2070
q 2001 The Society for the Study of Evolution. All rights reserved.
Evolution, 55(10), 2001, pp. 2070?2087
EVOLUTIONARY IMPLICATIONS OF DIVERGENT CLINES IN AN AVIAN
(MANACUS: AVES) HYBRID ZONE
ROBB T. BRUMFIELD,1,2,3 ROBERT W. JERNIGAN,4,5 DAVID B. MCDONALD,6,7 AND MICHAEL J. BRAUN1,2,8
1Laboratory of Molecular Systematics, National Museum of Natural History, Smithsonian Institution MRC 534,
Washington, D.C. 20560
2Department of Biology, University of Maryland, College Park, Maryland 20742
4Department of Mathematics and Statistics, American University, Washington, D.C. 20016
5E-mail: jernigan@american.edu
6Department of Zoology and Physiology, University of Wyoming, Laramie, Wyoming 82071
7E-mail: dbmcd@uwyo.edu
8E-mail:braun@lms.si.edu
Abstract. A previous study of the hybrid zone in western Panama between white-collared (Manacus candei) and
golden-collared manakins (M. vitellinus) documented the unidirectional introgression of vitellinus male secondary
sexual traits across the zone. Here, we examine the hybrid zone in greater genetic and morphological detail. Statistical
comparisons of clines are performed using maximum-likelihood and nonparametric bootstrap methods. Our results
demonstrate that an array of six molecular and two morphometric markers agree in cline position and width. Clines
for male collar and belly color are similar in width to the first eight clines, but are shifted in position by at least five
cline widths. The result is that birds in intervening populations are genetically and morphometrically very like parental
candei, but males have the plumage color of parental vitellinus. Neither neutral diffusion nor nonlinearity of color
scales appear to be viable explanations for the large cline shifts. Genetic dominance of vitellinus plumage traits is
another potential explanation that will require breeding experiments to test. Sexual selection remains a plausible
explanation for the observed introgression of vitellinus color traits in these highly dimorphic, polygynous, lek-mating
birds. Two other clines, including a nondiagnostic isozyme locus, are similar in position to the main cluster of clines,
but are broader in width. Thus, introgression at some loci is greater than that detected with diagnostic markers.
Assuming that narrow clines are maintained by selection, variation in cline width indicates that selection is not uniform
throughout the genome and that diagnostic markers are under more intense selective pressure. The traditional focus
on diagnostic markers in studies of hybrid zones may therefore lead to underestimates of average introgression. This
effect may be more pronounced in organisms with low levels of genetic divergence between hybridizing taxa.
Key words. Cline, hybrid zone, introgression, Manacus, manakin, sexual selection, speciation.
Received October 20, 2000. Accepted June 18, 2001.
Within a reproductively isolated group of organisms, the
majority of new alleles upon which natural selection can act
arise from mutation or recombination. Hybridization affords
another source of new alleles, one that is available only to
organisms with incomplete reproductive isolation from other
divergent taxa (Anderson 1949; Stebbins 1959; Arnold 1997).
Divergent eubacteria that have transferred plasmids harboring
antibiotic-resistance genes are examples of the important role
that hybridization can play in evolution (Spratt et al. 1989),
as are studies documenting the increased fitness of intro-
gressed populations that have successfully exploited new
habitats (Lewontin and Birch 1966) or habitats unsuitable to
either parental form (Anderson 1949; Stebbins 1959; Grant
1981; Arnold 1992).
Anderson and Hubricht (1938) coined the term ??intro-
gressive hybridization?? to describe the genetic intercon-
nectedness of hybridizing populations and proposed that
hybrids provide a means for advantageous genes to move
between differentiated taxa. Anderson (1949) emphasized
that the evolutionary significance of introgression may be
concealed by difficulty in detecting it. Without a detailed
understanding of the genetic architecture of the hybridizing
forms, introgressing genes that have little or no effect on
3 Present address: Department of Zoology, Box 351800, Univer-
sity of Washington, Seattle, Washington 98195; E-mail: brumfld@
u.washington.edu.
morphology will be difficult to detect (Barton 2001). In
extreme cases, hybridizing taxa whose morphological dif-
ferences are maintained by one or a few genes (e.g., but-
terflies, Sheppard et al. 1985) could be exchanging genes
freely across the rest of their genomes, yet maintaining sharp
boundaries between morphotypes (e.g., oaks, Dumolin-
Lape?gue et al. 1999).
Hybrid zones, where genetically divergent forms meet and
hybridize (Harrison 1990), are natural arenas in which to
examine the evolutionary potential of introgression (Riese-
berg et al. 1999). Cline theory provides a conceptual frame-
work to understand both the maintenance of hybrid zones
and the relevant evolutionary parameters involved in the
movement of traits across them (Haldane 1948; Slatkin 1973;
May et al. 1975; Endler 1977; Barton 1979; Szymura and
Barton 1986). Most hybrid zones are thought to be maintained
as tension zones by a balance between the dispersal of par-
entals into the zone and selection (Haldane 1948; Barton
2001). Both endogenous (e.g., intrinsic) and exogenous (e.g.,
environmental) selection can result in barriers to the move-
ment of genes across a hybrid zone (Hewitt 1988). Endog-
enous selection may be due to the breakdown of coadapted
gene complexes (Grant and Grant 1994), whereas exogenous
selection may result from a wide variety of environmental
gradients. To cross such a barrier, an allele must recombine
into a favorable, or at least viable, genetic background (An-
derson 1949; Stebbins 1959; Barton 1979). If selection
2071DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
FIG. 1. Map of western Bocas del Toro, Panama and extreme eastern Costa Rica indicating populations 2?12 sampled across the candei-
vitellinus hybrid zone. Detailed locality information is in the Appendix. Population 1 lies off this map, 138.25 km northwest of population
2. The northeastern extent of the mountains represents 500-m elevation contour.
against hybrids is strong or if genes that contribute to reduced
hybrid fitness are distributed throughout the genome (Kim
and Rieseberg 1999), such gene combinations may be rare
(Key 1968) or nonexistent for some loci. In addition, favor-
able combinations would initially arise at low frequencies
and could be lost by drift (Pia?lek and Barton 1997). However,
once a positively selected trait does cross a hybrid zone and
begins to increase in frequency, it should sweep rapidly
through the new population, leaving no easily detectable sig-
nature (Hewitt 1988). Thus, it is likely to be observed only
rarely in nature (Anderson 1949; Barton 2001).
An interesting case of apparent introgression was docu-
mented by Parsons et al. (1993) in their study of a hybrid
zone between the closely related white-collared (Manacus
candei) and golden-collared manakins (M. vitellinus). These
sexually dichromatic, lek-mating birds are divergent in a
number of morphological characters, but their most striking
differences lie in the color of the definitive male collar plum-
age (golden yellow in vitellinus; white in candei), and in the
color of their belly plumage (drab green in vitellinus; lemon
yellow in candei). Parsons et al. (1993) found that clines in
several male plumage traits were shifted approximately 40
km northwest of the clines for three molecular genetic mark-
ers and two morphometric markers. They interpreted the
shifted clines as evidence that the plumage traits of vitellinus
were introgressing unidirectionally across the hybrid zone.
The facts that the introgressing male plumage characters are
secondary sexual traits and that mating success is highly
skewed in these lek-mating birds (Lill 1974) led Parsons et
al. (1993) to suggest that introgression might be due to pos-
itive sexual selection for vitellinus plumage traits. Other than
the shifted plumage clines, the hybrid zone uniting candei
and vitellinus appeared typical of most described hybrid zones
(Moore 1977; Barton and Hewitt 1985, 1989), with numerous
character transition clines of the same approximate width and
position.
Butlin and Neems (1994) suggested that the Manacus
plumage clines might not be shifted in position at all, but
might simply be much wider than clines for molecular and
morphometric characters. They also suggested that if any of
the clines were indeed shifted, the shift might be due to
genetic dominance of vitellinus traits or to the nonlinear color
index used, with the underlying allele frequency clines ac-
tually coincident in position with the other clines. Butlin and
Neems preferred an interpretation of the data wherein male
plumage is irrelevant to mating success and the plumage
clines have become broader (or perhaps shifted slightly) due
to neutral diffusion. Parsons et al. (1994) rebutted these ar-
guments, while acknowledging that further evidence was
highly desirable.
To address these issues, we reexamined the genetic and
morphological structure of the hybrid zone in greater detail.
2072 ROBB T. BRUMFIELD ET AL.
TABLE 1. Summary of morphological measurements that differ significantly across the hybrid zone transect. Mean, standard deviation (SD),
and number of individuals measured (N) are presented for each population. Population 1 is the reference sample for candei and population 12
is the reference sample for vitellinus. See Figure 1 and Appendix for detailed locality information.
Population
Distance
(km)
Mass (g)
Mean SD N
Wing length (mm)
Mean SD N
Culmen length (mm)
Mean SD N
Bill depth (mm)
Mean SD N
1
2
3
4
5
0
138.25
151.75
159.5
182.25
18.2
19.4
18.7
19.8
19.6
0.88
0.42
1.18
0.78
1.21
21
5
11
21
18
53.3
52.6
52.6
52.9
53.6
1.04
0.79
0.52
0.90
1.00
22
4
9
9
15
9.6
9.5
9.3
9.6
9.9
0.49
0.45
0.31
0.31
0.35
20
5
10
15
17
3.6
3.4
3.4
3.4
3.6
0.22
0.13
0.25
0.17
0.17
22
5
9
15
12
6
7
8
9
188.25
198.5
201.25
210.0
20.0
20.5
19.7
18.9
1.20
0.75
1.14
0.74
12
5
14
13
53.5
53.1
52.6
52.2
0.38
0.71
0.77
0.86
5
11
20
19
9.5
9.7
9.6
10.0
0.60
0.18
0.53
0.49
11
5
14
9
3.4
3.2
3.4
3.2
0.28
0.19
0.17
0.13
11
4
15
9
10
11
12
230.75
319.5
569.5
20.0
19.4
17.7
0.76
0.66
1.15
16
4
3
51.1
51.2
51.6
0.63
0.80
0.81
10
5
21
10.0
10.2
10.1
0.64
0.35
0.46
7
2
20
3.3
3.4
4.0
0.21
0.35
0.20
7
2
19
Here, we present analyses based on an expanded data set of
seven molecular markers and 12 morphological characters.
We sampled new populations and increased the sample size
of others to measure the position and widths of the character
clines more accurately, and we refined the measurement of
color traits using reflectance spectrophotometry. We sur-
veyed wild populations for the homozygous recessive white-
collared birds expected under a hypothesis of genetic dom-
inance of vitellinus alleles. As a statistical framework for
comparing the structure of individual character clines, we
employ the maximum-likelihood methods of Barton and
Baird (1999) to estimate confidence intervals for the molec-
ular cline parameters and use the nonparametric bootstrap to
compare molecular and morphological clines.
MATERIALS AND METHODS
Population Samples
From 1989 to 1994, we collected 173 specimens along a
transect (Fig. 1, populations 2?11) through a narrow coastal
strip of humid, lowland, second-growth forest and abandoned
banana and cacao plantations in western Bocas del Toro,
Panama. Because Parsons et al. (1993) focused their sampling
on the area of plumage transition at the Changuinola River,
they lacked samples from the actual hybrid zone center. In
1994, we collected individuals from populations 8 and 9 (Fig.
1) to sample intensively in the center of the steep genetic
and morphometric character transitions identified by Parsons
et al. (1993). Thus, populations 10?12 of this paper corre-
spond to populations 8?10 of Parsons et al. (1993). Speci-
mens were deposited at the U.S. National Museum of Natural
History (USNM) and the Universidad de Panama. Skeletal
muscle, liver, and heart tissue samples were stored in liquid
nitrogen.
