9 Natural Selection, Genetic Variation, and Human Diversity
211
and are therefore outliers in the distribution of a statistic of interest (Akey et al.
2002).
Finally, human demographic changes have a significant impact on our ability
to correctly identify the signatures of selection in the first place. The statistical
power to detect selection depends on the interaction between local recombination
rate, the dominance and selection coefficients of the selected allele, and whether
the selected allele is new or selected from standing variation (Teshima et al. 2006).
Recent population bottlenecks reduce the power of haplotype homozygosity tests
to detect signatures of selection (Pickrell et al. 2009). Recent admixture, which
is an important consideration in modern human populations, decreases the power
of most neutrality test statistics but actually increases the power of Fay and Wu’s
H (Lohmueller et al. 2011b). Cryptic admixture can also obscure other signals of
selection (Lohmueller et al. 2011b). The unique demographic histories of different
populations also impart differing statistical power, meaning that we are better able
to detect selection in some populations than in others (Lohmueller et al. 2011b).
In addition to affecting statistical power, demography affects the false discovery
rate of tests for selection. Recessive selected alleles, alleles selected from standing
variation, and recent population bottlenecks all contribute to higher false discovery
rates (Teshima et al. 2006). New approaches are being developed to address these
issues, such as the composite of multiple signals (Grossman et al. 2010, 2013).
Likelihood approaches, machine learning, and approximate Bayesian computation
methods may also provide powerful ways to disentangle the effects of demography
and selection in the near future (Li et al. 2012). However, the severe tradeoff
between false-positive and false-negative rates in human populations means that
some selection events will always remain beyond our detection abilities (Teshima et
al. 2006; Li et al. 2012).
9.4
Empirical Studies of Positive Selection
9.4.1 Single Gene Studies
The earliest inferences of positive selection in humans focused on candidate gene
studies, where there was an a priori hypothesis that a particular gene may have
fitness consequences. Candidate gene studies were also the only practical way to
test hypotheses about selection given the limitations in DNA sequencing technology
at the time. Perhaps the most well-studied and understood example of positive
selection arising from candidate gene studies is that of LCT, which encodes for
the enzyme lactase and digests the sugar lactose (Fig. 9.2). In mammals, LCT is
expressed, and thus lactase is produced, during infancy and early childhood when
an individual is dependent on a milk diet, but the gene is turned off following
weaning (Harris and Meyer 2006). This pattern of LCT expression is also the
ancestral state in humans, but adaptive mutations have arisen in several pastoral
populations that allow lactase production to continue into and throughout adulthood.
Patterns of genetic variation around LCT show evidence for an incomplete or
211
and are therefore outliers in the distribution of a statistic of interest (Akey et al.
2002).
Finally, human demographic changes have a significant impact on our ability
to correctly identify the signatures of selection in the first place. The statistical
power to detect selection depends on the interaction between local recombination
rate, the dominance and selection coefficients of the selected allele, and whether
the selected allele is new or selected from standing variation (Teshima et al. 2006).
Recent population bottlenecks reduce the power of haplotype homozygosity tests
to detect signatures of selection (Pickrell et al. 2009). Recent admixture, which
is an important consideration in modern human populations, decreases the power
of most neutrality test statistics but actually increases the power of Fay and Wu’s
H (Lohmueller et al. 2011b). Cryptic admixture can also obscure other signals of
selection (Lohmueller et al. 2011b). The unique demographic histories of different
populations also impart differing statistical power, meaning that we are better able
to detect selection in some populations than in others (Lohmueller et al. 2011b).
In addition to affecting statistical power, demography affects the false discovery
rate of tests for selection. Recessive selected alleles, alleles selected from standing
variation, and recent population bottlenecks all contribute to higher false discovery
rates (Teshima et al. 2006). New approaches are being developed to address these
issues, such as the composite of multiple signals (Grossman et al. 2010, 2013).
Likelihood approaches, machine learning, and approximate Bayesian computation
methods may also provide powerful ways to disentangle the effects of demography
and selection in the near future (Li et al. 2012). However, the severe tradeoff
between false-positive and false-negative rates in human populations means that
some selection events will always remain beyond our detection abilities (Teshima et
al. 2006; Li et al. 2012).
9.4
Empirical Studies of Positive Selection
9.4.1 Single Gene Studies
The earliest inferences of positive selection in humans focused on candidate gene
studies, where there was an a priori hypothesis that a particular gene may have
fitness consequences. Candidate gene studies were also the only practical way to
test hypotheses about selection given the limitations in DNA sequencing technology
at the time. Perhaps the most well-studied and understood example of positive
selection arising from candidate gene studies is that of LCT, which encodes for
the enzyme lactase and digests the sugar lactose (Fig. 9.2). In mammals, LCT is
expressed, and thus lactase is produced, during infancy and early childhood when
an individual is dependent on a milk diet, but the gene is turned off following
weaning (Harris and Meyer 2006). This pattern of LCT expression is also the
ancestral state in humans, but adaptive mutations have arisen in several pastoral
populations that allow lactase production to continue into and throughout adulthood.
Patterns of genetic variation around LCT show evidence for an incomplete or