Molecular Ecology - 2014 - Cruickshank

Molecular Ecology (2014) 23, 3133–3157 doi: 10.1111/mec.
12796
INVITED REVIEWS AND SYNTHESES

Reanalysis suggests that genomic islands of speciation
are due to reduced diversity, not reduced gene flow
T A M I E . C R U I C K S H A N K * and M A T T H E W W . H A H N * †
*Department of Biology, Indiana University, Bloomington, IN 47405, USA, †School of Informatics and Computing, Indiana
University, Bloomington, IN 47405, USA
Abstract
The metaphor of ‘genomic islands of speciation’ was first used to describe heteroge-
neous differentiation among loci between the genomes of closely related species. The
biological model proposed to explain these differences was that the regions showing
high levels of differentiation were resistant to gene flow between species, while the
remainder of the genome was being homogenized by gene flow and consequently
showed lower levels of differentiation. However, the conditions under which such dif-
ferentiation can occur at multiple unlinked loci are restrictive; additionally, essentially,
all previous analyses have been carried out using relative measures of divergence,
which can be misleading when regions with different levels of recombination are com-
pared. Here, we test the model of differential gene flow by asking whether absolute
divergence is also higher in the previously identified ‘islands’. Using five species pairs
for which full sequence data are available, we find that absolute measures of diver-
gence are not higher in genomic islands. Instead, in all cases examined, we find
reduced diversity in these regions, a consequence of which is that relative measures of
divergence are abnormally high. These data therefore do not support a model of differ-
ential gene flow among loci, although islands of relative divergence may represent loci
involved in local adaptation. Simulations using the program IMa2 further suggest that
inferences of any gene flow may be incorrect in many comparisons. We instead present
an alternative explanation for heterogeneous patterns of differentiation, one in which
postspeciation selection generates patterns consistent with multiple aspects of the data.
Keywords: FST, hitchhiking, IM, recombination, sympatric speciation
Received 11 December 2013; revision received 5 May 2014; accepted 7 May 2014
researchers have turned to large-scale sequencing of

Introduction
natural populations in order to identify regions
Understanding the genetic basis of speciation is a major involved in speciation. By considering naturally hybrid-
goal of evolutionary biology. However, identifying the izing taxa, studies of DNA polymorphism and diver-
loci responsible for reproductive isolation using linkage gence make it possible to quickly reveal many aspects
mapping is a challenging task, and only a handful of of the speciation process. Such studies hold the promise
genes have been directly implicated in either prezygotic of identifying the number and location of regions
isolation (e.g. Hopkins & Rausher 2011) or postzygotic involved in isolation, and possibly even the individual
isolation (e.g. Wittbrodt et al. 1989; Barbash et al. 2003; mutations underlying important differences between
Presgraves et al. 2003). Because traditional genetic map- populations (e.g. Turner et al. 2010). More intriguingly,
ping is so laborious – and may miss both loci of small despite the seemingly stringent requirements for evolv-
effect and loci contributing to phenotypes that are not ing new species in the face of gene flow (Felsenstein
identified as important to reproductive isolation – 1981), these studies could support widespread sympat-
ric or parapatric speciation.
Correspondence: Matthew W. Hahn, Fax: 812-855-6705; Accumulating analyses of recently diverged species
E-mail: [email protected] pairs have found highly heterogeneous patterns of
© 2014 John Wiley & Sons Ltd

1365294x, 2014, 13, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mec.12796 by CochraneArgentina, Wiley Online Library on [29/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3134 T . E . C R U I C K S H A N K and M . W . H A H N
genetic differentiation across the genome (e.g. Harr the evolutionary mechanisms that generate islands of
2006; Carneiro et al. 2010; Nadeau et al. 2012). Some differentiation.
regions show little genetic differentiation, while others
show high levels of differentiation and may even con-
Models of species and sequence divergence
tain fixed differences that distinguish species. Although,
in most cases, the highly differentiated regions consist For our purposes, models of species divergence can be
of only a small proportion of the genome (e.g. Turner usefully separated into those with and without gene
et al. 2005), in others a large fraction of the genome flow. For those scenarios including gene flow, we must
may either be differentiated (e.g. Garrigan et al. 2012) or further distinguish between models where there is no
show some association with key isolating traits (e.g. initial period of allopatry or reduced gene flow (i.e.
Michel et al. 2010). sympatric speciation) and those where gene flow occurs
A common inference when a small number of differ- after a substantial period of independent evolution
entiated regions are found by these genome scans is between diverging taxa (i.e. secondary contact).
that they represent loci involved in reproductive isola- Although both models are often lumped together as
tion or ecological specialization and that there is strong ‘speciation-with-gene-flow’, secondary contact models
selection against their introgression between species do not pose the same challenges for the evolution of
(Wu 2001). Loci not involved in isolation will experi- isolating barriers as do models of sympatric speciation
ence the homogenizing effects of migration and will (cf. Felsenstein 1981). In this study, we refer to these as
show little-to-no differentiation. In this scenario, small ‘primary’ and ‘secondary’ models of speciation-
regions of elevated divergence within a genome are sur- with-gene-flow and separately consider the predictions
rounded by regions that have low divergence, and this of each model as well as the data supporting those
difference is thought to reflect differences in the effec- predictions.
tive amount of gene flow experienced by the two types There is strong evidence supporting both primary
of loci. Because genes within regions of elevated diver- and secondary models of speciation-with-gene-flow in
gence may be involved in reproductive isolation nature. Convincing examples of sympatric speciation
between species, these regions have been referred to as have been identified, complete with heterogeneous dif-
‘genomic islands of speciation’ (Turner et al. 2005). This ferentiation among the limited number of loci examined
phrase has largely been replaced with the slightly less (e.g. Savolainen et al. 2006). Lower levels of differentia-
loaded terms ‘genomic islands of differentiation’ (Harr tion in sympatric vs. allopatric populations of separate
2006) and ‘genomic islands of divergence’ (Nosil et al. species have provided good examples of gene flow
2009), although the latter term can itself connote pat- between species initially separated by isolating barriers
terns of species differences not actually present in the (e.g. Kulathinal et al. 2009; Martin et al. 2013). In addi-
data (see below). While there are many conceptual pre- tion, studies of hybrid zones formed after secondary
decessors to the ideas represented by these terms (cf. contact have provided major insights into the resistance
Harrison 2012), they do succinctly capture both the con- of loci to introgression across such zones (e.g. Payseur
figuration of species differences and the presumed et al. 2004; Kronforst et al. 2006; Teeter et al. 2008; Maro-
underlying model. ja et al. 2009). These studies generally find a small num-
Here, we consider multiple biological models that can ber of differentiated loci and a large number of loci
explain the observed patterns of heterogeneous diver- with evidence of introgression. Results from taxa
gence, as well as the unique predictions made by each. experiencing both primary and secondary speciation-
Based on a reanalysis of published data sets, we find with-gene-flow suggest that we may be able to identify
little evidence to support models in which islands of the reproductive isolating genes underlying ‘islands of
divergence contain loci resistant to introgression. We speciation’.
therefore subsequently re-examine the basis for claims Assuming introgression between species, differentia-
that the species under consideration are exchanging tion is expected at loci where the coefficient of selection
genes. Despite multiple lines of evidence clearly sup- is greater than the rate of migration (Haldane 1930;
porting gene flow between many species pairs, we find Wright 1931; Bulmer 1972). While this would appear to
high rates of false positives when applying the most be promising for the identification of narrow windows
widely used tests for gene flow to closely related of differentiation, ‘divergence hitchhiking’ hypothesizes
species. Finally, we elaborate on an alternative model that divergent natural selection causes reduced effective
that can explain patterns of heterogeneous genomic gene flow near a selected locus during primary differ-
differentiation in terms of divergence without differen- entiation (Via & West 2008; Via 2009, 2012; Feder & No-
tial gene flow among loci. As more whole-genome data sil 2010); this idea is supported by theoretical work on
sets are collected, it will become possible to determine the effects of migration, selection and linkage (e.g.

R E A N A L Y S I S O F G E N O M I C I S L A N D S 3135
Charlesworth et al. 1997; Nordborg 1997; Akerman & of genome scans conducted across taxa. Due to the ease
B€urger 2014). Overtime, larger regions are ‘captured’ by with which patterns of molecular variation can be que-
the selected locus, and therefore, larger regions begin to ried, scans for outlier loci (i.e. those loci with higher-
show high levels of differentiation. Divergence hitchhik- than-expected differentiation) continue to be conducted
ing is expected to mainly affect small populations when in a wide variety of species, across different timescales
migration is low (Feder & Nosil 2010), and it predicts of species divergence (e.g. Nadeau et al. 2013) and
that differentiation should decline with distance from across varying geographic relationships (e.g. Renaut
the selected locus. As the speciation process proceeds, et al. 2013). Almost without exception, studies identify-
this model proposes that a progressively larger propor- ing outlier markers have concluded that they represent
tion of the genome will become differentiated, finally loci underlying isolating barriers or locally adapted
resulting in two completely distinct genomes (Feder alleles and that they are resistant to gene flow that
et al. 2012). A similar model applies to cases of second- homogenizes the majority of the genome. The conclu-
ary contact, where neutral alleles with no effect on sion reached by many researchers has therefore been
fitness but that are linked to selected alleles will also that speciation-with-gene-flow (in this case, sympatric
become, or remain, differentiated across a hybrid zone speciation) is not only possible, but potentially wide-
(Barton & Bengtsson 1986). Overall, for models of both spread (e.g. Nosil 2008; Via 2012). A growing body of
primary and secondary speciation-with-gene-flow, pat- theoretical work has arisen based on the idea that
terns of heterogeneous differentiation have been islands of differentiation are resistant to introgression
interpreted as revealing loci resistant to introgression, (e.g. Feder & Nosil 2010; Flaxman et al. 2012, 2013), and
with the size of differentiated regions correlated with wholly constructed histories of the speciation process
the strength of selection and the stage of species have been proposed based on the same interpretation
divergence. of the data (Feder et al. 2012). The apparent widespread
Heterogeneous patterns of genomic differentiation acceptance of sympatric speciation based on patterns of
can also be explained by an alternative model in which molecular differentiation argues strongly for the close
there is no gene flow between recently diverged species examination of the various evolutionary forces that
(Noor & Bennett 2009; Turner & Hahn 2010; Hahn et al. could produce such patterns.
2012). Under this model, reproductive isolation is In the next section, we consider a further prediction
instantaneous and complete, although similar models of both speciation-with-gene-flow models, namely that
do not require the complete absence of introgression. absolute measures of species divergence (as opposed to
Species pairs show low overall levels of differentiation relative measures, see next section) should also be
because they have only recently split, and shared alleles higher in regions resistant to gene flow. Surprisingly,
are due to ancestrally inherited variation, not introgres- absolute measures have either not been considered by
sion. Heterogeneity in the level of differentiation among most researchers (but see Noor & Bennett 2009; Nach-
loci in this model is due to stochastic variation in coa- man & Payseur 2012), or relative measures were mistak-
lescent times (Barton 2006) and, more notably, natural enly thought to be absolute (a mistake one of us has
selection: those loci experiencing strong selection will made; Turner et al. 2005). We believe that comparisons
appear to be more differentiated and will share less among loci using absolute measures of divergence will
ancestral variation (see next section). The targets of nat- be especially revealing with respect to the contribution
ural selection may be directly involved in the ecological, of each mode of divergence between species.
morphological or behavioural specialization of each spe-
cies, or they may be unrelated to any trait involved in
Interpreting measures of sequence divergence
species divergence and simply represent the ‘back-
ground’ level of selection found in any organism. Most
Relative measures of divergence
importantly, however, in this model, there is no differ-
ential gene flow among loci, and the regions of highest There are a number of widely used statistics that enable
differentiation do not necessarily indicate the location researchers to quantify the genetic distance between spe-
of genes underlying isolating traits (‘incidental islands’: cies, either at individual sites or averaged over multiple
Turner & Hahn 2010; White et al. 2010). sites. One common class of measures is based on
Determining the contribution of these models – two Wright’s FST (Wright 1931, 1943), which measures the
with gene flow helping to outline the boundaries of normalized difference in allele frequencies between
regions involved in speciation and one with selection populations. There are many different ways of calculat-
defining targets of adaptation that are not necessarily ing FST and its related statistics (Charlesworth 1998;
involved in the speciation process – has important con- Excoffier 2007), and many different ways of interpreting
sequences for the interpretation of the growing number these statistics (Holsinger & Weir 2009). Regardless of

