Sympatric speciation in birds is rare: insights from range data and simulations.

Abstract

Sympatric speciation is now accepted as theoretically plausible and a likely explanation for divergence in a handful of taxa, but its contribution to large-scale patterns of speciation remains contentious. A major problem is that it is difficult to differentiate between alternate scenarios of geographic speciation when species ranges have shifted substantially in the past. Previous studies have searched for a signal of the geographic mode of speciation by testing for a correlation between time since speciation and range overlap. Here we use simulations to show that the proportion of species showing zero or complete range overlap are more reliable indicators of the geography of speciation than is the correlation between time since speciation and overlap. We then apply these findings to the distributions of 291 pairs of avian sister species. Although 49% of pairs show some overlap in their ranges, our simulations show that this is not surprising under allopatric models of speciation. More revealingly, less than 2% show complete range overlap. Our simulations demonstrate that the observed patterns are most consistent with a model in which allopatric speciation is dominant but in which sympatric speciation is also present and contributes 5% of speciation events.

Full text

Page 1

vol. 171, no. 5the american naturalistmay 2008
?
Sympatric Speciation in Birds Is Rare: Insights from
Range Data and Simulations
Albert B. Phillimore,1,*C. David L. Orme,1,†Gavin H. Thomas,1,‡Tim M. Blackburn,2,§Peter M. Bennett,3,k
Kevin J. Gaston,4,#and Ian P. F. Owens1,**
1. Division of Biology and Natural Environment Research Council
Centre for Population Biology, Imperial College London, Silwood
Park Campus, Ascot, Berkshire SL5 7PY, United Kingdom;
2. Institute of Zoology, Zoological Society of London, Regents
Park, London NW1 4RY, United Kingdom;
3. Durrell Institute of Conservation and Ecology, Marlowe
Building, University of Kent, Canterbury, Kent CT2 7NR, United
Kingdom;
4. Biodiversity and Macroecology Group, Department of Animal
and Plant Sciences, University of Sheffield, Alfred Denny Building,
Western Bank, Sheffield S10 2TN, United Kingdom
Submitted June 22, 2007; Accepted October 22, 2007;
Electronically published March 12, 2008
Online enhancements: appendix tables.
abstract: Sympatric speciation is now accepted as theoretically
plausible and a likely explanation for divergence in a handful of taxa,
but its contribution to large-scale patterns of speciation remains
contentious. A major problem is that it is difficult to differentiate
between alternate scenarios of geographic speciation when species
ranges have shifted substantially in the past. Previous studies have
searched for a signal of the geographic mode of speciation by testing
for a correlation between time since speciation and range overlap.
Here we use simulations to show that the proportion of species
showing zero or complete range overlap are more reliable indicators
of the geography of speciation than is the correlation between time
since speciation and overlap. We then apply these findings to the
distributions of 291 pairs of avian sister species. Although 49% of
* Corresponding author; e-mail: albert.phillimore@imperial.ac.uk.
†E-mail: d.orme@imperial.ac.uk.
‡E-mail: g.thomas@imperial.ac.uk.
§E-mail: tim.blackburn@ioz.ac.uk.
kE-mail: p.m.bennett@kent.ac.uk.
#E-mail: k.j.gaston@sheffield.ac.uk.
** E-mail: i.owens@imperial.ac.uk.
Am. Nat. 2008. Vol. 171, pp. 646–657. ? 2008 by The University of Chicago.
0003-0147/2008/17105-42685$15.00. All rights reserved.
DOI: 10.1086/587074
pairs show some overlap in their ranges, our simulations show that
this is not surprising under allopatric models of speciation. More
revealingly, less than 2% show complete range overlap. Our simu-
lations demonstrate that the observed patterns are most consistent
with a model in which allopatric speciation is dominant but in which
sympatric speciation is also present and contributes 5% of speciation
events.
Keywords: speciation, allopatric, sympatric, geographic range, birds,
phylogeny.
Sympatric speciation has been at the center of one of the
most protracted debates within evolutionary biology
(Walsh 1864; Jordan 1905; Mayr 1942; Bush 1975; Fu-
tuyma and Mayer 1980; Berlocher and Feder 2002; Coyne
and Orr 2004). Although sympatric speciation had been
largely dismissed in many earlier reviews of speciation
(e.g., Mayr 1942; Futuyma and Mayer 1980), recent theory
has demonstrated that it is possible under certain con-
ditions (e.g., Dieckmann and Doebeli 1999; Gavrilets and
Waxman 2002), and empirical studies have provided ex-
amples for which a sympatric origin of species appears to
be the most parsimonious explanation (e.g., Sorenson et
al. 2003; Savolainen et al. 2006). The argument has there-
fore shifted from a debate on the plausibility of sympatric
speciation to an examination of its frequency relative to
other geographic modes of speciation (Jiggins 2005).
Biologists have long suspected that contemporary geo-
graphic distributions of closely related species offer a win-
dow into the historical processes that have generated those
species (Jordan 1905, 1908; Jordan and Kellogg 1907; Mayr
1942). In many groups, recently diverged sister species
tend to have allopatric rather than sympatric distributions,
a classic pattern (sometimes referred to as Jordan’s law)
that has traditionally been interpreted as evidence for the
predominance of allopatric speciation (Wagner 1873; Jor-
dan1905, 1908; Allen1907; Jordan and Kellogg1907;Mayr
1942; but see Anderson and Evensen 1978). However, one
of the difficulties of using range overlap to study the geo-
graphic mode of speciation is that because of changes in
distribution over time (e.g., Lessa et al. 2003), current