To serve as reference parental samples for the genetic
markers, we collected 20 individuals of vitellinus approxi-
mately 250 km east of the hybrid zone (population 12), and
borrowed from the Louisiana State University Museum of
Natural Science (LSUMNS) frozen tissues from six individ-
uals of candei collected 140 km west of the hybrid zone
(population 1). We used 16 blood samples collected from
birds that were banded and released to augment the popu-
lation 2 sample, which was composed of only three individ-
uals in Parsons et al. (1993). Thus, for the genetic markers,
we assayed a total of 215 specimens, consisting of defini-
tively plumaged males ($2 years old), first-year males, and
females.
For the morphological markers, we measured a total of 122
definitively plumaged male specimens from the transect (Fig.
1, populations 2?11). The parental sample of candei was
composed of 22 specimens collected in Costa Rica (19 spec-
imens collected between the years 1893 and 1928 and three
specimens collected in 1990). A parental sample of vitellinus
was composed of 22 specimens collected in the Canal Zone
between the years 1911 and 1994. The overall sample size
for the morphological characters (N 5 166) was less than for
the genetic characters and more variable among populations,
primarily because some specimens (e.g., first-year males, fe-
males, skeletons) could not be assayed for the morphological
characters. Catalog numbers for all specimens appear in the
Appendix.
Morphometric Characters
The mass of each specimen was measured in the field prior
to specimen preparation, or, for historical specimens, was
noted from the museum tag. On definitively plumaged males,
R. T. Brumfield measured seven morphometric characters
with dial calipers (to the nearest 0.1 mm): wing length (chord
of unflattened wing from bend of wing to longest primary);
culmen length (exposed culmen); nare length (from anterior
edge of nare to tip of bill); bill depth (at its base); bill width
(at its base); tail length (measured from point of insertion of
central rectrices to tip of longest rectrix); and tarsus length
(from the joint of tarsometatarsus and tibiotarsus to the lateral
edge of last undivided scute). Two plumage characters were
measured with a ruler (to the nearest 0.5 mm): epaulet width
(from bend of folded wing to posterior edge of white or
yellow epaulet feathers) and beard length (measured from
insertion to tip of longest feather). Throughout the paper, we
use the terms ??morphometric?? characters to refer to the fore-
going 10 characters and ??plumage color?? characters in ref-
erence to male collar color and belly color measurements
2073DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
TABLE 1. Extended.
Bill width (mm)
Mean SD N
Tail length (mm)
Mean SD N
Tarsus length (mm)
Mean SD N
Epaulet width (mm)
Mean SD N
6.0
5.7
6.0
5.7
5.4
5.3
0.36
0.05
0.25
0.27
0.28
0.25
21
5
10
17
18
11
35.0
35.6
35.1
35.2
35.9
35.0
0.67
0.54
0.78
0.54
0.99
0.51
22
5
17
13
15
5
20.2
20.2
19.8
20.1
20.9
20.5
0.63
0.43
0.60
0.43
0.40
0.89
19
5
10
19
20
11
11.4
11.25
11.0
10.25
8.9
8.2
1.66
0.96
1.77
2.73
1.33
0.57
22
5
17
13
15
5
5.2
5.7
5.6
5.7
5.8
6.2
0.12
0.21
0.29
0.26
0.12
0.29
5
15
9
9
3
20
34.9
34.6
31.3
29.7
29.7
29.5
0.60
1.15
1.9
0.70
0.63
0.92
11
20
21
10
5
22
20.9
20.3
20.4
20.8
21.0
20.9
0.51
0.51
0.26
0.74
0.76
0.52
5
14
7
9
3
21
8.2
8.75
6.0
5.15
4.9
5.0
1.44
0.99
1.49
0.88
0.74
0.91
11
20
21
10
5
22
(described below). ??Morphological?? data refers to all of the
above characters.
We performed statistical and multivariate analyses of the
morphological data (Table 1) using the JMP software package
(SAS Institute 2000). Parental reference samples had signif-
icantly different means for all of the morphometric characters
except nare length (Brumfield 1999). Variation in some of
the morphometric characters, such as mass, wing length, tail
length, and tarsus length, likely reflect correlated differences
in overall size. Variation in these characters was summarized
by means of a principal components analysis (PCA) of the
correlation matrix. It seems reasonable to assume that vari-
ation in plumage characters, such as beard length, epaulet
width, collar color, and belly color, do not reflect variation
in overall size, and so these characters were treated inde-
pendently in all subsequent analyses. Specimens missing data
for more than two characters (N 5 24) were excluded from
the PCA. We estimated missing values for the remaining
specimens using multiple regression.
The first principal components axis (PC1) explained 37%
of the variation, with roughly equivalent loadings (0.39?0.50)
among mass, wing length, tail length, and tarsus length. Load-
ings on PC2, which explained 24% of the variation, were
highest (0.35?0.65) among the three bill characters (culmen
length, bill width, bill depth). Bill morphology, described by
the second component, did not exhibit obvious clinal vari-
ation, and was not analyzed further. Variation at PC3 was
also not clinal. We repeated the PCA excluding the bill char-
acters, reducing the number of excluded individuals from 24
to 19 (excluded individuals listed in the Appendix). PC1
exhibited strong clinal variation and explained 59% of the
variation, with similar loadings among all four characters
(mass, wing length, tarsus length, and tail length). Variation
at PC2 and PC3 was not clinal. Loadings on PC2, which
explained 20% of the variation, were similar in magnitude
among the four characters.
Plumage Color Measurements
We measured male collar and belly color using an Ultra-
scan (Hunter Labs, Inc., Reston, VA) integrating sphere re-
flectance spectrophotometer. Percentage reflectance at each
wavelength was measured relative to a pure white reflecting
surface (100% reflectance). The average of three independent
measurements of the reflectance spectrum was taken for each
specimen. We repositioned the specimen at the sample ap-
erture before each replicate measurement. Collar and belly
color measurements were taken near the center of the throat
and belly, respectively, of each specimen. Thus, these rep-
resent more rigorous quantification of the semi-quantitative
variables ??throat color?? and ??underparts color?? of Parsons
et al. (1993).
Several methods of summarizing color variation across the
hybrid zone were considered (Endler 1990). A PCA of the
entire reflectance spectrum was not adequate, because the
most heavily weighted wavelengths were those with the high-
est variance instead of those that were most divergent be-
tween the parental forms. There are a number of color scales
that summarize reflectance spectra in relation to human per-
ception (Hunter 1975). Although they describe the color var-
iation quantitatively, these scales do not have a biologically
interpretable unit of measurement. For the full analysis, we
opted to use the raw percent reflectance at 490 nm and 665
nm to represent collar color and belly color, respectively
(Table 1). These wavelengths had the greatest differences in
means between the two parental samples in otherwise rather
simple spectra (Fig. 2). Other divergent wavelengths were
examined, and all gave similar results. We also performed a
discriminant function analysis, and this too gave similar re-
sults.
Molecular Genetic Data
Protein electrophoresis was performed on Titan III cellu-
lose acetate plates (Helena Laboratories, Inc., Beaumont,
TX). We identified the isozyme loci Ada (E.C. 3.1.3.2), Ak-
2 (E.C. 2.7.4.3), Gsr (E.C. 1.6.4.2), and Pgm-2 (E.C. 2.7.5.1)
as having significant genic differentiation between vitellinus
and candei (Brumfield and Braun 2001). The presence of
female heterozygotes indicates that all these loci are auto-
somal. Ak-2 gels were run for 3 h at 125 V in a pH 7.5 Tris-
citrate buffer (100 mM Tris, 36 mM citric acid). To discrim-
inate Ak-2, the comigrating Ak-1 locus was inactivated by
incubating homogenates on ice with 1?2 mM AgNO3 for 10
min before electrophoresis (Harris and Hopkinson 1976).
Running conditions for other loci and histochemical staining
methods are presented in Brumfield (1999). Electromorphs
were coded alphabetically in order of relative mobility from
2074 ROBB T. BRUMFIELD ET AL.
TABLE 1. Extended.
Population
Distance
(km)
Beard length (mm)
Mean SD N
Collar color
(% reflectance at 490 nm)
Mean SD N
Belly color
(% reflectance at 665 nm)
Mean SD N
PC1
Mean SD N
1
2
3
4
5
6
0
138.25
151.75
159.5
182.25
188.25
12.4
11.75
11.9
12.2
13.2
12.5
0.90
0.96
0.81
0.83
0.98
0.5
22
5
17
13
15
5
57.2
67.2
64.3
11.4
9.8
6.3
5.14
7.59
6.25
3.59
3.93
1.39
22
4
17
13
15
5
67.7
81.5
78.4
44.7
36.2
29.2
5.79
1.90
4.63
11.76
9.45
7.03
22
4
17
13
15
5
0.75
0.77
0.71
0.62
0.73
0.81
0.16
0.10
0.16
0.12
0.15
0.06
20
3
9
9
15
5
7
8
9
10
11
12
198.5
201.25
210.0
230.75
319.5
569.5
12.7
12.2
16.3
17.4
18.8
18.7
1.80
1.22
2.30
1.26
1.92
1.61
11
20
21
10
5
22
5.7
6.3
7.81
3.9
3.4
3.4
1.55
2.16
12.771
1.60
1.23
1.09
11
20
211
10
5
22
25.5
26.9
25.8
23.4
23.2
25.2
3.35
4.27
5.04
3.27
3.49
2.70
9
20
21
10
5
22
0.71
0.69
0.47
0.23
0.34
0.31
0.19
0.11
0.15
0.14
0.12
0.15
11
20
19
10
5
21
1 Measurements for collar color in population 9 excluding the aberrant white specimen (USNM 608981) are: mean5 5.0, SD 5 2.54, N 5 20.
the origin with the most anodally migrating allele as ??a.??
Among-gel comparisons of all alleles were made by rerun-
ning selected individuals on cross-correlation gels. All four
samples from population 6 and one sample (tissue catalog
number B487) from population 7 were preserved in ethanol
and could not be scored for the isozyme loci. In addition,
the 16 blood samples used to augment the sample size of
population 2 were not scored for the isozyme loci. Frozen
specimens that did not include heart and liver tissues were
not scored for Ak-2 and Pgm-2, respectively. Sample sizes
and allele frequencies are given in Table 2.
DNA was isolated by proteinase K digestion and phenol/
chloroform extraction followed by ethanol precipitation. Af-
ter digestion with restriction enzymes, DNA was transferred
to nylon membranes (Southern 1975). Random priming
(Feinberg and Vogelstein 1993) was used to label probes to
high specific activity (108?109 dpm/mg) with a-32P dATP.
After probing and washing, wet membranes were wrapped
in cellophane and exposed to Kodak (Rochester, NY) XRP
film for 4?96 h at 2808C.
Parsons et al. (1993) identified three probes that detect
diagnostic restriction fragment length variants in candei and
vitellinus. l5 is a randomly cloned fragment of manakin DNA
used to probe Bgl II digests. A 5.6-kb band in vitellinus is
cleaved into 3.7-kb and 1.9-kb bands in candei. The second
probe, pSCN3, is a randomly cloned fragment of manakin
DNA used to probe Ava II digests. A 5.5-kb band in candei
is cleaved into 4.8-kb and 0.7-kb bands in vitellinus. The
presence of female heterozygotes indicates that both l5 and
pSCN3 are nuclear, autosomal markers. A third probe is mi-
tochondrial DNA (mtDNA) purified by cesium chloride gra-
dient centrifugation (Dowling et al. 1996). MtDNA from vi-
tellinus, an introgressed vitellinus-candei individual, and Car-
podacus mexicanus (house finch) gave equivalent results
when used as a probe of Bgl II (;28 kb, presumably uncut
band in vitellinus with ;16-kb band in candei), Ava I (;28-
kb band in vitellinus with ;23-kb, ;3-kb, and ;2-kb bands
in candei), and EcoR V (;9.4-kb, ;5-kb, and ;4-kb bands
in candei with ;9.4-kb, ;5-kb, ;2.8-kb, and 1.1-kb in vi-
tellinus) digests. All birds had either the vitellinus or candei
patterns for all three mtDNA markers. A few individuals
showed minor length variations in a single ;5-kb EcoR V
band, but haplotypes of all birds were clearly related to one
or the other parental type (Parsons et al. 1993). Sample sizes
and allele frequencies are given in Table 2.
