Detecting Natural Selection in Genomic Data

GE47CH05-Sabeti ARI 29 October 2013 12:12

Detecting Natural Selection

in Genomic Data

Joseph J. Vitti,

1,2

Sharon R. Grossman,

2,3,4

and Pardis C. Sabeti

1,2

Department of Organismic and Evolutionary Biology, Harvard University, Cambridge,

Massachusetts 02138; email: [email protected], [email protected]

Broad Institute of MIT and Harvard, Cambridge, Massachusetts 02142

Department of Systems Biology, Harvard Medical School, Boston, Massachusetts 02115

Department of Biology, Massachusetts Institute of Technology, Cambridge,

Massachusetts 02139

Annu. Rev. Genet. 2013. 47:97–120

The Annual Review of Genetics is online at

genet.annualreviews.org

This article’s doi:

10.1146/annurev-genet-111212-133526

 2013 by Annual Reviews.

Keywords

population genetics, adaptation, selective sweeps, genome scans,

evolutionary genomics

Abstract

The past ﬁfty years have seen the development and application of nu-

merous statistical methods to identify genomic regions t hat appear to

be shaped by natural selection. These methods have been used to in-

vestigate the macro- and microevolution of a broad range of organisms,

including humans. Here, we provide a comprehensive outline of these

methods, explaining their conceptual motivations and statistical inter-

pretations. We highlight areas of recent and future development in

evolutionary genomics methods and discuss ongoing challenges for re-

searchers employing such tests. In particular, we emphasize the impor-

tance of functional follow-up studies to characterize putative selected

alleles and the use of selection scans as hypothesis-generating tools for

investigating evolutionary histories.

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

Click here for quick links to

Annual Reviews content online,

including:

• Other articles in this volume

• Top cited articles

• Top downloaded articles

• Our comprehensive search

Further

ANNUAL

REVIEWS

GE47CH05-Sabeti ARI 29 October 2013 12:12

INTRODUCTION

As humans and other organisms moved to

inhabit every part of the world, they were ex-

posed to myriad new environments, diets, and

pathogens, and forced to adapt, leading to the

great diversity we observe today. Uncovering

the mechanism of this diversiﬁcation has for

years fascinated scientists and nonscientists

alike. In 1858, Darwin and Wallace gave

grounds for species evolution when they

articulated the principle of natural selection,

the idea that beneﬁcial traits—those that

improve an individual’s chances to survive and

reproduce—tend to become more frequent in

populations over time.

Scientists have continued to search for

evidence of evolution and for the speciﬁc

adaptations that underlie it. Animal and plant

breeders were some of the ﬁrst to identify

traits that are evolving, as they witnessed

dramatic changes in their stock through

artiﬁcial selection. Haldane uncovered the ﬁrst

adaptive trait in humans when he observed

that many diseases of red blood cells seemed

to be distributed in regions where malaria was

endemic (48). Haldane’s malaria hypothesis

was conﬁrmed by Allison a few years later,

when he demonstrated that the sickle cell

mutation in the Hemoglobin-B gene (HBB) was

the target of selection for malaria resistance (4).

The ability to assess evidence for selection

at the genetic level represented a breakthrough

for this pursuit. Computational analysis of pop-

ulation genetic data sets provides a statistically

rigorous way to infer the action of natural selec-

tion; in this way, the ﬁeld of evolutionary genet-

ics represents an antidote to the preponderance

of speculative just-so stories that some biolo-

gists have lamented (42). Moreover, it demon-

strates the full realization of the modern syn-

thesis: Darwinian concepts of selection have

been rendered quantitative and measurable in

real populations, thanks to methodological and

technological advances (1).

Through evolutionary genetics, many adap-

tive traits have been elucidated, from lactase

persistence and skin pigmentation in humans

(90, 125) to coat color in ﬁeld mice (81) to

armored plates in stickleback ﬁsh (64). These

instances were all identiﬁed using a forward

genetics approach, in which a phenotype was

ﬁrst hypothesized to be adaptive and the un-

derlying loci were then identiﬁed. With on-

going advancements in genomic technology,

we can now go further, from testing evidence

for selection on putative adaptive traits to un-

covering candidate genetic regions through

genome scans. This transition from hypothesis-

testing to hypothesis-generating science has

been made possible both by the new data (e.g.,

genome sequences from increasing numbers of

species and genome-wide variation data) and by

increasingly sophisticated tools that allow us to

make sense of this deluge of data and to ﬁne-

map evidence of selection to individual candi-

date variants.

Identifying such candidates is signiﬁcant not

only because they demonstrate evolution and

shed light on species histories but also because

they represent biologically meaningful varia-

tion. Given that selection operates at the level of

the phenotype, alleles showing evidence of se-

lection are likely to be of functional relevance.

Thus, alleles implicated in selection studies are

often linked either to resistance to infectious

diseases, as pathogens are believed to represent

one of the strongest selective pressures acting

on humans (40), or to noninfectious genetic

diseases, such as those associated with autoim-

mune diseases or metabolic disorders (54).

Further breakthroughs in genomic anno-

tation, genome manipulation technology, and

high-throughput molecular biology are be-

ginning to allow researchers to progress from

candidate variants to functionally elucidated

instances of evolution. Taken together, all of

these advancements present a path to realizing

the full potential of evolutionary genomics

in shedding light on species histories and

uncovering biologically meaningful variation.

Modes of Selection

Natural selection is based on the simple obser-

vation that ﬁtness-enhancing traits, i.e., those

98 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Heterozygote

advantage: atrendin

which the ﬁtness of a

heterozygote is greater

than that of either

homozygote. Also

referred to as

overdominance

Frequency-

dependent selection:

a trend in which the

ﬁtness of a given

genotype is correlated

with its prevalence in

the population (e.g., if

an allele is

advantageous when it

is rare)

Codominance:

condition in which

multiple alleles are

dominant; the

heterozygote expresses

phenotypes associated

with both alleles

that improve an organism’s chance of survival

or reproductive success in its environment, are

more likely to be passed on to that organism’s

offspring and therefore increase in prevalence

in the population over time. I n the genomic

era, selection refers to any nonrandom, dif-

ferential propagation of an allele as a conse-

quence of its phenotypic effect. There are many

speciﬁc modes of selection that have been de-

scribed, some of which share conceptual over-

lap, and some of which are referred to by multi-

ple names. In this section, we brieﬂy deﬁne the

different modes of selection that we employ in

our discussion (85).

Most simply, selection may act in a direc-

tional manner, in which an allele is favored and

so propagated (positive selection) or disfavored

(negative selection, also called purifying selec-

tion). Random mutations are more likely to be

deleterious than beneﬁcial, so many novel al-

leles are immediately subject to negative selec-

tion and become removed from the gene pool

before they can achieve detectable frequency

within the population. This ongoing removal

of deleterious mutations is a form of negative

selection referred to as background selection.

In genetic regions under strong background se-

lection, mutations are quickly removed from

the gene pool, resulting in highly conserved

stretches of the genome (i.e., regions where

variation is not observed).

More subtle conﬁgurations of positive and

negative selection give rise to other common

evolutionary trends, particularly (although not

exclusively) in diploid and polyploid organisms,

where the phenotype depends on the interac-

tion of multiple alleles at the same locus. One

such phenomenon is balancing selection, in

which multiple alleles are maintained at an ap-

preciable frequency within the gene pool. This

may happen as the result of, for example, het-

erozygote advantage (i.e., overdominance) or

frequency-dependent selection (20). If the alle-

les being maintained conduce to opposing phe-

notypic effects—for example, if large and small

body sizes are maintained within the population

to the exclusion of intermediate sizes—then the

trend is often further described as diversifying

or disruptive selection. By contrast, when inter-

mediate phenotypic values are favored, whether

by balancing selection of codominant alleles or

by positive selection of alleles that underlie in-

termediate phenotypes, the trend is called sta-

bilizing selection.

This diversity of modes of selection notwith-

standing, much research in recent years has fo-

cused on the development of genomic methods

to identify positive selection. One reason for

this emphasis on positive selection is practical:

Whereas negative selection is primarily observ-

able in highly conserved regions and balancing

selection’s effect on the genome is often subtle,

positive selection leaves a more conspicuous

footprint on the genome that can be detected

using a number of different approaches. An-

other reason for the interest in positive selection

is theoretical: Positive selection is understood

to be the primary mechanism of adaptation

(i.e., the genesis of phenotypes that are apt for

a speciﬁc environment or niche), which in turn

poses great theoretical interest to researchers

(1).

Here, we discuss the various approaches that

have been used to identify positive selection

while also indicating the ways that these meth-

ods may be used to detect and classify instances

of other modes of selection (Table 1). These

approaches typically use summary statistics

to compare observed data with expectations

under the null hypothesis of selective neutrality

(see sidebar, Selection and Neutrality).

We begin by discussing methods based

on comparisons of different species and their

relative rates of genetic change. These methods

are most often used to identify selective events

that took place within the deep past and that

reﬂect macroevolutionary trends that occur as a

result of selection between, rather than within,

species. We then turn our attention to pop-

ulation genetics methods used to identify mi-

croevolutionary selective events within species.

Variants identiﬁed by these latter methods are

believed to underlie local adaptations in hu-

mans following the out-of-Africa migration and

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 99

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Table 1 An overview of common approaches for detecting selection

Approach Intuition Representative tests References

Methods for

macroevolution

(alsoreferredtoas

or ω)

(43, 60)

Gene-based methods Synonymous substitutions are

(assumed to be) selectively neutral.

Thus, they tell us about the

background rate of evolution. If the

rate of nonsynonymous substitution

differs signiﬁcantly, it is suggestive

of selection.

McDonald-Kreitman test

(MKT)

(27, 78)

Other rate-based

methods

Levels of polymorphism and

divergence should be correlated

(because both are primarily

functions of the mutation rate)

unless selection causes one to

exceed the other.

Hudson-Kreitman-

Aguad

e (HKA) test

MKT

(59, 135)

Regions that undergo accelerated

change in one lineage but are

conserved in related lineages are

probable candidates for selection.

Identiﬁcation of

accelerated regions

(14, 77, 100,

102, 116)

Methods for

microevolution

Ewens-Watterson test (30, 133)

Tajima’s D and derivatives (38, 39, 122,

123)

Frequency-based

methods

In a selective sweep, a genetic variant

reaches high prevalence together

with nearby linked variants (high-

frequency derived alleles). From

this homogenous background, new

alleles arise but are initially at low

frequency (surplus of rare alleles).

Fay & Wu’s H (33)

Long-range haplotype

(LRH) test

(111, 141)

Long-range haplotype

similarity test

(52)

Integrated haplotype

score (iHS)

(131)

Cross-population

extended haplotype

homozygosity

(XP-EHH)

(113)

Linkage disequilibrium

decay (LDD)

(132)

Linkage

disequilibrium–

based

methods

Selective sweeps bring a genetic

region to high prevalence in a

population, including the causal

variant and its neighbors. The

associations between these alleles

deﬁne a haplotype, which persists in

the population until recombination

breaks these associations down.

