A phylogeny of Darwin's ?nches based on
microsatellite DNA length variation
Kenneth Petren
*
, B. Rosemary Grant and Peter R. Grant
Department of Ecology and Evolutionary Biology, Princeton University, Princeton, NJ 08544-1003, USA
Allele length variation at 16 microsatellite loci was used to estimate the phylogeny of 13 out of the 14
species of Darwin's ?nches. The resulting topology was similar to previous phylogenies based on
morphological and allozyme variation. An unexpected result was that genetic divergence among
Gala
?
pagos Island populations of the warbler ?nch (Certhidea olivacea) predates the radiation of all other
Darwin's ?nches. This deep split is surprising in view of the relatively weak morphological di?erentiation
among Certhidea populations and supports the hypothesis that the ancestor of all Darwin's ?nches was
phenotypically similar to Certhidea. The results also resolve a biogeographical problem: the Cocos Island
?nch evolved after the Gala
?
pagos ?nch radiation was under way, supporting the hypothesis that this
distant island was colonized from the Gala
?
pagos Islands. Monophyletic relationships are supported for
both major groups, the ground ?nches (Geospiza) and the tree ?nches (Camarhynchus and Cactospiza),
although the vegetarian ?nch (Platyspiza crassirostris) appears to have diverged prior to the separation of
ground and tree ?nches. These results demonstrate the use of microsatellites for reconstructing
phylogenies of closely related species and interpreting their evolutionary and biogeographic histories.
Keywords: phylogenetic; biogeography; simple sequence repeats; dinucleotide; Cocos; Gala
?
pagos
1. INTRODUCTION
Adaptive radiations are a major source of information
about the evolutionary origins of biological diversity
(Givnish & Sytsma 1997; Grant 1998). Darwin's ?nches
are one of a few classical examples of such radiations
(Lack 1947; Grant 1986; Givnish & Sytsma 1997). Species
in this group show adaptive variation in beak size, beak
shape and body size that is more typical of di?erences
among families of birds (Sushkin 1929), yet the entire
radiation is believed to have occurred in less than three
million years (Grant 1994).While much has been learned
about adaptation and speciation in the group, their
phylogenetic relationships remain poorly known. Lack
(1947) o?ered a phylogenetic reconstruction for the group
based on a non-quantitative comparison of morphological
characteristics (plumage, size and shape, see ?gure 1a; see
also Schluter 1984). Yang & Patton (1981) produced a
phylogeny from allozyme variation among 11 out of the 14
currently recognized species, but support for the tree
?nch was limited and the results di?ered according to the
methodology used for analysis (Stern & Grant 1996;
?gure 1b). Variation in mitochondrial (mt) and nuclear
DNA sequences appears to be insu?cient for resolving
relationships among the more closely related members of
the group (Freeland 1997; Sato et al. 1999a).
We have estimated the evolutionary history of Darwin's
?nches using microsatellite DNA length variation. Micro-
satellites are multilocus genetic markers with high
mutation rates that have been used frequently to test
parentage, assess population di?erentiation and detect
hybridization (MacDonald & Potts 1997). Although it has
been suggested that allele length polymorphism at these
loci may be useful for resolving phylogenetic relationships
(Takezaki & Nei 1996; MacDonald & Potts 1997), few
interspeci?c microsatellite phylogenies have been recon-
structed to date (Pollock et al. 1998; Primmer & Ellegren
1998; but see Roy et al. 1994).
We analysed microsatellite length variation among 13
out of the 14 currently recognized species of Darwin's
?nches including the Cocos Island ?nch (Pinaroloxias). The
missing species from the analysis is the rarest, Cactospiza
heliobates (mangrove ?nch), but it is extremely similar
morphologically to its congener Cactospiza pallida (wood-
pecker ?nch) (Lack 1947; Grant 1986). Two continental
species, Tiaris olivacea (yellow-faced grassquit) and
Sporophila aurita (variable seedeater), are included in the
analysis. Tiaris is among a small group of emberizines
(seedeaters and tanagers) which are believed to be the
closest mainland relatives of Darwin's ?nches (Lack 1947;
Baptista & Trail 1988; Sato et al. 1999b).
2. MATERIALS AND METHODS
(a) Sampling
Blood was collected (Petren et al. 1999) and standard labora-
tory protocols were used for DNA extraction, genomic library
screening and genotype determination (Sambrook et al. 1989;
Primmer et al. 1995). Speci?c protocols and primer sequences
are available elsewhere (Petren 1998). All loci contained a pure
(CA)
4 13
core motif except for two that contained a (GA)
4 13
motif. Eight of these loci have been used to test parentage in
159 Geospiza scandens o?spring (Petren et al. 1999) without
detection of a single `null' allele (Callen et al. 1993). Alleles at
Proc. R. Soc. Lond. B (1999) 266, 321^329 321 & 1999 The Royal Society
Received 28 September 1998 Accepted 9 November 1998
*
Author for correspondence (petren@princeton.edu).
one locus were inherited in a sex-linked (Z-linked) fashion
(Petren et al. 1999).
To estimate genetic distances among taxa, we use Nei's (1972)
unbiased genetic distance (G
ST
). We also present results using
(dm)
2
, a distance measure recently developed for microsatellites
(Goldstein et al. 1995). G
ST
makes no assumption about the
mechanism of mutation whereas (dm)
2
assumes stepwise
mutation.We expect G
ST
to be better than (dm)
2
at shorter time
intervals, while (dm)
2
should perform better than G
ST
at longer
time-scales because it is expected to remain more linear
(Goldstein et al. 1995).