To construct a summary measure of the degree of multilo-
cus genetic admixture across the hybrid zone, a maximum-
likelihood genetic hybrid index was estimated from the seven
molecular marker loci using a Mathematica (Wolfram 1996)
program written by S. J. E. Baird (Rieseberg et al. 1998).
The index is calculated as the sum, over each individual?s
alleles, of the probability that the allele was derived from a
vitellinus parental population instead of from a candei pa-
rental population (Rieseberg et al. 1998), scaled by the num-
ber of loci.
Cline-Fitting Analysis
The software package GENEPOP (Raymond and Rousset
1995) was used to perform probability tests for deviations
from Hardy-Weinberg equilibrium (HWE) at each locus (us-
ing complete enumeration). For loci having more than four
alleles, an unbiased approximation of the probability test for
HWE was performed using a Markov-chain procedure. The
computer program Analyse (Barton and Baird 1999) was used
for analyses of cline structure. The shape of a cline can be
modeled by combining three equations (Szymura and Barton
1986, 1991):
X(decay )ce intercept , (1a)c
1
, and (1b)
24X1 1 e
2X(decay )v1 2 e intercept , (1c)v
where decayc 5 2 , decayv 5 2 , interceptc 5 tv/(tc 1?u ?uc v
tv 1 tctv), interceptv 5 tc/(tc 1 tv 1 tctv), tc 5 Bcdecayc, tv 5
Bvdecayv, and X 5 (x 2 c)/w is the distance X along the cline
scaled relative to the center and width. Equation (1b) is a
sigmoid in the center, and equations (1a) and (1c) are ex-
ponential decay curves leading away from either side of the
center. The change from the central curve (1b) to the ex-
ponential tails occurs at the point where the decay curves (1a
and 1c) cross the sigmoidal curve (1b). This crossing point
changes depending on the parameters. uc and uv are param-
2075DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
FIG. 2. Reflectance spectra for (A) collar color and (B) belly color in the reference samples of candei (population 1) and vitellinus
(population 12). Mean measurements are presented along with standard deviations (error bars). We used the most divergent peaks, at
percent reflectance of 490 nm and 665 nm, to analyze collar- and belly-color variation across the hybrid zone. The dashed lines represent
the spectra for an aberrant white-collared male (USNM 608981) collected in population 9, showing that it differs sharply from the typical
candei spectra.
eters that describe the rate of exponential decay, and Bc and
Bv describe the size of the central barrier to gene flow (Szy-
mura and Barton 1986) in candei and vitellinus, respectively.
The width of the cline (w) is defined as 1/maximum slope,
c is the central position of the cline where pest 5 0.5, and x
is the distance (km) from population 1 (Table 1). Two ad-
ditional estimated parameters, pmin and pmax, represent the
frequencies at the two ends of the cline, and are used by
Analyse to scale pest:
p 5 p 1 p (p 2 p ).est9 min est max min (2)
Calculation of the likelihood of a fitted cline assumes a
binomial sampling distribution for the molecular genetic
markers. For each population, the support or goodness of fit
for a fitted cline is expressed as a log-likelihood and is cal-
culated as:
p 1 2 pobs obsln L 5 n p 1 (1 2 p ) ln , (3)obs obs
1 2 1 2[ ]p 1 2 pest est
where n is the number of alleles sampled in the population,
pobs is the observed frequency of the modeled allele and pest
is the estimated frequency of the modeled allele based on
Equation (1). The support for the entire fitted cline is the
sum of the population log-likelihoods.
Goodness-of-fit tests were performed to identify, for each
locus, a cline model that best describes the data with the
fewest number of estimated parameters. In model I, only c
(center of cline) and w (width) were estimated from the data.
uc/v and Bc/v were assumed to be one and zero, respectively,
on both sides of the cline, and the ends of the cline were
fixed to the maximum (pmax) and minimum (pmin) observed
allele frequencies. Model II included the c and w parameters
plus estimates of the allele frequencies (pmin and pmax) at the
two ends of the cline. In Model III, uc/v and Bc/v were also
allowed to vary. Model parameters were estimated by a
bounded, directional search of the parameter space using the
Metropolis-Hastings algorithm (Metropolis et al. 1953; Has-
tings 1970) implemented in Analyse. Three thousand itera-
tions were run from 20 different starting points. The test
statistic was calculated as two times the absolute difference
in log-likelihood between the models under comparison. Sig-
nificance (a 5 0.05) was determined by comparison to the
x
2
-distribution, with the degrees of freedom equivalent to the
difference in the number of estimated parameters between
the two models (Table 3).
Model II provided the best fit for all loci except Adaa and
l5b. The best fit for Adaa was provided by model I, for which
only center (c) and cline width (w) are estimated. Model III
provided the best fit for l5b, primarily because the cline
decays exponentially in frequency on the eastern side of the
hybrid zone. Using the best-fit model for each locus (Table
3), parameters were estimated by running 30,000 iterations
of the Metropolis-Hastings algorithm with 20 different start-
ing points. Except for Ada, all 20 searches converged on the
2076 ROBB T. BRUMFIELD ET AL.
TABLE 2. Allele frequencies and FIS estimates for each population at each locus. Sample sizes are presented for diploid (2n) and haploid (n)
loci.
Population
Distance
(km)
Ada
a b 2n FIS
Ak-2
a b 2n FIS
Gsr
a b c d e 2n FIS
1
2
3
4
5
6
0
138.25
151.75
159.5
182.25
188.25
0.100
0.357
0.200
0.450
0.310
?
0.900
0.643
0.800
0.550
0.690
?
10
14
40
40
42
0
?
10.143
10.397
10.216
10.243
?
0.000
0.143
0.025
0.025
0.024
?
1.000
0.857
0.975
0.975
0.976
?
10
14
40
40
42
0
?
20.091
?
?
?
?
0.000
0.000
0.025
0.075
0.048
?
0.100
0.000
0.025
0.075
0.024
?
0.900
0.929
0.950
0.850
0.905
?
0.000
0.000
0.000
0.000
0.024
?
0.000
0.071
0.000
0.000
0.000
?
10
14
40
40
42
0
?
?
20.013
10.087
20.046
?
7
8
9
10
11
12
198.5
201.25
210.0
230.75
319.5
569.5
0.545
0.568
0.442
0.625
0.425
0.675
0.455
0.432
0.558
0.375
0.575
0.325
22
44
52
40
40
40
20.429
10.189
20.150
20.365
10.105
10.227
0.000
0.091
0.615
0.975
0.895
1.000
1.000
0.909
0.385
0.025
0.105
0.000
20
44
52
40
38
40
?
10.468
10.206
?
20.091
?
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.023
0.173
0.225
0.450
0.225
1.000
0.955
0.731
0.425
0.375
0.725
0.000
0.023
0.096
0.350
0.175
0.050
0.000
0.000
0.000
0.000
0.000
0.000
22
44
52
40
40
40
?
20.012
10.028
10.250
20.173
10.194
same parameters, consistent with a single likelihood peak.
Two-unit log-likelihood support values (ln Lmax 2 2) were
obtained for each fitted cline by an additional 2000 iterations
of the Metropolis-Hastings search.
Equation (1) was used to fit clines for the morphological
characters (Butlin et al. 1991). Log-likelihood support values
(ln L) were not calculated, because the sampling distribution
of the morphological characters could not be reliably assumed
normal, as required by Analyse for treatment of quantitative
traits. Because goodness-of-fit model tests were not possible,
the most general model (model III) was used for all mor-
phological cline fits. Before clines were fit to the morpho-
logical character population means, individual measurements
were scaled to values between zero and one. All fitted clines
are illustrated so that vitellinus has the largest character value.
For example, because the percent reflectance at 490 nm is
greater in white-collared candei, a cline was fit to 100 2 %
reflectance so that vitellinus had a larger value on the eastern
side of the cline. This provides a cosmetic improvement for
qualitatively comparing cline shapes.
RESULTS
Molecular Variation and Cline Shape
Parental populations showed strong allele frequency dif-
ferences at all molecular marker loci (Table 2), but only
mtDNA and Ak-2 reached fixation for alternative alleles in
both parental populations. l5 and pSCN3 were fixed in pa-
rental vitellinus (population 12) and nearly fixed in parental
candei (population 1), so we treat them tentatively as diag-
nostic herein. FIS-values were not significantly different from
zero in any population, consistent with HWE. Individual ge-
notypic data indicate that the majority of individuals in pop-
ulation 9 were of hybrid origin (Brumfield 1999).
All molecular loci except Ada have steep clines of the same
approximate width (w 5 2.9?11.1 km) that are centered in
the same general position (c 5 206.5?209.9), a few kilo-
meters west of population 9 (Fig. 3, Table 4). Patterns of
variation were most straightforward at Ak-2, l5, pSCN3, and
mtDNA, the four loci having fixed or nearly fixed alternate
alleles in candei and vitellinus. Each of these loci was char-
acterized by steep transitions near population 9, with more
or less monotonic change across the hybrid zone (Fig. 3A).
Low frequencies of the ??wrong?? alleles in populations 1?6
and 10?11 probably represent shallow tails of introgression
at these loci, because the frequencies decline to near zero in
populations 1 and 12. Gsr and Pgm-2 also underwent steep
transitions near population 9 (Fig. 3B). Relatively high fre-
quencies of several alleles at these loci in populations 10?
12 may be due to polymorphism in parental vitellinus pop-
ulations rather than introgression. In contrast, Ada exhibited
a broad cline across the entire transect (Fig. 3B).
The transition in genetic hybrid index scores of individuals
(Fig. 4A) reflects a steep gradient in allele frequencies near
population 9 (Table 2). Population 10, situated 21 km south-
east of population 9, is entirely composed of individuals
whose genetic index scores are similar to those found in
parental vitellinus (population 12) more than 350 km farther
east. Population 8, located 9 km to the north of population
9, is composed of candei-like individuals, with the exception
of two individuals having intermediate scores. As expected,
there were more vitellinus-like than candei-like individuals
in population 9, because the sample is situated 0.5?3.5 km
southeast of the estimated hybrid zone center.
Statistical Comparisons of Molecular Clines
Six molecular clines agreed remarkably well in position
and width, but the maximum-likelihood estimates for Ada
differed substantially from all other loci (Table 4). Maximum
likelihood provides a framework in which to compare clines
statistically when the mode of inheritance is known (Edwards
1972). We performed a G-test of the null hypothesis that all
loci are compatible with a single null cline center position
(Rao 1973). To determine the null log-likelihood, log-like-
lihoods were calculated over a series of 13 potential null
points chosen to span the observed range in cline centers
(Table 5). Separate Metropolis-Hastings searches were per-
formed for each locus using the appropriate optimized model
and parameter values (Table 4), but with c fixed at one of
the 13 potential null cline centers. The null log-likelihood
used in the test was the maximum log-likelihood of the
2077DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
TABLE 2. Extended.
Pgm-2
a b c d 2n FIS
l5
a b 2n FIS
pSCN3
a b 2n FIS
mtDNA
a b n
Maximum-
likelihood
genetic
hybrid index
Mean SD
0.000
0.000
0.000
0.000
0.000
?
0.000
0.000
0.000
0.000
0.048
?
1.000
1.000
1.000
0.975
0.952
?
0.000
0.000
0.000
0.025
0.000
?
10
14
40
40
42
0
?
?
?
?
20.026
?
0.917
0.891
1.000
0.825
0.952
1.000
0.083
0.109
0.000
0.175
0.048
0.000
12
46
40
40
42
8
?
20.100
?
20.187
20.026
?
0.917
0.848
0.825
0.825
0.833
0.875
0.083
0.152
0.175
0.175
0.167
0.125
12
46
40
40
42
8
?
10.179
20.187
20.187
10.504
?