the specific way in which FST is calculated, the key aspect

haplotype from X and the jth haplotype from Y (Nei
that all FST-like measures share in common is that their
& Li 1979; Nei 1987, equation 10.20). The statistic dXY
values are inflated when diversity within populations is
is not affected by current levels of within-population
low: loci with lower levels of within-species variation are
diversity, although it is affected by both ancestral
expected to show higher values of FST (Nei 1973;
levels of diversity (see below) and the substitution
Charlesworth et al. 1997; Charlesworth 1998; Jakobsson
rate. For instance, coding regions will often show
et al. 2013). Because of this dependence on levels of
lower levels of absolute divergence simply because
within-species polymorphism, FST is referred to as a ‘rel-
there are fewer possible neutral mutations; con-
ative’ measure of divergence. Additional relative mea-
versely, at loci with a large number of substitutions
sures of divergence (da and df) are described in Box 1.
fixed by positive selection, dXY may actually be
higher.
There are also a number of relative measures of
Box 1. Measures of sequence divergence differentiation based on dXY that are in wide usage.
There are multiple different ways to measure Nei & Li (1979) defined a measure of the ‘net’ nucle-
sequence divergence when more than just a single otide differences between two populations, da (which
individual is sampled from each population or spe- they called d), as:
cies. One of the most widely used statistics is FST, a
da ¼ dXY ðpX þ pY Þ=2:
normalized measure of allele frequency differences
between populations (Wright 1931, 1943). There are This statistic (also commonly called Da or Dm) is
many different ways to calculate FST and FST-like sta- intended to capture only the differences that have
tistics (Charlesworth 1998; Excoffier 2007), although accumulated between populations since they split
many of the subtle distinctions between them are not (Fig. 1). It does so by subtracting out the differences
relevant here. A common way to calculate FST is: that had accumulated before this split, assuming that
the level of ancestral variation was equal to the aver-
pT pS
FST ¼ age of the variation found in the two current popula-
pT
tions. da is therefore a relative measure of divergence
where pT is the expected heterozygosity in the total because it can be strongly affected by the amount of
sample and pS is the average expected heterozygosity within-population variation. The number of fixed dif-
in each population. [Note that this statistic is some- ferences between populations or species, df, is a com-
times called cST (Nei 1982), and for single sites is plex function of the time since divergence between
equivalent to GST (Nei 1973).] One very important populations and the time to coalescence within each
aspect of all FST-like statistics is that they are population (Hey 1991) but is also a relative measure
strongly influenced by within-population levels of of divergence due to its reliance on levels of within-
variation – represented by pS in the above formula – population variation.
and because of this, we refer to them as relative mea- Finally, note that dXY and da are more informative
sures of differentiation. for full sequence data; that is, for data in which at
To get around the dependence on within-popula- least moderate length sequences have been collected
tion variation, Nei & Li (1979) calculated the average from homologous loci in two species. dXY can be cal-
number of pairwise differences between sequences culated for a single site, but it no longer has the
from two populations, excluding all comparisons property of being an absolute measure and offers lit-
between sequences within populations. Here, we tle precision. This is most easily demonstrated by
refer to this statistic as dXY, although it is also considering the expectation and variance for dXY
referred to in the literature as pXY (Nei & Li 1979), under a strictly allopatric model (Nei 1987, equation
DXY (Nei 1987) and pB (Charlesworth 1998). This 13.83):
absolute measure of differentiation is independent of EðdXY Þ ¼ 2lt þ hAnc
the levels of diversity within the two populations
being compared and is calculated as: ð2lt þ hAnc Þ
VarðdXY Þ ¼ þ hAnc ;
X L
dXY ¼ xi yj dij
where l is the neutral mutation rate, t is the time
ij
since the species split, hAnc is the level of diversity in
where, in two populations, X and Y, dij measures the the ancestral (presplit) population, and L is the
number of nucleotide differences between the ith length of locus being considered. We can see that the

to higher between-population differentiation (Kaplan

variance in dXY goes down with increasing sequence
et al. 1989; Nordborg et al. 1996; Slatkin & Wiehe 1998).
length, such that we can have more confidence in
As a consequence, variation in the effects of selection
estimates that come from longer loci. This means that
along a chromosome can produce islands of relative
we can calculate dXY from reduced representation
divergence even in the complete absence of gene flow.
methods that produce sequences of hundreds of nu-
Accordingly, presumed speciation islands are often
cleotides, such as RAD-seq (Baird et al. 2008). How-
found in centromeric or rearranged regions where
ever, these expressions also indicate that for
recombination is reduced (e.g. Turner et al. 2005;
estimates of dXY from single SNPs, the variance is
Kulathinal et al. 2009). Because higher levels of relative
necessarily greater than the expectation, so that it
divergence in regions of low recombination are a pre-
will be difficult to differentiate values of dXY among
diction of both speciation-with-gene-flow models and
loci. All of this implies that for many kinds of mark-
models without gene flow (Nachman & Payseur 2012),
ers – SNPs, AFLPs, microsatellites – only relative
neither the presence nor the location of islands alone
measures of divergence (i.e. only FST) will be infor-
can support one model over the other. Instead, we can
mative, and therefore, that no comparison between
use predictions about the behaviour of alternative
relative and absolute measures can be made for these
measures of species divergence.
markers. Unfortunately, SNPs, AFLPs and microsat-
ellites are still the most common types of markers
used in studies of between-species divergence. When Absolute measures of divergence
full sequence data are collected, relative measures
To circumvent the dependence of relative measures of
including FST, da and df can be compared to dXY.
divergence on within-species polymorphism, we can
calculate the average number of pairwise differences
between sequences from two species, excluding all com-
Because relative measures of divergence rely on parisons between sequences within species (Fig. 1;
within-species diversity, markers sampled from parts of Box 1); we refer to this statistic as dXY. It does not mat-
the genome with more or less diversity will provide ter exactly how many chromosomes are sampled in
very different views on levels of differentiation. For each species, as all chromosomes from one species will
instance, FST from regions of reduced recombination – show a similar level of divergence to any chromosome
which often have reduced diversity due to linked selec- sampled from the other species. In fact, even if only a
tion – will be higher than FST from regions of normal single sequence is sampled from each species – as is
recombination for no other reason than that total levels commonly done when calculating species divergence
of nucleotide diversity are different (Charlesworth et al. using the statistic d – this is still an estimate of dXY (Nei
1997; Charlesworth 1998; Noor & Bennett 2009; Nach- 1987). Because dXY is independent of the levels of diver-
man & Payseur 2012). As a consequence, Charlesworth sity within the two populations or species being com-
(1998) recommended that relative measures of pared, we refer to it as an ‘absolute’ measure of
differentiation ‘are not necessarily appropriate if we divergence.
wish to compare loci with very different levels of Absolute measures of divergence capture the number
within-population variation’. Differences in levels of of sequence differences at a locus since the most recent
variation are not always associated with rates of recom- common ancestor (MRCA) of the sampled chromo-
bination, although this is a common correlate (Hahn somes (Fig. 1). Because the MRCA almost certainly
2008; Cutter & Payseur 2013). existed in the ancestral population before the split into
With respect to the interpretation of genomic islands, the two descendent species, dXY measures both the dif-
these considerations suggest that regions may differ in ferences that have accumulated since the species split
values of FST (or other relative measures of divergence) and the differences that were present at the time of the
not because of differences in the amount of gene flow, split (Gillespie & Langley 1979). As such, variation in
but simply because of differences in levels of within- dXY can be due to variation in levels of ancestral poly-
species polymorphism. Extreme reductions in within- morphism, but is not affected by variation in levels of
species polymorphism at specific loci can be due to current polymorphism. Consequently, this means that
selection directly on adaptive traits encoded by such linked selection in extant populations will not cause the
regions, or due to linked selection on either advanta- appearance of ‘islands’ of high divergence in dXY,
geous (‘hitchhiking’; Maynard Smith & Haigh 1974) or although linked selection in ancestral populations can
deleterious mutations (‘background selection’; Charles- actually lower the value of dXY (Begun et al. 2007).
worth et al. 1993). Both background selection and To see how absolute and relative measures can give
hitchhiking reduce within-population diversity and lead dissimilar results in the presence of linked selection,

(a) (b)
speciation
df
df da
dXY da dXY
X, Y
X, Y
A B C D A B C D
Population X Population Y Population X Population Y
Fig. 1 Relative and absolute measures of divergence, with the effect of linked selection. Demonstrating the differences between pX,
pY, da, df and dXY. Panels (a) and (b) both show example genealogies relating four sampled chromosomes (A, B, C and D) from two
populations or species (X and Y). The statistics pX and pY measure the average number of nucleotide differences between samples in
population X and Y, respectively. dXY measures the average number of nucleotide differences between each sample in population X
and each sample in population Y, with no comparisons made within a population. da uses the average current levels of polymor-
phism as a measure of ancestral polymorphism, and subtracts this value from the total divergence (= dXY [pX + pY]/2). df represents
the total number of fixed differences between the two populations. For ease of comparison, the genealogies in the two panels have
the same height, as do the genealogies for populations X and Y within each figure. The important distinction between panels a) and
b) is that due to linked selection, there is a difference in pX, pY, da and df between the two panels, but no difference in dXY.
consider the two genealogies shown in Fig. 1. Figure 1a we take advantage of the theory laid out in Charles-
shows one hypothetical genealogical history of two spe- worth et al. (1997). This mathematical framework allows
cies for a locus in a region that has not experienced any us to make predictions not only about divergence at the
selection. Levels of within-species diversity are unaf- selected site, but also at linked neutral alleles for mod-
fected by linked selection and are therefore high rela- els in which there is gene flow between species, via
tive to the between-species portion of their history. In either primary divergence or secondary contact. Sur-
this case, p (a measure of within-population diversity) prisingly, even though the theoretical machinery pre-
is high in both species, and da and df, two measures of sented in Charlesworth et al. (1997) has been used in
between-species relative divergence similar to FST multiple studies to elaborate on patterns of relative
(Box 1; Fig. 1), are commensurately low because they divergence expected at equilibrium (e.g. Via & West
are dependent on within-species variation. Figure 1b, 2008; Nosil et al. 2009; Feder & Nosil 2010), to our
on the other hand, shows a hypothetical genealogical knowledge it has not been used to clarify predictions
history for a locus affected by linked selection in both about absolute divergence.
species (either hitchhiking or background selection), It is relatively easy to derive expectations for dXY
sampled from exactly the same individuals as in under this model directly from the equations presented
Fig. 1a. The time to the most recent common ancestor in Charlesworth et al. (1997). The within- and among-
within populations in this scenario is extremely recent population components of sequence diversity at neutral
due to linked selection. This means that p in each spe- polymorphisms linked to a locus under selection in a
cies is low and that da and df are commensurately high subdivided population are summarized in Table 1 of
(FST would show the same pattern). Most importantly, Charlesworth et al. (1997) and in Nordborg (1997). We
dXY is exactly the same in the two panels. In this way, make a number of the same simplifying assumptions as
relative and absolute measures can provide very differ- in the original papers (including large equally sized de-
ent pictures about the level of between-species diver- mes with symmetric migration) and sum the compo-
gence from locus to locus, simply due to linked nents [i.e. the variation within populations (Nordborg
selection. 1997; equation 65) and the expected neutral site differ-
entiation among populations (Charlesworth et al. 1997;
Table 1)] to find that:
The effect of gene flow on measures of divergence
r þ 4m~
To demonstrate the clear predictions of models with EðdXY Þ ¼ ð1 qÞa þ
~
8Ne mr
gene flow on both relative and absolute divergence at
loci that are resistant to gene flow (possibly because Here, q is the frequency of a locally deleterious allele,
alternative alleles confer a locally adapted phenotype), and variation is reduced due to background selection in

Table 1 Nucleotide divergence in islands and nonislands for five pairs of recently diverged taxa
References for sequence

Islands Nonislands Bin type data
Oryctolagus cuniculus dXY 0.0368 0.0409 N = 27,17; P = 0.40 X/A Carneiro et al. (2009, 2010);
cuniculus and O. c. algirus FST 0.4401 0.1456 P = 0.0002 Geraldes et al. (2006)
dXY 0.0358 0.0411 N = 23,21; P = 0.91 Migration*
FST 0.615 0.103 P < 0.0001
Mus musculus musculus and dXY 0.007 0.0079 N = 8,19; P = 0.75 High/low r Geraldes et al. (2011)
M. m. domesticus FST 0.7762 0.6931 P = 0.21
dXY 0.0105 0.008 N = 8,6; P = 0.47 High/low FST Harr (2006)
FST 0.6387 0.2417 P = 0.0002
Heliconius melpomene aglaope dXY 0.0145 0.0155 (B/D region) 5 islands; P = 0.80 High/low FST Nadeau et al. (2012)
and H. m. amaryllis FST 0.4284 0.184 P < 0.0001
dXY 0.0197 0.0140 (Yb region) 7 islands; P = 0.87 High/low FST
FST 0.3422 0.132 P < 0.0001
Anopheles coluzzii (M form) dXY 0.006 0.0096 N = 17,12; P = 0.17 high/low FST Turner et al. (2005);
and A. gambiae (S form) FST 0.5242 0.2138 P = 0.008 Turner & Hahn (2007);
White et al. (2010)
Ficedula albicollis and dXY 0.0036 0.0044 ~50 islands; N/A High/low FST Ellegren et al. (2012)
F. hypoleuca FST 0.742 0.357 N/A
*Based on results Sousa et al. (2013).
proportion to a (see equation 24 in Nordborg 1997). The
1.0
d XY
5
effective migration rate, m,~ accounts for differences in da

F ST
fitness between migrants and nonmigrants (see equation
0.8
A3 in Charlesworth et al. 1997). This expression for dXY
4
is equivalent to the average pairwise divergence

between chromosomes sampled from two subpopula-
0.6
3
tions. Expectations for da in this scenario can be easily

dXY/a
FST
derived from the expectation for dXY. Equations for cal-
culating all statistics from data can be found in Box 1,
0.4
2
as can the expectation for dXY under a purely allopatric

model.
0.2
1
As Fig. 2 shows, under this model at equilibrium FST,

da and dXY are expected to be high at loci that do not
introgress and to decline sharply as one moves away
0.0
0
from the selected locus. Gene flow reduces both abso-

0.000 0.002 0.004 0.006 0.008 0.010
lute and relative measures of species divergence, lead- r
ing to a clear difference in dXY at loci that experience
different levels of effective gene flow. Notably, the mag- Fig. 2 Analytical predictions for divergence with gene flow.
nitude of differences in both FST and dXY between intro- Analytical predictions for absolute divergence (dXY) and rela-
tive divergence (FST and da) for neutral sites linked to a
gressing and nonintrogressing regions is of the same
selected site in a subdivided population. The selected site lies
scale when there is differential gene flow, so that loci all the way at the left end of the x-axis, with increasing recom-
with relative divergence that is ten times higher than bination distance (r) moving away from this site. The popula-
background are also expected to have absolute diver- tions are assumed to have a total size of Ne = 5000, with
gence that is this much higher. The slight difference migration among demes occurring at rate m = 0.0001. The left-
observed in Fig. 2 in the rate at which FST and dXY drop hand y-axis measures the values of dXY and da, while the right-
off with distance from the selected site is not consistent hand y-axis measures the values of FST. All measures are
expected to decrease quickly with distance from a selected
across all combinations of parameters and cannot be
locus (s = 0.01).
expected to be a distinguishing characteristic of these
two measures. When populations are not at equilibrium results showing the differences in these statistics with
– such as shortly after speciation – expectations for FST and without gene flow when loci are sampled shortly
and dXY are more complex. Box 2 contains simulation after a species split.