Populations were grouped into species according to current
taxonomic classi?cation, which is based primarily on
morphology (Lack 1947; Grant 1986). Certhidea olivacea occurs on
all of the 17 major islands of the Gala? pagos (Grant 1986). The six
populations analysed here were divided into two groups,
C. olivacea and Certhidea fusca, because the mean G
ST
between
these groups (2.06) was larger than any other distance among
Darwin's ?nch species.The mean G
ST
distances among C. olivacea
(0.62) and C. fusca (0.52) are comparable to distances among
populations of other taxa (below). Following Swarth (1931), we
use the name C. fusca to refer to the populations from the outer
islands, while retaining C. olivacea for the central-island
populations (Santa Cruz and Santiago). Di?erentiation among
populations is the focus of a separate study (Grant et al. 1999).
Excluding individual populations from the analysis resulted in
only minor rearrangements that do not a?ect any of our
conclusions.
Sample sizes and locations for each species are as follows
(abbreviated name, number of populations and mean G
ST
among populations): Geospiza fuliginosa (G. fu., 6, 0.14), Dm-18,
Sc-14, So-9, Ra-10, Es-10, Pi-10; Geospiza fortis (G. fo., 4, 0.30),
Dm-36, Sc-24, Ra-3, So-2; Geospiza magnirostris (G. ma., 3, 0.22),
Dm-14, Sc-4, Ge-19; G. scandens (G. sc., 4, 0.15), Dm-68, Sc-15,
Ra-5, So-4; Geospiza conirostris (G. co., 2, 0.62), Ge-49, Es-23;
Geospiza di?cilis (G. di., 6, 0.69), So-14, Ge-30, Wo-10, Da-12, Pi-
23, Fe-8; C. pallida (C. pl., 1), Sc-16; Platyspiza crassirostris (P. cr., 4,
0.15), Sc-23, So-3, Ma-7, Pi-20; Camarhynchus parvulus (C. pv., 2,
0.03), Sc-11, Fl-22; Camarhynchus psittacula (C. ps., 3, 0.10), Sc-5,
Ma-3, Pi-8; Camarhynchus pauper (C. pp., 1), Fl-19; Pinaroloxias
inornata (P. in., 1), Co-30; C. olivacea (C. ol., 2, 0.62), Sc-13, So-20;
C. fusca (C. fu., 4, 0.52), Ge-13, Es-20, Ma-8, Pi-10. (Locations:
Sc, Santa Cruz; So, Santiago; Dm, Daphne Major; Ge,
Genovesa; Pi, Pinta; Fl, Floreana; Wo, Wolf; Co, Cocos; Da,
Darwin; Fe, Fernandina; Ra, Ra
?
bida; Es, Espa?ola; Ma,
Marchena.) Tiaris olivacea and S. aurita were collected in
Panama. Genetic distances among species are given in table 1.
(b) Phylogenetic analysis
There is no obvious single method of analysis because micro-
satellite loci have higher mutation rates and a fundamentally
di?erent mechanism of mutation than other phylogenetic
markers such as nucleotide sequences or allozymes (Goldstein &
Pollock 1997). Given uncertainty regarding the mechanisms of
mutation, we present phylogenetic reconstructions based on four
methods that make di?erent assumptions: UPGMA (Sokal &
Sneath 1963), the Fitch
^
Margoliash least-squares method (Fitch
& Margoliash 1967), the Fitch
^
Margoliash method with
contemporaneous taxa (KITCH; Felsenstein 1984, 1993) and
maximum likelihood (CONTML; Edwards & Cavalli-Sforza
1964; Felsenstein 1981, 1993).
The FITCH and KITCH methods (with power set to 2.0;
Felsenstein 1993) minimize the weighted squared distances along
branches of the tree. The UPGMA and KITCH methods
assume a molecular clock, so the tips of the tree are constrained
to be contemporaneous (Felsenstein 1984, 1993). CONTML is
based on a model of Brownian motion di?usion which is a
questionable assumption because large mutational jumps are not
uncommon in microsatellites (Primmer & Ellegren 1998).
Therefore we expect CONTML to be less reliable for compari-
sons across longer time-scales. Analyses and bootstraps were
performed with PHYLIP (Felsenstein 1993). The hypotheses we
address do not depend on uncovering a single exact tree
topology. Therefore, we discuss the implications of the consensus
topology among the di?erent methods and then consider the
bootstrap support for alternative topologies that would lead to
di?erent conclusions.
(c) Microsatellite variation
We examined microsatellite length variation for indications of
development bias and homoplasy. Development bias is evident
when microsatellite primers developed in one species fail to
produce a product or show little or no variation in other species
(Ellegren et al. 1995). A decrease in allele size is generally accom-
panied by reduced polymorphism (Primmer et al. 1995): smaller
repeat regions are known to have substantially reduced mutation
rates (Weber & Wong 1993). Reduced mutation rates at longer
genetic distances will lead to non-linear distance estimates for
both G
ST
and (dm)
2
. Among Darwin's ?nches, polymorphism
remains relatively high and allele size declines only slightly as
genetic distance increases from G. fortis, the species used for
microsatellite development (table 2). Allele size and variation
decline more abruptly in the mainland taxa, suggesting that at
this time-scale genetic distances may be non-linear.
Homoplasy obscures phylogenetic signal. It occurs when
characters scored as the same are identical by convergence and
not by common descent. In microsatellites, homoplasy can be
caused by mutational length changes in regions ?anking the
microsatellite repeat (Orti et al. 1997). The low frequency of odd-
sized alleles (table 2) implies minimal homoplasy due to
insertions and deletions in regions ?anking the repeat region
because insertions and deletions should be equally likely to
involve odd and even numbers of bases. In addition, the core
repeat region of alleles sequenced in both G. fortis and C. olivacea
were the expected length for all 11 loci tested to date (K. Petren,
unpublished data).