1.000
1.000
1.000
1.000
1.000
1.000
0.000
0.000
0.000
0.000
0.000
0.000
10
14
20
20
21
4
0.984
0.908
0.973
0.912
0.936
0.968
0.04
0.14
0.05
0.10
0.10
0.06
0.000
0.000
0.000
0.000
0.000
0.025
0.000
0.068
0.654
0.875
0.800
0.625
1.000
0.932
0.308
0.075
0.033
0.300
0.000
0.000
0.038
0.050
0.167
0.050
20
44
52
40
30
40
?
20.050
10.131
20.080
10.028
10.153
0.875
0.864
0.327
0.125
0.000
0.000
0.125
0.136
0.673
0.875
1.000
1.000
24
44
52
40
40
40
20.100
20.135
10.232
20.188
?
?
0.833
0.841
0.288
0.075
0.025
0.000
0.167
0.159
0.712
0.925
0.975
1.000
24
44
52
40
40
40
20.158
20.167
10.176
20.056
?
?
1.000
0.900
0.385
0.050
0.100
0.000
0.000
0.091
0.615
0.950
0.900
1.000
12
22
26
20
20
20
0.923
0.867
0.325
0.056
0.043
0.005
0.07
0.16
0.28
0.08
0.07
0.02
TABLE 3. Log-likelihood scores for fitted clines under different mod-
els. See Materials and Methods for descriptions of models and good-
ness-of-fit tests. An asterisk denotes the model that provides the best
fit with the fewest number of estimated parameters.
Locus
ln L
Model I
(parameters 5 2)
ln L
Model II (ln L)
(parameters 5 4)
ln L
Model III (ln L)
(parameters 5 8)
Adaa
Ak-2a
Pgm-2b
Gsrb
l5b
pSCN3b
mtDNAb
29.6*
243.9
217.7
215.7
236.9
217.9
219.1
28.3
26.2*
26.7*
25.8*
212.2
22.6*
21.8*
27.9
26.2
26.7
25.0
26.6*
21.2
21.8
searches from this set, in this case 243.6 (c 5 208.2 km).
If there are significant differences in cline position, this null
log-likelihood will be significantly worse than the overall
log-likelihood (238.7) calculated from an unconstrained
search. The test statistic is calculated as the absolute differ-
ence between the two log-likelihoods (constrained and un-
constrained), multiplied by two. Significance is evaluated by
comparison to a x2-table, with the degrees of freedom equiv-
alent to (number of loci 2 1). The null hypothesis that all
loci have a common center was not rejected (x2 5 9.8, df 5
6), indicating that the Ada cline is consistent with the position
of the other genetic clines, given available data. The fact that
its estimated center is approximately 40 km northwest of the
others (Table 4) may simply be due to variability in the Ada
data.
Using the same methodology, we performed a test to detect
significant differences in cline width. To determine the null
log-likelihood, log-likelihoods were calculated over a range
of cline widths that encompassed the observed variation (Ta-
ble 6). The null hypothesis that all loci have the same cline
width was rejected (x2 5 44.8, df 5 6). To determine whether
Ada was responsible for the significant test, we performed a
second test, but with Ada excluded. In this test, the null was
not rejected (x2 5 4.8, df 5 5). Thus, we infer that cline
width is not significantly different among loci except for Ada,
which is wider. Maximum-likelihood confidence limits on
the Ada cline width estimate are very large (Table 4), again
due to variability in the data.
Morphological Variation and Cline Shape
The clines for PC1 (Fig. 4B) and beard length (Fig. 5A)
were similar in position (Table 7) to the genetic clines (Table
4). Because it is composed of four mensural variables with
similar factor loadings, PC1 is closely related to overall size.
Average PC1 scores changed from candei-like in populations
1?8 to vitellinus-like in populations 10?12 (Fig. 4B). PC1
scores in populations 1?8 were not significantly different
(one-way ANOVA, F 5 1.2, P . 0.05). An ANOVA of
populations 1?9 was significant (F 5 6.5, P , 0.0001) due
to the morphological intermediacy of hybrid individuals in
population 9. The epaulet width cline was broader (Fig. 5A)
and its center several kilometers northwest of the other cline
centers (Table 7), but the estimate was near the minimum
two log-likelihood unit support limit for some of the genetic
clines (Table 4).
In contrast to the molecular and morphometric character
clines, the two plumage color clines (Fig. 5B) have steep tran-
sitions centered about 50 km northwest of the other cline cen-
ters. The clines are narrow, with estimated widths of 3.0 and
4.4 km (Table 7). In both cases, the most dramatic change in
color occurs between populations 3 and 4, which are only 8
km apart, but which lie on opposite banks of the Changuinola
River. Eighty-five percent of the change in mean collar color
occurs across the Changuinola River, with the remaining 15%
between populations 4 and 10. Qualitatively, collar color
changes abruptly from white in population 3 to lemon yellow
in population 4, then subtly over the next 70 km to golden
yellow in population 10. Belly color changes from lemon yel-
low at population 3 to a drab green at population 10, with
much the same trajectory. Fifty-four percent of the total change
in mean belly color occurs across the Changuinola River.
Statistical Evaluation of Cline Shifts
Statistical significance of the color cline shifts was eval-
uated with a nonparametric bootstrap approach (Efron and
2078 ROBB T. BRUMFIELD ET AL.
FIG. 3. Clines, fit by maximum likelihood, for (A) the four genetic markers with fixed or nearly fixed alternate alleles in each form
and (B) the three genetic markers that differed in allele frequency between the two forms. Symbols indicate observed allele frequencies.
The vertical dashed line indicates the position of the Changuinola River. Railroad bars on the x-axis denote a nonlinear break in geographic
distance.
Tibshirani 1993) using smoothing splines (Schluter 1988).
This has the advantages of speed and flexibility, and, more
importantly, is not constrained by any particular sampling
distribution. For individual collar and belly color measure-
ments, 200 bootstrap samples were taken at each population.
Generalized cross-validated smoothing splines were then fit
using S-plus (MathSoft 1999) to the resulting bootstrap sam-
ple means weighted by the inverse of the bootstrap sample
variances. The 200 resulting curves were used to estimate
200 sample cline centers as the point of steepest slope on
each curve. The distribution of these 200 cline centers models
the variability in position among the color bootstrap samples.
To compare the color clines to the molecular clines, 200
bootstrap samples of each genetic marker were generated
using a binomial model with probability of success being the
observed gene frequency at each population. Generalized
cross-validated smoothing splines were then fit to the re-
sulting logit transformed bootstrap sample means weighted
by the inverse of the estimated logit variances. The fitted
smoothing splines were transformed back to a cline of esti-
mated gene frequencies. As in the color data, these 200 re-
sulting curves were used to estimate 200 sample cline po-
sitions as the point of steepest slope of each curve. The dis-
tribution of these 200 cline centers models the variability in
position among the molecular bootstrap samples.
The statistical significance of the shifted positions of the
color clines was assessed using P-values given by the per-
centage of the bootstrap color cline positions that occur out-
side the distribution of the bootstrap genetic cline positions.
We found that all of the 200 bootstrap samples for both collar
and belly color had cline positions outside the distribution
of the 200 bootstrap samples of all genetic markers except
Ada (Fig. 6). Thus, sampling variability alone cannot account
for the shifted plumage color clines. The broad distribution
of Ada is consistent with the large confidence intervals on
its cline position estimate (Table 4). An analogous bootstrap
analysis of the epaulet width cline indicated that its position
is not significantly shifted.
DISCUSSION
Our results demonstrate that an array of six molecular and
two morphometric markers agree in cline position and width
and define the center of the hybrid zone within a few kilo-
meters of population 9. These include markers of the nuclear
and mtDNA genome, assayed by both allozyme and restric-
2079DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
TABLE 4. Likelihood support and parameter estimates for the genetic clines. Two log-likelihood unit support limits are presented in parentheses.
For comparison, two measures of cline width are presented: 1/maximum slope (w), and distance between 20% and 80% points (w1). Parameter
c is the cline center measured in distance (km) from population 1, pmin is the minimum estimated frequency at the west end of the cline, pmax
is the maximum estimated frequency at the eastern end, B and u are shape parameters for the exponential decay curves. Numbers in bold are
parameters reoptimized after excluding population 6 (no value is presented in cases where the reoptimized parameter was identical).
Adaa Ak-2a Pgm-2b Gsrb l5b pSCN3b mtDNAb
ln L
w
29.6
262.5
(110.0?856.4)
26.2
9.9
(0.9?15.3)
26.7
7.7
(0.5?12.0)
25.8
2.9
(0.2?37.9)
26.6
25.9
10.4
(7.2?14.4)
22.6
2.8
(0.6?15.4)
3.9
21.9
11.1
(6.9?19.0)
w1
c
204.9
171.0
(108.7?200.1)
6.9
208.6
(207.3?210.2)
5.50
206.5
(201.5?209.5)
4.4
209.9
(202.1?223.1)
11.6
208.2
(206.2?208.4)
208.3
4.9
209.4
(206.7?209.9)
209.3
7.7
208.3
(206.4?210.3)
208.2
pmin
pmax
Bc
0.1
(fixed)
0.675
(fixed)
0
0.0
(0.0?0.1)
1.0
(1.0?1.0)
0
0.0
(0.0?0.0)
0.8
(0.7?0.8)
0
0.0
(0.0?0.1)
0.3
(0.2?0.4)
0
0.1
(0.1?0.1)
1.0
(1.0?1.0)
1535.9
(1520?1556)
0.2
(0.1?0.2)
1.0
(0.9?1.0)
0
0.0
(0.0?0.0)
0.9
(0.9?1.0)
0
uc 1 1 1 1
3217.5
1.0
(0.2?1.0)
0.9
1 1
Bv
uv
0
1
0
1
0
1
0
1
3.1
(2.6?13.7)
0.1
(0.0?0.1)
0
1
0
1
tion-fragment-length-polymorphism analysis, and one mark-
er, PC1, that integrates four morphometric characters. Two
other clines, Ada and epaulet width, are similar in position,
but broader in width. Finally, two male plumage color clines,
collar color and belly color, are similar in width to the first
eight clines, but are shifted in position by at least five cline
widths. These results confirm and extend the observations of
Parsons et al. (1993). The two phenotypic clines in male
plumage color are shifted by about 50 km away from the
genetically and morphometrically defined hybrid zone to the
banks of the Changuinola River, between populations 3 and
4. The result is that birds in populations 4?8 are genetically
and morphometrically very like parental candei, but their
yellow collars and greenish bellies make them appear su-
perficially like parental vitellinus.
Butlin and Neems (1994) suggested that the apparent shift
of Manacus color clines might be an artifact of (1) greater
cline width; (2) nonlinear color scales; or (3) genetic dom-
inance of vitellinus traits. To address the first two possibil-
ities, we measured color quantitatively using a reflectance
spectrophotometer, sampled two new populations (8 and 9)
near the center of the hybrid zone, increased the sample size
of other populations, and made statistical tests of cline shifts.
The spectrophotometric measurements show even more dra-
matic shifts in color than was evident in the semiquantitative
scale previously employed. Fully 85% of the change in mean
collar color and 54% of the change in mean belly color occurs
across the Changuinola River. The increased sampling pro-
vides greater resolution near the center of the hybrid zone,
revealing that most of the genetic and morphometric clines
are in fact narrower than could be demonstrated with the
previous data. The statistical comparisons show unequivo-
cally that the collar and belly color cline centers are shifted
significantly. The net effect of these refinements is to reveal
that the collar and belly color clines are actually shifted about
10 km further west than the minimum estimated by Parsons
et al. (1993).
We did not examine a third potentially shifted cline, that
for collar width (Parsons et al. 1993), because we found that
the character was unduly influenced by the style of specimen
preparation. Instead, we substituted epaulet width, which ap-
pears to be correlated with collar width (at least in parental
populations) and can be measured more accurately (Brum-
field 1999). We found the epaulet width cline to be broader,
but not shifted significantly away from the genetic and mor-
phometric clines. This may indicate that the concerns of But-
lin and Neems (1994) about the collar width cline were well
founded. However, collar width is a relatively minor char-
acter; collar and belly color have much greater influence on
male appearance.