Identity-by-descent (IBD)

analyses

(15, 50)

Lewontin-Krakauer test

(LKT)

(11, 31, 73, 129)

Locus-speciﬁc branch

length (LSBL)

(117)

Population

differentiation–

based

methods

Selection acting on an allele in one

population but not in another

creates a marked difference in the

frequency of that allele between the

two populations. This effect of

differentiation stands out against

the differentiation between

populations with respect to neutral

(i.e., nonselected) alleles.

hapFLK (32)

(Continued )

100 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Table 1 (Continued)

Approach Intuition Representative tests References

Composite methods Composite likelihood

ratio (CLR)

(67, 68, 87, 89)

Combining test scores for multiple

sites across a contiguous region

can reduce the rate of false

positives.

Cross-population

composite likelihood

ratio (XP-CLR)

(22)

Combining multiple independent

tests at one site can improve

resolution and distinguish causal

variants. Different tests can

provide complementary

information.

DH test (138, 139)

Composite of multiple

signals (CMS)

(44, 45)

Homologs: traits or

sequences that are

similar in disparate

groups because of

common ancestry

Synonymous: a

change in the protein-

coding region of a

gene that does not

change the amino acid

encoded

thus have become the subject of much research

toward understanding human evolution and

history (112).

DETECTING SELECTION AT

THE MACROEVOLUTIONARY

LEVEL

Methods to detect selection at the macroevolu-

tionary level typically hinge on comparisons of

homologous traits or sequences among related

taxa (Figure 1a). These methods identify se-

quences that are likely to be functional (either

because they code for proteins or because they

are conserved among different species) and

then search for lineage-speciﬁc accelerations

in the rate of evolution. Such accelerations are

indicated by an excess of substitutions relative

to the baseline mutation rate, which can be

calculated either from the rate of synonymous

mutations (which are generally considered

neutral, but see Reference 21) or from the

overall rate of substitutions between species.

Gene-Based Methods

Perhaps the best-known statistic for detecting

selection is K

,alsoreferredtoasd

ω (Figure 1b). This statistic compares the rate

of nonsynonymous substitutions per site (i.e.,

per potential nonsynonymous change) with the

rate of synonymous substitutions per site (i.e.,

per potential synonymous change) (60). Be-

cause synonymous changes are assumed to be

functionally neutral (silent), their substitution

rate provides a baseline against which the rate

of amino acid alterations can be interpreted. A

relative excess of nonsynonymous substitutions

indicates ongoing (or recently ended) positive

selection favoring novel protein structures (or

else a cessation of negative selection against

protein alterations; see section, Challenges

in Applying Statistical Tests for Selection).

This is summarized by a value of K

greater than 1, whereas smaller values indicate

SELECTION AND NEUTRALITY

Kimura’s neutral theory of molecular evolution held that the vast

majority of genetic change is attributable to genetic drift rather

than Darwinian selection (69). However, as researchers began to

develop methods to distinguish neutral from adaptive change in

the genome, many came to reject the stronger versions of the

neutral theory and turned their attention toward quantifying the

relative contributions of drift and selection to molecular evolution

(71, 120).

Importantly, however, the neutral theory enabled the devel-

opment of tests for selection by assisting in the sophistication

of models of genetic drift. In many tests for selection (neutrality

tests), r esearchers compare empirical data against data generated

by simulations of drift, which serve as a null hypothesis. Other

neutrality tests may use background rates of change inferred from

whole-genome analyses to furnish a null hypothesis.

In this review, we focus our discussion on the wide range of

tests for selection that have been developed and their applica-

tions. Readers interested in the selectionist-neutralist debate are

encouraged to consult recent reviews on the subject (7, 33, 83).

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 101

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

TGCAGAGTAAGACCT

TGTAGCG CCAGACAT

Species 1

Species 2

Species 3

Species 4

Ancestral sequence

Species 4

Nonsynonymous substitutions

Synonymous substitutions

> 1?

Time

Genomic position

Positive selection

Divergence >

polymorphism

Substitution Mutation

Neutrality

Divergence

polymorphism

TCCAGAATGAGACGT

Figure 1

Methods for detecting selection at the macroevolutionary level. (a) Traits that are conserved across many

clades of a phylogeny but that show extreme differentiation in one or a few lineages are likely candidates for

selection. (b) Metrics such as K

compare the rate of nonsynonymous (i.e., amino acid–altering)

substitutions in a lineage to the rate of synonymous substitutions, which are assumed to be selectively

neutral. (c) The McDonald-Kreitman test and the Hudson-Kreitman-Aguad

e test hinge on the intuition that

levels of interspecies divergence and of intraspecies polymorphism are governed by the mutation rate and are

correlated unless selection or some other force (e.g., ﬂuctuations in population size) is at play.

Nonsynonymous: a

change in the protein-

coding region of a

gene that alters the

amino acid encoded

ongoing negative selection against deleterious

mutations and the consequent preservation

of protein structure. These methods may

also be applied across an entire open reading

frame or some subdivision thereof (down to

an individual codon), as different regions of a

protein may be subject to different selective

pressures (136). Various models for calculating

synonymous and nonsynonymous substi-

tution rates take into account the different

probabilities of different mutations (e.g.,

transitions are more likely than transversions)

102 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Codon usage bias:

the tendency of an

organism’s genome to

more commonly have

a certain codon for a

given amino acid than

any of its synonymous

counterparts

as well as the possibility of unobserved changes

(e.g., if one species undergoes two sequential

mutations at the same site) and codon usage

bias (43).

The McDonald-Kreitman test (MKT)

builds upon this method by utilizing not

only interspecies divergence data but also in-

traspecies diversity data (78). Essentially, the

MKT compares two K

values, one be-

tween species and one within species. Under

neutrality, these rates should be equal, given

constant rates of mutation and substitution. If

the between-species ratio signiﬁcantly exceeds

the within-species ratio, the null hypothesis

can be rejected, suggesting positive selection

between species. Conversely, a larger within-

species value suggests balancing selection or

else a surplus of maladaptive variants (e.g., re-

cessive disease alleles) under weak negative se-

lection within the species (see section, Detect-

ing Selection at the Microevolutionary Level).

Other Rate-Based Methods

Similar to the MKT, the Hudson-Kreitman-

Aguad

e (HKA) test uses both divergence and di-

versity data to compare relative rates of change

(Figure 1c). Speciﬁcally, the HKA test exam-

ines the ratios of ﬁxed interspeciﬁc differences

(D; i.e., substitutions) to within-species poly-

morphisms (P) across loci (59). The test hinges

on the supposition that, for a neutral site, both

D and P are functions of the site’s mutation

rate, which is assumed to have been roughly

constant at least since the point of species di-

vergence. Using a goodness-of-ﬁt test (e.g., χ

one can check individual sites for deviation from

the neutral D/P ratio, which allows rejection

of the null hypothesis and therefore can be in-

terpreted as evidence for selection. Relatively

large D/P values indicate either that change

contributing to speciation was accelerated (di-

rectional selection between species) or that di-

versity within the species is reduced (directional

selection within species; see section, Detecting

Selection at the Microevolutionary Level). Rel-

atively small values suggest balancing selection

between species.

One advantage of the HKA approach is that

it can be applied to any genetic region, not

just those that code for proteins. In practice,

however, the rate of neutral evolution in

protein-coding regions is much easier to infer

(i.e., by examining the synonymous substitu-

tion rate). The variability of the mutation rate

across different loci, coupled with a lack of

any a priori understanding of which sites (or,

indeed, what percentage of sites) are neutral,

has historically made application of the HKA

test challenging (140). In recent years, how-

ever, researchers have expanded this approach

in a maximum likelihood framework to allow

more efﬁcient multilocus comparisons (135).

By examining multiple sites, one can derive

the expected neutral D/P ratio for a lineage

while accounting for variation in the mutation

rate.

Other studies have used comparative ge-

nomic data to identify elements in the genome

that are highly conserved between disparate

species but show a signiﬁcantly accelerated rate

of substitution in a particular species or lin-

eage (14, 100, 102). For example, the gene

HAR1F, a noncoding RNA expressed during

brain development, is highly conserved be-

tween chimpanzees and other vertebrates but

has 40 times more substitutions in humans

than expected under neutrality (101). This ap-

proach has been used to identify several hun-

dred human-speciﬁc and primate-speciﬁc re-

gions (77). Similar relative-rate methods have

also been employed in understanding bacterial

evolution (116).

Phenotypic Methods

The idea of comparing related species and iden-

tifying striking differences can also be applied

to phenotypes. Traits that are conserved across

many closely related species (and thus likely

to be functional) but show extreme differenti-

ation in just one or a few of these species are

strong candidates for natural selection (110).

This approach has been used recently in com-

parative studies of gene expression (13, 97).

The gene SDR16C5, for example, regulates the

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 103

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

metabolism of retinol, a form of vitamin A

that is common in tree exudates. Slow lorises

and marmosets, which feed on tree bark, show

highly elevated expression levels of SDR16C5 in

the liver compared with their close evolution-

ary cousins, suggesting selection on regulatory

elements as a preventative measure against vi-

tamin A toxicity (97).

Alleles or traits that repeatedly arise in in-

dependent lineages suggest the action of con-

vergent evolution. This signature has been ob-

served in morphological traits, e.g., the loss

of pelvic structures in stickleback ﬁsh (18) and

wing pigmentation patterns in Drosophila (105).

It is also seen in viral and bacterial evolution, in

particular in the emergence of drug resistance

(12, 58).

DETECTING SELECTION AT

THE MICROEVOLUTIONARY

LEVEL

Positive selection causes a beneﬁcial allele to

sweep to high prevalence or ﬁxation (100%

prevalence) rapidly within a population. When

a beneﬁcial allele and surrounding variants on

the same haplotype reach high prevalence to-

gether, it produces a population-wide reduc-

tion in genetic diversity (sometimes referred to

as heterozygosity, polymorphism, or variabil-

ity) surrounding the causal allele (119). This

reduction, which persists until recombination

and mutation restore diversity to the popula-

tion at the selected locus, is t he hallmark of a

selective sweep (Figure 2a). There are various

ways of quantifying and detecting this signal,

which we discuss in the upcoming two sections.

We then discuss methods based on the envi-

ronmentally speciﬁc nature of selection, which

compare populations in which selection is or is

not hypothesized to be at play. We then turn

our attention to methods that combine the re-

sults of multiple tests to provide greater power

and resolution.

Frequency Spectrum–Based Methods

As a selected allele and its nearby hitchhiker

genetic region sweep toward ﬁxation, they shift

the distribution of alleles in the population

(Figure 2b). The sweep causes a population-

wide reduction in the genetic diversity around

the selected locus. New mutations appear on

this homogenous background, but they are

initially rare because they have only recently

appeared in the population. This creates a

surplus of rare alleles (i.e., many sites near

the selected variant have alleles that segregate

at low frequencies). Although the frequency

spectrum shifts back to baseline over time, the

distortion persists for thousands of generations

(several hundred t housand years in the case of

humans). Tajima’s D was the ﬁrst, and is the

most commonly used, test to detect this signal

(122).