Homoplasy may also arise because microsatellite repeat
regions are bounded in size (Garza et al. 1995; Goldstein &
Pollock 1997). These size constraints allow only a ?nite number
of character states and, at longer time intervals, homoplasy is
expected to rise and the phylogenetic signal will be obscured as
the signal becomes saturated (Takezaki & Nei 1996; Pollock et
al. 1998). In Darwin's ?nches, the mean allele size range for
each species considered separately (across all loci) is 22.2 bp,
while the mean range for all species combined is over twice this
?gure (46.4 bp). The large di?erence in allele size range suggests
saturation of phylogenetic signal through this type of homoplasy
has not been extensive. The wide range of genetic distances
(table 1) suggests that genetic distances are not likely to be satu-
rated, at least among the more closely related taxa.
3. RESULTS
(a) Tree comparisons
The phylogenetic reconstructions (?gure 2) share a
number of common elements. Every method supports
322 K. Petren and others Amicrosatellite phylogeny of Darwin's ?nches
Proc. R. Soc. Lond. B (1999)
(i) the monophyly of Darwin's ?nches; (ii) the basal
placement of C. olivacea; (iii) the non-sister relationship
between the two Certhidea; (iv) the derivation of the
Cocos Island ?nch (Pinaroloxias) from the Gala
?
pagos
?nches; (v) monophyly of the ground ?nches (Geospiza);
and (vi) monophyly of the combined tree ?nch genera
Camarhynchus and Cactospiza. The FITCH tree topology is
largely congruent with the KITCH topology. This shows
that the conclusions are not dependent upon assuming a
molecular clock. The main di?erence between these two
methods lies in the placement of Platyspiza, which is basal
to the ground ?nch
^
tree ?nch split in all but the FITCH
tree. We refer primarily to the UPGMA tree (?gure 2),
because it captures the elements most commonly observed
among the other methodologies and because this method
consistently revealed the same topology when subsets of
populations and species were analysed.
The microsatellite tree (?gure 3) is in general
agreement with the morphological tree (?gure 1). The
sections of the microsatellite tree that disagree most with
Lack's (1947) tree occur towards the tips of the branches
among the ground ?nches (Geospiza) and the tree ?nches
(Camarhynchus) and in the placement of Platyspiza. The
microsatellite tree also generally agrees with the allozyme
tree and the G
ST
distance matrices are signi?cantly
similar by the Mantel (1967) test (R
M
? 0.66 and
p5 0.005). However, the microsatellite tree provides
higher resolution and greater concordance among
di?erent methods of analysis than the allozyme tree
(?gure 1b; Stern & Grant 1996).
(b) Monophyly of Darwin's ?nches
The microsatellite data support monophyly of Darwin's
?nches not only by showing agreement across all methods
of analysis, but also because the patterns of allelic
variation (table 2) show a marked change in allele size,
variation and ampli?cation ability in Tiaris and
Sporophila. G
ST
does not incorporate information about
allele size or non-amplifying loci which clearly set the
outgroup taxa apart. The (dm)
2
distance provides high
bootstrap support for monophyly (?gure 2e). Similarly,
(dm)
2
better re?ects the pattern of relationships evident in
Amicrosatellite phylogeny of Darwin's ?nches K. Petren and others 323
Proc. R. Soc. Lond. B (1999)
Table 1. Genetic distances among Darwin's ?nches and two mainland relatives based on microsatellite length variation
(Below diagonal G
ST
(Nei 1972; Felsenstein 1993) and above diagonal (dm)
2
(Goldstein et al. 1995).)
G. fu. G. fo. G. ma G. sc. G. co. G. di. C. pl. P. cr.
G. fuliginosa
?
1.4 3.6 4.8 4.5 3.9 10.1 22.7
G. fortis 0.11
?
5.5 3.8 6.9 6.5 13.9 29.4
G. magnirostris 0.25 0.19
?
8.7 5.1 7.4 13.7 23.2
G. scandens 0.35 0.35 0.47
?
12.1 11.7 22.3 33.2
G. conirostris 0.39 0.36 0.35 0.37
?
5.8 5.1 14.6
G. di?cilis 0.24 0.38 0.54 0.56 0.56
?
6.7 13.7
C. pallida 0.57 0.60 0.90 0.56 0.61 0.66
?
12.4
P. crassirostris 0.82 1.02 1.26 1.05 1.06 0.91 0.71
?
C. parvulus 0.50 0.56 0.84 0.65 0.60 0.69 0.18 0.83
C. psittacula 0.59 0.69 1.00 0.71 0.62 0.68 0.22 0.85
C. pauper 0.53 0.56 0.88 0.66 0.59 0.66 0.12 0.80
P. inornata 1.08 1.31 1.64 1.16 1.22 0.88 0.99 1.25
C. fusca 0.96 1.02 1.32 1.14 0.90 0.96 1.08 1.32
C. olivacea 1.32 1.35 1.41 1.50 1.52 1.41 2.25 1.73
T. olivacea 2.47 2.42 3.35 2.79 2.48 2.56 2.45 2.26
S. aurita 1.55 1.61 1.74 1.66 1.74 1.73 1.89 2.78
C. pv. C. ps. C. pp. P. in. C. fu. C. ol. T. ol. S. au.
G. fuliginosa 18.0 23.3 14.1 10.3 10.6 17.5 85.4 131.0
G. fortis 21.8 27.3 16.8 13.1 11.5 20.0 102.3 150.6
G. magnirostris 16.7 23.4 14.0 12.0 15.4 16.8 82.3 131.9
G. scandens 31.7 37.2 26.8 10.4 14.2 31.1 97.7 145.2
G. conirostris 10.5 14.5 7.4 14.4 15.1 12.9 81.6 127.3
G. di?cilis 11.4 13.2 9.0 10.4 8.5 16.1 80.3 122.6
C. pallida 7.4 9.2 4.1 16.2 12.0 14.6 72.8 122.1
P. crassirostris 9.9 8.4 11.9 18.9 24.2 21.4 60.1 90.9
C. parvulus
?
1.6 1.4 18.3 15.0 9.0 69.4 100.8
C. psittacula 0.04
?