Page 2

Sympatric Speciation in Birds Is Rare647
patterns of range overlap may not be representative of
range overlap at the time of speciation (Brown and Gibson
1983). Great caution must therefore be taken when at-
tempting to use contemporary range information to infer
past speciation processes (Chesser and Zink 1994; Losos
and Glor 2003).
Using methods that allow for recent geographic range
changes, several authors have suggested that under an al-
lopatric model, the correlation between node age and
range overlap after speciation should be positive, and the
intercept should be close to 0. Conversely, under a sym-
patric model, a negative correlation and an intercept close
to 1 are expected (Barraclough et al. 1998; Berlocher 1998;
Barraclough and Vogler 2000; Ribera et al. 2001). The few
studies that have applied this approach to contemporary
distribution data have generally suggested that allopatric
speciation is the predominant mode of speciation, but in
situations where there may have been large changes in
geographic ranges, it has typically proven difficult to reject
the possibility that a substantial proportion of speciation
events could have been sympatric (Barraclough and Vogler
2000; Fitzpatrick and Turelli 2006; Perret et al. 2007).
The overall aims of this study were twofold: (i) to use
spatial simulations to test the ability of several indices of
range overlap to differentiate between alternate scenarios
of speciation and (ii) to apply these indices to avian dis-
tribution data in order to reveal the relative contributions
made by allopatric and sympatric speciation. One of the
problems associated with phylogenetic tests is the low in-
formation content that can be derived from ancient phy-
logenetic nodes (Berlocher and Feder 2002; Fitzpatrickand
Turelli 2006). To overcome this problem, we therefore re-
strict our analyses to recently diverged pairs of sister spe-
cies (Stephens and Wiens 2003; Coyne and Orr 2004). We
test whether the contemporary distributions of 291 avian
species pairs depart from the null expectations generated
under: (a) solely allopatric models of speciation or (b)
models in which sympatric speciation constitutes (i) 5%,
(ii) 10%, or (iii) 25% of speciation events. There are cer-
tain logistical advantages of conducting these analyses on
birds, namely, the abundance of well-resolved genus-level
phylogenetic information and a global database on geo-
graphical breeding ranges (Orme et al. 2005). While pre-
vious reviews of speciation in birds have tended to favor
a predominantly allopatric mode ofspeciation(Mayr1942,
1963; Cracraft 1982; Chesser and Zink 1994; Friesen and
Anderson 1997; Barraclough and Vogler 2000; Coyne and
Price 2000; Johnson and Cicero 2002; Price 2008), recent
empirical evidence consistent with sympatric speciation is
now emerging, notably in the brood-parasitic Vidua
finches (Sorenson et al. 2003; Balakrishnan and Sorenson
2006) and band-rumped storm petrels (Oceanodrama cas-
tro; Harris 1969; Monteiro and Furness 1998; Smith and
Friesen 2007). This suggests that a variety of geographic
modes of speciation is plausible in birds. Birds also rep-
resent a challenging case because of their great mobility
compared to that of other groups (Newton 2003),meaning
range changes are likely to have been particularly extensive
(Fitzpatrick and Turelli 2006). Consequently, if it is pos-
sible to reject a high incidence of sympatric speciation in
birds, in spite of frequent range change, then our simu-
lation approach is likely to be applicable to other less
mobile taxa.
Methods
Simulations
In order to explore how different degrees of geographic
range overlap between pairs of sister species can arise un-
der different historical scenarios, we developed a series of
simulations. Each simulation replicate was generated by
randomly dividing an initial geographic range of known
size into two “daughter” ranges, with the positions of these
daughter ranges in relation to one another being depen-
dent on the speciation model employed. These two daugh-
ter ranges, which represent the geographic ranges of a pair
of sister species after a speciation event, were then allowed
to move independently of each other for a number of time
steps. For each replicate, there were four main parameters
that could be varied: (i) the geographic mode ofspeciation,
(ii) the number of time steps, (iii) the starting size of the
ranges, and (iv) the rate at which ranges move. At the end
of each replicate, we recorded the extent of range overlap
between the two ranges and calculated a series of indices,
with the aim of discovering whether any of these indices
was able to recover the original mode of speciation under
a wide range of conditions.
We simulated the spatial movements of ranges, includ-
ing changes in both position and size, under several geo-
graphic and speciation scenarios. All simulations were per-
formed on a square plane of 100 units # 100 units. Our
simulation approach was similar in many ways to that of
Barraclough and Vogler (2000), although there were sev-
eral important differences, the most significant being that
we modeled range change and overlap between just two
species (for a simulation approach that shares several sim-
ilarities with ours, see Waldron 2007), whereas Barra-
clough and Vogler modeled these processes through mul-
tiple successive speciation events.
The simulations were used to model vicariant(allopatric
speciation via a barrier bisecting a range), peripatric (al-
lopatric speciation via dispersal at the periphery of a
range), parapatric (speciation in nonoverlapping space but
abutting areas in the face of gene flow), and sympatric
(speciation in completely overlappingspace)modesofspe-