Tajima’s D quantiﬁes this phenomenon by

comparing the number of pair-wise differences

between individuals with the total number

of segregating polymorphisms. Because low-

frequency alleles contribute less to the number

of pair-wise differences in a sample set than do

alleles of moderate frequency, a surplus of rare

−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→

Figure 2

Methods for detecting selective sweeps at the microevolutionary level. (a) Beneﬁcial mutations bring nearby hitchhiker variants to high

frequency, causing a population-wide reduction in the genetic diversity around the selected locus. This trough in diversity may be

shallower and/or narrower if the sweep is incomplete or if the mutation is not subject to immediate selection (i.e., selection on standing

variation or soft sweep). (b) A beneﬁcial mutation brings nearby derived alleles to high frequency. After the sweep is complete, novel

mutations against a homogenous background create a surplus of rare alleles. (c) A selective sweep causes extended haplotype

homozygosity (EHH), which is a measure of linkage disequilibrium, to rise across the haplotype that contains the selected allele. The

plateau of high EHH begins to break down when novel mutations and recombination gradually restore diversity to the population.

(d ) Differences in allele frequencies, reﬂecting the population-speciﬁc action of selection, cause Wright’s ﬁxation index (F

) between

two populations to increase. (e) Composite methods that integrate information from multiple signals of selection can provide ﬁner

resolution and help pinpoint causal variants.

104 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Ancestral: an allele

that was pre-existing

in a population and

from which a derived

allele may arise

alleles inﬂates the latter value disproportion-

ately to the former value. Thus, smaller (i.e.,

more negative) values of D suggest a surplus of

rare alleles, which may be indicative of positive

selection or population expansion (see section,

Challenges in Applying Statistical Tests for Se-

lection). Several variations on t his method have

been developed to take into account the polarity

of each allele (i.e., which one is derived or ances-

tral based on comparisons with an evolutionary

a Population diversity

Genomic position

Complete sweep de novo

Complete sweep from standing variation

Incomplete sweep

Soft sweep

d Population dierentiation

c Linkage disequilibriumb Frequency spectrum

e Composite methods

P(selection)

Genomic position

TIME

Population A Population B

~0.5

Small

(~0)

EHH

Genomic position

Reaches

xation

High-

frequency

derived

alleles

Diversity

returns

Surplus

of rare

alleles

Rises in

frequency

Novel

variant

Derived

allele

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 105

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Derived: an allele

that arises via a novel

mutation and does not

achieve ﬁxation in a

population (as

contrasted with an

ancestral allele)

Genetic drift: change

in allele frequencies

over time due to

chance (e.g. random

sampling)

Linkage

disequilibrium (LD):

tendency of certain

variants on the same

chromosome to be

coinherited at above

chance rates within a

population (e.g., owing

to selection or founder

effects)

outgroup) and to measure the abundance of rare

alleles in different ways (38, 39).

Selective sweeps also distort the frequency

spectrum by increasing the frequency of

derived alleles. Under genetic drift, it takes

many generations to bring neutral mutations

to moderate or high prevalence. However, in a

selective sweep, any derived alleles that reside

near the causal allele also hitchhike to high fre-

quency. Using a similar approach as Tajima’s

D, Fay & Wu’s (34) H compares the number

of pair-wise differences between individuals

to the number of individuals homozygous for

the derived allele. Small values of H indicate

an excess of high-frequency derived alleles,

suggestive of positive selection in the region

examined.

Site frequency spectrum analysis can also be

very useful for other modes of selection, such

as balancing selection, in which an excess of

intermediate-frequency alleles distorts metrics

like Tajima’s D (20). Andr

es et al. (5) sought ev-

idence for long-term balancing selection in the

human genome by leveraging frequency spec-

trum methods together with a modiﬁcation of

the HKA test to detect an excess of diversity in

regions linked to the selected variants. Long-

term balancing selection results in greater co-

alescence times than expected under neutrality

and thus fewer rare alleles.

Linkage Disequilibrium–Based

Methods

As it sweeps through the population, a selected

allele persists in strong linkage disequilibrium

(LD) with its neighboring hitchhiker variants

until recombination causes these associations

to break down. Together, the causal allele and

its linked neighbor variants deﬁne a haplotype.

Thus, a third suite of methods for detecting

positive selection looks for extended regions of

strong LD (or, equivalently, long haplotypes)

relative to their prevalence within a population

(Figure 2c). The thought is that such regions

must have swept to high prevalence quickly, or

else recombination would have caused LD to

break down and the haplotype to shorten.

LD-based approaches are particularly useful

for identifying variants that have undergone

a partial or incomplete selective sweep (see

section, Selection on Standing Variation and

Soft Sweeps), in which a new mutation has

risen to a modest frequency in the population

rather than reaching ﬁxation. This is useful

in many species, including humans, as most

novel alleles s ince the out-of-Africa migration

with realistic selection coefﬁcients are unlikely

to have yet reached ﬁxation (Figure 3). For

example, the causal allele of lactase persistence

in Europeans, which has a dominant effect, is

expected to take roughly 50,000 years to reach

Dominant

Recessive

Additive

s = 0.039

Time (ky)

1.0

12.5 25 50

37.5

0.8

0.6

0.4

0.2

0.0

Out-of-Africa

Time (ky)

25 50 75 100

s = 0.01

Allele frequency

s = 0.005

Time (ky)

25 50 75 100

Out-of-Africa

Figure 3

Trajectories of beneﬁcial alleles with realistic selection coefﬁcients simulated in human populations. The fate of a beneﬁcial allele

depends on many factors, including the strength of selection and the extent of the allele’s phenotypic inﬂuence (i.e., whether it is

dominant, recessive, etc.). Most alleles with realistic selection coefﬁcients that have arisen since the out-of-Africa migration are

expected to have not yet reached ﬁxation in their respective populations.

106 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Single nucleotide

polymorphism

(SNP): individual

base-pair sites in the

genome of an

organism where

multiple variants exist

ﬁxation, far longer than it has been in existence.

Thus, despite offering one the strongest known

selective advantages (with a selection coefﬁ-

cient estimated at 0.039) in humans, the allele

frequency is only 80% in Europeans (125).

Beneﬁcial mutations that arose more recently

or were under less extreme selective pressure

are even more likely to remain polymorphic in

the selected population, and many will never

reach ﬁxation because selective pressures can

change greatly over tens of thousands of years.

LD-based approaches can also be used to iden-

tify short-term balancing selection, where the

signal is comparable with that of an incomplete

sweep. For example, a number of papers have

demonstrated long-haplotype signals at the

sickle cell mutation in West Africa (51, 52).

One suite of widely used LD-based tests f or

selection centers around the extended haplo-

type homozygosity statistic (111). One deﬁnes

extended haplotype homozygosity (EHH) from

a core region (e.g., a putatively selected allele)

to a speciﬁed distance out in both directions

and calculates the probability that any two ran-

domly chosen chromosomes within the pop-

ulation carrying the core region are identical

by descent for the entire region. Thus, as one

travels further from the core region, EHH de-

creases, reﬂecting the action of recombination

whittling down the haplotype within the pop-

ulation. The long-range haplotype (LRH) test

compares a haplotype’s frequency to its relative

EHH at various distances, looking for haplo-

types that are extended as well as common, sug-

gesting that they rose to high prevalence quickly

enough that recombination has not had time to

break down the haplotype. Zhang et al. (141)

adapted this test by focusing on derived alle-

les (which are believed to be more likely candi-

dates for selective sweeps) as well as by intro-

ducing a genome-wide score. Hanchard et al.

(52) provided the long-range haplotype similar-

ity test, which utilizes a sliding window analysis

to quantify the population-wide homogeneity

of haplotypes.

The integrated haplotype score (iHS) (131)

is an inﬂuential variation on EHH. This statis-

tic compares the area under the curve deﬁned

by EHH for the derived and ancestral variants

as one travels further in genetic distance from

the core region. By calculating the area under

the curve deﬁned by EHH, this test captures

the intuition that both extreme EHH for a

short distance and moderate EHH for a longer

distance are suggestive of positive selection.

Another variation is the cross-population

extended haplotype homozygosity (XP-EHH)

statistic (113). This method compares haplo-

type lengths between populations to control

for local variation in recombination rates.

These two methods are complementary in

terms of their scope: Whereas iHS has more

power to detect incomplete sweeps, XP-EHH

is useful when the sweep is near ﬁxation within

one population (99).

Other LD-based tests include the LD decay

(LDD) test, which circumvents the need to de-

termine haplotypes (i.e., by phasing) by limiting

its scope to homozygous single nucleotide poly-

morphism (SNP) sites and inferring the fraction

of recombinant chromosomes at adjacent poly-

morphisms (132). Recently, Wiener & Pong-

Wong (134) developed a new test that ﬁts a re-

gression to heterozygosity data as a function of

genomic position: Selection is inferred on the

basis of the goodness-of-ﬁt to the reduction in

heterozygosity as predicted in a selective sweep.

The strength of this test is that whereas tradi-

tional LD-based approaches are designed for

analysis of SNP data, their regression test can

be used with any genetic marker.

In recent years, a number of researchers

have adapted identity-by-descent (IBD) anal-

yses to selection mapping, invoking essentially

the same conceptual motivations as earlier

EHH-based approaches (15, 50). IBD analyses,

which have been employed in a number of

population history analyses (142), search for

regions in which a set of individuals share a

long stretch of DNA, a pattern that presumably

can only be due to shared ancestry. Although

IBD- and EHH-based methods look for the

same pattern in genomic data, differences

in their computational implementation give

IBD-based approaches the advantage of being

able to detect selection on standing variation

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 107

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

(see section, Selection on Standing Variation

and Soft Sweeps) with greater power than

EHH-based approaches (3).

Population Differentiation–Based

Methods

An allele’s selective valence is dependent on

the particular environment in which it exists.

Different populations are subject to different

environmental pressures, and as a result, the

traits that would be adaptive in each may be

different. If selection is acting on a locus within

one population but not within other related

populations, then the allele frequencies at that

locus among the populations can differ signiﬁ-

cantly (Figure 2d ). This principle is the foun-

dation of a set of tests that rely on population

differentiation to detect evidence of selection.

The most commonly used metric for popu-

lation differentiation is Wright’s ﬁxation index

), which compares the variance of allele

frequencies within and between populations

(57). Comparatively large values of F

at a locus

(i.e., relative to neutral regions) indicate stark

differentiation between populations, which is

suggestive of directional selection. Compara-

tively small values indicate that the populations

being compared are homogenous, which may

be indicative of balancing or directional selec-

tion in both. Unlike other methods, population

differentiation–based approaches can detect

many types of selection, including classic

sweeps, sweeps on standing variants, and nega-

tive selection. In recent years, a n umber of alter-

native statistics and variations on F

have also

been proposed (for review, see Reference 79).