Box 2. Statistical power to detect differences in effective migration among loci.

An important assumption of the comparisons made in the main text is that we have the same statistical power to
detect regions with elevated relative and absolute divergence. While Fig. 2 suggests that this is true for populations
that have reached an equilibrium between migration, drift and selection, this will not necessarily be the case shortly
after a speciation event because populations will not be at equilibrium. Lacking theoretical expectations for non-
equilibrium conditions of primary speciation-with-gene-flow, Fig. B1 shows the results for different measures of
divergence from simulated populations shortly after speciation (where the species split happened instantaneously
at time 0). In each simulation, we measured relative and absolute divergence at loci with and without migration
between the two species. The results from the simulations lead to several important conclusions about our ability
to detect differences in the effective migration rate among loci.
First, in all conditions, those loci not experiencing gene flow between species diverge faster than those loci that
are being homogenized by gene flow, as expected. However, in all cases, we also lack power to distinguish
between these two classes of loci for a period of time after speciation (Fig. B1; the vertical line in each panel marks
the position where 95% confidence intervals between these classes do not overlap). In the best-case scenario (using
FST when there is high background migration), we are unable to identify those loci that are not introgressing
between species until ~0.1 Ne generations after the speciation event. This can be a substantial amount of time, even
for species with relatively small effective population sizes.
Fig. B1 Divergence, estimated by the statistics FST (top) and dXY (bottom), is plotted as a function of time as two populations split
(measured in Ne generations). Loci are sampled from populations simulated with zero (Nem = 0) and nonzero migration rates
(Nem = 1 on the left, or Nem = 5 on the right). Within each condition and for each measure, divergence with and without migra-
tion are indicated by grey and black lines, respectively. The width of each line represents the 95% confidence interval for 500
simulated data sets, and the vertical line within each panel shows the point in time at which there is no overlap between the con-
fidence intervals with and without migration. At this point, loci with and without migration would be expected to show signifi-
cantly different values of the relevant statistic.

Second, and as noted in the main text, in non-equilibrium populations, there is lower power to detect differences
among loci using dXY than there is using FST for very short divergence times (Fig. B1; compare top and bottom
panels in each column). This means that FST is more sensitive to a lack of gene flow than is dXY and will begin to
differ between loci with and without migration more quickly after speciation. The lag is due to the fact that muta-
tions must arise in order for dXY to increase, while FST only requires changes in allele frequencies. If there are loci
that do show an increase in FST due to a lack of gene flow, these results imply that they will not necessarily also
show a statistically significant difference in dXY. Note, however, that on average, these loci will show a higher
value of dXY (significant or not) and are not expected to have significantly lower levels of polymorphism
(cf. Tables 1 and 2).
Finally, the simulations also demonstrate that there is more power to distinguish between loci with and without
migration when background migration rates are high (Fig. B1; compare left and right panels in each row). That is,
differences between these classes of loci can be detected much more quickly after the species split with high gene
flow. In addition, the difference in the power of FST and dXY is magnified under conditions of high migration, with
a greater time period expected in which the relative measure (FST) is significant and the absolute measure (dXY) is
not.
In the previous section, we demonstrated that there are two contrasting explanations for islands of
different expectations for the levels of relative and abso- differentiation by examining both relative and absolute
lute measures of divergence in a strictly allopatric model measures of divergence in five systems that have
with linked selection. In contrast, models of primary and become widely cited as examples of speciation-with-
secondary speciation-with-gene-flow make similar pre- gene-flow (Table 1). Three of the species pairs are clear
dictions about all measures of divergence (see Table 1 in examples of gene flow after secondary contact (Ficedula,
Nachman & Payseur 2012). If islands represent regions Mus, Oryctolagus), while two are hypothesized to be
resistant to introgression that are surrounded by loci examples of primary speciation-with-gene-flow (Anophe-
homogenized by gene flow because they are not involved les, Heliconius). Although recent studies have also
in species isolation, then regions of high absolute diver- reviewed the evidence for islands using absolute diver-
gence should co-occur with regions of high relative gence in a subset of these species (Noor & Bennett 2009;
divergence. The convergence of measures that are and Nachman & Payseur 2012), here, we have used additional
are not affected by within-species diversity is a clear pre- data and additional methods for identifying islands.
diction of the speciation-with-gene-flow models, one that The data sets vary in the technologies used for
eliminates the possibility that reduced intraspecific varia- sequencing and in the proportion of the genome cov-
tion has created peaks of divergence independent of the ered. For one system (Ficedula), a whole-genome data
speciation process. set is available, with multiple individual birds
sequenced separately from each species (Ellegren et al.
2012). For another system (Heliconius), the sequences
Re-analysis of published sequence data
come from a few large regions of the genome that were
obtained by sequence-capture technology (Nadeau et al.
Patterns of sequence divergence do not support
2012). Sequences from moderate numbers of indepen-
differential introgression among loci
dently sequenced loci were available for Anopheles
The fact that linked selection affects relative measures of (Turner et al. 2005; Turner & Hahn 2007; White et al.
divergence but does not affect absolute measures of 2010), Mus (Harr 2006; Geraldes et al. 2011) and Oryctol-
divergence – and that the expectations are the same for agus (Geraldes et al. 2006; Carneiro et al. 2009, 2010).
all measures under a model with differential gene flow In each data set considered, previous analyses have
across loci – suggests a straightforward way to distin- identified islands of differentiation (or simply distinct
guish which model has had the greatest contribution to loci that were highly differentiated) based on relative
species divergence. With few exceptions (Noor & Ben- measures of divergence – either FST, da, df, or all three
nett 2009; Nachman & Payseur 2012), this difference has together – even if absolute measures of divergence were
not been examined in the many studies assessing levels also used (e.g. Carneiro et al. 2010; Geraldes et al. 2011;
of differentiation between closely related species. In part Ellegren et al. 2012). Using a definition of ‘islands’
this is due to the fact that absolute divergence is not based on the boundaries or divisions identified in the
informative for many kinds of markers (see Box 1), and original studies, we calculated absolute divergence
may also simply be due to the growing acceptance of (dXY) for the same regions. In none of the five species
speciation-with-gene-flow models. Here, we evaluate the pairs was absolute divergence significantly higher in

the islands of relative divergence than it was in

Anopheles (mosquitoes)
matched flanking regions or the genome as a whole
(Table 1). Despite the many-fold higher values of rela- Patterns of differentiation between the ‘M’ and ‘S’ form
tive divergence reported for these regions (Table 1), in mosquitoes in the species Anopheles gambiae (since
four of the five studies absolute divergence was in fact renamed A. coluzzii and A. gambiae, respectively; Coet-
lower in the islands than in nonislands, a result that is zee et al. 2013) were some of the first to be examined
never predicted by models that interpret the islands genomewide (Wang et al. 2001; Stump et al. 2005;
simply as regions resistant to introgression between Turner et al. 2005). Based on relative measures of diver-
hybridizing species. On the contrary, lower values of gence, at least three islands of divergence are consis-
dXY likely indicate that the regions identified as islands tently found across populations, and these coincide
have been experiencing recurrent bouts of selection with pericentromeric regions (Stump et al. 2005; Turner
since before the species split (see below). The fact that et al. 2005; Turner & Hahn 2007; Neafsey et al. 2010;
many of these islands of relative divergence also coin- White et al. 2010).
cide with regions of low recombination only serves to Here, we have focused on reanalyzing the data from
emphasize this expected pattern of low absolute diver- loci sequenced in three earlier studies (Table 1; Turner
gence (see below, and Begun et al. 2007). et al. 2005; Turner & Hahn 2007; White et al. 2010).
While a global view of sequence divergence is infor- [Upon reanalysis of whole-genome data from Anopheles
mative about the contributions of models of speciation (Lawniczak et al. 2010), we did not find patterns of rela-
(or at least models of introgression among loci), examin- tive sequence divergence supporting the previously
ing individual species pairs may offer a more nuanced identified islands and therefore did not use these data.]
understanding of what drives species divergence. Below, As shown in Fig. 3a, while FST in the islands is higher
we discuss what we consider to be key features – or than nonislands – as expected by the definition of
important caveats – in interpreting patterns of sequence islands – the values of dXY are not. Absolute divergence
evolution in each of the systems re-analysed here. is lower in the pericentromeric islands (Table 1), as is
(a) (b) (c)
(d) (e) (f)
Fig. 3 Reanalysis of published data sets. Box plots are drawn from two measures of divergence – FST (left) and dXY (right) – for genes
sampled from pairs of recently diverged taxa. For each measure and each species pair, a comparison is made between islands and
nonislands. Islands are regions of high FST and are defined differently for each system (see text for details). For panel d), only a sin-
gle genome-wide measure was available for islands. Central horizontal lines represent medians.

expected in regions of low recombination that experi-

Heliconius (butterflies)
ence recurrent selection. As shown here (Table 2) and
reported earlier from SNP-chip data (Neafsey et al. Butterflies in the genus Heliconius have undergone a
2010), there is a strong reduction in polymorphism in recent radiation that has created a large number of clo-
islands of relative divergence. This decrease in within- sely related species and races (Jiggins 2008). Heliconius
species polymorphism is expected under models of butterflies have conspicuous colour patterns and behav-
linked selection and will cause increased values of rela- iours that differ between species, but previous phyloge-
tive divergence measures unrelated to levels of effective netic analyses have revealed the adaptive introgression
gene flow. of colour-pattern genes even between nonsister species
Conclusions about the level of ongoing gene flow within the genus (Pardo-Diaz et al. 2012; The Heliconius
between M and S are mixed, with clear evidence of Genome Consortium 2012). Peaks of relative differentia-
hybridization (though at very low levels; Tripet et al. tion between two races in the species H. melpomene,
2001), including the introgression of highly advanta- H. m. amaryllis and H. m. aglaope were found in two
geous insecticide resistance alleles (Djogbenou et al. genomic regions that contain these colour-pattern genes
2008; Etang et al. 2009), but also a large deficit of recom- (Nadeau et al. 2012, 2013). The low values of FST in
binant genotypes (White et al. 2010). Whether M and S flanking regions surrounding these loci and across the
represent largely nonhybridizing species at an advanced genome have been interpreted as evidence for gene
stage of reproductive isolation (Lawniczak et al. 2010; flow, and the small size of selected regions indicates
Turner & Hahn 2010; White et al. 2010; Hahn et al. particularly high gene flow according to the divergence
2012) or are still exchanging genes across most of the hitchhiking hypothesis (Nadeau et al. 2012).
genome (Reidenbach et al. 2012; Lee et al. 2013), these Using targeted resequencing data from two known
results do not support a model in which islands of colour-pattern loci (Nadeau et al. 2012), we calculated
differentiation represent loci that have been resistant to dXY across both regions. As can be seen in Fig. 3b, c,
introgression over long periods of time. dXY is not higher within the peaks of FST relative to
Table 2 Nucleotide diversity in islands and nonislands for five pairs of recently diverged taxa
p islands p nonislands Bin type
N = 27,17 X/A
Oryctolagus c. cuniculus 0.004238 0.006328 P = 0.039
O. c. algirus 0.003772 0.006171 P = 0.013
N = 23,21 Migration*
O. c. cuniculus 0.003459 0.006436 P = 0.0018
O. c. algirus 0.002993 0.006404 P = 0.0002
N = 8,19; High/low r
Mus m. musculus 0.00087 0.00199 P = 0.025
M. m. domesticus 0.00139 0.00244 P = 0.16
N = 8,6; High/low FST
M. m. musculus 0.000625 0.002916 P = 0.022
M. m. domesticus 0.0015 0.003867 P = 0.04
5 islands; High/low FST
Heliconius m. aglaope (Yb) P = n.s.†
H. m. amaryllis (Yb) P < 0.05†
7 islands; High/low FST
H. m. aglaope (B/D) P < 0.05†
H. m. amaryllis (B/D) P < 0.05†
N = 17,12; High/low FST
Anopheles coluzzii (M form) 0.00338 0.01267 P = 0.195
A. gambiae (S form) 0.00529 0.01866 P = 0.123
50 islands, High/low FST
Ficedula albicollis 0.00067 0.00219 P < 0.001†
F. hypoleuca 0.00132 0.0037 P < 0.001†
All data come from the same references as in Table 1.