Page 3

648The American Naturalist
ciation. We simulated vicariant allopatric speciation by
placing the larger of the two ranges randomly on the grid,
subject to the constraint that there had to be a minimum
of 2 units plus the square root of the smaller species range
to the right of the larger species range. This was to allow
room for the smaller species range, which was always
placed 2 units to the right of the first, although its position
on the vertical axis was permitted at any point along the
vertical extent of the larger range. Under peripatric spe-
ciation, the larger of the two ranges was placed randomly
on the grid by using two sets of random uniform coor-
dinates, with the condition that the entire range had to
fall within the confines of the grid. The second species
range was then placed randomly on the grid, with the
additional constraint that the two species ranges could not
overlap. Parapatric speciation was simulated in a fashion
identical to that of vicariant speciation, except for the
omission of the 2-unit spatial separation of ranges. We
simulated sympatric speciation by placing the larger range
randomly on the grid and then placing the smaller species
range randomly so that its range was completely over-
lapped by the species with the larger range.
Under the vicariant speciation model,thepostspeciation
range sizes of species were obtained by randomly breaking
a stick of r units2(where r was one of 100, 1,000, or 2,000)
into two parts (MacArthur 1957; Anderson and Evensen
1978; Barraclough and Vogler 2000; but for criticism of
this approach, see Waldron 2007). On average, the smaller
of the two ranges generated by stick breaking was one-
third the size of the larger. To minimize the differences
between alternative speciation models, we used stick
breaking to simulate postspeciation range sizes under sym-
patric and parapatric speciation. Under peripatric speci-
ation, the range size of the peripheral isolate is expected
to be considerably smaller than that of its parent range.
To simulate this scenario, we obtained the smaller range
size by multiplying r by a value drawn from the g distri-
bution with shape and rate parameters of 0.1 and 10,
respectively. The larger range size was obtained as r minus
the smaller range. Under this peripatric speciation model,
the smaller range was !10% of r in 197% of simulations.
All the initial species ranges were square. Earlier simula-
tions of range overlap through time conducted by Fitz-
patrick and Turelli (2006) found that circular and square
ranges returned qualitatively similar results.
We simulated range change by adding a different ran-
dom normal deviate of a mean of 0 and standard deviation
of 0.5 to vectors corresponding to the top, bottom, right,
and left extents of each species range. This process was
repeated at each time step and allowed each of the four
sides of the species range to move in two directions in-
dependently of one another; this meant that unlike Bar-
raclough and Vogler’s (2000) simulations, range change
was not related to range size. However, ranges were not
allowed to expand beyond the limits of the grid and were
always rectangular. We also explored a faster rate of range
change with a mean of 0 and standard deviation of 2. We
conducted simulations of varying average duration: short
(obtained as rounded values from a random normal dis-
tribution with andmean p 100
a random normal distribution with
), and long (from a random normal distributionSD p 50
with andmean p 1,000 SD p 250
ranges contracted to 0 (or less) at any stage of the sim-
ulations (extinction), we started a new simulation with the
same simulation duration as the failed one (although the
starting range sizes and coordinates would be reallocated
randomly). Most simulations were conducted with the vi-
cariant speciation model (combined with varying fre-
quencies of sympatric speciation), using a combined initial
range size of 1,000 units2, a range change rate with a ran-
dom normal deviate of 0.5, and a simulation duration
averaging 100 time steps. We make it clear when simu-
lations depart from these default settings.
At the end of each simulation replicate, the proportion
of overlap between the two species ranges was calculated
as the area of overlap divided by the area of the smaller
range, following Chesser and Zink (1994). The spatial sim-
ilarity between the starting and the finishing position of
each range was assessed for each simulation using Jaccard’s
similarity coefficient; that is, the area of overlap between
the beginning and the ending range was divided by the
sum of the area covered by the two ranges. We then cal-
culated the average of the Jaccard similarity coefficients
for the two species ranges for each simulation.
To explore the influence of the geographic mode of
speciation, the proportion of speciation events that were
allopatric (vicariant, peripatric, or parapatric) versus sym-
patric was varied from 0 to 1 in iterations of 0.1. For each
selected combination of parameter values, we conducted
10,000 separate simulations, after which we estimated sev-
eral range overlap parameters. Using Kendall’s t, we es-
timated the correlation between range overlap and (i) sim-
ulation duration and (ii) the smaller range size. Kendall’s
t was preferred to parametric methods because of the very
high incidences of 0s and/or 1s in the overlap data. We
calculated the proportion of replicates that showed zero
overlap and complete overlap and the degree to which
simulations produced bimodal distributions, that is, high
instances of zero and complete overlap cases. We quan-
tified bimodality as(z # c)/(0.5 # n)
the number of individual simulations that resulted in zero
and complete overlap, respectively. The number of sim-
ulations (in this case 10,000) was denoted as n. The re-
sulting value lies between 0 and 1, where a value of 0
indicates no bimodality and a value of 1 corresponds to
), medium(from
mean p 200
SD p 25
and
). If either ofthespecies
, where z and c are

Page 4

Sympatric Speciation in Birds Is Rare 649
complete bimodality; that is, the cases are divided evenly
between those showing zero and complete overlap. Finally,
we calculated the median Jaccard similarity between the
range position at the outset and at the end of simulations
across each set of replicates. R code for running the sim-
ulations is available from A.B.P. on request.
Speciation in Birds
We identified 568 pairs of avian sister species, defined here
as the two most closely related extant descendants of a
common ancestor, by using two approaches. The first ap-
proach involved identifying monophyletic sister species
from published molecular phylogenies that included more
than 80% of the species in a clade. An 80% completeness
threshold was selected as a compromise between identi-
fying true sister taxa and maximizing the number of avail-
able pairs. The second approach was taxonomy based; two
species were recognized as sisters if their genus included
just two species. We used the taxonomy of Sibley and
Monroe (1990, 1993). A corollary of using this taxonomy
is that our species delimitation should generally corre-
spond to the biological species concept. Where the phy-
logeny and taxonomy conflicted, we adopted the phylo-
genetically identified sister pairs. Our expectation was that
although it is likely that a small fraction of the pairs of
species we identify in this manner may not be true sister
species, they would be close relatives. Furthermore, all
species-pair comparisons will be phylogenetically inde-
pendent of one another (Harvey and Pagel 1991).
Information on the geographical distribution of the
pairs of sister species was obtained from a database of
breeding range extent ArcGIS shapefiles (Orme et al. 2005,
2006). The range sizes of the focal species and the extent
of range overlap between sister species (both in km2) were
estimated from these data by using ArcGIS code written
by C.D.L.O. Across all 568 pairs of sister species, we also
calculated the incidences of both zero and complete over-
lap.
We conducted more thorough range overlap analyses
on a smaller sister pair data set that was designed to min-
imize the problem of major recent range shifts (as rec-
ommended by Coyne and Orr [2004]). We excluded spe-
cies found in the Holarctic, where major vegetation shifts
and concomitant avian range shifts occurred during the
past 22,000 years following the last ice age (Rand 1948;
Pianka 1966; Pielou 1991; Lessa et al. 2003). We also ex-
cluded pairs where at least one member was restricted to
islands smaller than 500,000 km2because in such cases
our unconstrained range change simulations would be
inappropriate.
For pairs of species in the reduced data set, we estimated
the genetic distance between sister species from mito-
chondrial sequence data stored in GenBank. The program
Geneious (Drummond et al. 2006) was used to search for
sequence data from two frequently sequenced mitochon-
drial protein coding genes, cytochrome b (cyt b) and
NADH dehydrogenase subunit II (ND2). Each species was
represented by a maximum of six sequences, and efforts
were made to sample from multiple subspecies if available.
Sequences were aligned using CLUSTAL W (Thompson
et al. 1994) in MEGA3.1 (Kumar et al. 2004) and checked
by eye. For each pair of species, the maximum overlapping
region of nucleotide base pairs was calculated, and se-
quences that spanned more than 75% of this region were
included in the analysis of genetic distance. For both cyt
b and ND2, the mean proportional pairwise differences in
sequence bases were estimated between species. We esti-
mated relative time since divergence of sister species by
using the proportional distances and made the assumption
that the rates of nucleotide substitution in cyt b and ND2
were similar and clocklike. The mean genetic distance es-
timates across the two genes were calculated from the pro-
portional distances after arcsine transformation.
Across the pairs of avian sister species, we calculated
the range overlap indices described for the simulations
above; these were (i) the correlation between time since
speciation (estimated using genetic distance) and propor-
tion of range overlap, (ii) the correlation between the
smaller range size and the proportion of range overlap,
(iii) the proportion of cases showing zero overlap, (iv) the
proportion of cases showing complete overlap, and (v) the
degree to which the range overlap data are bimodal. To
allow for mapping and rounding error, we defined com-
plete overlap as those cases where overlap was 10.9999.
Sympatric species that hybridize can be misidentified as
sister species on the basis of their mitochondrial gene trees
(Funk and Omland 2003). To minimize the potential for
hybridization inflating the frequency of apparent sympatry
(Fitzpatrick and Turelli 2006), we identified those species
pairs known from phylogenetic sources for which natural
hybridization has been inferred or reported (McCarthy
2006). We conducted analyses both including and exclud-
ing phylogenetic sister species that are known to hybridize.
A related problem is that hybridization may reduce the
estimated genetic distance between sister species that are
in parapatry or sympatry. Consequently, we excluded hy-
bridizing pairs from the analysis of the relationship be-
tween genetic distance and range overlap.
To identify the most likely mode of speciation in birds,
we examined whether the observed value for each of the
five indices lay within the expected distributions generated
for (i) a high incidence of sympatric speciation (25% of
all speciation events), (ii) a low incidence of sympatric
speciation (10% of all speciation events), (iii) a low in-
cidence of sympatric speciation (5% of all speciation