-based tests for selection have a long his-

tory, originating with the Lewontin-Krakauer

test (LKT) in 1973 (73). This method uses the

(then limited) available data to estimate F

multiple loci within n populations and evaluates

the neutrality of this distribution on the basis

of its goodness-of-ﬁt to a χ

distribution with

n − 1 degrees of freedom or on the comparison

of this distribution’s variance with a theoretical

predicted value. The production of large

genetic data sets in recent years has made fea-

sible a more robust application of this test, in

which researchers compare the genome-wide

distribution of F

to individual loci (2).

Although such outlier approaches are

believed to mitigate the confounding effect

of demographic events—operating on the

understanding that such events affect the

genome in its totality, whereas selection acts

in a locus-speciﬁc manner—certain patterns of

migration and mutation within subpopulations

can still produce false positives (82). To correct

for these effects, new variations on this test have

also been developed that incorporate explicit,

user-speciﬁed assumptions about demographic

history (11, 31, 129). Bonhomme et al.’s (11)

FLK

statistic, for example, modiﬁes the LKT

(labeled T

by the authors) to incorporate a

kinship matrix (F) derived from prespeciﬁed

neutral loci to account for historical population

branching. Another line of development

reinterprets the F

metric within a Bayesian

framework, often implemented via Markov

chain Monte Carlo algorithms (9, 36, 109).

These approaches utilize F

-based statistics

to estimate the posterior probability of a given

allele being under selection.

Other metrics that derive from F

improve

its computational power by incorporating more

data. These data come from either a greater

number of populations or a greater number

of allelic sites. On the one hand, following

the former strategy, the locus-speciﬁc branch

length metric (LSBL) uses pair-wise calcula-

tions of F

from three or more populations

to isolate population-speciﬁc changes in allele

frequency relative to a broader genetic context

(117). On the other hand, the cross-population

composite likelihood ratio (XP-CLR) of allele

frequency differentiation extends F

to many

loci (22). This method, which is analogous to

the XP-EHH method discussed above, identi-

ﬁes genetic regions in which changes in allele

frequency over many sites occur too quickly

(as assessed by the size of the affected region,

which would gradually return to a neutral

distribution over time) to be due to genetic

drift. More recently, Fariello et al. (32) intro-

duced a new statistic, hapFLK, that examines

108 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Unfolded site

frequency spectrum:

spectrum of allele

frequencies that takes

into account only

derived and not

ancestral alleles

differentiation among populations on the basis

of haplotypes rather than individual alleles.

Composite Methods

As the above discussion suggests, natural se-

lection leaves a number of footprints on the

genome, and each test is designed to pick up

on a slightly different signal. Accordingly, re-

searchers sometimes combine multiple metrics

into composite tests toward the goal of pro-

viding greater power and/or spatial resolution.

These tests come in two distinct forms, both of

which are typically referred to as composite.

First, some methods form a composite score

for a genetic region rather than a single genetic

marker by combining individual scores at all the

markers within the region. The motivation for

such an approach is that, although false pos-

itives may occur at any one site by chance, a

contiguous region of positive markers is much

more likely to represent a bona ﬁde signal (16).

Indeed, because selective sweeps affect whole

haplotypes, one assumes that the signal of selec-

tion extends across a region. Thus, composite

methods that incorporate the same test across

multiple sites improve power and reduce the

false discovery rate. Several of the previously

discussed tests, including iHS, XP-EHH, and

XP-CLR, employ such window-based analyses.

One exemplar of this approach is Kim

& Stephan’s (68) CLR test, which evaluates

the probability of a selective event being

responsible for a surplus of derived alleles (i.e.,

a skew of the unfolded site frequency spectrum)

across multiple sites. Subsequent variations

also incorporated LD-based data (67) and a

goodness-of-ﬁt test to help distinguish selec-

tion from demographic events (63). These tests

calculate a composite likelihood by multiplying

marginal likelihoods for each site considered

within a sequence, and then compare the

composite likelihood under a model in which

a sweep has occurred with t he composite like-

lihood under a model in which no sweep has

occurred. In the above tests, the null hypothesis

was calculated on the basis of a population

genetics model, which Nielsen et al. (89) fur-

ther modiﬁed by deriving the null hypothesis

from background patterns of variation in the

data itself. In a later, separate composite test,

Nielsen et al. (87) created a two-dimensional

site frequency spectrum using allele frequen-

cies from two populations; analysis of this

table involved the combination of population

differentiation–based signatures (i.e., F

) with

measures for high-frequency derived alleles

and excesses of low-frequency alleles.

Whereas these methods combine the results

of one or a few tests for many variants, other

composite methods combine the results of

many tests at a single site. The purpose of these

composite methods is to utilize complementary

information from different signals in order to

provide better spatial resolution (Figure 2e).

One such line of composite test develop-

ment began with Zeng et al.’s (138) uniﬁcation

of Tajima’s D and Fay & Wu’s H, each of which

is sensitive to different demographic processes.

Zeng et al. later observed that by limiting them-

selves to site frequency spectrum–based meth-

ods, the power of their test in the presence of

high recombination rates was also limited, and

they opted to further incorporate the Ewens-

Watterson test, which compares the popula-

tion’s Hardy-Weinberg homozygosity to that

predicted under a neutral model (30, 133) and is

largely insensitive to recombination (139). An-

other composite test of this sort was developed

by Grossman et al. (45). This test, called the

composite of multiple signals (CMS) test, in-

corporates metrics from all three suites of meth-

ods discussed here. Speciﬁcally, CMS inte-

grates F

with iHS and XP-EHH as well as two

new site frequency spectra–based tests that the

authors developed: DAF, which tests for de-

rived alleles t hat are at high frequency relative

to other populations, and iHH, which mea-

sures the absolute rather than relative length of

the haplotype.

MORE COMPLEX MODELS

OF SELECTION

Although the sweep model has been a useful

approach for identifying evidence of selection

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 109

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

in diverse species, many selective events in hu-

mans and other organisms may not adhere to

this model, and devising new tests to identify

different forms of sweeps continues to be an

area of active research (56, 103). In the selec-

tive sweep model, a novel allele at a single locus

immediately confers a ﬁtness beneﬁt. Two ways

to update the model are to delay the ﬁtness ben-

eﬁt and to allow for multiple loci. In the below

sections, we discuss these two possibilities. We

then turn our attention to ways that extant tests

have been modiﬁed to identify different targets

of selection.

Selection on Standing Variation

and Soft Sweeps

Because mutations happen randomly and not in

response to speciﬁc selective pressures, alleles

may arise at a time when they are not imme-

diately beneﬁcial. Such neutral alleles might

reach a moderate frequency within the pop-

ulation simply as the result of genetic drift. If

environmental pressures later change to make

such a variant beneﬁcial, the scenario is termed

“selection on standing variation.” Notably, a

standing variant in the EDA signaling pathway

present in seawater ﬁsh has been shown to

be under positive selection in freshwater

stickleback ﬁsh. The variant, which is largely

hidden in the heterozygous state in seawater

populations, has emerged to cause loss of scales

in multiple distinct freshwater populations (24).

Selection on standing variation is likely to

occur in two scenarios: when the selection co-

efﬁcient and mutation rate are both high and

when the selection coefﬁcient is weak (93). This

latter possibility suggests a potential applica-

tion to complex organisms, such as humans in

particular. Selection on standing variation af-

fects the genome in ways that are compara-

ble to selection on novel variants (8) but can

be more subtle and therefore more difﬁcult to

detect. For example, LD between the stand-

ing variant and its neighbors persists as in a

classical (or hard) sweep; however, compared

with a hard sweep, the resulting trough in di-

versity is shallower, owing to the fact that the

standing variant has time to recombine and as-

sociate with different haplotype backgrounds

(106) (Figure 2a). This fact also distorts the fre-

quency spectrum in a distinctive manner: Com-

pared with a hard sweep, selection on stand-

ing variation creates a greater number of linked

neutral sites that have alleles at intermediate

frequency (106). As the distinction between sig-

natures of hard sweeps and selection on stand-

ing variation may be subtle, Peter et al. (98) offer

an approximate Bayesian computation (ABC)

framework for distinguishing standing variants

from de novo mutations.

A special instance of selection on standing

variation occurs when the standing variant

(or another allele that performs the same

biological function) appears on multiple dis-

tinguishable haplotype backgrounds, e.g., as a

result of recurrent mutation or migration. This

phenomenon is called a soft sweep (55, 93, 94).

Although the term soft sweep is sometimes

mistakenly used to indicate selection on stand-

ing variation more broadly, the two should

be distinguished, as the selective signature

that these trends leave, and consequently the

methods developed to detect them, differ (104).

Through computational simulations, Pen-

nings & Hermisson (94) demonstrated that the

signature of a soft sweep should be in many ways

comparable to that of a hard sweep. Although

frequency-based methods do not have predic-

tive power for soft sweeps—owing to the fact

that soft sweeps may involve an arbitrary num-

ber of distinct haplotypes—LD-based methods

are able to detect the signatures of soft sweeps,

albeit with diminished power. Similar to a hard

sweep, the locus under selection is situated at

the bottom of a trough of genetic diversity.

These results suggest that computational

methods to identify soft sweeps are within

reach; it remains for researchers to ﬁne-tune

current LD-based methods to detect them.

Polygenic Networks and

Ecological Methods

All of the methods discussed thus far assume

that selection acts on one or a few sites at a

110 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Copy number

variants (CNVs):

a form of structural

variant in which

multiple copies of a

genetic region exist

Microsatellites:

genetic regions that

consist of repeating

sequences of two to six

base pairs. Also

referred to as short

tandem repeats (STRs)

or simple sequence

repeats (SSRs)

Structural variants

(SVs): alterations in

the genome that affect

relatively large

chromosomal regions,

including deletions

and insertions (indels),

translocations,

inversions, and

duplications

time. However, given the known importance

of polygenic networks and of epistatic interac-

tions, researchers have suggested that selection

may more often act on multiple sites in tandem,

causing coordinated and distributed shifts in al-

lele frequencies (53, 104).

One way to identify polygenic groups of sites

under selection is to incorporate ecological in-

formation. By binning related populations ac-

cording to presumably relevant variables (e.g.,

habitat, climate, mode of sustenance, etc.), one

can seek shifts in allele frequency shared across

ecologically similar populations. Joost et al.

(65) formalized this approach as the spatial

analysis method (SAM), using multiple univari-

ate logistic regressions to test for association

between allele frequencies and environmental

variables. Jones et al. (64) use a similar approach

in their comparison of marine and freshwa-

ter sticklebacks from globally distributed pop-

ulations to identify loci consistently associated

with habitat, and Hancock et al. (53) perform a

similar analysis to identify ecologically relevant

loci in humans.