3.0 22.4 18.1 12.3 74.4 103.5
C. pauper 0.05 0.09
?
17.6 13.1 10.3 75.9 116.1
P. inornata 1.18 1.17 1.05
?
8.3 25.3 67.6 113.2
C. fusca 0.98 1.13 0.94 1.07
?
21.7 84.3 128.7
C. olivacea 2.17 2.32 2.31 1.80 1.81
?
71.5 110.4
T. olivacea 2.55 2.58 2.37 2.14 2.39 1.81
?
44.1
S. aurita 1.85 2.00 1.91 1.78 3.19 3.00 1.69
?
table 2, as Sporophila appears to be more distantly related
to Darwin's ?nches than Tiaris. Alternative topologies
that would contradict the monophyly hypothesis were not
observed (50.1%) in bootstrap re-samplings when (dm)
2
was used with KITCH.
(c) Divergence of Certhidea
An unexpected result is that populations of Certhidea,
currently classi?ed as a single species, C. olivacea, diverged
well before any other species of Darwin's ?nch arose
(?gure 1c). Regardless of the exact phylogenetic topology,
this deep split among Certhidea populations on central and
peripheral islands, coupled with relatively little diver-
gence in morphology (Lack 1947) and song (Bowman
1983), stands in stark contrast to the great diversity that
evolved among other descendent lineages (?gure 3). A
second result is that Certhidea are not depicted as sister
taxa in any reconstruction. Topologies depicting Certhidea
species as sister taxa were rarely observed in bootstrap
replicates (UPGMA55%, KITCH53%, FITCH51%,
CONTML514% and (dm)
2
/KITCH51%).
(d) The origin of Pinaroloxias
The microsatellite phylogeny resolves the bio-
geographic problem of determining the sequence of
colonization of the Gala
?
pagos and Cocos Islands (Grant
1986). Cocos Island is approximately midway between
mainland Costa Rica (500 km) and the Gala
?
pagos
Islands (630 km), while the Gala? pagos Islands are over
900 km from the coast of Ecuador. There are three main
324 K. Petren and others Amicrosatellite phylogeny of Darwin's ?nches
Proc. R. Soc. Lond. B (1999)
Table 2. Variation at 16 microsatellite loci among Darwin's ?nches and two mainland species
mean values for16 loci
species n
distance (G
ST
)
to G. fortis
expected
heterozygosity
number
of alleles
allele size
range
a
(bp)
G. fortis 65
?
0.74 10.9 27
G. fuliginosa 71 0.11 0.78 12.9 33
G. magnirostris 37 0.19 0.65 7.4 25
G. scandens 92 0.35 0.68 9.9 26
G. conirostris 72 0.36 0.68 8.1 24
G. di?cilis 97 0.38 0.70 10.2 27
C. parvulus 33 0.56 0.52 6.8 22
C. pauper 19 0.56 0.50 5.2 18
C. pallida 16 0.60 0.41 4.2 15
C. psittacula 16 0.69 0.45 4.4 15
P. crassirostris 53 1.02 0.52 6.7 17
C. fusca 51 1.02 0.52 5.7 21
P. inornata 30 1.31 0.47 4.6 14
C. olivacea 33 1.35 0.69 9.8 27
T. olivacea 17 2.42 0.44 4.1 13
S. aurita 8 1.61 0.34 2.7 8
mean values for16 loci
species n
allele size
s.d.
allele size
deviation
b
variable
loci (%)
odd-sized
alleles
c
(%)
G. fortis 65 7.7 0.7 100 5 0.1
G. fuliginosa 71 8.3 70.6 100 0.0
G. magnirostris 37 6.5 70.3 100 0.0
G. scandens 92 7.2 0.3 100 0.0
G. conirostris 72 6.5 70.2 100 0.3
G. di?cilis 97 6.8 72.0 100 5 0.1
C. parvulus 33 5.6 73.0 94 0.2
C. pauper 19 5.1 71.8 94 0.2
C. pallida 16 3.8 72.3 94 3.3
C. psittacula 16 4.2 73.2 94 0.6
P. crassirostris 53 3.5 73.0 94 0.0
C. fusca 51 4.1 73.0 100 0.1
P. inornata 30 3.7 72.5 88 0.0
C. olivacea 33 6.7 72.6 100 0.5
T. olivacea 17 4.0 711.8 75 0.0
S. aurita 8 2.4 717.4 63 1.2
a
Allele size range is calculated as the di?erence between the maximum and minimum allele sizes for each locus.
b
The deviation of a species'mean allele size from the mean allele size of all alleles across all species.
c
Alleles with lengths not multiples of two bp di?erent frommost other alleles at the same locus.
hypotheses. Either the Gala? pagos Islands were colonized
from the mainland, followed by colonization of Cocos
Island from the Gala
?
pagos Islands (Snodgrass 1903),
Cocos Island was colonized ?rst, followed by colonization
of the Gala
?
pagos Islands by emigrants from Cocos Island
(Harris 1973), or Cocos Island and the Gala
?
pagos Islands
were colonized independently from the mainland
(Steadman 1982).
All ?ve methods place Pinaroloxias within the Darwin's
?nch clade, which is consistent with the ?rst hypothesis
and inconsistent with the other two. The Cocos Island
?nch (Pinaroloxias) was derived from the Gala? pagos
Islands' lineage after radiation was underway (?gure 1c).
This ?ts with the geological evidence. Cocos Island
appears to be much younger (ca. 2Myr ago; Castillo et al.
1988) than the Gala
?
pagos Islands (410Myr ago;
Christie et al. 1992) and when formed it was closer to the
Gala
?
pagos Islands. The third hypothesis is highly unlikely
because, given the phylogenetic topology, a complex colo-
nization history would be required. Topologies depicting
Pinaroloxias basal to all ?nch lineages, which would
contradict this interpretation, were not commonly
observed in bootstrap re-samplings (UPGMA53%,
KITCH58%, FITCH57%, CONTML5 28% and
(dm)
2
/KITCH52%).