Page 5

650The American Naturalist
events), and (iv) a zero incidence of sympatric speciation
(i.e., a solely allopatric model). To allow for sampling error
in our collection of species pairs, each single range change
simulation hadprobability of being allopatric and s1 ? s
probability of being sympatric (where s was 0.25, 0.10,
0.05, or 0). We adopted parameter values that we consid-
ered to be biologically plausible and conservative with re-
spect to the null hypotheses involving sympatric specia-
tion. The null hypotheses were generated via 1,000
simulations, keeping the sample size equal to that in the
observed data sets. All tests were two tailed.
Results
Simulations
Our simulations showed that when speciation was solely
allopatric, approximately half of the cases showed zero
range overlap (under the vicariant mode of speciation and
default parameters), with most of the remainder showing
intermediate range overlap and very few cases showing
complete overlap (fig. 1A, 1D). Alternatively, when spe-
ciation was an equal mixture of allopatric and sympatric,
the distribution of overlap was bimodal, with many cases
of zero and complete overlap and fewer intermediate cases
(fig. 1B, 1E). Finally, under a purely sympatric speciation
model, there were many cases of complete or high overlap
(overlap 10.5) and very few cases of zero overlap (fig. 1C,
1F). In all speciation scenarios, however, the relationships
between the degree of range overlap and time since spe-
ciation (fig. 1A–1C) or smaller range size were noisy (fig.
1D–1F).
In general, the simulations demonstrated that the cor-
relation between time since speciation and the degree of
range overlap tends to be very weak (?0.11 ! Kendall’s
t ! 0.08; fig. 2A; table A1 in the online edition of the
American Naturalist). There was relatively little variation
in the estimated correlation coefficient when we consid-
ered different simulation durations, starting range sizes,
or range change rates. Under some speciation scenarios,
the analyses conducted on the simulations returned the
positive and negative correlations predicted under pre-
dominantly allopatric and sympatric modes, respectively.
However, given small initial range sizes, rapid range
change, or long simulation durations, the simulations
showed little evidence for the predicted negative correla-
tion under the solely sympatric model(Kendall’st 1?0.04
in all cases). When the modes of speciation were 50%
allopatric and 50% sympatric, both positive and negative
correlations between time since speciation and range over-
lap were obtained, depending on thesimulationconditions
employed (fig. 2A; table A1). Therefore, although a cor-
relation as high as 10.1 would indicate allopatricspeciation
and a correlation as low as ?0.04 would indicatesympatric
speciation, intermediate correlation values could arise
from several different geographic speciation scenarios.
Peripatric and parapatric speciation models returned age
range correlations that were qualitatively very similar to
those of the vicariant model (table A1).
Under the default simulation parameters (vicariant spe-
ciation, medium range size, slow range change, and short
simulation duration), the size of the smaller range pos-
sessed useful properties as a correlate of range overlap (fig.
2B), with allopatric and sympatric models of speciation
returning positive and negative correlations, respectively.
However, when range changes were more pronounced
(because of a faster rate of range change, a smaller starting
range size, or a longer simulation duration), the correla-
tion tended to be positive or only weakly negative, even
under the solely sympatric speciation scenario. Although
the parapatric model returned results similar to those of
the vicariant model, the correlations produced by the peri-
patric model were highest at intermediate frequencies of
sympatric speciation and produced negative correlations
only when the frequency of sympatric speciation was very
high. This pattern can be explainedby thedifferentstarting
range sizes generated under the two speciation scenarios;
the smaller of the two postspeciation ranges will be larger
under the sympatric mode, thus explaining the correlation
between sympatry and range size. A combination of these
factors implies that a positive correlation is not reliable
evidence for a primarily allopatric mode of speciation.
Conversely, a negative correlation may still be a useful
indication of the presence of substantial sympatric spe-
ciation, but this approach will lack power if range change
has been extensive.
As the proportion of sympatric speciation events in-
creased, the proportion of cases showing zero over-
lap tended to decline (fig. 2C). Under the default and
medium-duration conditions, an increase in the frequency
of sympatry resulted in a consistent decrease in the pro-
portion of cases showing zero overlap, from more than
0.55 to less than 0.05. A smaller starting range size resulted
in qualitatively similar proportions when allopatric spe-
ciation predominated, but when all speciation was sym-
patric, a considerable proportion of species pairs still
showed zero range overlap (approximately 20%). Fur-
thermore, when the rate of range change was increased,
there was a corresponding decrease in the proportion of
cases that showed zero overlap when sympatry was rare
and an increase in the frequency of zero overlap when
sympatry was common. The parapatric model again re-
turned results similar to those of the vicariant model, ex-
cept for a slight decline in the proportion of cases showing