An important limitation of ecological

approaches is their reliance on user-speciﬁed

variables (104). These methods run the risk

of being biased by the information put in or

left out. Polygenic selection can be detected

without the risk of this bias by examining

shared functional sets, such as quantitative trait

loci (QTLs), in which multiple genetic regions

contribute to a single trait. Selection acting on a

network of QTLs can be inferred on the basis of

a signiﬁcant bias in their directionality, i.e., the

tendency of a locus to either amplify or lessen

the magnitude of the phenotype (91). Although

under neutrality, the distribution of positive or

negative QTLs may be random, an overrepre-

sentation of one or the other type of loci within

a lineage is s uggestive of selection. Fraser et al.

(37) developed a framework in which this test

can be applied in a genome-wide scan, focusing

on regulatory elements [i.e., expression QTLs

(eQTLs)] in mice. Similarly, Simonson et al.

(118) performed a genome-wide scan with

attention to genic networks known to be

involved in an oxygen-carry capacity to reveal

adaptation to high altitudes in a Tibetan

population.

Alternative Targets of Selection

Most natural selection studies to date have fo-

cused on genetic changes at the single nu-

cleotide level, primarily because they have been

the most accessible from a technological stand-

point, through advances in protein analysis and

SNP genotyping. Given their mutation mech-

anism, which typically creates simple biallelic

changes of unique origin, they also can be more

easily incorporated into statistical tests for se-

lection. Moreover, SNPs are useful in such tests

because they can act as markers: Nearby vari-

ants in LD with a S NP can be detected by using

said SNP as a proxy.

Many other genetic alterations that affect

an organism’s phenotype may be subject to

selection, including copy number variants

(CNVs) (115), microsatellites (46), chromo-

somal rearrangements (e.g., indels, inversions,

and translocations) (35), polygenic networks

(discussed above), and epigenetic annotations

(127). One of the ﬁrst elucidated examples

of selection were CNVs of α-andβ-globin

genes implicated in thalassemia, which, along

with sickle cell anemia, confer resistance to

malaria (6, 137). More recently, increased

CNV counts of the gene for amylase have

also been demonstrated to be associated with

diets containing larger amounts of starch

(96). Another example is a major inversion on

chromosome 17 in humans that was shown to

be associated with greater reproductive success

in an Icelandic population (121) and contains

population genetics evidence of positive selec-

tion. Structural variants (SVs), such as CNVs

and inversions, are often subject to negative

selection (especially those that may cause

frameshifts in protein-coding regions) (76)

or can lead to relaxed evolutionary constraint

through gene duplication (70). The many tests

for selection described above may be applied

to SVs, although the broad diversity of variants

under the umbrella term SV and the large

effects they can have on genomic architecture

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 111

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Epigenome:

annotations to the

DNA molecule that

alter patterns of gene

expression but do not

change the sequence

make the systematic detection of selected

variants challenging (61).

The recent discovery that certain epigenetic

arrangements are heritable across many gen-

erations also raises the possibility of selection

acting on the epigenome (62, 108). Such neo-

Lamarckian selection has been detected in or-

chids using SAM (92). It remains to be clari-

ﬁed to what extent such modes of selection are

prevalent, but it is an area of active interest.

CHALLENGES IN APPLYING

STATISTICAL TESTS

FOR SELECTION

Although each approach has its own particular

strengths and limitations, there are a number

of challenges that are shared among these

tests, particularly in the interpretation of

their signiﬁcance. A neutrality test may allow

rejection of the null hypothesis, but there

are many possible explanations other than

selection for the genomic results observed. For

example, demographic events (e.g., migration,

expansions, and bottlenecks) can often create

selection-mimicking signals. Historically, most

studies have aimed to rule out this possibility by

comparing locus-speciﬁc data to genome-wide

data, as demographic events are understood

to affect the genome in its totality, whereas

selection acts in a more targeted manner (17).

In recent years, however, some have questioned

this outlier approach, arguing that if selection

is pervasive (as in Drosophila; see Reference

74), then distributed patterns of genetic hitch-

hiking would be misinterpreted as reﬂecting

demographic events (47). More generally,

the recognition that the effects of selection

and demography may be interconnected have

led some to adopt other approaches, such as

explicitly estimating demographic parameters,

including population structure, through various

computational frameworks and incorporating

these into subsequent analyses (for examples,

see Reference 31; for review, see Reference

75). Another related issue is that false positives

can be produced when tests implicate neutral

variants in strong LD with a causal allele (124).

Even when these confounding effects can be

ruled out, the interpretation of selection may

not be straightforward. For example, rate-based

tests implicate regions in which evolutionary

change has been accelerated: This may be

due to positive selection of novel variants, but

the relaxation of selective constraint (i.e., of

purifying selection) over a region may have

the same effect. Distinguishing between these

possibilities involves case-by-case analysis. In a

study of the evolution of CNVs in humans, for

example, Nguyen et al. (84) ruled out positive

selection in regions in which they observed an

inverse relationship between rates of change

and rates of recombination. More generally,

however, functional analysis of candidate

regions can help adjudicate between these

two possibilities: If the derived variant has

no potentially ﬁtness-enhancing variation of

function, relative to the ancestral, then the re-

laxation of selective constraint is the more likely

explanation.

Another recurrent challenge for researchers

is accounting for systematic biases that may

be present in genomic data. The majority of

selection studies to date have utilized SNP

data, which is collected using genotyping arrays

designed to detect known polymorphisms.

The practical limitations of SNP discovery

protocols mean that low-frequency alleles

may go undetected, in which case they are

excluded from these arrays. These arrays can

therefore generate data that may be unrepre-

sentative of the full extent of genetic diversity,

a phenomenon known as ascertainment bias

(23). This sampling of the data can artiﬁcially

distort allele frequency measures as well as

derivative statistics that include LD. When the

SNP discovery protocol is known, statistical

measures can be taken to counteract the effect

of ascertainment bias (86, 88, 107). In addition,

genotyping assays that incorporate variable in-

tensity oligonucleotide (VINO) probes can be

used to mitigate the number of polymorphisms

overlooked as a result of ascertainment bias

(26).

Another salient issue for researchers in-

vestigating natural selection, particularly for

112 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Table 2 Using selection scans to study human evolution

Gene under

selection

Population(s)

Genomic

evidence for

selection

Functional

evidence

Putative adaptive role References

FOXP2 All (selection predates

out-of-Africa

migration)

Accelerated

evolution in

coding region,

D, H

Mouse transgenic Affects development of

corticobasal ganglia

circuits; thought to be

involved in mechanics of

speech

(28, 29)

LCT Northern Europeans,

East Africans

(pastoralist societies)

EHH, iHS; F

analysis

Human association

study; in vitro

lactase expression

assay

Confers lactase

persistence; allows

digestion of lactose into

adulthood

(10, 125)

EDAR East Asians and Native

Americans

CMS Human association

study; mouse

transgenic

Affects morphology of

hair, sweat glands, and

mammary glands

(45, 66)

TLR5 West Africans CMS In vitro assay of

NF-κB pathway

activation

Modulates immune

response to bacterial

ﬂagellin

(44, 45)

DARC African populations in

malaria-endemic

regions

Human association

study

Heterozygosis reduces

susceptibility to malaria

(49, 66a, 80)

APOL1 African populations in

trypanosome-endemic

regions

CMS In vitro assay of

response to

trypanosome

invasion

Modulates susceptibility

to trypanosomiasis

(45, 95, 128)

HBB African populations in

malaria-endemic

regions

LRH

LRH similarity

Human association

study

Heterozygosis reduces

susceptibility to malaria

(4, 51, 72,

111)

EPAS1,

EGLN,

et al.

Tibetans iHS, XP-EHH Human association

study

Selected variants decrease

hemoglobin

concentration and

modulate hypoxia

response

(41, 118)

SLC24A5,

SLC45A2

Europeans F

analysis,

XP-EHH, CMS

Human association

study; in vitro assay

of melanocyte

cultures; zebraﬁsh

transgenic

Decreases melanin

pigmentation in skin

(25, 90, 126)

CBARA1,

VAV3, et al.

Ethiopian-highland

populations

LSBL, iHS,

XP-EHH

Human association

study

Selected variants decrease

hemoglobin

concentration and

modulate hypoxia

response

(114)

Abbreviations: CMS, composite of multiple signals; EHH, extended haplotype homozygosity; F

, Wright’s ﬁxation index; iHS, integrated haplotype

score; LRH, long-range haplotype; LSBL, locus-speciﬁc branch length metric; XP-EHH, cross-population extended haplotype homozygosity.

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 113

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

Pleiotropy: atrendin

which one genotype

affects multiple

phenotypes

those studying it in humans, is the potential for

misinterpretation of results and their societal

signiﬁcance. By attending to linguistic sub-

tleties and employing caution in disseminating

results, researchers can help prevent unethical

application of evolutionary research (130).

FROM GENOME SCANS TO

EVOLUTIONARY HYPOTHESES

The ultimate validation of genomic metrics

of selection is the demonstration that putative

selective variants have phenotypic effects

with import for organismal ﬁtness (Table 2).

Functional understanding of a candidate region

begins with the ﬁne-mapping of that region so

as to localize the signal. Until recently, localiz-

ing signals of selection was a major challenge,

but new composite methods and full-genome

sequence data provide stronger resolution

that can allow researchers to identify tractable

candidates for functional scrutiny (44, 45).

Once individual alleles have been identiﬁed for

experimentation, researchers can measure the

effects of said alleles as compared with their

wild-type analogs. Genomic annotation can be

informative for experimental design by sug-

gesting the most probable types of traits that

a variant may affect or by suggesting the types

of cells in which a variant is most commonly

expressed.

Phenotypic screening may then proceed

through an association study of various

traits in the organism in question, although

background genetic variation can introduce

noise into the data. To correct for this,

researchers may instead compare the derived

and ancestral variants against the same genetic

background introduced into a cell line in vitro

or into model organisms in vivo. Even in

such situations, however, the possibility that a

variant has pleiotropic effects makes it difﬁcult

to discern whether a functional follow-up study

correctly identiﬁes the selective signiﬁcance of

the variant in question (7). Although exhaustive

phenotype screens are not possible, researchers

can bolster the strength of their evidence by

screening through as comprehensive a list of

possible effects as possible. For example, Enard

et al. (28) introduced two human-speciﬁc

amino acid substitutions in the FOXP2 gene

into mice and screened approximately three

hundred traits, ultimately ﬁnding that only

a small fraction of these (those involving the

structure and function of corticobasal ganglia

circuits) were signiﬁcantly different between

humanized and wild-type mice.

Creating a case for selection necessitates a

combination of genomic and functional evi-

dence. With the availability of large population

genetics data sets, statistical methods to inter-

pret that data, and increasingly sophisticated

technologies for transgenesis and other func-

tional methods, researchers are moving into a

new era of natural selection studies, in which

both the causes and effects of changes to the

genomes of humans and other organisms can

be modeled and understood.

SUMMARY POINTS

1. The development of genotyping and sequencing technologies has allowed for the full

realization and application of methods to investigate selection on the basis of theory from

the ﬁelds of comparative genomics and population genetics.

