全 文 ：Journal of Systematics and Evolution 46 (3): 349–374 (2008) doi: 10.3724/SP.J.1002.2008.08056
(formerly Acta Phytotaxonomica Sinica) http://www.plantsystematics.com
Tracking character evolution and biogeographic history through time
in Cornaceae—Does choice of methods matter?
Qiu-Yun (Jenny) XIANG* David T. THOMAS
(Department of Plant Biology, North Carolina State University, Raleigh, NC 27695, USA)
Abstract This study compares results on reconstructing the ancestral state of characters and ancestral areas of
distribution in Cornaceae to gain insights into the impact of using different analytical methods. Ancestral charac-
ter state reconstructions were compared among three methods (parsimony, maximum likelihood, and stochastic
character mapping) using MESQUITE and a full Bayesian method in BAYESTRAITS and inferences of ancestral area
distribution were compared between the parsimony-based dispersal-vicariance analysis (DIVA) and a newly
developed maximum likelihood (ML) method. Results indicated that among the six inflorescence and fruit char-
acters examined, “perfect” binary characters (no homoplasy, no polymorphism within terminals, and no missing
data) are little affected by choice of method, while homoplasious characters and missing data are sensitive to
methods used. Ancestral areas at deep nodes of the phylogeny are substantially different between DIVA and ML
and strikingly different between analyses including and excluding fossils at three deepest nodes. These results,
while raising caution in making conclusions on trait evolution and historical biogeography using conventional
methods, demonstrate a limitation in our current understanding of character evolution and biogeography. The
biogeographic history favored by the ML analyses including fossils suggested the origin and early radiation of
Cornus likely occurred in the late Cretaceous and earliest Tertiary in Europe and intercontinental disjunctions in
three lineages involved movements across the North Atlantic Land Bridge (BLB) in the early and mid Tertiary.
This result is congruent with the role of NALB for post-Eocene migration and in connecting tropical floras in
North America and Africa, and in eastern Asia and South America. However, alternative hypotheses with an
origin in eastern Asia and early Trans-Beringia migrations of the genus cannot be ruled out.
Key words AReA, BAYESTRAITS, BEAST, biogeography, chromosome evolution, Cornaceae, fruit and inflores-
cence evolution, LAGRANGE, MESQUITE, divergence time.
Reconstructing ancestral states is a powerful
method to understand the pathways and pattern of
character evolution, which holds the key to our under-
standing of evolutionary processes (Cunningham,
1999; Maddison & Maddison, 1992, 2001; Ronquist,
2004). During the past decade, character mapping on a
phylogeny has become more and more a common
component in research articles to test evolutionary
hypotheses, as well as to understand evolution of
organism traits and behaviors, and more recently of
genes, genomes, and gene functions (Blanchette et al.,
2004; Ouzounis, 2005; e.g., Rossnes et al., 2005;
Renner et al., 2007; Kondo & Omland, 2007; Zhang et
al., 2008; Ekman et al., 2008; also see Ronquist,
2004). A number of analytical methods have been
developed for reconstructing ancestral states of char-
acters, which may be divided into four major catego-
ries: parsimony (MP) (Maddison & Maddison, 1992,
2001, 2007; Rossnes et al., 2005), maximum likeli-
hood (ML) (Harvey & Pagel, 1991; Schluter et al.,
1997; Pagel, 1999), Bayesian inference (Huelsenbeck
et al., 2000, 2001; Ronquist 2004; Pagel et al., 2004),
and stochastic character mapping (SCM) (Nielsen,
2002; Huelsenbeck et al., 2003). Until approximately
five years ago, the easy use of the parsimony method
implemented in MacClade 3.0 and 4.03 (Maddison &
Maddison, 1992, 2001) had made it the most useful
approach for tracking evolutionary history of mor-
phological characters (see Huelsenbeck et al., 2003;
Ronquist, 2004). The recent development of other
methods and their automation in computer packages
provided alternative methods that are more robust
and realistic for users to study character evolution and
overcome drawbacks of the parsimony method [e.g.,
ML and stochastic character mapping in MESQUITE
2.01 (Maddison & Maddison, 2007), the ML method
in BAYESTRAITS 1.0 (Pagel & Meade, 2006), the
Bayesian methods in BAYESTRAITS 1.0 (Pagel &
Meade, 2006) and SIMMAP (Bollback, 2006)]. Both
ML and SCM in MESQUITE 2.01 apply stochastic
models of character state change and can explicitly
accommodate uncertainty in ancestral states. The

———————————
Received: 17 April 2008 Accepted: 4 May 2008
* Author for correspondence. E-mail: jenny_xiang@ncsu.edu;
Tel.: 919-515-2728; Fax: 919-515-3436.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 350
Bayesian approach in BAYESTRAITS and SIMMAP
estimates the instantaneous rates of character change
using maximum likelihood and accommodates phy-
logenetic uncertainty by evaluating the ancestral