Page 6

Sympatric Speciation in Birds Is Rare 651
Figure 1: Relationship between the extent of overlap in the geographic ranges of sister species and time since speciation (A–C) and range size (D–
F) under different geographical speciation scenarios. Plots illustrate the results obtained from 10,000 simulations. A and D represent 100% allopatric
speciation (vicariant model), B and E represent 50% allopatric and 50% sympatric, and C and F represent 100% sympatric speciation. Simulations
were run for an average of 100 time units (short duration), the combined starting range size was 1,000 units2, and the standard deviation used for
simulating rate of range change was 0.5. Time since speciation represents the duration for which each single speciation and range change simulation
was run. Smaller range size is the size in arbitrary square units of the smaller of the two sister species ranges at the end of a simulation. For each
pair of species, proportion of range overlap was estimated as the proportion of the smaller range that overlapped with the larger range.
zero overlap when sympatric speciation was infrequent.
The peripatric model also departed from the vicariant
model when sympatric speciation was rare, but under this
mode of speciation the proportion of cases showing zero
overlap was much higher (approximately 0.9 when peri-
patric speciation was ubiquitous; table A1). Thus, low pro-
portions (!0.1) of zero overlap are robust indicators that
sympatric speciation is frequent. Depending on the dom-
inant forms of allopatric speciation, either high (10.5) or
very high (10.8) incidences of zero overlap may be re-
quired to provide robust evidence to reject frequent sym-
patric speciation. Again, intermediate proportions of cases
showing zero overlap arise under several speciation and
range change scenarios.
A large proportion of cases showing complete sympatry
were a robust indicator of frequent sympatric speciation
under a variety of speciation models and simulation pa-
rameters (fig. 2D; table A1). When initial range size was
small, range change was fast, or simulation duration was
long, the proportion of cases showing complete overlap
was low. This means that a low proportion of complete
sympatry (often considered indicative of allopatric spe-
ciation) can arise under a range of scenarios and is likely
to have low power to reject hypotheses postulating a role
for sympatric speciation.
A low bimodality score arose from a prevalence of either
allopatric or sympatric speciation or as a result of erosion
of the geographic signal of speciation, such as may occur

Page 7

652The American Naturalist
Figure 2: Relationships between the proportion of speciation events that are sympatric rather than allopatric and five indices of range overlap under
different model parameters. Points represent values obtained from 10,000 simulations. A, Kendall’s t correlation statistics between time since
speciation and the proportion of range overlap at different frequencies of sympatric speciation. B, Kendall’s t correlation between the smaller range
size and the proportion of range overlap. C, Proportion of cases where range overlap was 0. D, Proportion of cases where range overlap wascomplete.
E, Median bimodality score. Circles represent the results of simulations using the default parameters, that is, short simulation duration (100 time
units), a combined range size of 1,000 units2, and slow rate of range change (
(200 time units). Plus signs differ from the default in the rate of range change (
combined range size (100 units2).
). Triangles differ from the default only in simulation duration
). Crosses differ from the default in the use of a smallerSD p 2.0
SD p 0.5
through rapid range change or long simulation duration
(fig. 2E). A high bimodality score (10.15) will arise only
if speciation is a mixture of allopatric and sympatric and
the geographic signature of speciation is relatively well
preserved. At intermediate frequencies of sympatric spe-
ciation, the peripatric and parapatric models resulted in
higher and lower bimodality scores,respectively(tableA1).
As expected, the similarity between the starting post-
speciation range position and the range position at the
end of simulations was eroded when the simulation du-
ration was increased, the rate of range change was in-
creased, or the starting range size was reduced (table A1).
Similarity was also lower under the peripatric model of
range splitting, as a result of small ranges being more likely
to differ in position between the start and the end of the
simulation. Ideally, an observed similarity estimate could

Page 8

Sympatric Speciation in Birds Is Rare653
Figure 3: Correlation between range overlap in continental non-Holarctic sister species and genetic distance (A) and smaller range size (B). Genetic
distance is the back-transformed mean of the arcsine-transformed proportional distances obtained from cyt b and ND2 sequences.
be calculated by incorporatingpaleontologicalinformation
to compare the distribution of species at two time slices;
this could then be used to inform a study of the actual
rate of range change and suitable simulation parameters.
Speciation in Birds
We identified a total of 568 avian sister pairs (table A2 in
the online edition of the American Naturalist), of which
295 (51.9%) showed no range overlap and just eight pairs
(1.4%) showed complete range overlap. The low incidence
of complete overlap was reflected in a low bimodalityscore
of 0.029. The exclusion of 53 species pairs that had been
identified using phylogenies and that were known to hy-
bridize did not qualitatively change these patterns (results
not shown).
After restricting our database to continental species not
inhabiting the Holarctic, we were left with 291 species
pairs. Of these, 149 (51.2%) showed zero range overlap,
and four (1.4%) showed complete range overlap. Again,
the rarity of cases of complete sympatry resulted in a low
bimodality score (0.028). The correlation between genetic
distance and overlap was not significantly different from
0 (,,
t p 0.053 P p .48 n p 98
lation between smaller range size and range overlap was
positive and significant (
t p 0.187 P ! .001 n p 291
3B).
We then tested whether our results for the reduced data
set of species pairs not found in the Holarctic were sig-
; fig. 3A), but the corre-
,, ; fig.