*Based on results from Sousa et al. (2013).
†
As reported in the original paper.

flanking regions. Although the precise genomic coordi- pairs that are further along in the speciation process.
nates of the divergence islands were not specified – Even if the races of H. melpomene analysed here are con-
they were defined as regions with FST > 0.3 – we sidered separate species, they are almost certainly too
attempted to match the regions of maximal relative closely related to show significant signs of restricted
divergence shown as grey shaded bars in Fig. 2 of Na- gene flow by either FST or dXY (Box 2).
deau et al. (2012). In neither the HmB/D region nor the
HmYb region is dXY significantly higher in islands than
Ficedula (flycatchers)
nonislands (Table 1), and in fact dXY is lower in the
islands of relative divergence for the B/D region. Within Divergence between the collared flycatcher (F. albicollis)
both races, nucleotide diversity is lower in islands com- and the pied flycatcher (F. hypoleuca) was previously
pared with nonislands for the HmB/D regions; in HmYb, assessed by sequencing the genomes of 10 males from
diversity is lower in islands in only one of the two races each species (Ellegren et al. 2012). These species are
considered (H. m. amaryllis; Table 2). Absolute diver- only very recently split, with genome-wide estimates of
gence was higher overall in these two regions compared the average pairwise divergence between species
to three unlinked regions (Nadeau et al. 2012), but it is (dXY = 0.0046) only marginally higher than the average
not clear exactly what this indicates. FST ranges from pairwise differences within species (p = 0.0036 and
~0.1 to >0.7 in the HmB/D region (which is more than p = 0.0021 for F. albicollis and F. hypoleuca, respectively;
700 kb long), and the interpretation of such a pattern in Ellegren et al. 2012). Despite this recent split, there are
a speciation-with-gene-flow context is that there is dif- strong premating isolating barriers and almost complete
ferential gene flow among loci within this region. Dif- postzygotic isolation resulting from ‘prolonged periods
ferences in absolute divergence between very large of allopatric isolation’ (Sætre & Sæther 2010). Any gene
regions of the genome may indicate different muta- flow would therefore result from secondary contact
tional environments experienced by each region, but between these species.
further exploration of this pattern will require genome- Using relative measures of divergence, there are
wide data. strong signals of islands of differentiation between these
One further issue with interpreting the data from species, with more than 50 islands where FST > 0.5
these two races is whether this comparison relates to (Ellegren et al. 2012). Values of FST are more than twice
speciation at all. There is strong geographic structure as high in islands of divergence as the genomic back-
involving the wing colour patterns that define these ground (Table 1; Fig. 3d), and values of df are more
morphs as races, largely due to selection determined than six times as high in the islands (Ellegren et al.
by colour morphs in the M€ ullerian mimic, H. erato 2012). Extreme islands show up to 50-fold higher levels
(Mallet et al. 1990). But the races are not separate spe- of relative divergence than the background, and there
cies: they do not show evidence of hybrid sterility or are many signatures of hitchhiking events. However,
inviability and appear to be randomly mating in the the high relative divergence is accompanied by com-
narrow zone where the colour morphs overlap (Mallet mensurately strong reductions in polymorphism within
et al. 1990). This raises the possibility that the colour- islands (Table 2), which means that these statistical
patterning loci contain locally adapted alleles within a measures of divergence may be misleading. Indeed, the
largely panmictic (or at least continuously distributed) original study reported that ‘dXY did not exceed back-
population and that gene flow outside of these regions ground levels in divergence islands’ (Ellegren et al.
represents nothing more than the normal movement of 2012). Although the genotypes for each individual were
alleles within a species. In this case, there should be not available for analysis, the mean value of dXY in
high FST and dXY between allelic types determining col- islands of divergence was 0.0036 (H. Ellegren, personal
our, as the locus is behaving like a balanced polymor- communication). This level of absolute divergence is
phism within a species (Charlesworth 2006). It is much lower than the value for the genomic background
therefore possible that the higher value of dXY observed in equivalently sized windows (dXY = 0.0044) and is
in the region overall is consistent with this expectation, equal to levels of within-species polymorphism in F. al-
but the physical extent and level of divergence will be bicollis. Unfortunately, no formal statistical test can be
determined by the age and amount of selection on the performed for these reported values of absolute diver-
balanced polymorphism rather than selection against gence between islands and nonislands.
migrants. As with previous studies in Anopheles (Turner et al.
There may be other pairs of taxa within the genus He- 2005; Turner & Hahn 2007), the number of fixed differ-
liconius for which comparisons of relative and absolute ences between flycatcher species was mistakenly used
divergence can be made in order to test the speciation- as an alternative to relative measures of divergence like
with-gene-flow model, but they will have to come from FST. In fact, as is discussed above (see Fig. 1), df is

highly sensitive to levels of within-species diversity. mice (Table 1; Fig. 3f), but we did find significantly
Reduced levels of polymorphism due to linked selection lower levels of polymorphism in the islands (Table 2).
will cause large shifts in the proportion of fixed differ- One further caveat with using dXY in both the Mus
ences in a region for closely related species, although and Oryctolagus systems must be mentioned, largely
this will not change dXY. There are multiple signatures due to the fact that many of the islands are on the X
of strong selection in the Ficedula divergence islands – chromosome in both species. Because dXY is affected by
including reduced diversity, a skewed allele frequency both the substitution rate and levels of ancestral varia-
spectrum, and higher linkage disequilibrium (LD) – and tion, comparisons of divergence between X-linked and
they largely coincide with centromeres, regions known autosomal loci suffer from several inequalities. First, in
to have reduced recombination in syntenic bird ge- mammals, there is an increased rate of mutation in
nomes (e.g. Backstr€ om et al. 2010). Together with the males (Shimmin et al. 1993), leading to an increased rate
lower values of absolute divergence, the data better of substitution on the autosomes (which spend half
support a model in which regions of 50-fold higher their time in males) relative to the X (which spends
‘divergence’ instead indicate regions of 50-fold lower one-third its time in males). To avoid this problem, one
polymorphism and that many of these islands have can rescale levels of absolute divergence relative to an
been subject to recurrent selection since before the spe- outgroup (e.g. Feder et al. 2005), which controls for
cies split. unequal rates of substitution among loci or chromo-
somes. However, when such methods were applied to
the Mus and Oryctolagus data sets, no significant differ-
Oryctolagus (rabbits) and Mus (mice)
ence in divergence levels was found between islands
For both the rabbit subspecies (O. cuniculus cuniculus and nonislands (Nachman & Payseur 2012). Second, the
and O. c. algirus) and the mouse subspecies (M. muscu- X chromosome is expected to have three-quarters the
lus musculus and M. m. domesticus) considered here, the effective population size of the autosomes under a neu-
incipient species meet in a hybrid zone after periods of tral model, resulting in an expectation of three-quarters
allopatry (Biju-Duval et al. 1991; Branco et al. 2000; Fer- the level of ancestral polymorphism. Given the very
rand & Branco 2007; Teeter et al. 2008). Both cases small difference in current levels of X vs. autosomal
therefore represent scenarios of secondary speciation- polymorphism (e.g. 0.1% in Oryctolagus; Carneiro et al.
with-gene-flow, similar to the situation in Ficedula. Nev- 2010), this is unlikely to contribute to a lack of differ-
ertheless, despite the relative ease with which isolating ence in dXY. For both species pairs, therefore, it again
barriers may arise in allopatry, introgression upon sec- appears as though a model invoking postspeciation
ondary contact is expected to leave similar signatures linked selection – especially on genes in regions of
on relative and absolute divergence measures. reduced recombination – can explain the results at least
We used loci that were Sanger sequenced in multiple as well as a model with differential gene flow among
individuals, from either rabbits (Geraldes et al. 2006; loci.
Carneiro et al. 2009, 2010) or mice (Harr 2006; Geraldes
et al. 2011) for the analysis. Because multiple definitions
Reconciling studies of introgression with
are used in previous studies, we alternatively defined
estimates of absolute divergence
‘islands’ as regions that contained more fixed differ-
ences than expected by chance based on the genome- The results presented above suggest that there is little
wide average (Harr 2006); regions with low recombina- evidence for differential introgression among the loci
tion relative to the genome-wide average (Geraldes considered, for scenarios involving both primary and
et al. 2011); or highly differentiated regions on the X secondary speciation-with-gene-flow. Although the
chromosome (Carneiro et al. 2010). Despite values of power to detect differential gene flow using absolute
FST that ranged from 0.008 to 0.977 in rabbits, we found measures of divergence can be low for taxa that have
no significant difference in dXY between the two classes recently split (Box 2), the same should not be true for
of loci (Table 1; Fig. 3e). As with several previous data situations of secondary contact between species because
sets, dXY was lower in the islands of relative divergence there has been enough time for mutations to accumu-
than it was in nonislands, at least for some subsets of late. Under secondary contact, the divergence accumu-
the data (Table 1). Comparisons between other pairs of lated during periods of allopatry makes it easier to
Mus subspecies have found higher values of absolute distinguish linked selection from introgression,
measures of divergence in islands (Nachman & Payseur although determining the minimum time in allopatry
2012), although these were not significant. Likewise, we necessary requires further investigation.
found no significant difference in dXY between regions These considerations lead to an apparent contradic-
that did differ significantly in relative divergence in tion: for many examples of secondary contact, including

those analysed here, there is strong evidence for differ- and where it is clear that there is gene flow between
ential introgression of loci across hybrid zones (e.g. lineages after an initial period of isolation. However, for
Sætre et al. 2003; Payseur et al. 2004; Teeter et al. 2008, cases of primary speciation-with-gene-flow, it is often
2010; Carneiro et al. 2009). How then can it be that very difficult to determine whether lineages are
analyses of absolute sequence divergence fail to detect exchanging migrants at all. One popular way to infer
this differential introgression? whether gene flow is occurring is to apply a coalescent-
We believe that the apparently contradictory results based model for gene flow, referred to as the isolation-
are a function of the type of data collected from differ- with-migration (IM) model (Nielsen & Wakeley 2001;
ent studies of introgression and not a failure of any sin- Hey & Nielsen 2004, 2007; Hey 2010). In the next sec-
gle method for detecting introgression. Absolute tion, we examine the statistical properties of this model,
measures of sequence divergence have been used to finding that it has unacceptably high false positive rates
uncover differential gene flow among loci, most often for recently diverged taxa.
as ‘islands of introgression’ (Garrigan et al. 2012) – loci
with low absolute divergence set against a genomic
Questions about inferring gene flow using the
background of high absolute divergence. Examples of
isolation-with-migration (IM) model
such introgressed regions are found in many species,
including Drosophila (Garrigan et al. 2012; Brand et al. One of the most challenging tasks in molecular popu-
2013), Anopheles (Djogbenou et al. 2008; Etang et al. lation genetics is to distinguish between the causes of
2009), Heliconius (The Heliconius Genome Consortium shared polymorphisms among populations or species.
2012; Smith & Kronforst 2013) and Mus (Song et al. Because both migration and recent splitting events
2011). There is therefore little reason to believe that we result in shared polymorphisms, both of these pro-
could not also detect differential introgression using cesses can explain a wide range of values of differen-
absolute divergence in the data analysed here. tiation. The IM program and its successors, IMa (Hey
However, for the vast majority of studies examining & Nielsen 2007) and IMa2 (Hey 2010), are intended to
gene flow across hybrid zones – the main source of be used to estimate fundamental evolutionary parame-
evidence for differential introgression – introgression is ters between recently diverged species, including
detected by using markers that are fixed (or nearly migration rates. Comparisons between the probabilities
fixed) between species in populations far from the of models with and without migration can be used to
zone of overlap (e.g. Rieseberg et al. 1999; Sætre et al. conduct a likelihood-ratio test of the null hypothesis
2003; Payseur et al. 2004; Nolte et al. 2006; Teeter et al. of zero gene flow (Nielsen & Wakeley 2001; Pinho &
2008, 2010). Such markers are generally necessary to Hey 2010).
study gene flow across a hybrid zone (for an excep- The inference of parameters in the isolation-with-
tion, see Yatabe et al. 2007), but obviously cannot – migration model using IM offers a powerful way to
almost by definition – represent loci that have freely infer recent evolutionary processes, but is not without
introgressed across species boundaries and spread problems. Most difficulties arise from the assumptions
across species ranges. The loci that have introgressed made by the methods, including selective neutrality at
across species and that would therefore show reduced all loci (i.e. no positive or balancing selection), no
absolute divergence when estimated from sequences recombination within genes, no population structure
collected in allopatry, would not have any fixed differ- within the ancestral population, independence between
ences between species and would not show any differ- genes and no migration from unsampled populations.
ence between allopatric and sympatric comparisons. Although some of these assumptions can be relaxed
While recent secondary contact may not have allowed without affecting the inferences made in many cases
enough time for truly neutral markers to have intro- (Becquet & Przeworski 2009; Strasburg & Rieseberg
gressed across an entire species’ range, it is also possi- 2010), we suspected that violations of these assumptions
ble that weak selection against even the least might lead to false signals of migration, even between
differentiated loci that still show fixed differences in species that were completely reproductively isolated.
allopatry has not allowed them to introgress fully. Sim- For instance, the IM model favours a scenario with
ilar arguments apply to contrasts of differentiation ongoing gene flow over one with strict allopatric diver-
among sympatric and allopatric populations of differ- gence both between Drosophila melanogaster and D. simu-
ent species, at least for those species in which there is lans (Wang & Hey 2010) and between humans and
no isolation by distance. chimpanzees (Wang 2009; Mailund et al. 2012). Given
The above examples generally come from cases of the high degree of isolation between these species pairs,
secondary contact, where there may be more power to it appears that the IM model may be erroneously infer-
distinguish introgressing from nonintrogressing loci, ring gene flow when there is none.