Several methods have been used to test for differences in
cline structure. Simple approaches involve testing for ho-
mogeneity of allele frequencies among loci or among pop-
ulations on a transect (Dod et al. 1993; Parsons et al. 1993;
Jaarola et al. 1997). These approaches are limited in the as-
pects of cline structure examined and use only a fraction of
the relevant data. Several techniques attempt to use more
data by modeling cline structure with fitted curves (Szymura
and Barton 1986; Tucker et al. 1992; Barton and Baird 1999).
These approaches are constrained by the requirements of us-
ing a fixed mathematical form or assuming a particular mode
of inheritance. Smoothing splines provide a flexible approach
to curve fitting without requiring a fixed mathematical form.
In effect, our approach allows the data to determine the form
that best describes them. The nonparametric bootstrap allows
for statistical comparisons among clines where the mode of
2080 ROBB T. BRUMFIELD ET AL.
FIG. 4. Individuals scores in each population for (A) the maximum-likelihood genetic hybrid index of seven loci and (B) the first
principal component axis (PC1) integrating four morphometric characters (mass, wing length, tail length, tarsus length). For the genetic
index, the number of individuals having scores of 0 or 1 is in italics. The starred individual is a male (USNM 608981) from population
9 with aberrant plumage color. The vertical dashed line indicates the position of the Changuinola River. Bars on the x-axis denote a
nonlinear break in geographic distance.
inheritance is not known. This flexibility is crucial in the
present case, where we compare phenotypic plumage color
clines to allele frequency clines at molecular loci.
The possibility that genetic dominance of vitellinus color
traits could cause the phenotypic clines to be shifted, when
the underlying allele frequency clines are in fact coincident,
is more difficult to eliminate. In both Dendroica warbler
(Rohwer and Wood 1998) and Gymnorhina magpie hybrid
zones (Hughes 1982), genetic dominance at one or a few
hypothetical plumage trait genes were invoked as potential
explanations for the shifted clines observed in those zones.
Ultimately, crossing experiments would be required to ad-
dress directly the mode of inheritance of the plumage traits
in Manacus. Such experiments will be difficult to perform in
lek-mating birds (we have tried).
It is useful to consider the number of loci and allele fre-
quencies required by a dominance hypothesis to account for
the observed phenotypic frequencies (Parsons et al. 1994).
Under a model of dominant vitellinus color traits, white-col-
lared males would most likely be observed at population 4,
near the source of candei alleles. We have now observed
more than 200 males in the region of populations 4?8, many
individually marked and released, and have never detected a
typical white-collared male (Parsons et al. 1994; McDonald
et al. 2001). Thus, the observed frequency of white-collared
males is less than 0.005. If we assume a simple model in
which collar color is controlled by a series of unlinked loci,
and that dominant vitellinus alleles occur at all loci at a fre-
quency of 0.2, about 12 loci would be required to account
for the observed phenotypic frequency [(0.8)2]12 5 0.0047).
These are generous assumptions considering the average fre-
quency of diagnostic vitellinus molecular markers at popu-
lation 4 is 0.094 (range 0.000?0.175). Given the uniformity
in the widths of the clines for the other diagnostic molecular
and morphological markers, it seems more plausible to sup-
pose that the plumage color clines are also narrow, but shift-
ed. This interpretation is consistent with available studies
documenting that similar plumage traits in other birds are
under the control of relatively few genes (Johnson and Brush
1972; Buckley 1987; Rohwer and Wood 1998).
We have found only a single white-collared male in the
transect populations east of the Changuinola River that might
conceivably represent a homozygous recessive recombinant.
In 1994, a white-collared specimen (USNM 608981) was
2081DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
TA
B
L
E
5.
G
-
te
st
o
ft
he
n
u
ll
hy
po
th
es
is
th
at
se
v
en
lo
ci
ha
ve
th
e
sa
m
e
cl
in
e
ce
n
te
r.
M
at
rix
en
tr
ie
s
ar
e
th
e
lo
g-
lik
el
ih
oo
d
su
pp
or
tv
al
ue
s
(ln
L c
)f
or
an
an
al
ys
is
in
w
hi
ch
c
fo
ra
ll
cl
in
es
is
co
n
st
ra
in
ed
to
o
n
e
o
ft
he
13
n
u
ll
v
al
ue
s
(c
5
17
1.
0?
20
9.
9)
ch
os
en
to
sp
an
th
e
o
bs
er
v
ed
ra
n
ge
in
cl
in
e
ce
n
te
rs
.l
n
L u
is
th
e
u
n
co
n
st
ra
in
ed
lo
g-
lik
el
ih
oo
d
su
pp
or
t.
U
si
ng
th
e
m
ax
im
um
o
v
er
al
ll
n
L c
as
th
e
n
u
ll
v
al
ue
(i.
e.,
2
43
.6
),
th
e
n
u
ll
hy
po
th
es
is
th
at
al
lc
lin
e
ce
n
te
rs
ar
e
th
e
sa
m
e
is
n
o
t
re
jec
ted
(x
2
5
9.
8,
df
5
6)
.T
he
cr
iti
ca
lx
2 -
v
al
ue
fo
rs
ix
de
gr
ee
s
o
ff
re
ed
om
is
12
.6
.
C
lin
e
c
e
n
te
r
(c)
17
1.
0
17
5.
6
18
0.
3
18
4.
9
18
9.
5
19
4.
1
19
8.
8
20
3.
4
20
6.
5
20
7.
3
20
8.
2
20
9.
0
20
9.
9
ln
L u
Ad
a
Ak
-2
G
sr
Pg
m
-2
2
9.
6
2
10
1.
4
2
20
.2
2
73
.5
2
9.
7
2
94
.9
2
19
.4
2
67
.2
2
9.
9
2
87
.3
2
18
.4
2
60
.5
2
10
.2
2
78
.5
2
17
.4
2
53
.3
2
10
.5
2
68
.8
2
16
.3
2
45
.5
2
11
.0
2
58
.0
2
15
.1
2
37
.0
2
11
.4
2
44
.3
2
13
.9
2
27
.5
2
11
.8
2
21
.5
2
7.
4
2
7.
5
2
12
.2
2
9.
8
2
7.
1
2
6.
7
2
12
.2
2
7.
7
2
6.
7
2
7.
0
2
12
.3
2
6.
4
2
6.
3
2
8.
0
2
12
.4
2
6.
4
2
6.
0
2
8.
2
2
12
.5
2
7.
2
2
5.
8
2
8.
2
2
9.
6
2
6.
2
2
5.
8
2
6.
7
l
5 pS
C
N
3
m
tD
N
A
S
ln
L
2
33
.6
2
60
.7
2
58
.4
2
35
7.
4
2
33
.6
2
55
.8
2
54
.3
2
33
4.
9
2
24
.5
2
50
.3
2
49
.4
2
30
0.
3
2
22
.4
2
44
.3
2
44
.0
2
27
0.
1
2
18
.5
2
38
.2
2
37
.8
2
23
5.
6
2
13
.4
2
32
.6
2
30
.6
2
19
7.
7
2
8.
0
2
27
.2
2
21
.8
2
15
4.
1
2
6.
1
2
13
.6
2
8.
8
2
76
.7
2
6.
0
2
4.
7
2
3.
1
2
49
.6
2
6.
0
2
3.
4
2
2.
2
2
45
.2
2
5.
9
2
2.
8
2
1.
9
2
43
.6
*
2
6.
2
2
2.
6
2
2.
1
2
43
.9
2
7.
6
2
2.
6
2
2.
8
2
46
.7
2
5.
9
2
2.
6
2
1.
9
2
38
.7
D
5
[S
ln
L c
2
S
ln
L u
]
x
2
5
2D
31
8.
7
63
7.
4
29
6.
2
59
2.
4
26
1.
6
52
3.
2
23
1.
4
46
2.
8
19
6.
9
39
3.
8
15
9
31
8
11
5.
4
23
0.
8
38 76
10
.9
21
.8
6.
5
13
4.
9
9.
8*
5.
2
10
.4
8 16
collected at population 9, near the center of the genetic and
morphological clines. The bird is aberrant in that the reflec-
tance of its white collar plumage is uniformly higher than in
parental candei (Fig. 2), and its belly is drab green suffused
with yellow streaks, typical of other males in population 9,
but unlike the lemon yellow belly color of candei. Other than
its white collar, the specimen is characteristic of other intro-
gressed individuals in population 9, with intermediate PC1
and maximum-likelihood genetic hybrid index scores (see
starred individual in Fig. 4). Because the specimen was col-
lected far from population 4, where white-collared homo-
zygous recessive birds are most likely under the dominance
hypothesis, it is unclear whether the specimen is white-col-
lared because it is homozygous recessive or because it is a
mutant. The disruption of coadapted gene complexes that
could cause such a mutation (Woodruff 1989) is expected to
be highest in population 9, at the hybrid zone center.
At this point, the dominance hypothesis seems less likely
because the cline shifts are much greater (about five cline
widths) than would be expected under simple models of dom-
inance (Endler 1977) and because the homozygous recessive
white-collared birds expected under this hypothesis are not
observed at a plausible frequency in populations 4?8 (Parsons
et al. 1994). In addition, a potential behavioral correlate of
sexual selection on male plumage color has now been ob-
served (McDonald et al. 2001). Experiments involving the
presentation of taxidermic mounts to males on leks found
that lemon- or golden-collared males, regardless of location,
were significantly more aggressive than were white-collared
males. Heightened aggression could conceivably translate
into increased mating success for yellow-collared males in
mixed leks (McDonald et al. 2001).
Evolutionary Implications of Shifted Clines
The majority of hybrid zone studies report cline transitions
that coincide in position (Barton and Hewitt 1985). Relatively
few cases have been reported where clines are noncoincident,
although such reports seem to be increasing (Barton 1993;
Jaarola et al. 1997). This increase may be due to larger num-
bers of clines being characterized per study and the increased
use of molecular and chromosomal markers. Jaarola et al.
(1997) reviewed examples of noncoincident clines and dis-
cussed potential mechanisms that might cause cline displace-
ment. Such cases may be roughly divided into two types
(Barton 1993); those characterized by staggered clines, where
an array of individual character clines are shifted slightly
away from each other, and those due to asymmetric intro-
gression of one or a few characters beyond the limits of the
hybrid zone. The Manacus case is an example of the latter.
The large cline shifts found in cases of asymmetric intro-
gression might be explained by neutral introgression, move-
ment of the zone, founder effect, or selective advantage. In
the Manacus zone, the magnitude of the cline shift appears
to be greater than can be explained by neutral processes alone.
Although clines can be displaced by genetic drift given suf-
ficient population subdivision, the displacement is not ex-
pected to exceed one cline width (Barton 1983). Moreover,
there is probably little population subdivision among man-
akins of populations 4?8. Early successional forest and aban-
2082 ROBB T. BRUMFIELD ET AL.
TABLE 6. G-test of the null hypothesis that all loci have the same cline width. Matrix entries are the log-likelihood support values (ln Lc) for
an analysis in which w is constrained to one of the eight null values (w 5 2.0?62.0 km). These widths were examined because they span the
range of observed cline widths, with the exception of Ada. Larger cline widths were examined, but all produced increasingly worse log-
likelihoods. ln Lu is the unconstrained log-likelihood support. Using the maximum overall ln Lc as the null value (i.e., 261.1), the null hypothesis
that the seven cline widths are the same is rejected (x2 5 44.8, df 5 6). Repeating this analysis with Ada excluded results in a loss of significance
(x2 5 4.8, df 5 5), indicating that the cline for Ada is significantly wider. The critical x2-value for five degrees of freedom is 11.1.