Page 9

654The American Naturalist
Table 1: Tests of agreement between the values of five indices of the geographic mode of speciation observed in
analyses of avian sister species and the values expected under different geographic scenarios
Allopatric
mode of
speciation
Proportion of
sympatric
speciation
events
Simulation
duration
Indices of geographic mode of speciation
Proportion of
cases with
overlap p 0
Proportion of
cases with
overlap p 1
Age range
correlation
Correlation with
smaller range
size
Bimodality
score
Vicariant
Vicariant
Vicariant
Vicariant
Peripatric
Peripatric
Peripatric
Peripatric
Vicariant
Vicariant
Vicariant
Vicariant
0 100
100
100
100
100
100
100
100
1,000
1,000
1,000
1,000
?.089**
?.062*
?.029
.058
?.426**
?.381**
?.333**
?.196**
.131**
.144**
.151**
.182**
.007
?.007
?.020*
?.065**
.011*
?.003
?.017
?.062**
?.027**
?.031*
?.034**
?.041**
.017
.017
.025
.044
.030
.029
.040
.058
?.010
?.005
.004
.013
.065
.082
.094*
.135**
.119**
.019
?.037
?.122**
?.005
.001
.008
.022
.011
?.020
?.050*
?.115**
.015
?.034
?.080**
?.187**
?.037*
?.037*
?.039*
?.043**
.05
.1
.25
0
.05
.1
.25
0
.05
.1
.25
Note: All statistics represent the observed value minus the median expected value obtained from 1,000 simulations. For details regarding
the calculation of different statistics, see “Methods.”
* Two-tailed significance at.P ! .05
** Two-tailed significance at.P ! .01
nificantly different from the expectation under (i) a solely
allopatric, (ii) a 5% sympatric, (iii) a 10% sympatric, or
(iv) a 25% sympatric mode of speciation. We did this by
simulating the null expectation for a set of chosen param-
eter values. In the absence of substantial fossil evidence
regarding range change parameters, we compared the ob-
served parameter values to those expected under several
null hypotheses, including both vicariant and peripatric
modes of allopatric speciation.
The observed values from the pairs of aviansisterspecies
were most similar to those generated under the short-
duration vicariant/sympatric null model, with sympatric
speciation comprising 5% of events (table 1). Theobserved
data showed fewer cases of zero overlap than expected
under the solely allopatric scenario but also showed fewer
cases of complete overlap and a lower bimodality score
than expected under the 10%–25% sympatric simulations.
The peripatric/sympatric simulations showed very little
agreement with the observed data as a result of the ob-
served frequency of zero range overlaps being considerably
lower than thatexpected.Moreover,whenthecontribution
of sympatric speciation was high (25%), then the observed
bimodality and the incidence of zero overlap were both
significantly lower than expected. The observed results dif-
fer substantially from those expected under a vicariant/
sympatric model run for 1,000 time steps, as the incidence
of zero overlap and the bimodality score were consistently
greater and lower than the null expectation, respectively.
Discussion
Our simulations suggest that the most reliable indicators
of the geographic mode of speciation are the proportions
of cases showing zero overlap and complete overlap. If
speciation tends to be allopatric, a high proportion of cases
will show zero overlap, and a low proportion will show
complete overlap, while the reverse is true if speciation
tends to be sympatric. These two indices are less sensitive
to changes in range size or rate of range change than the
other indices we explored. When used in conjunction with
other indices, the simple bimodality score that we have
introduced represents a robust means of examining
whether the geographic mode of speciation varies or
whether a single mode predominates. Across a large data
set of the geographic ranges of avian sister species, we
found that the incidence of zero overlap is high while the
incidence of complete overlap is very low. Althoughalmost
half the sister species show some range overlap, this does
not depart from the expectation under several allopatric
speciation models. Rather, we find that the rarity of cases
of complete overlap and the low bimodality of the data
provide the strongest evidence for the low frequency of
sympatric speciation. Under the short-duration vicariant/
sympatric simulations, we can robustly reject a hypothesis
under which sympatric speciation comprises 25% or 10%
of speciation events. However, we also rejected the solely
allopatric model, anditwas in fact the5%sympatricmodel

