42
M. Slatkin
Recombination between the A and B loci will reduce that excess. Simple theory
shows that, if the selection coefficient in favor of A is s, then substantial LD will
be created by hitchhiking at neutral loci for which c < s. That is, neutral loci that
are very closely linked to the selected locus will remain in substantial LD with the
advantageous allele, while more distant neutral loci will not.
These theoretical results have useful practical applications. If there is substantial
LD in a region surrounding a functionally important allele, it is likely that the allele
has increased in frequency recently because of positive selection. For example, the
A– allele of the G6PD gene in a west African population has a frequency of 11%.
Loci as far away as 700 kb are in significant LD with the A– allele, a distance much
larger than the normal scale of LD in the human genome (Saunders et al. 2005).
Data of this type not only indicate that the A– allele increased because of positive
selection but also make it possible to infer that the selection coefficient in favor of
A– was at least 0.05 and that it arose by mutation between 3000 and 6000 years ago
(Slatkin 2008).
2.10 Population Subdivision
Population subdivision creates LD when there are local differences in allele
frequencies. We can see why by considering a simple example. Suppose that two
populations are fixed for different alleles at each of two loci, population 1 is fixed
for A and B, while population 2 is fixed for a and b. In this case, every individual in
both populations is doubly homozygous, either AABB or aabb. Next, suppose that
a researcher who is concerned with LD at these two loci samples individuals from
both populations. If the researcher does not realize that there are in fact two distinct
populations, individuals from both would be combined into a single sample. The
resulting sample would be a mixture of AABB or aabb individuals. In this sample,
only AB and ab haplotypes would be present, so there is apparently perfect LD
between these two loci (D = 1). Yet, that conclusion is obviously an artifact of
mixing individuals from two populations with quite different allele frequencies.
Although this example is an extreme case that can be understood without doing
any calculations, the conclusion is quite general. If allele frequencies at two loci
differ at all between two or more populations and if samples from those two
populations are combined, there will in general more LD in the mixture than in the
separate populations (Mitton et al. 1973; Nei and Li 1973). This effect is called the
“two-locus Wahlund effect” because of its similarity to the classic Wahlund effect,
which is the decrease in heterozygosity in a mixture of two or more populations. In
the simple example, there are no heterozygous individuals at either locus, which is
an extreme case of the Wahlund effect.
It is usually possible to distinguish the two-locus Wahlund effect from selection
as a cause of LD because the Wahlund effect affects all pairs of loci at which
allele frequencies differ among subpopulations, while selection will probably affect
only one genomic region. Still, the two-locus Wahlund effect is important for the
design and interpretations of genome-wide association studies (GWAS). GWAS
M. Slatkin
Recombination between the A and B loci will reduce that excess. Simple theory
shows that, if the selection coefficient in favor of A is s, then substantial LD will
be created by hitchhiking at neutral loci for which c < s. That is, neutral loci that
are very closely linked to the selected locus will remain in substantial LD with the
advantageous allele, while more distant neutral loci will not.
These theoretical results have useful practical applications. If there is substantial
LD in a region surrounding a functionally important allele, it is likely that the allele
has increased in frequency recently because of positive selection. For example, the
A– allele of the G6PD gene in a west African population has a frequency of 11%.
Loci as far away as 700 kb are in significant LD with the A– allele, a distance much
larger than the normal scale of LD in the human genome (Saunders et al. 2005).
Data of this type not only indicate that the A– allele increased because of positive
selection but also make it possible to infer that the selection coefficient in favor of
A– was at least 0.05 and that it arose by mutation between 3000 and 6000 years ago
(Slatkin 2008).
2.10 Population Subdivision
Population subdivision creates LD when there are local differences in allele
frequencies. We can see why by considering a simple example. Suppose that two
populations are fixed for different alleles at each of two loci, population 1 is fixed
for A and B, while population 2 is fixed for a and b. In this case, every individual in
both populations is doubly homozygous, either AABB or aabb. Next, suppose that
a researcher who is concerned with LD at these two loci samples individuals from
both populations. If the researcher does not realize that there are in fact two distinct
populations, individuals from both would be combined into a single sample. The
resulting sample would be a mixture of AABB or aabb individuals. In this sample,
only AB and ab haplotypes would be present, so there is apparently perfect LD
between these two loci (D = 1). Yet, that conclusion is obviously an artifact of
mixing individuals from two populations with quite different allele frequencies.
Although this example is an extreme case that can be understood without doing
any calculations, the conclusion is quite general. If allele frequencies at two loci
differ at all between two or more populations and if samples from those two
populations are combined, there will in general more LD in the mixture than in the
separate populations (Mitton et al. 1973; Nei and Li 1973). This effect is called the
“two-locus Wahlund effect” because of its similarity to the classic Wahlund effect,
which is the decrease in heterozygosity in a mixture of two or more populations. In
the simple example, there are no heterozygous individuals at either locus, which is
an extreme case of the Wahlund effect.
It is usually possible to distinguish the two-locus Wahlund effect from selection
as a cause of LD because the Wahlund effect affects all pairs of loci at which
allele frequencies differ among subpopulations, while selection will probably affect
only one genomic region. Still, the two-locus Wahlund effect is important for the
design and interpretations of genome-wide association studies (GWAS). GWAS