(e) Patterns of divergence within the tree and ground
?nches
All methods support the monophyletic relationship of
the ground ?nches (Geospiza) and the Camarhynchus and
Cactospiza tree ?nches (?gure 2). All methods except
CONTML support the monophyletic arrangement of
G. magnirostris, G. fortis and G. fuliginosa. Four out of ?ve
methods denote Camarhynchus as monophyletic with boot-
strap values in the range of 90^100% for three methods.
However, the vegetarian tree ?nch (Platyspiza) is placed
outside this clade (basal to the tree and ground ?nches) in
all but the FITCH reconstruction, which, however, has
low bootstrap support for this alternative arrangement.
Bootstrap support for the alternate grouping of Platyspiza
with the tree ?nches ranged from 17^27% among the
other four methods.
4. DISCUSSION
(a) Phylogenetic reconstruction
For the six main conclusions of the analysis, there was
complete agreement across ?ve di?erent methods of
phylogenetic reconstruction despite the fact that they
di?er in their underlying assumptions. The tree-building
methods which assume a molecular clock (UPGMA and
KITCH) revealed much higher bootstrap support
(re?ecting concordance among loci) than the FITCH
method, which makes no such assumption. A recent
simulation study revealed that UPGMA performs better
when evolutionary rates are high (Huelsenbeck &
Kirkpatrick 1996) and evolutionary rates are expected to
be high with microsatellites. The UPGMA method gives
greater weight to the genetic relationships among more
closely related taxa (Sokal & Sneath 1963). This may be
appropriate for microsatellite analysis because G
ST
is
expected to vary more linearly with shorter time-scales
(Goldstein et al. 1995; table 2). As expected, (dm)
2
provided strong support for conclusions involving deeper
nodes of the tree. Although (dm)
2
may be better with
these relatively longer time-scales, the use of this
measure may be limited because of nonlinearities
associated with changes in the mutation rate. The
CONTML method showed less support for the deeper
nodes; however, this maximum-likelihood method
unrealistically assumes that mutation is absent
(Felsenstein 1981, 1993).
Hybridization will hinder the recovery of true phylo-
genetic relationships if introgression occurs frequently
among taxa (Grant & Grant 1992; Avise 1994). This is a
problem that is not unique to microsatellites. We expect
that hybridizing species will tend to cluster together more
closely than if they did not hybridize (Grant 1986). Some
evidence of rare hybridization has been recorded among
most species pairs within the ground ?nch clade, within
the tree ?nch clade and between the warbler and tree
?nches (Grant 1986). However, in the best-studied case of
introgression in Darwin's ?nches, between G. fortis and
G. scandens on the island of Daphne Major (Grant 1993;
Grant & Grant 1994), the genetic a?nity of these
populations is still much closer to other populations of the
same species (K. Petren, B. R. Grant and P. R. Grant,
unpublished data) and these taxa do not cluster together
phylogenetically (?gure 3). Pinaroloxias on the isolated
Amicrosatellite phylogeny of Darwin's ?nches K. Petren and others 325
Proc. R. Soc. Lond. B (1999)
Figure 1. Phylogenetic hypotheses for Darwin's ?nches.
(a) The phylogenetic topology proposed by Lack (1947)
based on morphological characteristics. (b) The phylogenetic
topology based on allozyme variation (Yang & Patton 1980;
Stern & Grant 1996) using Nei's (1972) distance (G
ST
) and
UPGMA (Sokal & Sneath 1963). Numbers indicate the
percentage of bootstrap replicates (4 50%) that supported
the node.
326 K. Petren and others Amicrosatellite phylogeny of Darwin's ?nches
Proc. R. Soc. Lond. B (1999)
G
G
G
??
G. fu.
G. fo.
G. ma.
G. sc.
G. co.
G. di.
C. pv.
C. ps.
C. pp.
C. pl.
P. cr.
C. fu.
P. in.
C. ol.
T. ol.
S. au.
G. fu.
G. fo.
G. ma.
G. di.
G. co.
G. sc.
C. pv.
C. ps.
C. pp.
C. pl.
P. cr.
C. fu.
P. in.
C. ol.
S. au.
T. ol.
G. ma.
G. fo.
G. fu.
G. di.
G. sc.
G. co.
C. ps.
C. pv.
C. pa.
C. pa.
P. cr.
C. fu.
P. in.
C. ol.
T. ol.
G. co.
G. ma.
G. sc.
G. fu.
G. fo.
G. di.
C. ps.
C. pv.
C. pp.
C. pl.
P. cr.
C. fu.
P. in.
C. ol.
T. ol.
S. au. G. fu.
G. fo.
G. ma.
G. sc.
G. di.
G. co.
C. pv.
C. pp.
C. ps.
C. pl.
C. fu.
P. in.
P. cr.
C. ol.
T. ol.
S. au.
Figure 2. (a
^
e) Phylogenetic
reconstructions using microsatellite
length variation (see ? 2 for
descriptions of techniques). Primer
pairs developed in G. fortis (n? 16;
Petren 1998; Petren et al. 1999)
were used to obtain more than
11 000 genotypes from 710 birds.
Branch lengths are proportional
to genetic distance as indicated by
the scales. Numbers indicate the
percentage of 1000 bootstrap
replicates (4 50%) that supported
the node.
Cocos Island is immune from problems arising from
hybridization.
(b) Evolution in Darwin's ?nches
This is the ?rst molecular (or biochemical) study to
support the monophyletic classi?cation of Darwin's
?nches, as well as the placement of Pinaroloxias within the
group. These results agree with other studies that clearly
place Darwin's ?nches in a monophyletic group based on
morphology, plumage, song and other characteristics
(Bowman 1961, 1983; Grant 1986). Molecular sequence
and microsatellite analysis of other mainland taxa not
included here are consistent with these results (Freeland
1997; Sato et al. 1999a; K. Petren, unpublished data).