Page 10

Sympatric Speciation in Birds Is Rare655
that exhibited the fewest significant departures from the
null expectation.
Our estimate that approximately 49% of pairs of avian
sister species show some range overlap may be, however,
an overestimate of the frequency of syntopy. This is be-
cause species distribution maps reflect the extent of oc-
currence, a measure that is likely to overestimate the area
of occupancy (Orme et al. 2005, 2006). This does not affect
the main conclusion that sympatric speciation is rare.
However, it may serve to reduce the disagreement between
the null expectation under the 100% vicariant scenario
and the observed frequency of zero overlap.
In light of the results of our analyses of avian sister
species, we suggest that recent high-profile studies re-
porting potential instances of sympatric speciation in birds
(Sorenson et al. 2003; Smith and Friesen 2007) are unlikely
to represent a commonplace mechanism of avian speci-
ation (Price 2008). Given that a combination of peripatric,
vicariant, or parapatric speciation is likely to explain the
vast majority of avian speciation events, both species-level
studies (e.g., Irwin et al. 2005; Seddon and Tobias 2007)
and broader comparative analyses (Lynch 1989; Chesser
and Zink 1994; Barraclough and Vogler 2000; Fitzpatrick
and Turelli 2006) should make distinguishing among the
relative contributions a priority. However, the results of a
recent simulation study suggest that recovering the true
contribution of each mode is likely to be challenging
(Waldron 2007).
Our simulations suggest that the correlation between
species age and the degree of range overlap, the index
employed in a number of earlier studies (e.g., Barraclough
et al. 1998; Berlocher 1998; Barraclough and Vogler 2000),
lacks power to distinguish between alternative speciation
scenarios, meaning that the approach is likely to be of
limited utility in highly mobile species. This is largely in
agreement with earlier studies that observed that the age
range overlap correlation approach had low power to dis-
tinguish between alternative scenarios (Barraclough and
Vogler 2000; Fitzpatrick and Turelli 2006). An additional
issue with the age range correlation approach is that the
genetic distances estimated between sister species maycap-
ture different stages of the speciation process, depending
on the geography of speciation. If speciation takes place
in allopatry, genetic distance should relate to the timing
of geographical isolation, whereas if speciation takes place
in sympatry or parapatry, genetic distance should relate
to the point when reproductive isolation declined to zero.
The bias introduced here will depend on the time taken
for sympatric/parapatric speciation to be completed and
will be minimized if sympatric speciation occurs on very
short timescales. When range change was limited, the cor-
relation between the degree of range overlap and the size
of the smaller range showed useful properties as an in-
dicator of the mode of speciation. However, this index is
likely to be of limited practical value because if geograph-
ical distributions change substantially over time or if the
dominant mode of allopatric speciation is peripatry, the
correlation will always tend to be positive. A negative cor-
relation between range size and overlap nonetheless con-
stitutes strong evidence for substantial sympatric speci-
ation.
One consequence of using a simulation approach to
study speciation scenarios is that it requires assumptions
regarding the properties of ranges, the space available to
species, and range change over time. Fortunately, several
of the indices explored here, such as the proportion of
cases showing complete range overlap, appeared to be rel-
atively robust to changes in many of the parameters in-
cluded in the models. Biological scenarios exist that may
render the simple simulations presented here unrealistic.
For example, while we assume that, postspeciation, the
ranges of species move independentlyofoneanother,com-
petition and adaptation to particular environments and
niches are likely to be major determinants of species dis-
tributions. If competition between species that have orig-
inated in sympatry is strong, then this may lead to the
rapid transition to allopatric distributions (Barraclough
and Vogler 2000). Alternatively, species may originate in
allopatry, in which case, because of phylogenetic niche
conservatism, they are likely to be adaptedtosimilarniches
(Harvey and Pagel 1991). Thus, if one species’ rangemoves
such that it comes into contact with its sister species, then
the environment occupied by the sister species may be
more suitable and more likely to be invaded than sur-
rounding areas containing different environments. Al-
though these scenarios are possible, we have no reason to
believe that they are commonplace or that they undermine
our results.
The simulations presented in this article suggest that
under a wide range of situations, contemporary range in-
formation can be used to infer generalities regarding the
geography of speciation. Moreover, when this method was
applied to birds, an exceptionally mobile group of animals,
we were able to reject the hypothesis that sympatric spe-
ciation is commonplace. Instead, our analyses suggest that
allopatric speciation is the dominant geographic mode of
speciation in birds, with sympatric speciationbeinglimited
to 5% of cases.
Acknowledgments
We thank J. Weir and at least one anonymous reviewer
for many perceptive comments that greatly improved this
article; T. Barraclough for providing valuable discussion
and ideas; S. Meiri, V. Olson, A. Pigot, T. Price, N. Seddon,
and J. Tobias for comments on one or more versions of

Page 11

656 The American Naturalist
this manuscript; M. Burgess, F. Eigenbrod, and N. Pickup
for compilation of the ArcGIS range data; and the Natural
Environment Research Council for funding.
Literature Cited
Allen, J. A. 1907. Mutations and the geographic distributionofnearly
related species in plants and animals. American Naturalist41:653–
655.
Anderson, S., and M. K. Evensen. 1978. Randomness in allopatric
speciation. Systematic Zoology 27:421–430.
Balakrishnan, C. N., and M. D. Sorenson. 2006. Song discrimination
suggests premating isolation among sympatric indigobird species
and host races. Behavioral Ecology 17:473–478.
Barraclough, T. G., and A. P. Vogler. 2000. Detecting the geographical
pattern of speciation from species-level phylogenies. American
Naturalist 155:419–434.
Barraclough, T. G., A. P. Vogler, and P. H. Harvey. 1998. Revealing
the factors that promote speciation. Philosophical Transactions of
the Royal Society B: Biological Sciences 353:241–249.
Berlocher, S. H. 1998. Can sympatric speciation via host or habitat
shift be proven from phylogenetic and biogeographic evidence?
Pages 99–113 in S. H. Berlocher, ed. Endless forms: species and
speciation. Oxford University Press, New York.
Berlocher, S. H., and J. L. Feder. 2002. Sympatric speciation in phy-
tophageous insects: moving beyond controversy? Annual Review
of Entomology 47:773–815.
Brown, J. H., and A. C. Gibson. 1983. Biogeography.Mosby,St.Louis.
Bush, G. L. 1975. Modes of animal speciation. Annual Review of
Ecology and Systematics 6:339–364.
Chesser, R. T., and R. M. Zink. 1994. Modes of speciation in birds:
a test of Lynch’s method. Evolution 48:490–497.
Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer, Sunderland,
MA.
Coyne, J. A., and T. D. Price. 2000. Little evidence for sympatric
speciation in island birds. Evolution 54:2166–2171.
Cracraft, J. 1982.Geographicdifferentiation,cladistics,andvicariance
biogeography: reconstructing the tempo and mode of evolution.
American Zoologist 22:411–424.
Dieckmann, U., and M. Doebeli. 1999. On the origin of species by
sympatric speciation. Nature 400:354–357.
Drummond, A. J., M. Kearse, J. Heled, and R. Moir. 2006. Geneious,
version 1.0. Available at http://www.geneious.com/.
Fitzpatrick, B. M., and M. Turelli. 2006. The geography of mam-
malian speciation:mixedsignalsfromphylogeniesandrangemaps.
Evolution 60:601–615.
Friesen, V. L., and D. J. Anderson. 1997. Phylogeny and evolution
of the Sulidae (Aves: Pelecaniformes): a test of alternative modes
of speciation. Molecular Phylogenetics and Evolution 7:252–260.
Funk, D. J., and K. E. Omland. 2003. Species-level paraphyly and
polyphyly: frequency, causes, and consequences,withinsightsfrom
animal DNA. Annual Review of Ecology, Evolution, and System-
atics 34:397–423.
Futuyma, D. J., and D. J. Mayer. 1980. Non-allopatric speciation in
animals. Systematic Zoology 29:254–271.
Gavrilets, S., and D. Waxman. 2002. Sympatric speciation by sexual
conflict. Proceedings of the National Academy of Sciences of the
USA 99:10533–10538.
Harris, M. P. 1969. The biology of storm petrels in the Galapagos
Islands. Proceedings of the California Academy of Sciences 37:95–
165.
Harvey, P. H., and M. D. Pagel. 1991. The comparative method in
evolutionary biology. Oxford University Press, Oxford.
Irwin, D. E., S. Bensch, J. H. Irwin, and T. D. Price. 2005. Speciation
by distance in a ring species. Science 307:414–416.
Jiggins, C. D. 2005. Sympatric speciation: why the controversy? Cur-
rent Biology 16:R333–R334.
Johnson, N. K., and C. Cicero. 2002. The role of ecologic diversifi-
cation in sibling speciation of Empidonax flycatchers(Tyrannidae):
multigene evidence from mtDNA. Molecular Ecology 11:2065–
2081.
Jordan, D. S. 1905. The origin of species through isolation. Science
22:545–562.
———. 1908. The law of germinate species. American Naturalist 42:
73–80.
Jordan, D. S., and V. L. Kellogg. 1907. Evolution and animal life: an
elementary discussion of facts, processes, laws andtheoriesrelating
to the life and evolution of animals. Appleton, London.
Kumar, S., K. Tamura, and M.Nei. 2004.MEGA3:integratedsoftware
for molecular evolutionary genetics analysis and sequence align-
ment. Briefings in Bioinformatics 5:150–163.
Lessa, E. P., J. A. Cook, and J. L. Patton. 2003. Genetic footprints of
demographic expansion in North America, but not Amazonia,
during the Late Quaternary. Proceedings of the National Academy
of Sciences of the USA 100:10331–10334.
Losos, J. B., and R. E. Glor. 2003. Phylogenetic comparativemethods
and the geography of speciation. Trends in Ecology & Evolution
18:220–227.
Lynch, J. D. 1989. The gauge of speciation: on the frequencies and
modes of speciation. Pages 527–553 in J. A. Endler, ed. Speciation
and its consequences. Sinauer, Sunderland, MA.
MacArthur, R. H. 1957. On the relative abundance of bird species.
Proceedings of the National Academy of Sciences of the USA 43:
293–295.
Mayr, E. 1942. Systematics and the origin of species. Columbia Uni-
versity Press, New York.
———. 1963. Animal species and evolution. Harvard University
Press, Cambridge, MA.
McCarthy, E. M. 2006. Handbook of avian hybrids of the world.
Oxford University Press, New York.
Monteiro, L. R., and R. W. Furness. 1998. Speciation through tem-
poral segregation of Madeiran storm petrel (Oceanodroma castro)
populations in the Azores? Philosophical Transactions of theRoyal
Society B: Biological Sciences 353:945–953.
Newton, I. 2003. Speciation and biogeography of birds. Academic
Press, London.
Orme, C. D. L., M. Burgess, F. Eigenbrod, N. Pickup, R. G. Davies,
V. Olson, A. J. Webster, et al. 2005. Global hotspots of species
richness are not congruent with endemism or threat. Nature 436:
1016–1019.
Orme, C. D. L., R. G. Davies, V. A. Olson, G. H. Thomas, T.-S. Ding,
P. C. Rasmussen, R. S. Ridgely, et al. 2006. Global patterns of
geographic range size in birds. PLoS Biology 4, no. 7, doi:10.1371/
journal.pbio.0040208.
Perret, M., A. Chautems, R. Spichiger, T. G. Barraclough, and V.
Savolainen. 2007. The geographical pattern of speciation andfloral
diversification in the Neotropics: the tribe Sinningieae (Gesneri-
aceae) as a case study. Evolution 61:1641–1660.