A common approach to applying IM is to consider 2–50 loci, each 2500 bp in length with no recombination.
large numbers of individuals sampled at a small num- We simulated a total of 36 parameter combinations: six
ber of loci. The parameters of the IM model can then be different split times each simulated with six different
estimated by Markov chain Monte Carlo (MCMC) sam- numbers of loci. For each set of parameters, 100 inde-
pling, such as is implemented in IMa2 (Hey 2010) and pendent simulations were run, each producing a data
MIMAR (Becquet & Przeworski 2007). Studies using set of sequences by running the ms output through
these methods commonly find evidence for gene flow Seq-Gen (Rambaut & Grassly 1997) with an infinite sites
between closely related species (e.g. Geraldes et al. model specified.
2006; Nadachowska & Babik 2009; Runemark et al. 2012; Using IMa2, we attempted to estimate migration and
Mu~ noz et al. 2013), although the number of loci used in divergence time parameters for all simulated data sets.
such studies is often far less than the number recom- Despite the fact that all simulations were carried out
mended in order to get accurate results (see supplemen- without gene flow, IMa2 consistently reported signifi-
tary table 1 in Pinho & Hey 2010). We used simulations cant nonzero migration rates (Fig. 4). A data set was
to ask whether recent split times between species or considered to be a ‘false positive’ if either of the two
small numbers of sampled loci could lead to erroneous migration parameters (one into each species) was iden-
inferences of gene flow, even when all other IM model tified by IMa2 to be significant by a likelihood-ratio test;
assumptions are met. While a number of studies have in many cases, we found that the inferred migration
examined the consequences of violating IM’s assump- rates were highly asymmetric, often with only one
tions on parameter estimates (e.g. Becquet & Przeworski parameter significant (data not shown). Inconsistencies
2009; Hey 2010; Strasburg & Rieseberg 2010), to our between simulated and estimated migration rates were
knowledge, the effect of recent split times – those that more common in samples with few independent loci
are most relevant to possible cases of primary specia- (n = 2–10) and recent population split times (Ne < 1). In
tion-with-gene-flow – on inferring whether there is any extreme cases of few loci and very recent split times,
migration at all has not been reported. false positive rates exceeded 80% (Fig. 4).
We simulated populations with no migration using Our results are also consistent with previous simula-
the program ms (Hudson 2002). Sequences for 15 indi- tion results. For some conditions tested (two to three
viduals were generated from each of two species (30 populations, 10 loci, t = 2), Hey (2010) reported false
chromosomes total) with splitting times ranging from positive rates of 30% for individual migration parameters
less than 1 up to 8Ne generations (Fig. 4). We simulated despite the low proportion of false positives when aver-
aging over all parameters (see supplementary table 1 in
Number of loci Hey 2010). The same general pattern is reported for MI-
0.8
2 MAR: results are not reliable for fewer than 20 sampled

3
5 loci (Becquet & Przeworski 2007). However, neither of
10
20
these studies considered very recent split times, the fac-
tor that appears to have the greatest effect on our results.
0.6
50
Recently, Sousa et al. (2013) have relaxed the assump-

False positive rate
tion within IM that all loci share a single migration his-

tory. In order to identify loci at which there is selection
0.4
against gene flow, they developed a new version of

IMa2 that classifies loci by their inferred effective migra-
tion rates. When applied to the same Oryctolagus data as
analysed above, the authors identified two groups of
0.2
loci: those where the data are consistent with high

migration, and those with low migration. Even though
the latter group is enriched for highly differentiated loci
0.0
0.02 0.2 0.6 1 2 8 (i.e. those with high FST), dXY does not differ between
Divergence time (Ne generations) the two categories and is actually lower for the low-
migration set of loci (mean dXY = 0.0358 [low migration]
Fig. 4 Assessing the accuracy of IMa2 in inferring gene flow. vs. 0.0411 [high migration], P = 0.91; Table 1). Clearly,
Percentage of false positives observed for simulated popula-
support for differential gene flow among loci using the
tions with no gene flow. False positives occur when the likeli-
hood-ratio statistic indicates a significant nonzero migration
isolation-with-migration model can also be misleading.
rate in one or two directions using IMa2 (Hey 2010). 100 simu- Although it should be noted that we found the most
lations were run for all conditions except those representing 50 serious problems with IM to occur at very short
loci, where 50 simulations were run. divergence times – more recently, for instance, than the

cases of secondary contact considered above – our sim- Nachman & Payseur 2012), so here, we attempt to clarify
ulations have actually met all of the assumptions of the specific predictions of the model and to assess the fit of
IM model. When considering natural populations various aspects of the data to this alternative model.
undergoing selection and recombination, it may be that The main feature of the alternative model – variously
these violations of the underlying model cause false sig- referred to as the ‘incidental island’ model (Turner &
nals of gene flow regardless of the divergence times. Hahn 2010), the ‘selection at linked sites’ model (Nach-
This is likely to be the problem in the primate and Dro- man & Payseur 2012) or the ‘low-gene-flow’ model
sophila examples cited above. Similar problems may (Hahn et al. 2012) – is that heterogeneous levels of rela-
affect any methods that do not consider the effects of tive divergence are largely due to heterogeneous natu-
selection on population differentiation when fitting ral selection across the genome and not heterogeneous
models of gene flow (e.g. Duvaux et al. 2011; Nada- migration rates (Fig. 5b). Because any type of selection
chowska-Brzyska et al. 2013). that lowers levels of linked neutral diversity will result
in increased relative divergence (Charlesworth et al.
1997; Charlesworth 1998; Slatkin & Wiehe 1998), differ-
An alternative model: Divergence after
ences in the effects of selection across the genome will
speciation
result in different levels of relative divergence. Impor-
As discussed in the Introduction, the observed patterns tantly, linked selection will not increase absolute diver-
of heterogeneous differentiation across genomes of gence (Birky & Walsh 1988), and therefore, this model
closely related species have been interpreted as evidence predicts no islands of absolute divergence that overlap
for speciation-with-gene-flow, either primary or second- with the islands of relative divergence (though other
ary. To be specific, the model that is often invoked pro- processes can increase absolute divergence; Box 1). The
poses that low levels of differentiation are due to model further proposes that low levels of differentiation
ongoing gene flow homogenizing a majority of the gen- across the majority of the genome are largely due to
ome, while regions of high differentiation (‘islands’) shared ancestral polymorphism: the more recent the
contain loci refractory to introgression because they split between the taxa being compared, the more similar
underlie isolating traits between the species (Fig. 5a). allele frequencies will be at ancestrally inherited poly-
Under these models, we also expect higher absolute morphisms. The lineage-sorting process is a slow one,
divergence in islands shortly after speciation, a pattern and in the absence of selection, two species will not be
that was not found in the data sets considered here. The reciprocally monophyletic at 95% of loci (that is, they
lack of support for speciation-with-gene-flow models will no longer share ancestral variation) until 9-12Ne
begs the question: what model can explain the data? generations after speciation (Hudson & Coyne 2002). At
Although a number of alternative explanations are those loci where natural selection has acted in one or
surely possible, we emphasize one model in which pop- both taxa, however, lineage sorting is accelerated and
ulation divergence is not determined by differential ancestral variation is removed quickly. These regions of
migration across the genome. The outlines of this model reduced diversity also then have increased relative
have been proposed previously (Noor & Bennett 2009; divergence. The alternative model can include no gene
Turner & Hahn 2010; White et al. 2010; Hahn et al. 2012; flow or a low level of gene flow between the species
(a) (b) Fig. 5 Speciation-with-gene-flow model

vs. divergence-after-speciation model.
Panel (a) shows a cartoon of the specia-
tion-with-gene-flow model, with migra-
tion (arrows) homogenizing most of the
genome (white); grey areas represent
those loci resistant to introgression. Panel
(b) shows a cartoon of the divergence-
after-speciation model, with selection
(arrows) driving relative divergence in a
small number of regions (grey); white
regions are similar between the two
daughter species because of shared
ancestral polymorphism.
Speciation-with-gene-flow Divergence-after-speciation
Migration homogenizes much of the genome Selection differentiates some of the genome

being considered, and in fact the power to detect introgression and selection after speciation are likely to
regions resistant to introgression will be higher with be occurring in many taxa, below we consider a
higher levels of gene flow across the genome (Box 2). number of different features of such data that are
The key feature of the alternative model is not the explained equally well, or better, by a model of diver-
presence or absence of gene flow or selection against gence after speciation as compared to models of specia-
migrant genotypes – instead, it is that the heterogeneity tion-with-gene-flow.
in levels of divergence among loci is not due to
differential effective migration, but rather differential
Islands most often occur in regions of low
selection.
recombination
One further elaboration to the alternative model may
explain many of the patterns previously associated only Since it was first found in Drosophila (Begun & Aquadro
with the speciation-with-gene-flow model. Although 1992), one of the most ubiquitous patterns in molecular
much of the selection acting to reduce levels of diver- evolution is the reduction in diversity found in regions
sity in each descendant lineage may be due to adapta- of reduced recombination (Hahn 2008; Sella et al. 2009;
tion unrelated to the speciation process, or to selection Cutter & Payseur 2013). This reduction in diversity has
against recurrent deleterious mutations, it is also likely been ascribed to linked selection on both advantageous
that immediately after speciation, many new alleles that (Kaplan et al. 1989) and deleterious mutations (Charles-
increase adaptation are selected for. Any alleles that worth et al. 1993), and also lower mutation rates in
increase niche specialization, decrease direct competi- these regions (e.g. Hellmann et al. 2003; Kulathinal et al.
tion with congeners, or that decrease hybrid matings 2008). However, close analyses of the relationship
will be immediately advantageous and are therefore between recombination and nucleotide divergence (e.g.
likely to be some of the first targets of selection postspe- McGaugh et al. 2012), and related predictions about the
ciation. There are multiple signals of positive (adaptive) effects of linked selection in reducing Ne in these
selection in sequence data collected from islands (e.g. regions (Pease & Hahn 2013), all indicate that linked
Turner et al. 2005; Ellegren et al. 2012), and genes sug- selection (or processes that mimic selection, such as
gestive of species-specific adaptations are often found meiotic drive) is the major driver of reduced variability.
in islands (see below). The alternative model therefore In light of the universality of these patterns, it should
suggests that islands of relative divergence should not not be surprising that regions of reduced recombination
be dismissed as nonexistent or unimportant, but instead also show signals of increased relative divergence (e.g.
that some of these islands may contain loci closely tied Begun & Aquadro 1993; Stephan et al. 1998; Keinan &
to postspeciation adaptation. Reich 2010). For four of the data sets analysed here
(Anopheles, Ficedula, Mus, Oryctolagus), the islands of
divergence are found in pericentromeric or peritelomer-
What does the alternative model explain?
ic regions. Such regions often show reduced levels of
In the study thus far, we have largely concerned our- recombination (i.e. crossing-over) and can have reduced
selves with examining predictions concerning levels of diversity across megabase-size regions of individual
relative and absolute sequence differences under alter- chromosomes (Cutter & Payseur 2013). This same pat-
native models of speciation and divergence. For this tern is also seen in other data sets examining genome-
particular set of predictions, we have argued that the wide levels of relative divergence, at least when the
data do not support models of speciation-with-gene- recombination rate associated with individual markers
flow, where islands of high relative divergence repre- is known (e.g. Renaut et al. 2013).
sent loci that are not introgressing between species. The effects of linked selection are not limited to only
However, lack of support for speciation-with-gene-flow regions of low recombination. Although the length of
models does not immediately suggest that our favoured the region affected scales negatively with the recombi-
alternative model – divergence after speciation – is nation rate (Kaplan et al. 1989), individual adaptive sub-
strongly supported. There are multiple reasons why we stitutions in regions of normal recombination can also
may not have found patterns of sequence divergence reduce linked neutral variation (e.g. Kern et al. 2002;
supporting the original model (including low statistical Sattath et al. 2011; McGaugh et al. 2012); this reduction
power; Box 2), and possibly, there are alternative mod- in variation will affect measures of relative divergence.
els that we have not considered. But we believe that the As expected, when islands of relative divergence have
alternative model described in the previous section is a been identified in regions of normal recombination,
well-supported one and that it can explain multiple they have been on average smaller than those regions
aspects of the data collected on genomic divergence linked to centromeres and telomeres (e.g. Turner et al.
between closely related species to date. While both 2005). However, it is also the case that the length of the

region affected by selection is strongly dependent on speciation-with-gene-flow models have been invoked to
the strength of selection, so that even in regions of typi- explain this concordance, with the implication that
cal recombination selective sweeps may reduce varia- there is a lack of introgression of alternative alleles
tion over a wide area (e.g. Tishkoff et al. 2007). Larger found at loci controlling ‘speciation phenotypes’ (Shaw
regions of higher relative divergence are therefore per- & Mullen 2011).
fectly consistent with selection after divergence, and As argued above, however, it appears equally plausi-
‘divergence hitchhiking’ does not have to be invoked ble to argue that many of these traits – and ones like
(cf. Via & West 2008). The effects of linked selection are them – would be just as important if they arose post-
also relatively independent of standing levels of linkage speciation, when reproductive isolation is largely com-
disequilibrium, and as such, measures of LD cannot be plete. In this scenario selection on the underlying loci
expected to be predictive of the length of a region occurred after postzygotic or postmating reproductive
affected by linked selective events (Kaplan et al. 1989). isolation has already evolved, as any alleles increasing
The role for regions of reduced recombination in local adaptation or decreasing nonproductive matings
models of primary speciation-with-gene-flow is to (i.e. reinforcing already existing reproductive barriers)
ensure linkage disequilibrium between alleles for local would be favoured. Behaviours increasing host-plant
adaptation and alleles for nonrandom mating (Felsen- fidelity when there is local adaptation (e.g. Via 2012), or
stein 1981). In this model, therefore, there are predicted that increase within-species preferences for mates (e.g.
to be more than one gene involved in isolating barriers Pennetier et al. 2010), both fall into this category. Selec-
within each individual region of reduced recombination would again act to reduce within-species polymor-
tion. The fact that there are multiple, unlinked regions phism, consequently increasing between-species relative
of reduced recombination across the genome is not nec- divergence.
essarily expected under this model, as the free recombi- In other cases, no known QTL or overrepresented
nation among islands works against coadapted allelic class of genes has been found in islands of relative
combinations (see below). By contrast, in a model of divergence (e.g. Ellegren et al. 2012), or only classes of
divergence after speciation, there is nothing necessarily genes that are generally found to be rapidly evolving
special about such regions. They may or may not be are overrepresented in these regions (e.g. Harr 2006). It
involved with divergent selection between the species therefore seems likely that there are at least two pro-
and do not have to contain multiple targets of divergent cesses that lead to islands of divergence. In the first,
selection. It is simply that repeated bouts of linked specific adaptive alleles that increase the fit of newly
selection (of any kind) guarantee that signals of relative evolved species to new environments are favoured and
divergence are greatly increased. fix within one or both species. There is no a priori rea-
son to expect that selection in these regions has affected
both daughter species, but even reductions in polymor-
Colocalization of islands and species-defining traits
phism in one of them will cause an increase in relative
In systems where genetic crosses are possible, it has divergence. It may also be the case that such adaptive
been noted that there is a statistically significant corre- alleles are only favoured in a subset of populations
spondence between the location of quantitative trait loci within a species, leading to population-specific sweeps
(QTL) underlying traits that differ between closely and population-specific islands (e.g. Turner & Hahn
related species and the location of peaks of FST (Rogers 2007). Such adaptive alleles may arise multiple times,
& Bernatchez 2005, 2007; Via & West 2008; Via 2012). In leading to ‘parallel’ islands of divergence in multiple
some cases, the traits are species-defining behaviours or species (e.g. Nadeau et al. 2012). However, pairwise
phenotypes (e.g. Via 2012), while in others, they are dif- comparisons of FST values that use polymorphism from
ferences in gene expression between species (Derome a single population in all comparisons are also noninde-
et al. 2006; Cassone et al. 2008; Whiteley et al. 2008; pendent, as reduced polymorphism in one species will
Renaut et al. 2009). There is also a coarse correspon- cause increased relative divergence in all comparisons;
dence between QTL for reproductive isolation and truly, independent comparisons are needed to investi-
regions of high differentiation (e.g. Payseur et al. 2004), gate whether true hotspots of selection exist (e.g. Baxter
especially involving inversions (e.g. Rieseberg et al. et al. 2010). The second process creating islands repre-
1999; Yatabe et al. 2007; Kulathinal et al. 2009). Among sents the general, nonspecies-specific reduction in poly-
the systems considered here, the peaks in relative morphism seen in all genomes, especially the
divergence between races of Heliconius butterflies were reductions associated with regions of low recombina-
found within regions known to be involved in wing tion. In these cases, it may even be background selec-
colour patterning (Nadeau et al. 2012), a trait important tion that acts as the major force reducing
for mate choice (Jiggins et al. 2001). In such cases, polymorphism, and no association between adaptive