Cline width (w) 2.0 7.0 12.0 17.0 22.0 32.0 42.0 62.0 ln Lu
Ada
Ak-2
Gsr
Pgm-2
230.1
27.2
25.8
27.5
229.7
26.7
25.8
26.8
228.5
26.6
26.2
28.5
228.0
28.8
26.7
211.9
227.5
211.7
27.1
214.5
225.5
217.6
27.5
218.4
222.2
223.5
27.7
222.8
216.0
233.4
28.4
231.8
29.6
26.2
25.8
26.7
l5
pSCN3
mtDNA
26.0
22.6
24.8
26.0
22.6
23.6
26.0
23.4
21.9
26.5
25.1
23.0
29.3
26.6
24.4
213.5
28.9
27.5
215.8
211.8
210.7
221.6
217.2
216.7
25.9
22.6
21.9
S ln L (with Ada)
S ln L (w/o Ada)
D 5 [S ln Lc 2 S ln Lu] (with Ada)
D 5 [S ln Lc 2 S ln Lu] (w/o Ada)
x
2
5 2D (with Ada)
x
2
5 2D (w/o Ada)
264.0
233.9
25.3
4.8
50.6
9.6
261.2
231.5
22.5
2.4
45.0
4.8*
261.1
232.6
22.4
3.5
44.8*
7.0
270.0
242.0
31.3
12.9
62.6
25.8
281.1
253.6
42.4
53.6
84.8
107.2
298.9
273.4
60.2
73.4
120.4
146.8
2114.5
292.3
75.8
92.3
151.6
184.6
2145.1
2129.1
106.4
129.1
212.8
258.2
238.7
229.1
doned cacao plantations occur almost continuously through-
out the region, and manakins are abundant in these habitats.
It would be hard to exclude movement of the zone as an
explanation for the shifted clines without longer-term tem-
poral information. Analyses of historical specimens suggest
an eastward shift of the hybrid zone over the past 100 years
on the order of several kilometers, except for collar color,
which has shifted northwestward (R. T. Brumfield, unpubl.
data).
The Manacus case is an unusual example of asymmetric
introgression because it involves secondary sexual traits,
which are likely to play a role in mating success in these
highly sexually dichromatic lek-mating species. Selection for
vitellinus traits remains a plausible explanation for the shifted
clines. Females of both species could preferentially mate with
yellow-collared males, or yellow-collared males could out-
compete white-collared males within leks. The greater ag-
gression of both golden- and lemon- versus white-collared
males in male-male interactions may confer a selective ad-
vantage (McDonald et al. 2001). Thus, selection may be act-
ing directly on male aggression, with male plumage color
introgressing as a correlated trait (Rohwer et al. 2001).
Other possible advantages of yellow collars involve en-
vironmental gradients in light regime. Fitness differences
could relate to visibility differences among male plumage
types in the forest understory. Recent studies of other man-
akin species have shown that the conspicuousness of brightly
plumaged males varies with the background light regime (En-
dler and The?ry 1996). Under this hypothesis, a light regime
gradient along which male candei and vitellinus are not equal-
ly conspicuous could lead to displaced color clines (Hewitt
1988). Although forest composition is superficially similar
along the transect, this possibility needs to be examined.
If vitellinus color traits are under positive selection, it is
puzzling that they have not yet progressed to fixation. Be-
cause selected traits are expected to sweep rapidly to fixation,
it is unusual to observe a sweep in progress. One possibility
is that the Rio Changuinola, where the plumage color clines
are centered, is a barrier to the spread of the traits. Using
data from microsatellite loci, Brumfield (1999) found that
genetic differentiation among populations across the river is
greater than that on either side, suggesting that the river is
a significant barrier to the movement of individuals. The
barrier is clearly not impenetrable, however, because vitel-
linus alleles (l5b, pSCN3b) have been detected in populations
west of the river. The hypothesized selection for vitellinus
color traits might be frequency dependent, which would ac-
centuate the effect of a river barrier in obstructing intro-
gression.
Why the genetic and morphometric clines are centered
where they are also remains unclear. Although no obvious
environmental transitions occur in this region, a number of
avian species pairs replace each other in western Caribbean
Panama, so the region is a suture zone (sensu Remington
1968). However, none of the other contact zones occur in
the same exact location as the Manacus contact near popu-
lation 9 (Olson 1993). The lowland forest habitat of Manacus
in western Panama is restricted to a thin coastal strip sand-
wiched between the mountains and the sea. This ribbon of
habitat is especially narrow around populations 8 and 9, per-
haps creating a bottleneck for dispersal. Tension zones are
expected to move to regions where dispersal into the zone
is equivalent for the two parental species, which may explain
the location of the clines in this area. The Changuinola River
represents another possible region to which the zone would
be expected to move, because dispersal appears to be inter-
rupted by this barrier (Brumfield 1999). However, its effect
may be less pronounced. Morphometric analysis of historical
specimens from near population 5 suggests that those clines
have actually moved several kilometers eastward over the
past 100 years, away from the Changuinola River (Brumfield
unpubl. data).
Divergent Cline Widths
Assuming the hybrid zone is in equilibrium, additional
insights into the architecture of manakin speciation are pos-
sible through comparisons of the relative widths of indepen-
dent clines. Clines for six of the seven molecular loci did
not differ in position or width, a pattern that is consistent
2083DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
FIG. 5. Fitted clines for (A) beard length and epaulet width and (B) collar color and belly color as measured by percent reflectance at
490 nm and 665 nm, respectively. Error bars are standard deviations. The vertical dashed line indicates the position of the Changuinola
River. Bars on the x-axis denote a nonlinear break in geographic distance.
TABLE 7. Maximum-likelihood parameter estimates for morpholog-
ical character clines. See Table 4 for parameter definitions.
PC1
Beard
length
Epaulet
width
Collar
color
Belly
color
w
w1
c
pmin
pmax
13.9
9.7
208.7
0.3
0.7
10.3
7.2
208.8
0.6
0.8
65.2
44.8
200.0
0.3
0.7
4.4
3.0
157.6
0.4
1.0
3.0
2.1
157.8
0.3
0.9
Bc
Qc
Bv
Qv
61.
0.0
1000
0.8
1.3
0.8
0.9
0.2
7500
0.9
12.3
0.1
96.4
0.4
54.2
0.0
1041.2
0.2
8.2
0.0
with a tension zone model wherein hybrid zone stability is
mediated by a balance between the influx of parentals into
the zone and the removal of hybrids due to selection (Haldane
1948; Barton 1983). Both endogenous and exogneous selec-
tion would have the same general impact on cline structure
if ??speciation genes?? exist throughout the genome. Con-
cordance of cline widths reflects the sum of effects of loci
acting in association over the genome.
The Ada cline was significantly broader than the other
molecular clines, although we caution that this conclusion
assumes that the hyperbolic tangent function was an adequate
model of character transition. Ada was also the only locus
that was strongly polymorphic in both parental forms. We
might expect, a priori, that polymorphic loci are less likely
to be subject to incompatibility selection from the simple fact
that both alleles exist in both parental populations. Thus, the
greater degree of bidirectional introgression at Ada is con-
sistent with the idea that it is experiencing less selection and
more introgression than other loci. The same may be true of
the epaulet width cline. While selection against hybrids may
have limited introgression at some loci, it clearly has not
prevented introgression of Ada alleles or of vitellinus color
traits.
Although clines for the six concordant molecular loci are
on the whole very similar, they differ in their tails of low
level introgression. The mtDNA displayed the least amount
of introgression; the vitellinus mitochondrial haplotype was
not detected west of population 8, unlike other vitellinus
markers, which were all found at least as far west as popu-
lation 2. Less extensive sampling on the east side of the
hybrid zone prevents any strong conclusions regarding in-
trogression of the candei haplotype. It is noteworthy that the
difference among markers in the extent of low level intro-
gression is readily apparent in allele frequencies (Table 2),
but is not conveyed by either cline width measure (Table 4).
The apparently limited introgression of mtDNA in the man-
2084 ROBB T. BRUMFIELD ET AL.
FIG. 6. Histograms showing significant displacement of 200 boot-
strap cline centers for introgressing lemon collar color and greenish
belly color relative to most of the genetic markers. The position of
the Ada cline could not be estimated with confidence. The position
of the Changuinola River is indicated by the vertical dashed line.
akin hybrid zone may be a stochastic effect of the smaller
effective population size inherent in a uniparental haploid
locus. There may also be an effect related to Haldane?s rule
(Haldane 1922; Coyne and Orr 1989), given that mtDNA is
maternally inherited and females are the heterogametic sex
in birds. However, there are good reasons to suspect that
selection could contribute to this pattern as well. From a
purely biochemical perspective, the fact that many essential
genetic functions are packed onto the small mtDNA molecule
suggests that incompatibilities could arise readily. The lack
of recombination in mtDNA precludes the removal of in-
compatibilities from a haplotype. Finally, because the mi-
tochondrial genome is haploid, all variation is exposed to
selection in all individuals. Thus, its limited introgression
could be due to the same forces that produce narrow clines
at sex chromosome markers (Coyne and Orr 1989; Tucker et
al. 1992; Dod et al. 1993).
Marker Choice in Hybrid Zone Studies
The contrast between the sharp clines for diagnostic mark-
ers and the broad, bidirectional introgression observed at the
polymorphic locus Ada highlights a bias in hybrid zone stud-
ies that may be subtly influencing our views of speciation.
Evolutionary biologists typically characterize hybrid zones
using diagnostic traits that show fixed differences between
parental taxa. The choice seems natural, because hybrids can
be identified unambiguously with such markers and any oc-
currence of a foreign allele can be interpreted as introgres-
sion. However, if we are interested in gaining a quantitative
sense of introgression across a hybrid zone, diagnostic loci
are likely to give an underestimate. Diagnostic loci, on av-
erage, are likely to be under stronger selection than other
loci for the reasons given above. When they are, diagnostic
loci will yield underestimates of average introgression, be-
cause relevant modes of selection will oppose the movement
of alleles across a hybrid zone (Sattler and Braun 2000).
The strength of this bias will vary with the degree of ge-
nomic divergence between hybridizing taxa. When substan-
tial genomic divergence is present, there are likely to be many
genetic incompatibilities spread throughout the genome, and
many diagnostic markers all experiencing similar levels of
generalized selection against hybrids (Foltz 1997), because
physical linkage and gametic disequilibria can extend the
effects of selection at one locus to others. This is the classic
tension zone model, where the sum of the effects of these
associated loci creates a genomewide barrier to introgression
(Barton and Hewitt 1989). At a finer scale, the strength of
this barrier to introgression will vary among regions of the
genome depending on the number of loci under selection and
the degree of physical or epistatic linkage of marker loci to
them (Barton and Bengtsson 1986). By sampling enough
markers, molecular data can provide a detailed map of se-
lection intensities experienced across the entire genome (Kim
and Rieseberg 1999).
Compared to many organisms, bird species typically show
low levels of genetic divergence and low levels of hybrid
dysfunction (Avise 1994). Thus, the bias caused by using
diagnostic rather than nondiagnostic markers as estimators
of average introgression is likely to be large. Indeed, a rel-
ative dearth of genetic markers is often available for distin-
guishing closely related avian species. In studies of hybrid
zones, ornithologists have had no choice but to include mark-
ers that are not diagnostic, sometimes characterizing hybrid
zones without a single diagnostic genetic marker (e.g., Bar-
rowclough 1980). The serendipitous benefit is that these may
represent neutral or weakly selected markers (Sattler and
Braun 2000), loci that are more likely to reveal nuclear in-
trogression.