Page 12

Sympatric Speciation in Birds Is Rare 657
Pianka, E. R. 1966. Latitudinal gradients in species diversity: a review
of concepts. American Naturalist 100:33–46.
Pielou, C. 1991. After the ice age: the return of life to glaciated North
America. University of Chicago Press, Chicago.
Price, T. D. 2008. Speciation in birds. Roberts, Greenwood Village,
CO.
Rand, A. L. 1948. Glaciation, an isolating factor in speciation. Evo-
lution 2:314–321.
Ribera, I., T. G. Barraclough, and A. P. Vogler. 2001. The effect of
habitat type on speciation rates and range movements in aquatic
beetles: inferences from species-level phylogenies. MolecularEcol-
ogy 10:721–735.
Savolainen, V., M.-C. Anstett, C. Lexer, I. Hutton, J. J. Clarkson, M.
V. Norup, M. P. Powell, D. Springate, N. Salamin, and W. J. Baker.
2006. Sympatric speciation in palms on an oceanic island. Nature
441:210–213.
Seddon, N., and J. A. Tobias. 2007. Song divergence at the edge of
Amazonia: an empirical test of the peripatric speciation model.
Biological Journal of the Linnean Society 90:173–188.
Sibley, C. G., and B. L. Monroe. 1990. Distribution and taxonomy
of birds of the world. Yale University Press, New Haven, CT.
———. 1993. Distribution and taxonomy of birds of the world:
supplement. Yale University Press, New Haven, CT.
Smith, A. L., and V. L. Friesen. 2007. Differentiation of sympatric
populations of the band-rumped storm-petrel in the Galapagos
Islands: an examination of genetics, morphology, and vocaliza-
tions. Molecular Ecology 16:1593–1603.
Sorenson, M. D., K. M. Sefc, and R. B. Payne. 2003. Speciation by
host switch in brood parasitic indigobirds. Nature 424:928–931.
Stephens, P. R., and J. J. Wiens. 2003. Explaining species richness
from continents to communities: the time-for-speciation effect in
emydid turtles. American Naturalist 161:112–127.
Thompson, J. D., D. G. Higgins, and T. J. Gibson. 1994. CLUSTAL
W: improving the sensitivity of progressive multiple sequence
alignment through sequence weighting, position-specific gap pen-
alties and weight matrix choice. Nucleic Acids Research 22:4673–
4680.
Wagner, M. 1873. The Darwinian theory and the law of themigration
of organisms. Stanford, London.
Waldron, A. 2007. Null models of geographic range size evolution
reaffirm its heritability. American Naturalist 170:221–231.
Walsh, B. D. 1864. On phytophagic varieties and phytophagicspecies.
Proceedings of the Entomological Society of Philadelphia 3:403–
430.
Associate Editor: Stephen B. Heard
Editor: Michael C. Whitlock