There are indications of the deep split among Certhidea
populations from allozyme data (Polans 1983) as well as
recent mtDNA sequence analysis (Freeland 1997).
However, until now, the division of Certhidea has not been
placed in a phylogenetic context. Lack (1947, 1961)
argued for a Geospiza-like ancestor of all Darwin's ?nches
(but see Swarth 1931). If the non-sister relationship
among Certhidea is correct and these lineages represent
two independent branchings from the main lineage, the
argument for a more Certhidea-like ancestor to all of
Darwin's ?nches is strengthened. We cannot reject Lack's
(1947, 1961) view that the ancestor of all Darwin's ?nches
possessed Geospiza-like traits such as black plumage and a
blunt beak. Yet if this was the case then the two Certhidea
lineages would represent a remarkable case of conver-
gence in morphology, behaviour, plumage and song.
The Certhidea results suggest that, in some instances,
morphology may be a poor guide to the genetic distinct-
ness of populations. This is of particular relevance to
management strategies. Similar ?ndings have emerged
from molecular studies of birds (Avise & Nelson 1989)
and reptiles (Daugherty et al. 1990), but, to our knowl-
edge, discovery of an unsuspected divergence occurring at
the base of an adaptive radiation has not been previously
reported. There are no Certhidea populations currently in
danger of extinction, but, if they become threatened in
the future, more than one will deserve protection by
virtue of their genetic distinctiveness. This emphasizes the
importance of verifying the genetic ancestry of not only
threatened species, but also threatened populations.
Many phenotypic traits of Darwin's ?nches, such as
beak size and shape, body size and plumage coloration,
have been studied extensively (Lack 1947; Bowman 1961,
1983; Grant 1986). Reconstructing their evolution is not
straightforward and is not attempted here because many
of these traits (particularly beak shape) are subject to
strong ecological and genetic constraints (Grant & Grant
1999). Furthermore, non-parsimonious evolutionary
reconstructions are biologically plausible since evolution
can proceed very rapidly in this system (Grant & Grant
1995). However, given the improved phylogenetic resolu-
tion provided by microsatellites (?gure 3) we o?er two
relevant observations.
First, with the exception of Platyspiza, we note that
species that root basally on the tree and the basal
members of the tree and ground ?nches have relatively
long, pointed beaks. This beak form is generally
associated with an insectivorous diet (Bowman 1961). Two
novel blunt-beaked forms evolved later in the ground
Amicrosatellite phylogeny of Darwin's ?nches K. Petren and others 327
Proc. R. Soc. Lond. B (1999)
93
93
83
79
0.1
71
66
75
70
60
53
98
92
63
63
100
100
77
57
87
81
G. fuliginosa
G. fortis
G. magnirostris
G. scandens
G. conirostris
G. difficilis
C. parvulus
C. psittacula
C. pauper
C. pallida
P. crassirostris
C. fusca
P. inornata
C. olivacea
10 g
13 g
8 g
34 g
21 g
18 g
20 g
13 g
20 g
28 g
21 g
35 g
20 g
14 g
Figure 3. A phylogram of Darwin's ?nches based onmicrosatellite
length variation constructed usingG
ST
andUPGMA. Photographs
of birds are proportional to actual size. Themaximum amount of
black colouring inmale plumage and themean bodymass among
populations is indicated for each species (Grant 1986). Horizontal
branch lengths are proportional to units of genetic distance (G
ST
)
as indicated by the scale. Numbers indicate the percentage of 1000
bootstrap replicates (4 50%) that supported the node (UPGMA
method above andKITCHmethod below). Names of genera are
given in full in table 1. On the basis of these results, reclassi?cation
is justi?able but has not been done.
?nches (G. magnirostris, G. fortis and G. fuliginosa), and in
the tree ?nches (Camarhynchus). The ground ?nch beak is
e?cient for crushing seeds at the base, while the tree
?nch beak permits greater biting strength at the tip for
tearing vegetation (Bowman 1961). Once a novel beak
shape evolved in these two groups, beak shape remained
relatively constant while body size and beak size changed.
This supports the view that evolution most easily
proceeds along lines of allometry (Grant & Grant 1995):
once a novel shape is formed away from the line of
allometry, a rapid divergence in size along a new line
of allometry is possible.
Second, the microsatellite phylogeny generally supports
Lack's (1947) view that plumage is a relatively conserved
trait: species with similar plumage patterns generally
cluster together (?gure 3). Yet the tree also suggests that
multiple evolutionary transitions have taken place in
plumage coloration, as Geospiza and Pinaroloxias males
share black plumage, while Certhidea and Cactospiza males
share similar green
^
grey plumage. It is likely that one of
these instances of a shared trait is the result of conver-
gence. This is perhaps not surprising because melanism in
other birds is controlled by a small number of loci
(Buckley 1987).
In conclusion, we suggest that microsatellites may be
useful for reconstructing the evolutionary history of other
groups of organisms which, like Darwin's ?nches, have
radiated relatively recently and rapidly. Microsatellite
studies could complement DNA sequence and allozyme
studies by providing resolution at shorter time-scales.
Microsatellite variation is expected to be most useful
when divergence times are short and when populations
are small (Nauta & Weissing 1996; Takezaki & Nei 1996),
as is the case with Darwin's ?nches. All 16 loci used in
this study were simple (pure) dinucleotide repeats, which
show greater promise for phylogenetic reconstruction
than other microsatellite motifs (Primmer & Ellegren
1998). Nevertheless, in view of continuing uncertainty
over mutation mechanisms, homoplasy and the best
methods of analysis for microsatellite data, there is a need
for further theoretical and empirical investigation of their
use in estimating phylogenies.