By focusing strictly on diagnostic markers, we tend to
create a self-fulfilling paradigm of hybrid zones as black
holes: Genes go in but they do not come out. This view has
led some to suggest that studies of hybridization are irrelevant
to speciation (Gill et al. 1993). We encourage hybrid zone
researchers to examine the structure of as many genetic mark-
ers as possible, including both diagnostic and nondiagnostic
loci. Similar cline widths among many unlinked loci (in-
cluding those likely to be neutral) provide evidence for strong
genomewide selection, whereas differences in cline structure
reveal distinct forces acting at different loci. Although the
likelihood of introgression across hybrid zones will vary with
degree of genomic divergence (Barton and Bengtsson 1986;
Foltz 1997), in most cases these zones probably provide a
evolutionary conduit through which adaptive and neutral
traits can move (Pia?lek and Barton 1997; Barton 2001), en-
larging the pool of genetic variation available to both parental
populations (Parsons et al. 1993).
ACKNOWLEDGMENTS
We thank the late C. Handley for setting up fieldwork in
Panama; R. Clay, T. Glenn, J. Blake, L. Horth, A. Pineda
Sr., A. Pineda Jr., A. Ceden?o, and K. Gorricha?tegui for field
2085DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
assistance; and INRENARE for granting collecting permits
in Panama. M. Leone, G. Maggiori, E. Bermingham, and
others at the Smithsonian Tropical Research Institute facil-
itated our work in Panama. T. Glenn, C. Huddleston, G. Sat-
tler, T. Parsons, and others at the Laboratory of Molecular
Systematics provided technical assistance. S. J. E. Baird as-
sisted with cline-fitting analyses and provided criticisms of
the manuscript. E. Russek-Cohen helped design the likeli-
hood-ratio tests. J. Dean assisted in the preparation and cat-
aloging of specimens. For lending skins, we thank the cu-
ratorial staffs of the Louisiana State University Museum of
Natural Science (Baton Rouge), the Carnegie Museum of
Natural History (Pittsburgh), and the Museum of Vertebrate
Zoology (Berkeley). We thank the Louisiana State University
Museum of Natural Sciences Collection of Genetic Resources
for providing tissues from Costa Rica. M. Baker assisted with
reflectance spectrophotometry. D. Cole kindly prepared Fig-
ure 1. H. L. Gibbs and two anonymous reviewers provided
constructive comments on the manuscript. This research was
supported in part by a Smithsonian Predoctoral Fellowship
to RTB.
LITERATURE CITED
Anderson, E. 1949. Introgressive hybridization. John Wiley, New
York.
Anderson, E., and L. Hubricht. 1938. Hybridization in Tradescantia.
III. The evidence for introgressive hybridization. Am. J. Bot.
25:396?402.
Arnold, M. L. 1992. Natural hybridization as an evolutionary pro-
cess. Annu. Rev. Ecol. Syst. 23:237?261.
???. 1997. Natural hybridization and evolution. Oxford Univ.
Press, New York.
Avise, J. C. 1994. Molecular markers, natural history, and evolution.
Chapman and Hall, New York.
Barrowclough, G. F. 1980. Genetic and phenotypic differentiation
in a wood warbler (genus Dendroica) hybrid zone. Auk 97:
655?668.
Barton, N. H. 1979. Gene flow past a cline. Heredity 43:333?339.
???. 1983. Multilocus clines. Evolution 37:415?426.
???. 1993. Hybrid zones: Why species and subspecies? Curr.
Biol. 3:797?799.
???. 2001. The role of hybridization in evolution. Mol. Ecol.
10:551?568.
Barton, N. H., and S. J. E. Baird. 1999. Analyse: software for
analysis of geographic variation and hybrid zones. Ver. 1.03 for
Macintosh. Available via http://helios.bto.ed.ac.uk/evolgen/
Mac/Analyse/index.html.
Barton, N. H., and B. O. Bengtsson. 1986. The barrier to genetic
exchange between hybridising populations. Heredity 56:
357?376.
Barton, N. H., and G. M. Hewitt. 1985. Analysis of hybrid zones.
Annu. Rev. Ecol. Syst. 16:113?148.
???. 1989. Adaptation, speciation, and hybrid zones. Nature
341:497?503.
Brumfield, R. T. 1999. Evolution of brilliant male plumage traits
in Manacus. Ph.D. diss., University of Maryland, College Park,
MD.
Brumfield, R. T., and M. J. Braun. 2001. Phylogenetic relationships
in bearded manakins (Pipridae: Manacus) indicate that male
plumage color is a misleading taxonomic marker. Condor 103:
248?258.
Buckley, P. A. 1987. Mendelian genes. Pp. 1?44 in F. Cooke and
P. A. Buckley, eds. Avian genetics. Academic Press, San Diego,
CA.
Butlin, R. K., and R. M. Neems. 1994. Hybrid zones and sexual
selection. Science 265:122.
Butlin, R. K., M. G. Ritchie, and G. M. Hewitt. 1991. Comparisons
among morphological characters and between localities in the
Chorthippus parallelus hybrid zone (Orthoptera: Acrididae). Phi-
los. Trans. R. Soc. Lond. B 334:297?308.
Coyne, J. A., and H. A. Orr. 1989. Two rules of speciation. Pp.
180?207 in D. Otte and J. A. Endler, eds. Speciation and its
consequences. Sinauer, Sunderland, MA.
Dod, B., L. S. Jermiin, P. Boursot, V. H. Chapman, J. T. N. Nielsen,
and F. Bonhomme. 1993. Counterselection on sex chromosomes
in the Mus musculus European hybrid zone. J. Evol. Biol. 6:
529?546.
Dowling, T. E., C. Moritz, J. D. Palmer, and L. H. Rieseberg. 1996.
Nucleic acids. III. Analysis of fragments and restriction sites.
Pp. 249?320 in D. M. Hillis, C. Moritz, and B. K. Mable, eds.
Molecular systematics. Sinauer, Sunderland, MA.
Dumolin-Lape?gue, S., A. Kremer, and R. J. Petit. 1999. Are chlo-
roplast and mitochondrial DNA variation species independent
in oaks? Evolution 53:1406?1413.
Edwards, A. W. F. 1972. Likelihood. Johns Hopkins Univ. Press,
Baltimore, MD.
Efron, B., and R. J. Tibshirani. 1993. An introduction to the boot-
strap. Chapman and Hall, New York.
Endler, J. A. 1977. Geographic variation, speciation, and clines.
Princeton Univ. Press, Princeton, NJ.
???. 1990. On the measurement and classification of colour in
studies of animal colour. Biol. J. Linn. Soc. 41:315?352.
Endler, J. A., and M. The?ry. 1996. Interacting effects of lek place-
ment, display behavior, ambient light, and color patterns in three
Neotropical forest-dwelling birds. Am. Nat. 148:421?452.
Feinberg, A. P., and B. Vogelstein. 1993. A technique for radio-
labeling DNA restriction endonuclease fragments to high spe-
cific activity. Anal. Biochem. 132:6?13.
Foltz, D. W. 1997. Hybridization frequency is negatively correlated
with divergence time of mitochondrial DNA haplotypes in a sea
star (Leptasterias spp.) species complex. Evolution 51:283?288.
Gill, F. B., A. M. Mostrom, and A. L. Mack. 1993. Speciation in
North American chickadees. I. Patterns of mtDNA divergence.
Evolution 47:195?212.
Grant, P. R., and B. R. Grant. 1994. Phenotypic and genetic effects
of hybridization in Darwin?s finches. Evolution 48:297?316.
Grant, V. 1981. Plant speciation. Columbia Univ. Press, New York.
Haldane, J. B. S. 1922. Sex ratio and unisexual sterility in animal
hybrids. J. Genet. 12:101?109.
???. 1948. The theory of a cline. J. Genet. 48:277?284.
Harris, H., and D. A. Hopkinson. 1976. Handbook of electrophoresis
in human genetics. North-Holland Publishing, Amsterdam.
Harrison, R. G. 1990. Hybrid zones: windows on the evolutionary
process. Pp. 69?128 in D. Futuyma and J. Antonovics, eds. Ox-
ford surveys in evolutionary biology. Vol. 7. Oxford Univ. Press,
Oxford, U.K.
Hastings, W. K. 1970. Monte Carlo sampling methods using Markov
chains and their applications. Biometrika 57:97?109.
Hewitt, G. M. 1988. Hybrid zones: natural laboratories for evolu-
tionary studies. Trends Ecol. Evol. 3:158?167.
Hughes, J. M. 1982. An explanation for the asymmetrical ?hybrid
zone? between white-backed and black-backed magpies. Emu
82:50?53.
Hunter, R. S. 1975. The measurement of appearance. Wiley and
Sons, New York.
Jaarola, M., H. Tegelstro?m, and K. Fredga. 1997. A contact zone
with noncoincident clines for sex-specific markers in the field
vole (Microtus agrestis). Evolution 51:241?249.
Johnson, N. K., and A. H. Brush. 1972. Analysis of polymorphism
in the sooty-capped bush tanager. Syst. Zool. 21:245?262.
Key, K. H. L. 1968. The concept of stasipatric speciation. Syst.
Zool. 17:14?22.
Kim, S.-C., and L. H. Rieseberg. 1999. Genetic architecture of
species differences in annual sunflowers: implications for adap-
tive trait introgression. Genetics 153:965?977.
Lewontin, R. C., and L. C. Birch. 1966. Hybridization as a source
of new variation for adaptation to new environments. Evolution
20:315?336.
Lill, A. 1974. Sexual behavior of the lek-forming white-bearded
2086 ROBB T. BRUMFIELD ET AL.
APPENDIX
Specimen information for each locality. Asterisks denotes reference specimens (populations 1 and 12) for morphological and/or genetic
measurements. The m superscript denotes a definitively plumaged male skin for which all measurements were taken. The mf superscript denotes
a definitively plumaged male that was prepared as a flat skin. Only a subset of the morphological measurements were taken on these. Most
notably, wing length was not measured. Specimens excluded from the principal components analysis are denoted by the x superscript. The s
superscript denotes an individual for which only a skeleton preparation is available. These individuals were only assayed for the genetic
markers. The f superscript denotes a female bird, and the g superscript denotes a green, first-year male. The b superscript indicates a banded
bird, for which only the restriction-fragment-length-polymorphism markers were assayed. When numbers prefaced with a B appear in brackets,
these correspond to tissue catalog numbers, and indicate that the sample was used in genetic analyses. Specimens only having a B number
represent blood samples from banded birds. All of the reference specimens are definitively plumaged male birds prepared as skins. The mna
superscript indicates definitely plumaged males whose skins were not available for morphological analysis.
Population Specific locality and specimen catalog numbers
1 Costa Rica: prov. Limon; 11 km by road W Guapiles (LSUMNS 138736f [B16157, collected by S. J. Hackett], 138738f
[B16278, collected by J. P. O?Neill], B16282 (collected by SJH), 138739* [B16287, collected by SJH], 138740* [B16297,
collected by SJH], 138741* [B16298, collected by JPO]); Costa Rica: prov. Limon; 15 km NNE Turrialba (LSUMNS
30623*); Costa Rica: prov. Limon; Guapiles (CM 13210*, 13345*,x, 13381*, 13472*); Costa Rica: prov. Limon; Guacimo
(CM13517*, 23552*, LSUMNS 35404*); Costa Rica: prov. Limon; Jimenez, old line RR (CM25286*, 155471*, -2*);
Costa Rica: prov. Cartago; ,1.6 km NE Chitaria (LSUMNS 32524*); Costa Rica: prov. Heredia; Puerto Viejo (MVZ
155647*); Carrillo (CM25453*, 25990*); El Hogar (CM 26758*, -59*, -60*, 26801*, -2*,x, 28026*, -7*); Juan Miras
(CM 11102*).
2
98279N 828409W
Panama: prov. Bocas del Toro; hills on S bank of Rio Sixaola, 25 km W Guabito, (USNM 608191m [B1879], 608192m
[B1877], 608193mna,x [B1878]); Panama: prov. Bocas del Toro, 8 km SW Guabito (USNM 608988m,x [B4590], 608989m
[B4595], 608991f [B3246], 62013f [B3248], B4554, -5, B4557?61, B4567, B4573?77, B4579, B4582).
manakin (Manacus manacus trinitatis Hartert). Z. Tierpsychol.