traits and islands is necessarily expected. Even when help to explain this pattern, as there is free recombination
hitchhiking of adaptive substitutions is causing the between islands on different chromosomes. The prospect
decrease in diversity, this adaptation may be unrelated of strong transmission ratio distortion has been exam-
to the speciation process. ined in Anopheles, but none was found (Hahn et al. 2012).
Regardless of which process is acting, postspeciation In the case of secondary speciation-with-gene-flow,
selection does not require multiple tightly linked alleles standing hybrid zones make it much easier for neutral
in any single region of low recombination, as migration alleles unlinked to selectively disadvantageous alleles to
is not continuously introducing alternative arrange- introgress between species. The constant homogenization
ments via which recombination can break up locally of such neutral loci by gene flow will inevitably lead to
favourable combinations of alleles. Therefore, no clus- LD between alleles involved in isolating barriers. These
tering of selected loci is expected, whether such loci associations will continue at least until the number of
arise in situ or by rearrangement (cf. Yeaman 2013). genes involved in the barriers to exchange becomes so
great that effectively all loci are linked to them (Barton &
Bengtsson 1986); at this point, gene flow is substantially
Near-perfect linkage disequilibrium among unlinked
reduced across the genome (cf. Flaxman et al. 2013).
islands
In the divergence-after-speciation model, it is quite
One of the most compelling, and puzzling, patterns of straightforward to explain perfect LD among unlinked
divergence between closely related species is the exis- loci: alternate alleles are fixed by selection in each spe-
tence of multiple, unlinked islands. Many of the sys- cies, at one locus or at many loci. If these species are
tems considered here (e.g. Anopheles and Ficedula) show already completely reproductively isolated, then there
high numbers of fixed differences between species at is no opposing force that generates large numbers of
many loci throughout the genome (Turner et al. 2005; recombinant individuals. The alleles at each island are
Ellegren et al. 2012). These islands of divergence segre- in LD simply because of the population structure gener-
gate independently, but show near-perfect linkage dis- ated by the absence of gene flow, and no selection
equilibrium: that is, alleles associated with one species maintaining these specific associations must be invoked.
in one island are always found in the presence of alleles Again, the shared variation found across the rest of the
associated with that same species at all the other genome is due to shared ancestral variation and not
islands. In these cases, the linkage disequilibrium is recent introgression between species. No strong selec-
being measured in the two populations or species taken tion or strong transmission ratio distortion is needed to
together; no linkage disequilibrium is expected within explain these patterns.
either population, except under scenarios of incomplete
introgression after secondary contact.
Lower absolute divergence in islands due to recurrent
In the speciation-with-gene-flow models, introgression
selection
in islands is prevented by selection, while recombination
between the islands allows for the homogenization of the In our above analyses, we found that absolute diver-
rest of the genome. However, while it is fairly easy to gence was lower in most of the islands of increased rel-
imagine how this can occur for a single island, if alleles ative divergence (Table 1; Fig. 3). While linked selection
at multiple islands must remain associated with each in extant populations should increase relative diver-
other throughout hybridization, it is much more difficult gence, it is not expected to decrease absolute diver-
to get introgression across the genome, at least for mod- gence. The question therefore arises: why is absolute
els of primary speciation-with-gene-flow (Turner & divergence lower in these regions?
Hahn 2010; White et al. 2010). In order to maintain the It is important to recall that the value of dXY at the time
strong association among alleles at unlinked loci in a pri- of population splitting is equal to the amount of variation
mary speciation-with-gene-flow model, it must be the present in the single prespeciation population, hAnc. In
case that either a) a large fraction of offspring with other words, the absolute divergence between two spe-
recombinant genotypes does not survive or b) some form cies is made up of the accumulation of substitutions post-
of very strong transmission ratio distortion favours speciation plus the amount of variation that existed at
gametes with ‘matching’ alleles at all islands. With more speciation (Box 1; Gillespie & Langley 1979). This implies
than even a few islands that do not introgress because of that while linked selection in current-day populations
selection, the number of recombinant individuals con- will not cause a decrease in dXY, any form of selection
taining the correct combination of parental alleles at all that reduced variation in the ancestral population can
islands – and therefore that can pass material between decrease dXY (Fig. 6). The correlation found between low
species – becomes vanishingly small. The fact that there levels of current diversity and low absolute divergence
is low recombination within individual islands does not in islands therefore suggests that there have been

(a) (b)
A
A
dXY dXY
X, Y
X, Y
Population X Population Y Population X Population Y
No linked selection Recurrent selection
Fig. 6 The effect of recurrent selection on levels of absolute divergence. Panel (a) shows a scenario where there is no linked selection
in either the prespeciation ancestral population or current-day populations. In this case, measures of ancestral polymorphism (pA)
and current polymorphism (pX and pY) are at their neutral expectations. Panel (b) shows a scenario where there is recurrent linked
selection, such that selection has affected both ancestral and current-day populations. In this case, measures of both ancestral poly-
morphism and current polymorphism are lower than expected. Note that, overall, there is a correlation between dXY and ancestral
polymorphism levels.
recurrent bouts of linked selection at these loci, including introgression after the evolution of reproductive isola-
at times pre-dating speciation (cf. Begun et al. 2007). tion. Higher absolute divergence between species within
Prespeciation reductions in polymorphism also have inversions may constitute one stereotypical example of
an unfortunate effect on the power of our test for this phenomenon, especially as inversions can show
reduced gene flow in islands – they could result in even extreme patterns of local adaptation and sequence diver-
lower statistical power to detect high absolute divergence even within species (e.g. Cheng et al. 2012).
gence at these loci. In the worst-case scenario, a very
strong selective sweep preceding speciation removes all
Species trees are recapitulated in islands
variation, and dXY at the time of speciation is 0. In order
to detect the increase in dXY that is indicative of a Many studies that have conducted comparisons of rela-
reduction in gene flow at such a locus, there would tive divergence across the genomes of closely related
have to be very high levels of gene flow across the rest organisms have also been able to construct phylogenetic
of the genome. Even less-severe reductions in ancestral trees describing relationships among sampled popula-
diversity will have some effect on the power of our test tions (reviewed in Nosil et al. 2009). The main conclu-
(Box 2). These results apply solely to primary specia- sion from examining these trees is that those made
tion-with-gene-flow, as the ability to detect differences from loci within islands of high relative divergence
in dXY among loci after secondary contact will be deter- often reflect the known species relationships, while
mined predominantly by the time spent in allopatry. those made from loci outside islands do not (Via &
One additional caveat concerning the identification of West 2008; Nosil et al. 2009; Keller et al. 2013). In the
islands of divergence is suggested by these consider- context of speciation-with-gene-flow models, this differ-
ations, as well as the case of the Heliconius colour morphs ence is thought to reflect differences in gene flow at the
discussed earlier. Because dXY is affected by ancestral lev- two types of loci: selection maintains the ‘true’ species
els of diversity, loci that experience balancing selection in tree at loci within islands, while migration obscures the
the ancestral population can show increased values of true relationships at all other loci.
absolute divergence. While this likely will not lead to However, we may just as easily interpret this differ-
incorrect inferences when multiple balanced alleles are ence in tree topology as due to selection after speciation,
all inherited by both species after splitting, scenarios in with no effect at all of migration. As mentioned earlier,
which the balanced alleles segregate themselves between it takes a long time after an initial split for species to
species can generate a false signal of an island of diver- become reciprocally monophyletic (Hudson & Coyne
gence. Examples of this would involve locally adapted 2002). This is because the lineage-sorting process due to
alleles defining morphs that eventually become species drift alone is quite slow, and ancestral variation can be
(as could happen between Heliconius colour morphs), in shared among daughter lineages for thousands to
which case the ‘island of divergence’ is already present at millions of generations after speciation. As a result,
speciation, and need not have been formed by differential gene trees will not necessarily reflect current species

relationships until well after speciation has occurred supports inferences about gene flow. We have shown
and full reproductive isolation has evolved (Avise 2004). that a particular application of the isolation-with-migra-
In contrast, when selection acts at a locus, it greatly tion model to sequence data has unacceptably high false
accelerates lineage sorting, removing ancestral variation positive rates, such that migration is inferred even
and making species reciprocally monophyletic. There- when none exists. We caution the further use of such
fore, in the context of the divergence-after-speciation models with recently diverged taxa and/or small num-
model, trees constructed from islands of high relative bers of loci. A further piece of data often cited as evi-
divergence more accurately reflect species relationships dence for gene flow is the presence of hybrids, often
because selection has reduced the effective population only F1 hybrids. If these hybrids, or later-generation
size, accelerating the lineage-sorting process (cf. Pease & individuals, exhibit strong extrinsic or intrinsic incom-
Hahn 2013). Trees constructed from outside islands do patibilities, then their existence does not always indicate
not necessarily reflect species relationships because they actual gene flow. Most importantly, even in those cases
still contain large amounts of ancestral polymorphism; in which there is demonstrable introgression between
however, this does not indicate anything about the dif- recently separated species – whether or not there are
ferential effects of migration at such loci. different levels of introgression among loci – this does
In general, divergence leads to lineage sorting. Over not constitute evidence for sympatric speciation. While
time the genomes of closely related species diverge, gene flow shortly after speciation is certainly a chal-
with the accumulation of species-specific genetic lenge to the existence of two independently evolving
changes and the sorting out of ancestral variation taxa, it is the rare case in which the conditions neces-
among lineages. Both of these processes are expected to sary for truly sympatric speciation have been met (e.g.
be somewhat heterogeneous, with variation in ancestral Savolainen et al. 2006). In many cases, gene flow may
coalescence times, differential selection and differential simply be occurring after isolation has evolved in allop-
mutation all causing loci to diverge at different rates. atry, and secondary contact should not be confused
Species pairs examined after the initial split will show with primary ‘speciation-with-gene-flow’.
heterogeneous patterns of divergence – and divergence Finally, it should be noted that none of our results
will increase with the time since the split – but no novel preclude the existence of true islands of divergence in
processes or models of speciation must be invoked to other systems that have yet to be examined using abso-
explain the data (e.g. Feder et al. 2012). lute divergence (e.g. sticklebacks; Hohenlohe et al. 2010;
Jones et al. 2012). If two ‘good’ species are exchanging
genes, then it would be expected that loci underlying
Conclusions
ecological and/or reproductive isolation would not flow
We have argued here that a fuller understanding of het- freely between them. Although there may be further
erogeneous genomic divergence requires that we use alternative explanations for observed patterns of diver-
both relative and absolute measures of sequence divergence (e.g. Bierne et al. 2013), we expect that multiple
gence. In particular, although relative measures have systems will be found in the near future with the
been used in the vast majority of studies examining the expected pattern. Evidence that individual loci do show
possibility of speciation-with-gene-flow, elevated values reduced gene flow will simply require that absolute
of these statistics (e.g. FST, da and df) can be due to measures of sequence divergence agree with the pat-
selection rather than reduced gene flow. Relative and terns so often shown by FST.
absolute measures of sequence divergence are both
appropriate summaries of nucleotide differences, and Acknowledgements
either may be preferred under certain circumstances.
However, in weighing the claims regarding differential We thank N. Besansky, M. Lawniczak, S. Emrich, K. Reiden-
bach, H. Ellegren and B. Harr for sharing data; L. Moyle, G.
migration among loci, we expect the two types of mea-
Coop, M. Whitlock, J. Mallet, J. Hey, R. Guerrero, S. Yeaman,
sures to agree; in all of the cases examined here, they J.B.W. Wolf, B. Payseur and R. Harrison for helpful discus-
do not agree. Linked selection therefore appears to be a sions; and L. Bernatchez and three anonymous reviewers for
more likely driver of the observed patterns. their constructive comments. This work is funded by National
Our analyses should raise concerns about the existing Science Foundation Grant MCB-1127059.
evidence for primary speciation-with-gene-flow (i.e.
sympatric speciation), especially for systems without
References
clear evidence of secondary contact. For most of the
systems considered here, there is good external Akerman A, B€urger R (2014) The consequences of gene flow
evidence that the two species are truly exchanging for local adaptation and differentiation: a two-locus two-
deme model. Journal of Mathematical Biology, 68, 1135–1198.
genes, but in many other cases, more limited evidence