We thank the Gala
?
pagos and Costa Rica National Parks
Services and the Charles Darwin Research Station (Gala? pagos
Islands), the sta? on Cocos Island and the STRI for administra-
tive and logistical support, the National Science Foundation
(USA) and the NSF
^
Sloan Foundation for ?nancial support,
P. T. Boag, B. Fessl, G. Seutin, S. Tebbich, and many assistants
for help in the Gala
?
pagos Islands, M. Wikelski and M. Hau for
mainland samples, E. K. Monson, V. Schulz and V. A. Zakian
for laboratory advice, and L. F. Keller, T. D. Price, J. T. Bonner,
T. J. Case and ?ve anonymous reviewers for comments on ear-
lier drafts.
REFERENCES
Avise, J. 1994 Molecular markers, natural history and evolution. New
York: Chapman & Hall.
Avise, J. C. & Nelson,W. S. 1989 Molecular genetic relationships
of the extinct dusky seaside sparrow. Science 243, 646^648.
Baptista, L. F. & Trail, P. W. 1988 On the origin of Darwin's
?nches. Auk 105, 663^671.
Bowman, R. I. 1961 Morphological di?erentiation and
adaptation in the Gala? pagos ?nches. Univ. Calif. Publ. Zool. 58,
1^302.
Bowman, R. I. 1983 The evolution of song in Darwin's ?nches.
In Patterns of evolution in Gala
?
pagos organisms (ed. R. I. Bowman,
M. Berson & A. E. Leviton), pp. 237^537. San Francisco:
American Association for the Advancement of Sciences.
Buckley, P. A. (ed.) 1987 Mendelian genes. In Avian genetics (ed. F.
Cooke & P. A. Buckley), pp. 1^44. London: Academic Press.
Callen, D. F., Thompson, A. D. & Shen, Y. 1993 Incidence and
origin of `null' alleles in the (AC)
n
microsatellite markers. Am.
J. Hum. Genet. 52, 922^927.
Castillo, P., Batiza, R., Vanko, D., Malvassi, E., Basqueso, J. &
Fernandez, E. 1988 Anomalously young volcanoes on old hot-
spot traces. I. Geology and petrology of Cocos Island. Geol.
Soc. Am. Bull. 100, 1400^1414.
Christie, D. M., Duncan, R. A., McBirney, A. R., Richards,
M. A., White, W. M., Harpp, K. S. & Fosc, C. G. 1992
Drowned islands downstream from the Gala? pagos hotspot
imply extended speciation times. Nature 355, 246^248.
Daugherty, C. H., Cree, A. H., Hay, J. M. & Thompson, M. B.
1990 Neglected taxonomy and continuing extinctions of
Tuatara sphenodon. Nature 347, 177^179.
Edwards, A. F. W. & Cavalli-Sforza, L. L. 1964 Reconstruction
of evolutionary trees. In Phenetic and phylogenetic classi?cation
(ed. V. H. Heywood & J. McNeill), pp. 67^76. London:
Systematics Association.
Ellegren, H., Primmer, C. R. & Sheldon, B. C. 1995
Microsatellite e`volution': directionality or bias? Nat. Genet. 11,
360^362.
Felsenstein, J. 1981 Evolutionary trees from gene frequencies and
quantitative characters: ?nding maximum likelihood
estimates. Evolution 35, 1229^1242.
Felsenstein, J. 1984 Distance methods for inferring phylogenies:
a justi?cation. Evolution 38, 16^24.
Felsenstein, J. 1993 PHYLIP ( phylogeny inference package), version
3.5c. Zoology Department, University of Washington, Seattle.
Fitch, W. M. & Margoliash, E. 1967 Construction of phylo-
genetic trees. Science 155, 279^284.
Freeland, J. R. 1997 The genetic evolutionary history of the
Darwin's ?nches (Aves: Geospizinae). PhD thesis, Queen's
University, Kingston.
Garza, J. C., Slatkin, M. & Freimer, N. B. 1995 Microsatellite
allele frequencies in humans and chimpanzees, with
implications for constraints on allele size. Mol. Biol. Evol. 12,
594^603.
Givnish, T. J. & Sytsma, K. J. (ed.) 1997 Molecular evolution and
adaptive radiation. Cambridge University Press.
Goldstein, D. B. & Pollock, D. D. 1997 Launching micro-
satellites: a review of mutation processes and methods of
phylogenetic inference. J. Hered. 88, 335^342.
Goldstein, D. B., Linares, A. R., Cavalli-Sforza, L. L. &
Feldman, M. W. 1995 An evaluation of genetic distances for
use with microsatellite loci. Genetics 139, 463^471.
Grant, P. R. 1986 Ecology and evolution of Darwin's ?nches.
Princeton University Press.
Grant, P. R. 1993 Hybridization of Darwin's ?nches on Isla
Daphne Major, Gala? pagos. Phil. Trans. R. Soc. Lond. B 340,
127^139.
Grant, P. R. 1994 Population variation and hybridization:
comparison of ?nches from two archipelagos. Evol. Ecol. 8,
598^617.
Grant, P. R. (ed.) 1998 Evolution on islands. Oxford University
Press.
Grant, P. R. & Grant, B. R. 1992 Hybridization of bird species.
Science 256, 193^197.
Grant, P. R. & Grant, B. R. 1994 Phenotypic and genetic e?ects
of hybridization in Darwin's ?nches. Evolution 48, 297^316.
328 K. Petren and others Amicrosatellite phylogeny of Darwin's ?nches
Proc. R. Soc. Lond. B (1999)
Grant, P. R. & Grant, B. R. 1995 Predicting microevolutionary
responses to directional selection on heritable variation.
Evolution 49, 241^251.
Grant, P. R. & Grant, B. R. 1999 Quantitative genetic variation
in populations of Darwin's ?nches. In Adaptive variation in the
wild (ed. T. A. Mousseau, B. L. Sinervo & J. Endler). Oxford
University Press. (In the press.)
