Nielsen R, Yang Z (2003) Mol Biol Evol 20:1231–1239.
31. R Development Core Team (2005) R: A Language and Environment for Statistical
Computing (R Foundation for Statistical Computing, Vienna, Austria).
32. Eyre-Walker A, Woolfit M, Phelps T (2006) Genetics 173:891–900.
33. Hoekstra R (2000) Hum Reprod 15(Suppl 2):102–111.
34. Lynch M (1996) Evolution (Lawrence, Kans.) 13:209–220.
35. Lynch M (1997) Evolution (Lawrence, Kans.) 14:914–925.
36. Lynch M, Blanchard JL (1998) Genetica 102/103:29–39.
37. Loewe L (2006) Genet Res 87:133–159.
38. Alroy J (1998) Science 280:731–734.
39. Van Valkenburgh B, Wang X, Damuth J (2004) Science 306:101–104.
40. Carroll RL (1987) Vertebrate Paleontology and Evolution (Freeman, New York).
41. Lande R (1994) Evolution (Lawrence, Kans.) 48:1460–1469.
42. Lynch M, Burger R, Butcher D, Gabriel W (1993) J Hered 84:339–344.
43. Kondrashov A (1995) J Theor Biol 175:583–594.
44. Gabriel W, Lynch M, Bu ¨rger R (1993) Evolution (Lawrence, Kans.) 47:1744–1757.
45. McKinney ML (1997) Annu Rev Ecol Syst 28:495–516.
46. Purvis A, Gittleman JL, Cowlishaw G, Mace GM (2000) Proc R Soc London
Ser B 267:1947–1952.
47. Polishchuk LV (2002) Science 297:1123.
48. Cardillo M, Mace GM, Jones KE, Bielby J, Bininda-Emonds ORP, Sechrest W,
Orme CDL, Purvis A (2005) Science 309:1239–1241.
49. Thompson JD, Gibson TJ, Plewniak F, Jeanmougin F, Higgins DG (1997)
Nucleic Acids Res 25:4876–4882.
50. Guinden S, Gascuel O (2003) Syst Biol 52:696–704.
51. Yang ZH (1997) Comput Appl Biosci 13:555–556.
52. Murphy JW, Eizirik E, Johnson WE, Zhang YP, Ryder OA, O’Brien SJ (2001)
53. Kimura M (1980) J Mol Evol 16:111–120.
54. Yang Z, Goldman N, Friday AE (1995) Syst Biol 44:384–399.
55. Taylor WR (1986) J Theor Biol 119:205–218.
56. Smith FA, Lyons SK, Ernest SKM, Jones KE, Kauffman DM, Dayan T,
Marquet PA, Brown JH, Haskell JP (2003) Ecology 84:3403.
(Oxford Univ Press, Oxford).
58. Fisher DO, Owens IPF (2004) Trends Ecol Evol 19:391–398.
59. Felsenstein J (1985) Am Nat 125:1–15.
60. Ives AR, Zhu J (2006) Ecol Appl 16:20–32.
61. Diaz-Uriarte R, Garland T (1996) Syst Biol 45:27–47.
63. Hedeker D (2005) in Encyclopedia of Statistics in Behavioral Science, eds Everitt
B, Howell D (Wiley, Chichester, U.K.), pp 729–738.
64. Blackburn TM, Duncan RP (2001) Nature 414:195–197.
65. Brook BW, Traill LW, Bradshaw CJA (2006) Ecol Lett 9:375–382.
66. Pinheiro J, Bates D, DebRoy S, Sarkar D (2005) nlme: Linear and nonlinear
mixed effects models (R Foundation for Statistical Computing, Vienna, Aus-
tria), R package version 3.1-60.
Popadin et al. PNAS ?
August 14, 2007 ?
vol. 104 ?
no. 33 ?