2. Methods to detect selection in the genome may be categorized by their effective timescale

(i.e., macro- versus microevolutionary) as well as by the types of data they utilize (i.e.,

interspecies divergence data, intraspecies diversity data, or a combination of these), or

the type of selective signature they identify.

114 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

3. Tests to detect selection at the macroevolutionary level make interspeciﬁc comparisons,

often aided by phylogenetic considerations, of the rates of change at the nucleotide level

and look for genetic regions in species that have experienced accelerated change.

4. Tests for microevolutionary selection come in a broad range of formats but often aim

to detect regions of reduced genetic diversity, which is indicative of a selective sweep.

Other tests compare populations in which selection is or is not hypothesized to be at play

and measure the extent of differentiation between them. Combining multiple tests can

increase power and resolution.

5. An active area of research is the development of tests for modes of selection that do

not adhere to the selective sweep model. Among these are polygenic selection and soft

sweeps.

6. Genomic evidence for selection is suggestive but not conclusive. A combination of ge-

nomic and functional evidence constitutes the current standard for the ﬁeld.

FUTURE ISSUES

1. How can we make tests for alternative selective modes (soft sweeps, polygenic selection,

etc.) more robust?

2. How can we accurately quantify the prevalence of selection and the relative contribution

of drift in humans and other organisms?

3. Can we develop high-throughput assays for functional analysis and validation of candidate

variants?

DISCLOSURE STATEMENT

The authors are not aware of any afﬁliations, memberships, funding, or ﬁnancial holdings that

might be perceived as affecting the objectivity of this review.

LITERATURE CITED

1. Akey JM. 2009. Constructing genomic maps of positive selection in humans: Where do we go from here?

Genome Res. 19(5):711–22

2. Akey JM, Zhang G, Zhang K, Jin L, Shriver MD. 2002. Interrogating a high-density SNP map for

signatures of natural selection. Genome Res. 12(12):1805–14

3. Albrechtsen A, Moltke I, Nielsen R. 2010. Natural selection and the distribution of identity-by-descent

in the human genome. Genetics 186(1):295–308

4. Allison AC. 1954. Protection afforded by sickle-cell trait against subtertian malarial infection. Br. Med.

J. 1(4857):290–94

5. Andr

es AM, Hubisz MJ, Indap A, Torgerson DG, Degenhardt JD, et al. 2009. Targets of balancing

selection in the human genome. Mol. Biol. Evol. 26(12):2755–64

6. Barrai I, Rosito A, Cappellozza G, Cristofori G, Vullo C, et al. 1984. Beta-thalassemia in the Po Delta:

selection, geography, and population structure. Am. J. Hum. Genet. 36(5):1121–34

7. Barrett RDH, Hoekstra HE. 2011. Molecular spandrels: tests of adaptation at the genetic level. Nat. Rev.

Genet. 12(11):767–80

8. Barrett RDH, Schluter D. 2008. Adaptation from standing genetic variation. Trends Ecol. Evol. (Amst.)

23(1):38–44

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 115

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

9. Beaumont MA, Balding DJ. 2004. Identifying adaptive genetic divergence among populations from

genome scans. Mol. Ecol. 13(4):969–80

10. Bersaglieri T, Sabeti PC, Patterson N, Vanderploeg T, Schaffner SF, et al. 2004. Genetic signatures of

strong recent positive selection at the lactase gene. Am. J. Hum. Genet. 74(6):1111–20

11. Bonhomme M, Chevalet C, Servin B, Boitard S, Abdallah J, et al. 2010. Detecting selection in population

trees: the Lewontin and Krakauer test extended. Genetics 186(1):241–62

12. Boucher CAB, O’Sullivan E, Mulder JW, Ramautarsing C, Kellam P, et al. 1992. Ordered appearance of

zidovudine resistance mutations during treatment of 18 human immunodeﬁciency virus-positive subjects.

J. Infect. Dis. 165(1):105–10

13. Brawand D, Soumillon M, Necsulea A, Julien P, Cs

ardi G, et al. 2011. The evolution of gene expression

levels in mammalian organs. Nature 478(7369):343–48

14. Burbano HA, Green RE, Maricic T, Lalueza-Fox C, De la Rasilla M, et al. 2012. Analysis of human

accelerated DNA regions using archaic hominin genomes. PLoS ONE 7(3):e32877

15. Cai Z, Camp NJ, Cannon-Albright L, Thomas A. 2011. Identiﬁcation of regions of positive selection

using shared genomic segment analysis. Eur. J. Hum. Genet. 19(6):667–71

16. Carlson CS, Thomas DJ, Eberle MA, Swanson JE, Livingston RJ, et al. 2005. Genomic regions exhibiting

positive selection identiﬁed from dense genotype data. Genome Res. 15(11):1553–65

17. Cavalli-Sforza LL. 1966. Population structure and human evolution. Proc. R. Soc. Lond. Ser. B.

164(995):362–79

18. Chan YF, Marks ME, Jones FC, Villarreal G, Shapiro MD, et al. 2010. Adaptive evolution of pelvic

reduction in sticklebacks by recurrent deletion of a Pitx1 enhancer. Science 327(5963):302–5

19. Charlesworth B, Morgan MT, Charlesworth D. 1993. The effect of deleterious mutations on neutral

molecular variation. Genetics 134(4):1289–303

20. Charlesworth D. 2006. Balancing selection and its effects on sequences in nearby genome regions. PLoS

Genet. 2(4):e64

21. Chamary JV, Parmley JL, Hurst LD. 2006. Hearing silence: non-neutral evolution at synonymous sites

in mammals. Nat. Rev. Genet. 7(2):98–108

22. Chen H, Patterson N, Reich D. 2010. Population differentiation as a test for selective sweeps. Genome

Res. 20(3):393–402

23. Clark AG, Hubisz MJ, Bustamante CD, Williamson SH, Nielsen R. 2005. Ascertainment bias in studies

of human genome-wide polymorphism. Genome Res. 15(11):1496–502

24. Colosimo PF, Hosemann KE, Balabhadra S, Villarreal G Jr, Dickson M, et al. 2005. Widespread parallel

evolution in sticklebacks by repeated ﬁxation of ectodysplasin alleles. Science 307(5717):1928–33

25. Cook AL, Chen W, Thurber AE, Smit DJ, Smith AG, et al. 2009. Analysis of cultured human melanocytes

based on polymorphisms within the SLC45A2/MATP, SLC24A5/NCKX5, and OCA2/P loci. J. Investig.

Dermatol. 129(2):392–405

26. Didion JP, Yang H, Sheppard K, Fu C-P, McMillan L, et al. 2012. Discovery of novel variants in

genotyping arrays improves genotype retention and reduces ascertainment bias. BMC Genomics 13:34

27. Egea R, Casillas S, Barbadilla A. 2008. Standard and generalized McDonald-Kreitman test: a website to

detect selection by comparing different classes of DNA sites. Nucleic Acids Res. 36:W157–62

28. Enard W, Gehre S, Hammerschmidt K, H

olter SM, Blass T, et al. 2009. A humanized version of Foxp2

affects cortico-basal ganglia circuits in mice. Cell 137(5):961–71

29. Enard W, Przeworski M, Fisher SE, Lai CSL, Wiebe V, et al. 2002. Molecular evolution of FOXP2,a

gene involved in speech and language. Nature 418(6900):869–72

30. Ewens WJ. 1972. The sampling theory of selectively neutral alleles. Theor. Popul. Biol. 3(1):87–112

31. Excofﬁer L, Hofer T, Foll M. 2009. Detecting loci under selection in a hierarchically structured popu-

lation. Heredity (Edinb.) 103(4):285–98

32. Fariello MI, Boitard S, Naya H, San Cristobal M, Servin B. 2013. Detecting signatures of selection

through haplotype differentiation among hierarchically structured populations.

Genetics 193:929–41

33. Fay JC. 2011. Weighing the evidence for adaptation at the molecular level. Trends Genet. 27(9):343–49

34. Fay JC, Wu CI. 2000. Hitchhiking under positive Darwinian selection. Genetics 155(3):1405–13

35. Feuk L, Carson AR, Scherer SW. 2006. Structural variation in the human genome. Nat. Rev. Genet.

7(2):85–97

116 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

36. Foll M, Gaggiotti O. 2008. A genome-scan method to identify selected loci appropriate for both dominant

and codominant markers: a Bayesian perspective. Genetics 180(2):977–93

37. Fraser HB, Babak T, Tsang J, Zhou Y, Zhang B, et al. 2011. Systematic detection of polygenic cis-

regulatory evolution. PLoS Genet. 7(3):e1002023

38. Fu Y-X. 1997. Statistical tests of neutrality of mutations against population growth, hitchhiking and

background selection. Genetics 147(2):915–25

39. Fu YX, Li WH. 1993. Statistical tests of neutrality of mutations. Genetics 133(3):693–709

40. Fumagalli M, Sironi M, Pozzoli U, Ferrer-Admettla A, Pattini L, Nielsen R. 2011. Signatures of environ-

mental genetic adaptation pinpoint pathogens as the main selective pressure through human evolution.

PLoS Genet. 7(11):e1002355

41. Ge R-L, Simonson TS, Cooksey RC, Tanna U, Qin G, et al. 2012. Metabolic insight into mechanisms

of high-altitude adaptation in Tibetans. Mol. Genet. Metab. 106(2):244–47

42. Gould S. 1978. Sociobiology: the art of storytelling. New Sci. 80(1129):530–33

43. Graur D, Li W-H. 2000. Fundamentals of Molecular Evolution. Sunderland, MA: Sinauer Assoc.

44. Grossman SR, Andersen KG, Shlyakhter I, Tabrizi S, Winnicki S, et al. 2013. Identifying recent adap-

tations in large-scale genomic data. Cell 152(4):703–13

45. Grossman SR, Shylakhter I, Karlsson EK, Byrne EH, Morales S, et al. 2010. A composite of multiple

signals distinguishes causal variants in regions of positive selection. Science 327(5967):883–86

46. Haasl RJ, Payseur BA. 2012. Microsatellites as targets of natural selection. Mol. Biol. Evol. 30(2):285–98

47. Hahn MW. 2008. Toward a selection theory of molecular evolution. Evolution 62(2):255–65

48. Haldane JBS. 2006. Disease and evolution. In Malaria: Genetic and Evolutionary Aspects,ed.KR

Dronamraju, P Arese, pp. 175–87. New York: Springer

49. Hamblin MT, DiRienzo A. 2000. Detection of the signature of natural selection in humans: evidence

from the Duffy blood group locus. Am. J. Hum. Genet. 66(5):1669–79

50. Han L, Abney M. 2012. Using identity by descent estimation with dense genotype data to detect positive

selection. Eur. J. Hum. Genet. 21(2):205–11

51. Hanchard N, Elzein A, Trafford C, Rockett K, Pinder M, et al. 2007. Classical sickle β-globin haplotypes

exhibit a high degree of long-range haplotype similarity in African and Afro-Caribbean populations. BMC

Genet. 8(1):52

52. Hanchard NA, Rockett KA, Spencer C, Coop G, Pinder M, et al. 2006. Screening for recently selected

alleles by analysis of human haplotype similarity. Am. J. Hum. Genet. 78(1):153–59

53. Hancock AM, Witonsky DB, Ehler E, Alkorta-Aranburu G, Beall C, et al. 2010. Colloquium paper:

human adaptations to diet, subsistence, and ecoregion are due to subtle shifts in allele frequency. Proc.

