82
D. Enard
1991). The aim of the test is to quantify the rate of adaptive substitutions with
a potential functional impact, like amino acid substitutions in a coding sequence
or substitutions affecting a regulatory segment. The test can be conducted from
scales ranging from a single gene to the entire genome. The idea is to compare
functional diversity and divergence with nonfunctional diversity and divergence
to measure an excess of functional divergence. The test has been most frequently
used to quantify the rate of amino acid changes in coding sequences (Bustamante
et al. 2005; Boyko et al. 2008), where it requires to know the numbers of nonsynonymous and synonymous polymorphic sites P n and P s and the number of
non-synonymous and synonymous divergent sites with a closely related species
D n and D s . Recurrent adaptive amino acid changes are expected to increase D n
while leaving P n unaffected. Indeed, adaptive mutations get fixed rapidly and do
not contribute to polymorphism. Recurrent adaptive amino acid changes thus make
D n higher than expected given P n . Variations in the mutation rate are then controlled
by comparing the ratios D n /D s and P n /P s instead of just comparing D n and P n . The
rate of adaptive amino acid changes can then be easily computed as a function of
D n , D s , P n , and P s . Note that the category “non-synonymous” can be replaced with
any other functional category and that “synonymous” can also be replaced by any
other neutral category.
The McDonald–Kreitman test has been adapted for quantifying adaptation
genome-wide (Boyko et al. 2008) or at single genes (Bustamante et al. 2005).
Although the approach is elegant and looks simple at first, it suffers from many
biases. The most important one is deleterious mutations (Eyre-Walker and Keightley
2009; Messer and Petrov 2013a). Recessive deleterious mutations can reach relatively high frequencies before they get eliminated from a population. This means
that they have a significant contribution to polymorphism, but no contribution to
divergence. In coding sequences, non-synonymous deleterious mutations inflate
P n while leaving D n unaffected, which biases estimates of the rate of adaptive
substitutions downward. This problem is especially severe after a long population bottleneck (Eyre-Walker 2002). Indeed, reduced population size means that
purifying selection is less efficient at weeding out deleterious mutations, which
increases P n . Fluctuations in selective constraint are also an issue, as D n represents
the integration of selective constraint over a long period of time while P n is affected
only by the recent level of constraint. Despite these limitations, the McDonald–
Kreitman approach has been remarkably useful to quantify rates of adaptation in
the human genome.
4.3
How to Choose a Specific Test of Selection?
The choice of a test of positive selection depends on a number of parameters:
– The type of selection. Different types of selection often require different tests,
although the same tests can sometimes be used to study both hitchhiking and
balancing selection. For example, SFS-based tests like Tajima’s D test (Tajima
D. Enard
1991). The aim of the test is to quantify the rate of adaptive substitutions with
a potential functional impact, like amino acid substitutions in a coding sequence
or substitutions affecting a regulatory segment. The test can be conducted from
scales ranging from a single gene to the entire genome. The idea is to compare
functional diversity and divergence with nonfunctional diversity and divergence
to measure an excess of functional divergence. The test has been most frequently
used to quantify the rate of amino acid changes in coding sequences (Bustamante
et al. 2005; Boyko et al. 2008), where it requires to know the numbers of nonsynonymous and synonymous polymorphic sites P n and P s and the number of
non-synonymous and synonymous divergent sites with a closely related species
D n and D s . Recurrent adaptive amino acid changes are expected to increase D n
while leaving P n unaffected. Indeed, adaptive mutations get fixed rapidly and do
not contribute to polymorphism. Recurrent adaptive amino acid changes thus make
D n higher than expected given P n . Variations in the mutation rate are then controlled
by comparing the ratios D n /D s and P n /P s instead of just comparing D n and P n . The
rate of adaptive amino acid changes can then be easily computed as a function of
D n , D s , P n , and P s . Note that the category “non-synonymous” can be replaced with
any other functional category and that “synonymous” can also be replaced by any
other neutral category.
The McDonald–Kreitman test has been adapted for quantifying adaptation
genome-wide (Boyko et al. 2008) or at single genes (Bustamante et al. 2005).
Although the approach is elegant and looks simple at first, it suffers from many
biases. The most important one is deleterious mutations (Eyre-Walker and Keightley
2009; Messer and Petrov 2013a). Recessive deleterious mutations can reach relatively high frequencies before they get eliminated from a population. This means
that they have a significant contribution to polymorphism, but no contribution to
divergence. In coding sequences, non-synonymous deleterious mutations inflate
P n while leaving D n unaffected, which biases estimates of the rate of adaptive
substitutions downward. This problem is especially severe after a long population bottleneck (Eyre-Walker 2002). Indeed, reduced population size means that
purifying selection is less efficient at weeding out deleterious mutations, which
increases P n . Fluctuations in selective constraint are also an issue, as D n represents
the integration of selective constraint over a long period of time while P n is affected
only by the recent level of constraint. Despite these limitations, the McDonald–
Kreitman approach has been remarkably useful to quantify rates of adaptation in
the human genome.
4.3
How to Choose a Specific Test of Selection?
The choice of a test of positive selection depends on a number of parameters:
– The type of selection. Different types of selection often require different tests,
although the same tests can sometimes be used to study both hitchhiking and
balancing selection. For example, SFS-based tests like Tajima’s D test (Tajima