Avise JC (2004) Molecular Markers, Natural History, and Evolu- gus cuniculus): Islands of differentiation on the X
tion. Sinauer Associates, Sunderland, Massachusetts. chromosome and autosomes. Evolution, 64, 3443–3460.
Backstr€om N, Forstmeier W, Schielzeth H et al. (2010) The Cassone BJ, Mouline K, Hahn MW et al. (2008) Differential
recombination landscape of the zebra finch Taeniopygia gutta- gene expression in incipient species of Anopheles gambiae.
ta genome. Genome Research, 20, 485–495. Molecular Ecology, 17, 2491–2504.
Baird NA, Etter PD, Atwood TS et al. (2008) Rapid SNP discov- Charlesworth B (1998) Measures of divergence between popu-
ery and genetic mapping using sequenced RAD markers. lations and the effect of forces that reduce variability. Molec-
PLoS ONE, 3, e3376. ular Biology and Evolution, 15, 538–543.
Barbash DA, Siino DF, Tarone AM, Roote J (2003) A rapidly Charlesworth D (2006) Balancing selection and its effects on
evolving MYB-related protein causes species isolation in sequences in nearby genome regions. PLoS Genetics, 2, e64.
Drosophila. Proceedings of the National Academy of Sciences, 100, Charlesworth B, Morgan MT, Charlesworth D (1993) The effect
5302–5307. of deleterious mutations on neutral molecular variation.
Barton NH (2006) Evolutionary biology: how did the human Genetics, 134, 1289–1303.
species form? Current Biology, 16, R647–R650. Charlesworth B, Nordborg M, Charlesworth D (1997) The
Barton N, Bengtsson BO (1986) The barrier to genetic exchange effects of local selection, balanced polymorphism and back-
between hybridising populations. Heredity, 57, 357. ground selection on equilibrium patterns of genetic diversity
Baxter SW, Nadeau NJ, Maroja LS et al. (2010) Genomic hot- in subdivided populations. Genetical Research, 70, 155–174.
spots for adaptation: the population genetics of M€ ullerian Cheng C, White BJ, Camdem C et al. (2012) Ecological genomics
mimicry in the Heliconius melpomene clade. PLoS Genetics, 6, of Anopheles gambiae along a latitudinal cline in Cameroon: a
e1000794. population resequencing approach. Genetics, 190, 1417–1432.
Becquet C, Przeworski M (2007) A new approach to estimate Coetzee M, Hunt RH, Wilkerson R et al. (2013) Anopheles col-
parameters of speciation models with application to apes. uzzii and Anopheles amharicus, new members of the Anopheles
Genome Research, 17, 1505–1519. gambiae complex. Zootaxa, 3619, 246–274.
Becquet C, Przeworski M (2009) Learning about modes of spe- Cutter AD, Payseur BA (2013) Genomic signatures of selection
ciation by computational approaches. Evolution, 63, 2547– at linked sites: unifying the disparity among species. Nature
2562. Reviews Genetics, 14, 262–274.
Begun DJ, Aquadro CF (1992) Levels of naturally occurring Derome N, Duchesne P, Bernatchez L (2006) Parallelism in gene
DNA polymorphism correlate with recombination rates in transcription among sympatric lake whitefish (Coregonus clu-
Drosophila melanogaster. Nature, 356, 519–520. peaformis Mitchill) ecotypes. Molecular Ecology, 15, 1239–1249.
Begun DJ, Aquadro CF (1993) African and North American Djogbenou L, Chandre F, Berthomieu A et al. (2008) Evidence
populations of Drosophila melanogaster are very different at of introgression of the ace-1R mutation and of the ace-1 dupli-
the DNA level. Nature, 365, 548–550. cation in West African Anopheles gambiae s. s. PLoS ONE, 3,
Begun DJ, Holloway AK, Stephens K et al. (2007) Population e2172.
genomics: whole-genome analysis of polymorphism and Duvaux L, Belkhir K, Boulesteix M, Boursot P (2011) Isolation
divergence in Drosophila simulans. PLoS Biology, 5, e310. and gene flow: inferring the speciation history of European
Bierne N, Gagnaire P-A, David P (2013) The geography of house mice. Molecular Ecology, 20, 5248–5264.
introgression in a patchy environment and the thorn in the Ellegren H, Smeds L, Burri R et al. (2012) The genomic land-
side of ecological speciation. Current Zoology, 59, 72–86. scape of species divergence in Ficedula flycatchers. Nature,
Biju-Duval C, Ennafaa H, Dennebouy N et al. (1991) Mitochon- 491, 756–760.
drial DNA evolution in lagomorphs: origin of systematic het- Etang J, Vicente JL, Nwane P et al. (2009) Polymorphism of
eroplasmy and organization of diversity in European rabbits. intron-1 in the voltage-gated sodium channel gene of Anophe-
Journal of Molecular Evolution, 33, 92–102. les gambiae s.s. populations from Cameroon with emphasis
Birky CW, Walsh JB (1988) Effects of linkage on rates of molec- on insecticide knockdown resistance mutations. Molecular
ular evolution. Proceedings of the National Academy of Sciences, Ecology, 18, 3076–3086.
85, 6414–6418. Excoffier L (2007) Analysis of population subdivision. In: Hand-
Branco M, Ferrand N, Monnerot M (2000) Phylogeography of book of Statistical Genetics, vol. 2 (eds Balding DJ, Bishop M,
the European rabbit (Oryctolagus cuniculus) in the Iberian Cannings C), pp. 980–1020. Wiley, West Sussex.
peninsula inferred from RFLP analysis of the cytochrome b Feder JL, Nosil P (2010) The efficacy of divergence hitchhiking
gene. Heredity, 85, 307–317. in generating genomic islands during ecological speciation.
Brand CL, Kingan SB, Wu L, Garrigan D (2013) A selective Evolution, 64, 1729–1747.
sweep across species boundaries in Drosophila. Molecular Biol- Feder JL, Xie X, Rull J et al. (2005) Mayr, Dobzhansky, and
ogy and Evolution, 30, 2177–2186. Bush and the complexities of sympatric speciation in Rhag-
Bulmer MG (1972) Multiple niche polymorphism. The American oletis. Proceedings of the National Academy of Sciences, 102,
Naturalist, 106, 254–257. 6573–6580.
Carneiro M, Ferrand N, Nachman MW (2009) Recombination Feder JL, Egan SP, Nosil P (2012) The genomics of speciation-
and speciation: loci near centromeres are more differenti- with-gene-flow. Trends in Genetics, 28, 342–350.
ated than loci near telomeres between subspecies of the Felsenstein J (1981) Skepticism towards Santa Rosalia, or why
European rabbit (Oryctolagus cuniculus). Genetics, 181, 593– are there so few kinds of animals? Evolution, 35, 124–138.
606. Ferrand N, Branco M (2007) The evolutionary history of the
Carneiro M, Blanco-Aguiar JA, Villafuerte R, Ferrand N, Nach- European rabbit (Oryctolagus cuniculus): major patterns of
man MW (2010) Speciation in the European rabbit (Oryctola- population differentiation and geographic expansion inferred

from protein polymorphism. In: Phylogeography of Southern Hudson RR (2002) Generating samples under a Wright-Fisher
European Refugia (eds Weiss S, Ferrand N), pp. 207–235. neutral model of genetic variation. Bioinformatics, 18, 337–
Springer. 338.
Flaxman S, Feder J, Nosil P (2012) Spatially explicit models of Hudson RR, Coyne JA (2002) Mathematical consequences of
divergence and genome hitchhiking. Journal of Evolutionary the genealogical species concept. Evolution, 56, 1557–1565.
Biology, 25, 2633–2650. Jakobsson M, Edge MD, Rosenberg NA (2013) The relationship
Flaxman SM, Feder JL, Nosil P (2013) Genetic hitchhiking and between FST and the frequency of the most frequent allele.
the dynamic buildup of genomic divergence during specia- Genetics, 193, 515–528.
tion with gene flow. Evolution, 67, 2577–2591. Jiggins CD (2008) Ecological speciation in mimetic butterflies.
Garrigan D, Kingan SB, Geneva AJ et al. (2012) Genome BioScience, 58, 541–548.
sequencing reveals complex speciation in the Drosophila sim- Jiggins CD, Naisbit R, Coe R, Mallet J (2001) Reproductive isola-
ulans clade. Genome Research, 22, 1499–1511. tion caused by colour pattern mimicry. Nature, 411, 302–305.
Geraldes A, Ferrand N, Nachman MW (2006) Contrasting pat- Jones FC, Grabherr MG, Chan YF et al. (2012) The genomic
terns of introgression at X-linked loci across the hybrid zone basis of adaptive evolution in threespine sticklebacks. Nature,
between subspecies of the European rabbit (Oryctolagus cuni- 484, 55–61.
culus). Genetics, 173, 919–933. Kaplan NL, Hudson R, Langley C (1989) The “hitchhiking
Geraldes A, Basset P, Smith KL, Nachman MW (2011) Higher effect” revisited. Genetics, 123, 887–899.
differentiation among subspecies of the house mouse (Mus Keinan A, Reich D (2010) Human population differentiation is
musculus) in genomic regions with low recombination. Molec- strongly correlated with local recombination rate. PLoS
ular Ecology, 20, 4722–4736. Genetics, 6, e1000886.
Gillespie JH, Langley CH (1979) Are evolutionary rates really Keller I, Wagner CE, Greuter L et al. (2013) Population genomic
variable? Journal of Molecular Evolution, 13, 27–34. signatures of divergent adaptation, gene flow and hybrid
Hahn MW (2008) Toward a selection theory of molecular evo- speciation in the rapid radiation of Lake Victoria cichlid
lution. Evolution, 62, 255–265. fishes. Molecular Ecology, 22, 2848–2863.
Hahn MW, White BJ, Muir CD, Besansky NJ (2012) No Kern AD, Jones CD, Begun DJ (2002) Genomic effects of nucle-
evidence for biased co-transmission of speciation islands in otide substitutions in Drosophila simulans. Genetics, 162, 1753–
Anopheles gambiae. Philosophical Transactions of the Royal 1761.
Society of London. Series B, Biological Sciences, 367, 374–384. Kronforst MR, Young LG, Blume LM, Gilbert LE (2006) Mul-
Haldane JBS (1930) A mathematical theory of natural and arti- tilocus analyses of admixture and introgression among
ficial selection (Part VI, Isolation). Proceedings of the Cambridge hybridizing Heliconius butterflies. Evolution, 60, 1254–1268.
Philosophical Society, 26, 220–230. Kulathinal RJ, Bennett SM, Fitzpatrick CL, Noor MA (2008)
Harr B (2006) Genomic islands of differentiation between Fine-scale mapping of recombination rate in Drosophila
house mouse subspecies. Genome Research, 16, 730–737. refines its correlation to diversity and divergence. Proceedings
Harrison RG (2012) The language of speciation. Evolution, 66, of the National Academy of Sciences, 105, 10051–10056.
3643–3657. Kulathinal RJ, Stevison LS, Noor MA (2009) The genomics of
Hellmann I, Ebersberger I, Ptak SE, P€a€abo S, Przeworski M speciation in Drosophila: diversity, divergence, and introgres-
(2003) A neutral explanation for the correlation of diversity sion estimated using low-coverage genome sequencing. PLoS
with recombination rates in humans. The American Journal of Genetics, 5, e1000550.
Human Genetics, 72, 1527–1535. Lawniczak MKN, Emrich SJ, Holloway AK et al. (2010) Wide-
Hey J (1991) The structure of genealogies and the distribution spread divergence between incipient Anopheles gambiae species
of fixed differences between DNA sequence samples from revealed by whole genome sequences. Science, 330, 512–514.
natural populations. Genetics, 128, 831–840. Lee Y, Marsden CD, Norris LC et al. (2013) Spatiotemporal
Hey J (2010) Isolation with migration models for more than dynamics of gene flow and hybrid fitness between the M
two populations. Molecular Biology and Evolution, 27, 905–920. and S forms of the malaria mosquito, Anopheles gambiae. Pro-
Hey J, Nielsen R (2004) Multilocus methods for estimating ceedings of the National Academy of Sciences, 110, 19854–19859.
population sizes, migration rates and divergence time, with Mailund T, Halager AE, Westergaard M et al. (2012) A new
applications to the divergence of Drosophila pseudoobscura isolation with migration model along complete genomes
and D. persimilis. Genetics, 167, 747–760. infers very different divergence processes among closely
Hey J, Nielsen R (2007) Integration within the Felsenstein equa- related great ape species. PLoS Genetics, 8, e1003125.
tion for improved Markov chain Monte Carlo methods in Mallet J, Barton N, Lamas G et al. (1990) Estimates of selection
population genetics. Proceedings of the National Academy of and gene flow from measures of cline width and linkage
Sciences, 104, 2785–2790. disequilibrium in Heliconius hybrid zones. Genetics, 124,
Hohenlohe PA, Bassham S, Etter PD et al. (2010) Population ge- 921–936.
nomics of parallel adaptation in threespine stickleback using Maroja LS, Andres JA, Harrison RG (2009) Genealogical discor-
sequenced RAD tags. PLoS Genetics, 6, e1000862. dance and patterns of introgression and selection across a
Holsinger KE, Weir BS (2009) Genetics in geographically struc- cricket hybrid zone. Evolution, 63, 2999–3015.
tured populations: defining, estimating and interpreting FST. Martin SH, Dasmahapatra KK, Nadeau NJ et al. (2013) Gen-
Nature Reviews Genetics, 10, 639–650. ome-wide evidence for speciation with gene flow in Helico-
Hopkins R, Rausher MD (2011) Identification of two genes nius butterflies. Genome Research, 23, 1817–1828.
causing reinforcement in the Texas wildflower Phlox drum- Maynard Smith J, Haigh J (1974) The hitch-hiking effect of a
mondii. Nature, 469, 411–414. favourable gene. Genetical Research, 23, 23–35.