Natl. Acad. Sci. USA 107(Suppl. 2):8924–30

54. Hancock AM, Witonsky DB, Gordon AS, Eshel G, Pritchard JK, et al. 2008. Adaptations to climate in

candidate genes for common metabolic disorders. PLoS Genet. 4(2):e32

55. Hermisson J, Pennings PS. 2005. Soft sweeps: molecular population genetics of adaptation from standing

genetic variation. Genetics 169(4):2335–52

56. Hernandez RD, Kelley JL, Elyashiv E, Melton SC, Auton A, et al. 2011. Classic selective sweeps were

rare in recent human evolution. Science 331(6019):920–24

57. Holsinger KE, Weir BS. 2009. Genetics in geographically structured populations: deﬁning, estimating

and interpreting F(ST). Nat. Rev. Genet. 10(9):639–50

58. Holt KE, Parkhill J, Mazzoni CJ, Roumagnac P, Weill F-X, et al. 2008. High-throughput sequencing

provides insights into genome variation and evolution in Salmonella Typhi. Nat. Genet. 40(8):987–93

59. Hudson RR, Kreitman M, Aguad

e M. 1987. A test of neutral molecular evolution based on nucleotide

data. Genetics 116(1):153–59

60. Hurst LD. 2002. The Ka/Ks ratio: diagnosing the form of sequence evolution. Trends Genet. 18(9):486

61. Iskow RC, Gokcumen O, Lee C. 2012. Exploring the role of copy number variants in human adaptation.

Trends Genet. 28(6):245–57

62. Jablonka E, Raz G. 2009. Transgenerational epigenetic inheritance: prevalence, mechanisms, and im-

plications for the study of heredity and evolution. Q. Rev. Biol. 84(2):131–76

63. Jensen JD, Kim Y, DuMont VB, Aquadro CF, Bustamante CD. 2005. Distinguishing between selective

sweeps and demography using DNA polymorphism data. Genetics 170(3):1401–10

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 117

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

64. Jones FC, Grabherr MG, Chan YF, Russell P, Mauceli E, et al. 2012. The genomic basis of adaptive

evolution in threespine sticklebacks. Nature 484(7392):55–61

65. Joost S, Bonin A, Bruford MW, Despr

es L, Conord C, et al. 2007. A spatial analysis method (SAM)

to detect candidate loci for selection: towards a landscape genomics approach to adaptation. Mol. Ecol.

16(18):3955–69

66. Kamberov YG, Wang S, Tan J, Gerbault P, Wark A, et al. 2013. Modeling recent human evolution in

mice by expression of a selected EDAR variant. Cell 152(4):691–702

66a. Kasehagen LJ, Mueller I, Kinboro B, Bockarie MJ, Reeder JC, et al. 2007. Reduced Plasmodium vivax

erythrocyte infection in PNG Duffy-negative heterozygotes. PLoS ONE 2(3):e336

67. Kim Y, Nielsen R. 2004. Linkage disequilibrium as a signature of selective sweeps. Genetics 167(3):1513–

68. Kim Y, Stephan W. 2002. Detecting a local signature of genetic hitchhiking along a recombining chro-

mosome. Genetics 160(2):765–77

69. Kimura M. 1985. The Neutral Theory of Molecular Evolution. Cambridge: Cambridge Univ. Press

70. Kondrashov FA. 2012. Gene duplication as a mechanism of genomic adaptation to a changing environ-

ment. Proc. R. Soc. Lond. Ser. B 279(1749):5048–57

71. Kreitman M, Akashi H. 1995. Molecular evidence for natural selection. Annu. Rev. Ecol. Syst. 26:403–22

72. Kwiatkowski DP. 2005. How malaria has affected the human genome and what human genetics can teach

us about malaria. Am. J. Hum. Genet. 77(2):171–92

73. Lewontin RC, Krakauer J. 1973. Distribution of gene frequency as a test of the theory of the selective

neutrality of polymorphisms. Genetics 74(1):175–95

74. Li H, Stephan W. 2006. Inferring the demographic history and rate of adaptive substitution in Drosophila.

PLoS Genet. 2(10):e166

75. Li J, Li H, Jakobsson M, Li S, Sj

odin P, Lascoux M. 2012. Joint analysis of demography and selection

in population genetics: Where do we stand and where could we go? Mol. Ecol. 21(1):28–44

76. Li Y, Zheng H, Luo R, Wu H, Zhu H, et al. 2011. Structural variation in two human genomes mapped

at single-nucleotide resolution by whole genome de novo assembly. Nat. Biotechnol. 29(8):723–30

77. Lindblad-Toh K, Garber M, Zuk O, Lin MF, Parker BJ, et al. 2011. A high-resolution map of human

evolutionary constraint using 29 mammals. Nature 478(7370):476–82

78. McDonald JH, Kreitman M. 1991. Adaptive protein evolution at the Adh locus in Drosophila. Nature

351(6328):652–54

79. Meirmans PG, Hedrick PW. 2011. Assessing population structure: FST and related measures. Mol. Ecol.

Resour. 11(1):5–18

80. Miller LH, Mason SJ, Clyde DF, McGinniss MH. 1976. The resistance factor to Plasmodium vivax in

blacks. N. Engl. J. Med. 295(6):302–4

81. Mullen LM, Vignieri SN, Gore JA, Hoekstra HE. 2009. Adaptive basis of geographic variation:

genetic, phenotypic and environmental differences among beach mouse populations. Proc. Biol. Sci.

276(1674):3809–18

82. Nei M, Maruyama T. 1975. Letters to the editors: Lewontin-Krakauer test for neutral genes. Genetics

80(2):395

83. Nei M, Suzuki Y, Nozawa M. 2010. The neutral theory of molecular evolution in the genomic era. Annu.

Rev. Genomics Hum. Genet. 11:265–89

84. Nguyen D-Q, Webber C, Hehir-Kwa J, Pfundt R, Veltman J, Ponting CP. 2008. Reduced purify-

ing selection prevails over positive selection in human copy number variant evolution. Genome Res.

18(11):1711–23

85. Nielsen R. 2005. Molecular signatures of natural s election. Annu. Rev. Genet. 39:197–218

86. Nielsen R, Hubisz MJ, Clark AG. 2004. Reconstituting the frequency spectrum of ascertained single-

nucleotide polymorphism data.

Genetics 168(4):2373–82

87. Nielsen R, Hubisz MJ, Hellmann I, Torgerson D, Andr

es AM, et al. 2009. Darwinian and demographic

forces affecting human protein coding genes. Genome Res. 19(5):838–49

88. Nielsen R, Signorovitch J. 2003. Correcting for ascertainment biases when analyzing SNP data: appli-

cations to the estimation of linkage disequilibrium. Theor. Popul. Biol. 63(3):245–55

118 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

89. Nielsen R, Williamson S, Kim Y, Hubisz MJ, Clark AG, Bustamante C. 2005. Genomic scans for selective

sweeps using SNP data. Genome Res. 15(11):1566–75

90. Norton HL, Kittles RA, Parra E, McKeigue P, Mao X, et al. 2007. Genetic evidence for the convergent

evolution of light skin in Europeans and East Asians. Mol. Biol. Evol. 24(3):710–22

91. Orr HA. 1998. Testing natural selection versus genetic drift in phenotypic evolution using quantitative

trait locus data. Genetics 149(4):2099–104

92. Paun O, Bateman RM, Fay MF, Hedr

en M, Civeyrel L, Chase MW. 2010. Stable epigenetic effects

impact adaptation in allopolyploid orchids (Dactylorhiza: Orchidaceae). Mol. Biol. Evol. 27(11):2465–73

93. Pennings PS, Hermisson J. 2006. Soft sweeps II: molecular population genetics of adaptation from

recurrent mutation or migration. Mol. Biol. Evol. 23(5):1076–84

94. Pennings PS, Hermisson J. 2006. Soft sweeps III: the signature of positive selection from recurrent

mutation. PLoS Genet. 2(12):e186

95. P

erez-Morga D, Vanhollebeke B, Paturiaux-Hanocq F, Nolan DP, Lins L, et al. 2005. Apolipoprotein

L-I promotes trypanosome lysis by forming pores in lysosomal membranes. Science 309(5733):469–72

96. Perry GH, Dominy NJ, Claw KG, Lee AS, Fiegler H, et al. 2007. Diet and the evolution of human

amylase gene copy number variation. Nat. Genet. 39(10):1256–60

97. Perry GH, Melsted P, Marioni JC, Wang Y, Bainer R, et al. 2012. Comparative RNA sequencing reveals

substantial genetic variation in endangered primates. Genome Res. 22(4):602–10

98. Peter BM, Huerta-Sanchez E, Nielsen R. 2012. Distinguishing between selective sweeps from standing

variation and from a de novo mutation. PLoS Genet. 8(10):e1003011

99. Pickrell JK, Coop G, Novembre J, Kudaravalli S, Li JZ, et al. 2009. Signals of recent positive selection

in a worldwide sample of human populations. Genome Res. 19(5):826–37

100. Pollard KS, Salama SR, King B, Kern AD, Dreszer T, et al. 2006. Forces shaping the fastest evolving

regions in the human genome. PLoS Genet. 2(10):e168

101. Pollard KS, Salama SR, Lambert N, Lambot M-A, Coppens S, et al. 2006. An RNA gene expressed

during cortical development evolved rapidly in humans. Nature 443(7108):167–72

102. Prabhakar S, Noonan JP, P

abo S, Rubin EM. 2006. Accelerated evolution of conserved noncoding

sequences in humans. Science 314(5800):786

103. Pritchard JK, Di Rienzo A. 2010. Adaptation: not by sweeps alone. Nat. Rev. Genet. 11(10):665–67

104. Pritchard JK, Pickrell JK, Coop G. 2010. The genetics of human adaptation: hard sweeps, soft sweeps,

and polygenic adaptation. Curr. Biol. 20(4):R208–15

105. Prud’homme B, Gompel N, Rokas A, Kassner VA, Williams TM, et al. 2006. Repeated morphological

evolution through cis-regulatory changes in a pleiotropic gene. Nature 440(7087):1050–53

106. Przeworski M, Coop G, Wall JD. 2005. The signature of positive selection on standing genetic variation.

Evolution 59(11):2312–23

107. Ram

ırez-Soriano A, Nielsen R. 2009. Correcting estimators of θ and Tajima’s D for ascertainment biases

caused by the single-nucleotide polymorphism discovery process. Genetics 181(2):701–10