36:1?36.
MathSoft. 1999. S-plus 2000. MathSoft, Inc., Seattle, WA.
May, R. M., J. A. Endler, and R. E. McMurtrie. 1975. Gene fre-
quency clines in the presence of selection opposed by gene flow.
Am. Nat. 109:659?676.
McDonald, D. B., R. P. Clay, R. T. Brumfield, and M. J. Braun.
2001. Sexual selection on plumage and behavior in an avian
hybrid zone: experimental tests of male-male interactions. Evo-
lution 55:1443?1451.
Metropolis, N., A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller,
and E. Teller. 1953. Equations of state calculations by fast com-
puting machines. J. Chem. Phys. 21:1087?1092.
Moore, W. S. 1977. An evaluation of narrow hybrid zones in ver-
tebrates. Q. Rev. Biol. 52:263?277.
Olson, S. L. 1993. Contributions to avian biogeography from the
archipelago and lowlands of Bocas del Toro, Panama. Auk 110:
100?108.
Parsons, T. J., S. L. Olson, and M. J. Braun. 1993. Unidirectional
spread of secondary sexual plumage traits across an avian hybrid
zone. Science 260:1643?1646.
???. 1994. Hybrid zones and sexual selection: response to Butlin
and Neems. Science 265:122?123.
Pia?lek, J., and N. H. Barton. 1997. The spread of an advantageous
allele across a barrier: the effects of random drift and selection
against heterozygotes. Genetics 145:493?504.
Rao, C. R. 1973. Linear statistical inference and its applications.
John Wiley and Sons, New York.
Raymond, M., and F. Rousset. 1995. GENEPOP (version 3.1c):
population genetics software for exact tests and ecumenicism.
J. Hered. 86:248?249.
Remington, C. L. 1968. Suture-zones of hybrid interaction between
recently joined biotas. Pp. 321?428 in T. Dobzhansky, M. K.
Hecht, and W. C. Steere, eds. Evolutionary biology. Appleton-
Century-Crofts, New York.
Rieseberg, L. H., S. J. E. Baird, and A. M. Desrochers. 1998. Pat-
terns of mating in wild sunflower hybrid zones. Evolution 52:
713?726.
Rieseberg, L. H., J. Whitton, and K. Gardner. 1999. Hybrid zones
and the genetic architecture of a barrier to gene flow between
two sunflower species. Genetics 152:713?727.
Rohwer, S., and C. Wood. 1998. Three hybrid zones between hermit
and townsend?s warblers in Washington and Oregon. Auk 115:
284?310.
Rohwer, S., E. Bermingham, and C. Wood. 2001. Plumage and
mitochondrial DNA haplotype variation across a moving hybrid
zone. Evolution 55:405?422.
SAS Institute. 2000. JMP: statistical discovery software. SAS In-
stitute, Inc., Cary, NC.
Sattler, G. D., and M. J. Braun. 2000. Morphometric variation as
an indicator of genetic interactions between black-capped and
Carolina chickadees at a contact zone in the Appalachian Moun-
tains. Auk 117:427?444.
Schluter, D. 1988. Estimating the form of natural selection on a
quantitative trait. Evolution 42:849?861.
Sheppard, P. M., J. R. G. Turner, K. S. Brown, W. W. Benson, and
M. C. Singer. 1985. Genetics and the evolution of Mullerian
mimicry in Heliconius butterflies. Philos. Trans. R. Soc. Lond.
B 308:433?613.
Slatkin, M. 1973. Gene flow and selection in a cline. Genetics 75:
733?756.
Southern, E. M. 1975. Detection of specific sequences among DNA
fragments separated by gel electrophoresis. J. Mol. Biol. 98:
503?517.
Spratt, B. G., Q.-Y. Zhang, D. M. Jones, A. Hutchison, J. A. Bran-
nigan, and C. G. Dowson. 1989. Recruitment of a penicillin-
binding protein gene from Neisseria flavescens during the emer-
gence of penicillin resistance in Neisseria meningitidis. Proc.
Natl. Acad. Sci. USA 86:8988?8992.
Stebbins, G. L. 1959. The role of hybridization in evolution. Proc.
Am. Philos. Soc. 103:231?251.
Szymura, J., and N. H. Barton. 1986. Genetic analysis of a hybrid
zone between the fire-bellied toads, Bombina bombina and B.
variegata, near Cracow in southern Poland. Evolution 40:
1141?1159.
???. 1991. The genetic structure of the hybrid zone between the
fire-bellied toads Bombina bombina and B. variegata: compari-
sons between transects and between loci. Evolution 45:237?261.
Tucker, P. K., R. D. Sage, J. Warner, A. C. Wilson, and E. M.
Eicher. 1992. Abrupt cline for sex chromosomes in a hybrid
zone between two species of mice. Evolution 46:1146?1163.
Wolfram, S. 1996. The Mathematica book. Cambridge Univ. Press,
Cambridge, U.K.
Woodruff, D. S. 1989. Genetic anomalies associated with Cerion
hybrid zones: the origin and maintenance of new electrophoretic
variants called hybrizymes. Biol. J. Linn. Soc. 36:281?294.
Corresponding Editor: H. L. Gibbs
2087DIVERGENT CLINES IN A MANAKIN HYBRID ZONE
APPENDIX.
Continued.
Population Specific locality and specimen catalog numbers
3
98229N 828349500W
Panama: prov. Bocas del Toro; N bank Rio Teribe 3 km SW Charagre, Quebrada Carbon (USNM 614056mf,x [B1880],
614057mf,x [B1882], 608194m [B1883], 608195m [B1884], 614058mf,x [B1885], 614059mf,x [B1886], 608196g [B1887],
608197m [B1888], 614060mf,x [B1889], 608198m [B1890], 614061mf,x [B1891], 608199m [B1892], 608200f [B1893],
608201m [B1894], 608202mna [B1895], 608203g [B1920], 608204m [B1921], 614062mf,x [B1922], 614063mf,x [B1923],
608205m [B1924]).
4
98249N 828319W
Panama: prov. Bocas del Toro; 3/5 km SE RR bridge E (5 S) bank Changuinola River (USNM 608158m [B1896],
608162m [B1897], 614036mf,x [B1898], 608161m [B1899], 608160m [B1900], 608165g [B1902], 608163m [B1903],
608164mna [B1904], 608169f [B1905], 608166m [B1907], 608159m [B1908], 614037mf,x [B1909], 614038mf,x [B1910],
614039mf,x [B1911], 614040g [B1912], 614041f [B1913], 614042f [B1914], 608170f [B1915], 608167m [B1948], 608168m
[B1949]).
5
98159N 828259W
Panama: prov. Bocas del Toro; 0.5 km N bridge on Rio Oeste (USNM 606941m [B500], 608171m [B1925], 608172m
[B1926], 608173m [B1927], 608174m [B1928], 608175m [B1929], 608176m [B1930], 608177mna [B1931], 608178m
[B1932], 608179m [B1933], 614043mf [B1934], 608180m [B1936], 608181m [B1937], 614044mf [B1938], 608182m
[B1941], 608183m [B1942], 608184m [B1943], 608185m [B1944], 608186m [B1945], 608188f [B1946], 608187g [B1947].
6
98129N 828229W
Panama: prov. Bocas del Toro; Quebrada Pastores (USNM 606937m [B484], 606939m [B485], 606940m [B486], 606936m
[B488], 606938m).
7
98109N 828159W
Panama: prov. Bocas del Toro; Tierra Oscura (USNM 606932f [B411], 612389f [B473], 606920g [B439], 606934f
[B437], 606922m [B444], 606919g [B432], 612388f [B474], 606921m [B445], 606915m [B410], 606916m, 606917g,
606918f , 612392f [B507], 606930m [B565], 606931f , 606933f , 606935f , 606924m [B487], 606925m, 606926m, 606927m,
606928m, 606929m).
8
9879N 828179300W
Panama: prov. Bocas del Toro; 2 km SW confluence Rio Uyama and Quebrada Garza (USNM 608936m [B3173], 608937m
[B3174], 608941m [B3175], 608938m [B3176], 608942m [B3177], 608939m [B3178], 608943m [B3179], 608940m [B3180],
608944m [B3181], 608945m [B3182], 608946m [B3183], 608947m [B3184], 608948g [B3185], 608949m [B3186], 608950m
[B3187], 608951f [B3188], 608952m [B3189], 608953m [B3190], 608954m [B3191], 608955m [B3192], 608956m [B3193],
608957m [B3194]).
9
98029N 828179W
Panama: prov. Bocas del Toro; 1 km W of Palma Real (above Rio Robalo), 50 m (USNM 608958m [B3195], 608959m
[B3196], 608960f [B3197], 608961f [B3198], 608962g [B3199], 608963f [B3200], 608964m [B3201], 608965m [B3202],
608966m [B3203], 608967m [B3204], 608968m [B3205], 608969m [B3206], 608970m [B3207], 608971m [B3209], 608972m
[B3210], 608973m [B3211], 608974m [B3212], 608975mf,x [B3213], 608976m [B3214], 608977m [B3215], 608978m
[B3216], 608979m [B3217], 608980m [B3218], 608981m [B3219]; Panama: prov. Bocas del Toro; 3 km NE Palma Real
(above Rio Robalo), 30 m (608982mf,x [B3220], 608983g [B3221]).
10
88559N 828089W
Panama: prov. Bocas del Toro; 6 km SW Chiriqui Grande (USNM 608139m [B1965], 608138m [B1966], 608149m [B1978],
608148m [B1979], 608140g [B1981], 608145f [B1983], 608143f [B1984], 608144f [B1985], 608147m [B1987], 608155g
[B1988], 608146m [B1989], 608151m [B1991], 608152m [B1992], 608150m [B1993], 608141g [B1994], 608142g [B1995],
608156f [B1996], 608153m [B2037], 608154mna [B2038], 608157g [B2039]).
11
98089N 818549W
Panama: prov. Bocas del Toro; Valiente Peninsula, Playa Verda, NW of Muturi (607823m [B1391], 607824f [B1370],
607825f [B1369]; Panama: prov. Bocas del Toro; Valiente Peninsula, Punta Alegre (USNM 607817f [B1224], 607818f
[B1251]); 607819 [B1257], 607820f [B1263], 613471 [B1284], 607812m [B1299], 607821f [B1311], 607813m [B1330],
613469 [B1331], 607814m [B1332], 607815m [B1334], 613472 [B1355], 607822f [B1359], 613474 [B1371], 613470g
[B1380], 613473 [B1383], 613477 [B1393]).
12 Panama: Canal Zone; Rio Indio, near Gatun (USNM 206829*, 206830*, 206831*, 206832*, 206835*); Panama: Canal
Zone; Gatun (USNM 206836*, 206837*x); Panama: Canal Zone; Lion Hill (USNM 206845*, 206846*, 229485*);
Panama: Canal Zone; Tabernilla (USNM 206848*); Panama: Canal Zone; Frijoles (USNM 229483*); Panama: Canal
Zone; R. R. (Frijole) (USNM 16880*); Panama: Canal Zone; Gamboa (USNM 534054*, 608985*, 608987*); Panama:
Canal Zone; Juan Mina, Rio Chagre (USNM 456205*, 458696*); Panama: Canal Zone; Juan Mina, Quebrada Agua
Clara (USNM 474494*); Panama: Canal Zone; Gamboa, Las Cruces trail (USNM 470154*); Panama: Canal Zone; start
of Pipeline Road, near Gamboa (USNM 608135* [B1860], 608136* [B1865], USNM 614019s [B1857], 614020s [B1858],
614021s [B1859], 614022s [B1861], 614023s [B1862], 614024s [B1863], 614025s [B1864], 614026s [B1866], 614027s
[B1867], 614028s [B1868], 614029s [B1869], 614030s [B1870], 614032s [B1871], 614031s [B1872], 614033s [B1873],
608137mna [B1874], 614034s [B1875], 614035s [sex unknown; B1876]).