McGaugh SE, Heil CS, Manzano-Winkler B et al. (2012) Recom- Nosil P, Funk DJ, Ortiz-Barrientos D (2009) Divergent selection
bination modulates how selection affects linked sites in Dro- and heterogeneous genomic divergence. Molecular Ecology,
sophila. PLoS Biology, 10, e1001422. 18, 375–402.
Michel AP, Sim S, Powell THQ et al. (2010) Widespread geno- Pardo-Diaz C, Salazar C, Baxter SW et al. (2012) Adaptive
mic divergence during sympatric speciation. Proceedings of introgression across species boundaries in Heliconius butter-
the National Academy of Sciences, 107, 9724–9729. flies. PLoS Genetics, 8, e1002752.
Mu~ noz MM, Crawford NG, McGreevy TJ et al. (2013) Diver- Payseur BA, Krenz JG, Nachman MW (2004) Differential
gence in coloration and ecological speciation in the Anolis patterns of introgression across the X chromosome in a
marmoratus species complex. Molecular Ecology, 22, 2668– hybrid zone between two species of house mice. Evolution,
2682. 58, 2064–2078.
Nachman MW, Payseur BA (2012) Recombination rate varia- Pease JB, Hahn MW (2013) More accurate phylogenies inferred
tion and speciation: theoretical predictions and empirical from low-recombination regions in the presence of incom-
results from rabbits and mice. Philosophical Transactions of plete lineage sorting. Evolution, 67, 2376–2384.
the Royal Society of London. Series B, Biological Sciences, 367, Pennetier C, Warren B, Dabire KR, Russell IJ, Gibson G (2010)
409–421. “Singing on the wing” as a mechanism for species recogni-
Nadachowska K, Babik W (2009) Divergence in the face of tion in the malarial mosquito Anopheles gambiae. Current Biol-
gene flow: the case of two newts (Amphibia: Salamandridae). ogy, 20, 131–136.
Molecular Biology and Evolution, 26, 829–841. Pinho C, Hey J (2010) Divergence with gene flow: models and
Nadachowska-Brzyska K, Burri R, Olason PI et al. (2013) data. Annual review of ecology, evolution, and systematics, 41,
Demographic divergence history of pied flycatcher and col- 215–230.
lared flycatcher inferred from whole-genome re-sequencing Presgraves DC, Balagopalan L, Abmayr SM, Orr HA (2003)
data. PLoS Genetics, 9, e1003942. Adaptive evolution drives divergence of a hybrid inviabil-
Nadeau NJ, Whibley A, Jones RT et al. (2012) Genomic islands ity gene between two species of Drosophila. Nature, 423,
of divergence in hybridizing Heliconius butterflies identified 715–719.
by large-scale targeted sequencing. Philosophical Transactions Rambaut A, Grassly NC (1997) Seq-Gen: an application for the
of the Royal Society of London. Series B, Biological Sciences, 367, Monte Carlo simulation of DNA sequence evolution along
343–353. phylogenetic trees. Computer applications in the biosciences:
Nadeau NJ, Martin SH, Kozak KM et al. (2013) Genome-wide CABIOS, 13, 235–238.
patterns of divergence and gene flow across a butterfly radi- Reidenbach KR, Neafsey DE, Costantini C et al. (2012) Patterns
ation. Molecular Ecology, 22, 814–826. of genomic differentiation between ecologically differentiated
Neafsey D, Lawniczak M, Park D et al. (2010) SNP genotyping M and S forms of Anopheles gambiae in West And Central
defines complex gene-flow boundaries among African Africa. Genome Biology and Evolution, 4, 1202–1212.
malaria vector mosquitoes. Science, 330, 514–517. Renaut S, Nolte A, Bernatchez L (2009) Gene expression diver-
Nei M (1973) Analysis of gene diversity in subdivided popula- gence and hybrid misexpression between lake whitefish spe-
tions. Proceedings of the National Academy of Sciences, 70, 3321– cies pairs (Coregonus spp. Salmonidae). Molecular Biology and
3323. Evolution, 26, 925–936.
Nei M (1982) Evolution of human races at the gene level. In: Renaut S, Grassa C, Yeaman S et al. (2013) Genomic islands of
Human Genetics, Part A: The Unfolding Genome (ed. Bonne- divergence are not affected by geography of speciation in
Tamir B), pp. 167–181. Alan R. Liss, New York. sunflowers. Nature Communications, 4, 1827.
Nei M (1987) Molecular Evolutionary Genetics. Columbia Univer- Rieseberg LH, Whitton J, Gardner K (1999) Hybrid zones and
sity Press, New York. the genetic architecture of a barrier to gene flow between
Nei M, Li W-H (1979) Mathematical model for studying two sunflower species. Genetics, 152, 713–727.
genetic variation in terms of restriction endonucleases. Pro- Rogers S, Bernatchez L (2005) Integrating QTL mapping and
ceedings of the National Academy of Sciences, 76, 5269–5273. genome scans towards the characterization of candidate loci
Nielsen R, Wakeley J (2001) Distinguishing migration from iso- under parallel selection in the lake whitefish (Coregonus clu-
lation: a Markov chain Monte Carlo approach. Genetics, 158, peaformis). Molecular Ecology, 14, 351–361.
885–896. Rogers S, Bernatchez L (2007) The genetic architecture of eco-
Nolte AW, Freyhof J, Tautz D (2006) When invaders meet logical speciation and the association with signatures of
locally adapted types: rapid moulding of hybrid zones selection in natural lake whitefish (Coregonus sp. Salmonidae)
between sculpins (Cottus, Pisces) in the Rhine system. Molec- species pairs. Molecular Biology and Evolution, 24, 1423–1438.
ular Ecology, 15, 1983–1993. Runemark A, Hey J, Hansson B, Svensson EI (2012) Vicariance
Noor MAF, Bennett SM (2009) Islands of speciation or mirages divergence and gene flow among islet populations of an
in the desert? Examining the role of restricted recombination endemic lizard. Molecular Ecology, 21, 117–129.
in maintaining species. Heredity, 103, 439–444. Sætre G-P, Sæther SA (2010) Ecology and genetics of speciation
Nordborg M (1997) Structured coalescent processes on differ- in Ficedula flycatchers. Molecular Ecology, 19, 1091–1106.
ent time scales. Genetics, 146, 1501–1514. Sætre G-P, Borge T, Lindroos K et al. (2003) Sex chromosome
Nordborg M, Charlesworth B, Charlesworth D (1996) The evolution and speciation in Ficedula flycatchers. Proceedings of
effect of recombination on background selection. Genetical the Royal Society of London. Series B: Biological Sciences, 270,
Research, 67, 159–174. 53–59.
Nosil P (2008) Speciation with gene flow could be common. Sattath S, Elyashiv E, Kolodny O, Rinott Y, Sella G (2011)
Molecular Ecology, 17, 2103–2106. Pervasive adaptive protein evolution apparent in diversity

patterns around amino acid substitutions in Drosophila simu- Arabidopsis lyrata to serpentine soils. Nature Genetics, 42, 260–
lans. PLoS Genetics, 7, e1001302. 263.
Savolainen V, Anstett M-C, Lexer C et al. (2006) Sympatric spe- Via S (2009) Natural selection in action during speciation.
ciation in palms on an oceanic island. Nature, 441, 210–213. Proceedings of the National Academy of Sciences, 106, 9939–
Sella G, Petrov DA, Przeworski M, Andolfatto P (2009) Perva- 9946.
sive natural selection in the Drosophila genome? PLoS Genet- Via S (2012) Divergence hitchhiking and the spread of genomic
ics, 5, e1000495. isolation during ecological speciation-with-gene-flow. Philo-
Shaw KL, Mullen SP (2011) Genes versus phenotypes in the sophical Transactions of the Royal Society of London. Series B,
study of speciation. Genetica, 139, 649–661. Biological Sciences, 367, 451–460.
Shimmin LC, Chang BH, Li W-H (1993) Male-driven evolution Via S, West J (2008) The genetic mosaic suggests a new role for
of DNA sequences. Nature, 362, 745–747. hitchhiking in ecological speciation. Molecular Ecology, 17,
Slatkin M, Wiehe T (1998) Genetic hitch-hiking in a subdivided 4334–4345.
population. Genetical Research, 71, 155–160. Wang Y (2009) Estimating the Process of Speciation for Humans
Smith J, Kronforst MR (2013) Do Heliconius butterfly species and Chimpanzees. PhD Thesis, Rutgers University.
exchange mimicry alleles? Biology Letters, 9, 20130503. Wang Y, Hey J (2010) Estimating divergence parameters with
Song Y, Endepols S, Klemann N et al. (2011) Adaptive introgres- small samples from a large number of loci. Genetics, 184,
sion of anticoagulant rodent poison resistance by hybridiza- 363–379.
tion between Old World mice. Current Biology, 21, 1296–1301. Wang R, Zheng L, Toure YT, Dandekar T, Kafatos FC (2001)
Sousa VC, Carneiro M, Ferrand N, Hey J (2013) Identifying loci When genetic distance matters: measuring genetic differenti-
under selection against gene flow in isolation-with-migration ation at microsatellite loci in whole-genome scans of recent
models. Genetics, 194, 211–233. and incipient mosquito species. Proceedings of the National
Stephan W, Xing L, Kirby DA, Braverman JM (1998) A test of Academy of Sciences, 98, 10769–10774.
the background selection hypothesis based on nucleotide White BJ, Cheng C, Simard F, Costantini C, Besansky NJ (2010)
data from Drosophila ananassae. Proceedings of the National Genetic association of physically unlinked islands of geno-
Academy of Sciences, 95, 5649–5654. mic divergence in incipient species of Anopheles gambiae.
Strasburg JL, Rieseberg LH (2010) How robust are “Isolation Molecular Ecology, 19, 925–939.
with Migration” analyses to violations of the IM model? A Whiteley AR, Derome N, Rogers SM et al. (2008) The
simulation study. Molecular Biology and Evolution, 27, 297– phenomics and expression quantitative trait locus mapping
310. of brain transcriptomes regulating adaptive divergence in
Stump AD, Fitzpatrick MC, Lobo NF et al. (2005) Centromere- lake whitefish species pairs (Coregonus sp.). Genetics, 180,
proximal differentiation and speciation in Anopheles gambiae. 147–164.
Proceedings of the National Academy of Sciences, 102, 15930– Wittbrodt J, Adam D, Malitschek B et al. (1989) Novel putative
15935. receptor tyrosine kinase encoded by the melanoma-inducing
Teeter KC, Payseur BA, Harris LW et al. (2008) Genome-wide Tu locus in Xiphophorus. Nature, 341, 415–421.
patterns of gene flow across a house mouse hybrid zone. Wright S (1931) Evolution in Mendelian populations. Genetics,
Genome Research, 18, 67–76. 16, 97–159.
Teeter KC, Thibodeau LM, Gompert Z et al. (2010) The variable Wright S (1943) Isolation by distance. Genetics, 28, 114–138.
genomic architecture of isolation between hybridizing spe- Wu C-I (2001) The genic view of the process of speciation. Jour-
cies of house mice. Evolution, 64, 472–485. nal of Evolutionary Biology, 14, 851–865.
The Heliconius Genome Consortium (2012) Butterfly genome Yatabe Y, Kane NC, Scotti-Saintagne C, Rieseberg LH (2007)
reveals promiscuous exchange of mimicry adaptations Rampant gene exchange across a strong reproductive barrier
among species. Nature, 487, 94–98. between the annual sunflowers, Helianthus annuus and H.
Tishkoff SA, Reed FA, Ranciaro A et al. (2007) Convergent petiolaris. Genetics, 175, 1883–1893.
adaptation of human lactase persistence in Africa and Eur- Yeaman S (2013) Genomic rearrangements and the evolution of
ope. Nature Genetics, 39, 31–40. clusters of locally adaptive loci. Proceedings of the National
Tripet F, Toure Y, Taylor C et al. (2001) DNA analysis of trans- Academy of Sciences, 110, E1743–E1751.
ferred sperm reveals significant levels of gene flow between
molecular forms of Anopheles gambiae. Molecular Ecology, 10,
1725–1732.
Turner TL, Hahn MW (2007) Locus- and population-specific M.W.H. conceived of the project. T.E.C. carried out the
selection and differentiation between incipient species of analyses. T.E.C. and M.W.H. wrote the paper.
Anopheles gambiae. Molecular Biology and Evolution, 24, 2132–
2138.
Turner TL, Hahn MW (2010) Genomic islands of speciation or
genomic islands and speciation? Molecular Ecology, 19, 848–850. Data accessibility
Turner T, Hahn M, Nuzhdin SV (2005) Genomic islands of spe-
ciation in Anopheles gambiae. PLoS Biology, 3, e285. Data available from the Dryad Digital Repository:
Turner TL, Bourne EC, Von Wettberg EJ, Hu TT, Nuzhdin SV http://dx.doi.org/10.5061/dryad.cc2sk.
(2010) Population resequencing reveals local adaptation of

Molecular Ecology - 2014 - Cruickshank

Uploaded by

Copyright:

Available Formats

Molecular Ecology - 2014 - Cruickshank

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Molecular Ecology - 2014 - Cruickshank

Uploaded by

Copyright:

Available Formats

Molecular Ecology (2014) 23, 3133–3157 doi: 10.1111/mec.

INVITED REVIEWS AND SYNTHESES

researchers have turned to large-scale sequencing of

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

the specific way in which FST is calculated, the key aspect

© 2014 John Wiley & Sons Ltd

to higher between-population differentiation (Kaplan

© 2014 John Wiley & Sons Ltd

Population X Population Y Population X Population Y

© 2014 John Wiley & Sons Ltd

References for sequence

*Based on results Sousa et al. (2013).

proportion to a (see equation 24 in Nordborg 1997). The

effective migration rate, m,~ accounts for differences in da

is equivalent to the average pairwise divergence

tions. Expectations for da in this scenario can be easily

as can the expectation for dXY under a purely allopatric

As Fig. 2 shows, under this model at equilibrium FST,

from the selected locus. Gene flow reduces both abso-

© 2014 John Wiley & Sons Ltd

Box 2. Statistical power to detect differences in effective migration among loci.

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

the islands of relative divergence than it was in

(a) (b) (c)

(d) (e) (f)

© 2014 John Wiley & Sons Ltd

expected in regions of low recombination that experi-

p islands p nonislands Bin type

All data come from the same references as in Table 1.

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

2 MAR: results are not reliable for fewer than 20 sampled

Recently, Sousa et al. (2013) have relaxed the assump-

tion within IM that all loci share a single migration his-

against gene flow, they developed a new version of

loci: those where the data are consistent with high

© 2014 John Wiley & Sons Ltd

(a) (b) Fig. 5 Speciation-with-gene-flow model

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

Population X Population Y Population X Population Y

No linked selection Recurrent selection

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

© 2014 John Wiley & Sons Ltd

You might also like