108. Richards EJ. 2011. Natural epigenetic variation in plant species: a view from the ﬁeld. Curr. Opin. Plant

Biol. 14(2):204–9

109. Riebler A, Held L, Stephan W. 2008. Bayesian variable selection for detecting adaptive genomic differ-

ences among populations. Genetics 178(3):1817–29

110. Romero IG, Ruvinsky I, Gilad Y. 2012. Comparative studies of gene expression and the evolution of

gene regulation. Nat. Rev. Genet. 13(7):505–16

111. Sabeti PC, Reich DE, Higgins JM, Levine HZP, Richter DJ, et al. 2002. Detecting recent positive

selection in the human genome from haplotype structure. Nature 419(6909):832–37

112. Sabeti PC, Schaffner SF, Fry B, Lohmueller J, Varilly P, et al. 2006. Positive natural selection in the

human lineage. Science 312(5780):1614–20

113. Sabeti PC, Varilly P, Fry B, Lohmueller J, Hostetter E, et al. 2007. Genome-wide detection and char-

acterization of positive selection in human populations. Nature 449(7164):913–18

114. Scheinfeldt LB, Soi S, Thompson S, Ranciaro A, Woldemeskel D, et al. 2012. Genetic adaptation to

high altitude in the Ethiopian highlands. Genome Biol. 13(1):R1

115. Sebat J, Lakshmi B, Troge J, Alexander J, Young J, et al. 2004. Large-scale copy number polymorphism

in the human genome. Science 305(5683):525–28

www.annualreviews.org

•

Detecting Natural Selection in Genomic Data 119

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47CH05-Sabeti ARI 29 October 2013 12:12

116. Shapiro BJ, Alm EJ. 2008. Comparing patterns of natural selection across species using selective signa-

tures. PLoS Genet. 4(2):e23

117. Shriver MD, Kennedy GC, Parra EJ, Lawson HA, Sonpar V, et al. 2004. The genomic distribution of

population substructure in four populations using 8,525 autosomal SNPs. Hum. Genomics 1(4):274–86

118. Simonson TS, Yang Y, Huff CD, Yun H, Qin G, et al. 2010. Genetic evidence for high-altitude adaptation

in Tibet. Science 329(5987):72–75

119. Smith JM, Haigh J. 1974. The hitch-hiking effect of a favourable gene. Genet. Res. 23(1):23–35

120. Smith NGC, Eyre-Walker A. 2002. Adaptive protein evolution in Drosophila. Nature 415(6875):1022–24

121. Stefansson H, Helgason A, Thorleifsson G, Steinthorsdottir V, Masson G, et al. 2005. A common

inversion under selection in Europeans. Nat. Genet. 37(2):129–37

122. Tajima F. 1989. Statistical method for testing the neutral mutation hypothesis by DNA polymorphism.

Genetics 123(3):585–95

123. Tajima F. 1993. Simple methods for testing the molecular evolutionary clock hypothesis. Genetics

135(2):599–607

124. Teshima KM, Coop G, Przeworski M. 2006. How reliable are empirical genomic scans for selective

sweeps? Genome Res. 16(6):702–12

125. Tishkoff SA, Reed FA, Ranciaro A, Voight BF, Babbitt CC, et al. 2007. Convergent adaptation of human

lactase persistence in Africa and Europe. Nat. Genet. 39(1):31–40

126. Tsetskhladze ZR, Canﬁeld VA, Ang KC, Wentzel SM, Reid KP, et al. 2012. Functional assessment of

human coding mutations affecting skin pigmentation using zebraﬁsh. PLoS ONE 7(10):e47398

127. Turner BM. 2009. Epigenetic responses to environmental change and their evolutionary implications.

Philos. Trans. R. Soc. Lond. B 364(1534):3403–18

128. Vanhamme L, Paturiaux-Hanocq F, Poelvoorde P, Nolan DP, Lins L, et al. 2003. Apolipoprotein L-I

is the trypanosome lytic factor of human serum. Nature 422(6927):83–87

129. Vitalis R, Dawson K, Boursot P. 2001. Interpretation of variation across marker loci as evidence of

selection. Genetics 158(4):1811–23

130. Vitti JJ, Cho MK, Tishkoff SA, Sabeti PC. 2012. Human evolutionary genomics: ethical and interpretive

issues. Trends Genet. 28(3):137–45

131. Voight BF, Kudaravalli S, Wen X, Pritchard JK. 2006. A map of recent positive selection in the human

genome. PLoS Biol. 4(3):e72

132. Wang ET, Kodama G, Baldi P, Moyzis RK. 2006. Global landscape of recent inferred Darwinian

selection for Homo sapiens. Proc. Natl. Acad. Sci. USA 103(1):135–40

133. Watterson GA. 1978. The homozygosity test of neutrality. Genetics 88(2):405–17

134. Wiener P, Pong-Wong R. 2011. A regression-based approach to selection mapping. J. Hered. 102(3):294–

305

135. Wright SI, Charlesworth B. 2004. The HKA test revisited. Genetics 168(2):1071–76

136. Yang, Bielawski. 2000. Statistical methods for detecting molecular adaptation. Trends Ecol. Evol.

15(12):496–503

137. Yokoyama S. 1983. Selection for the α-thalassemia genes. Genetics 103(1):143–48

138. Zeng K, Fu Y-X, Shi S, Wu C-I. 2006. Statistical tests for detecting positive selection by utilizing

high-frequency variants. Genetics 174(3):1431–39

139. Zeng K, Shi S, Wu C-I. 2007. Compound tests for the detection of hitchhiking under positive selection.

Mol. Biol. Evol. 24(8):1898–908

140. Zhai W, Nielsen R, Slatkin M. 2009. An investigation of the statistical power of neutrality tests based

on comparative and population genetic data. Mol. Biol. Evol. 26(2):273–83

141. Zhang C, Bailey DK, Awad T, Liu G, Xing G, et al. 2006. A whole genome long-range haplo-

type (WGLRH) test for detecting imprints of positive selection in human populations. Bioinformatics

22(17):2122–28

142. Zhuang Z, Gusev A, Cho J, Pe’er I. 2012. Detecting identity by descent and homozygosity mapping in

whole-exome sequencing data. PLoS ONE 7(10):e47618

120 Vitti

Grossman

Sabeti

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47-FrontMatter ARI 2 November 2013 9:9

Annual Review of

Genetics

Volume 47, 2013

Contents

Causes of Genome Instability

Andr´es Aguilera and Tatiana Garc´ıa-Muse ppppppppppppppppppppppppppppppppppppppppppppppppp1

Radiation Effects on Human Heredity

Nori Nakamura, Akihiko Suyama, Asao Noda, and Yoshiaki Kodama ppppppppppppppppppp33

Dissecting Social Cell Biology and Tumors Using Drosophila Genetics

Jos´e Carlos Pastor-Pareja and Tian Xu ppppppppppppppppppppppppppppppppppppppppppppppppppppp51

Estimation and Partition of Heritability in Human Populations Using

Whole-Genome Analysis Methods

Anna A.E. Vinkhuyzen, Naomi R. Wray, Jian Yang, Michael E. Goddard,

and Peter M. Visscher pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp75

Detecting Natural Selection in Genomic Data

Joseph J. Vitti, Sharon R. Grossman, and Pardis C. Sabeti pppppppppppppppppppppppppppppppp97

Adaptive Translation as a Mechanism of Stress Response

and Adaptation

Tao Pan ppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp121

Organizing Principles of Mammalian Nonsense-Mediated

mRNA Decay

Maximilian Wei-Lin Popp and Lynne E. Maquat ppppppppppppppppppppppppppppppppppppppp139

Control of Nuclear Activities by Substrate-Selective

and Protein-Group SUMOylation

Stefan Jentsch and Ivan Psakhye ppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp167

Genomic Imprinting: Insights From Plants

Mary Gehring pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp187

Regulation of Bacterial Metabolism by Small RNAs

Using Diverse Mechanisms

Maksym Bobrovskyy and Carin K. Vanderpool ppppppppppppppppppppppppppppppppppppppppppp209

Bacteria and the Aging and Longevity of Caenorhabditis elegans

Dennis H. Kim pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp233

The Genotypic View of Social Interactions in Microbial Communities

Sara Mitri and Kevin Richard Foster ppppppppppppppppppppppppppppppppppppppppppppppppppppp247

SIR Proteins and the Assembly of Silent Chromatin in Budding Yeast

Stephanie Kueng, Mariano Oppikofer, and Susan M. Gasser pppppppppppppppppppppppppppp275

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.

GE47-FrontMatter ARI 2 November 2013 9:9

New Gene Evolution: Little Did We Know

Manyuan Long, Nicholas W. VanKuren, Sidi Chen, Maria D. Vibranovski ppppppppppp307

RNA Editing in Plants and Its Evolution

Mizuki Takenaka, Anja Zehrmann, Daniil Verbitskiy, Barbara H¨artel,

and Axel Brennicke pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp335

Expanding Horizons: Ciliary Proteins Reach Beyond Cilia

Shiaulou Yuan and Zhaoxia Sun ppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp353

The Digestive Tract of Drosophila melanogaster

Bruno Lemaitre and Irene Miguel-Aliaga ppppppppppppppppppppppppppppppppppppppppppppppppp377

RNase III: Genetics and Function; Structure and Mechanism

Donald L. Court, Jianhua Gan, Yu-He Liang, Gary X. Shaw, Joseph E. Tropea,

Nina Costantino, David S. Waugh, and Xinhua Ji pppppppppppppppppppppppppppppppppppp405

Modernizing the Nonhomologous End-Joining Repertoire:

Alternative and Classical NHEJ Share the Stage

Ludovic Deriano and David B. Roth ppppppppppppppppppppppppppppppppppppppppppppppppppppppp433

Enterococcal Sex Pheromones: Signaling, Social Behavior,

and Evolution

Gary M. Dunny pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp457

Control of Transcriptional Elongation

Hojoong Kwak and John T. Lis ppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp483

The Genomic and Cellular Foundations of Animal Origins

Daniel J. Richter and Nicole King ppppppppppppppppppppppppppppppppppppppppppppppppppppppppp509

Genetic Techniques for the Archaea

Joel A. Farkas, Jonathan W. Picking, and Thomas J. Santangelo ppppppppppppppppppppppp539

Initation of Meiotic Recombination: How and Where? Conservation

and Speciﬁcities Among Eukaryotes

Bernard de Massy pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp563

Biology and Genetics of Prions Causing Neurodegeneration

Stanley B. Prusiner ppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp601

Bacterial Mg

Homeostasis, Transport, and Virulence

Eduardo A. Groisman, Kerry Hollands, Michelle A. Kriner, Eun-Jin Lee,

Sun-Yang Park, and Mauricio H. Pontes ppppppppppppppppppppppppppppppppppppppppppppppp625

Errata

An online log of corrections to Annual Review of Genetics articles may be found at

http://genet.annualreviews.org/errata.shtml

vi Contents

Annu. Rev. Genet. 2013.47:97-120. Downloaded from www.annualreviews.org

by Harvard University on 11/26/13. For personal use only.