Grant, P. R., Grant, B. R. & Petren, K. 1999 The allopatric
phase of speciation: the sharp-beaked ground ?nch (Geospiza
di?cilis) on the Gala
?
pagos Islands. (In preparation.)
Harris, M. P. 1973 Coereba ?aveola and the Geospizinae. Bull. Br.
Ornithol. Club 92, 164^168.
Huelsenbeck, J. P. & Kirkpatrick, M. 1996 Do phylogenetic
methods produce trees with biased shapes? Evolution 50,
1418^1424.
Lack, D. 1947 Darwin's ?nches. Cambridge University Press.
Lack, D. 1961 Darwin's ?nches, 2nd edn. Cambridge University
Press.
McDonald, D. B. & Potts, W. K. 1997 DNA microsatellites as
genetic markers at several scales. In Avian molecular evolution
and systematics (ed. D. P. Mindell), pp. 29^49. New York:
Academic Press.
Mantel, N. 1967 The detection of disease clustering and a gener-
alized regression approach. Cancer Res. 27, 209^220.
Minch, E., Ruiz-Linares, A., Goldstein, D. B., Feldman, M. W.
& Cavalli-Sforza, L. L. 1996 Microsat program, version 1.5e.
Biology Department, University of Stanford.
Nauta, M. J. & Weissing, F. J. 1996 Constraints on allele size at
microsatellite loci: implications for genetic di?erentiation.
Genetics 143, 1021^1032.
Nei, M. 1972 Genetic distance between populations. Am. Nat.
106, 283^292.
Orti, G., Pearse, D. E. & Avise, J. 1997 Phylogenetic assessment
of length variation at a microsatellite locus. Proc. Natl Acad.
Sci. USA 94, 10745^10749.
Petren, K. 1998 Microsatellite primers from Geospiza fortis and
cross-species ampli?cation in Darwin's ?nches. Mol. Ecol. 7,
1782^1784.
Petren, K., Grant, B. R. & Grant, P. R. 1999 Low extra-pair
paternity in the cactus ?nch (Geospiza scandens). Auk 116,
252^256.
Polans, N. O. 1983 Enzyme polymorphisms in Gala
?
pagos
?nches. In Patterns of evolution in Galapa
?
gos organisms (ed. R. I.
Bowman, M. Berson & A. E. Leviton), pp. 219^236. San
Francisco: American Association for the Advancement of
Sciences.
Pollock, D. D., Bergman, A., Feldman, M. W. & Goldstein,
D. B. 1998 Microsatellite behavior with range constraints:
parameter estimation and improved distances for use in
phylogenetic reconstruction.Theor. Popul. Biol. 53, 256^271.
Primmer, C. R. & Ellegren, H. 1998 Patterns of molecular evolu-
tion in avian microsatellites. Mol. Biol. Evol. 15, 997^1008.
Primmer, R., M?ller, A. P. & Ellegren, H. 1995 Resolving
genetic relationships with microsatellite markers: a parentage
testing system for the swallow Hirundo rustica. Mol. Ecol. 4,
493^498.
Roy, M. S., Ge?en, E., Smith, D., Ostrander, E. A. & Wayne,
R. K. 1994 Patterns of di?erentiation and hybridization in
North American wol?ike canids, revealed by analysis of
microsatellite loci. Mol. Biol. Evol. 11, 553^570.
Sambrook, J., Fritsch, E. F. & Maniatis, T. 1989 Molecular
cloning: a laboratory manual, 2nd edn. New York: Cold Spring
Harbor Laboratory Press.
Sato, A., O'Huigin, C., Figueroa, F., Grant, P. R., Grant, B. R.,
Tichy, H. & Klein, J. 1999a Phylogeny of Darwin's ?nches as
revealed by mitochondrial DNA sequences. (In preparation.)
Sato, A., Tichy, H., O'Huigin, C., Grant, P. R., Grant, B. R. &
Klein, J. 1999b On the origin of Darwin's ?nches. Proc. Natl
Acad. Sci. USA. (Submitted.)
Schluter, D. 1984 Morphological and phylogenetic relations
among the Darwin's ?nches. Evolution 38, 921^930.
Snodgrass, R. E. 1903 Notes on the anatomy of Geospiza,
Cocornis, and Certhidea. Auk 20, 402^417.
Sokal, R. R. & Sneath, P. H. 1963 Principles of numerical taxonomy.
San Francisco: Freeman.
Steadman, D. 1982 The origin of Darwin's ?nches (Fringil-
lidae, Passeriformes). Trans. San Diego Soc. Nat. Hist. 19,
279^296.
Stern, D. L. & Grant, P. R. 1996 A phylogenetic reanalysis of
allozyme variation among populations of Gala
?
pagos ?nches.
Zool. J. Linn. Soc. 118, 119^134.
Sushkin, P. P. 1929 On some peculiarities of adaptive radiation
presented by insular faunae. In Proceedings of the Sixth
International Ornith. Congress, Kopenhagen. pp. 375^378.
Swarth, H. S. 1931 The avifauna of the Gala? pagos Islands. Occ.
Pap. Calif. Acad. Sci. 18, 1^299.
Takezaki, N. & Nei, M. 1996 Genetic distances and recon-
struction of phylogenetic trees from microsatellite DNA.
Genetics 144, 389^399.
Weber, J. & Wong, C. 1993 Mutation of human short tandem
repeats. Hum. Mol. Genet. 2, 1123^1128.
Yang, S. Y. & Patton, J. L. 1981 Genetic variability and di?eren-
tiation of Gala
?
pagos ?nches. Auk 98, 230^242.
As this paper exceeds the maximum length normally permitted,
the authors have agreed to contribute to production costs.
Amicrosatellite phylogeny of Darwin's ?nches K. Petren and others 329
Proc. R. Soc. Lond. B (1999)