character state on trees sampled from the posterior
distribution. In previous practice, most authors applied
one or the other method in their studies, and few used
multiple approaches to explore the differences of
results derived from different methods. Recently,
Pedersen et al. (2007) employed maximum parsimony
in MacClade 4.03 (Maddison & Maddison, 2001) and
Bayesian construction with BAYESTRAITS 1.0 (Pagel
& Meade, 2006) in the moss family Bryaceae and
found largely congruent results. In contrast, Ekman et
al. (2008) found striking differences among parsi-
mony, maximum likelihood, stochastic character
mapping, and Bayesian reconstructions in their study
of the fungi group Ascus (Lecanorales). It is unclear to
what extent the congruence or incongruence is af-
fected by phylogenetic uncertainty and the nature of
the characters under consideration. Explicit knowl-
edge of how reconstructions change under different
methods (and how this depends on the type of charac-
ter) would be useful in guiding the choice of methods
for a study. It is also particularly important to
re-evaluate our previous understanding of character
evolution based on a single method, especially using
parsimony alone. Inconsistent results from different
methods would suggest that many previous studies on
character evolution based on a single method may
need to be reevaluated. This is especially relevant to
studies published before the newer methods (e.g., ML,
Bayesian, and SCM) became available. In the present
study, we used BAYESTRAITS v 1.0 (Pagel &
Meade, 2006) to reconstruct evolutionary histories of
six key discrete morphological traits in Cornaceae and
compared the results with those from MP, ML, and
SCM implemented in MESQUITE 2.01(Maddison &
Maddison, 2007) to gain more insight into the sensi-
tivity of “character mapping” to the choice of meth-
ods.
Inferring the biogeographic histories (e.g., origin,
persistence, dispersal, and extinction) of lineages is
analogous to reconstruction of character evolutionary
histories in many ways (but see below) and is funda-
mental to understanding the origin and evolution of
the modern distribution of biodiversity. However,
comparing to the study of character histories, there are
fewer quantitative statistical methods available for
historical biogeography. The present practice of
tracking biogeographic histories on phylogeny is, in
this regard, facing more challenges than studying
character evolution. During the past decade, the
parsimony-based dispersal-vicariance analysis (DIVA;
Ronquist, 1996, 1997) has been the most widely used
approach in phylogenetic biogeography (e.g., Xiang &
Soltis, 2001; Donoghue & Smith, 2004; Sanmartín &
Ronquist, 2004; Xiang et al., 2006; Hines, 2008) due
to many advantages of the method, e.g., permitting
inference of explicit vicariant and dispersal events,
their relative timing, and dispersal directions, as well
as the availability of fast and user-friendly software
(DIVA 1.1, Ronquist, 1997). However, the method
inherits the intrinsic principle of parsimony, and
therefore it may underestimate character transforma-
tion events (i.e., dispersal). Although some studies
applied the character model of likelihood methods for
biogeographic studies to estimate the likelihood of
ancestral areas (Nepokroeff et al., 2003), the models
of character evolution are not appropriate for geo-
graphic ranges (see Ree & Smith, 2008). As a re-
sponse to the need of more and better analytical
biogeographic methods, Ree and colleagues recently
developed a likelihood-based approach for inference
of ancestral geographic areas (Ree et al., 2005; Ree &
Smith, 2008). The newly developed method represents
a significant advance in biogeographic methodology
by implementing a maximum likelihood statistical
model that includes information from biological and
abiotic factors in calculating the likelihood of bio-
geographic pathways for a given phylogenetic tree and
the distributions of taxa. For example, the rate of
dispersal and local extinction, lineage surviving time,
and the probabilities of dispersal between geographic
ranges at different geological times (Ree et al., 2005;
Moore et al., 2006) can all be used in the estimation.
Their methods model geographic range evolution by
stochastic dispersal, local extinction, and speciation in
a set of discrete areas in continuous time. Unlike the
character models, a single taxon is allowed to have
more than one character state (e.g., a wide distribution
in two or more areas), as a result of direct dispersal
(Ree et al., 2005). A recent version of the method of
Ree and Smith (2008) uses instantaneous rates of
dispersal and local extinction to significantly improve
the computation time for parameter estimation previ-
ously done using a simulation approach in Ree et al.
(2005). This makes it feasible for ancestral area
reconstruction using ML optimization analogous to
character mapping. In the present study, we apply the
methods of Ree et al. (2005) and Ree & Smith (2008)
to Cornaceae and compare the results from those
derived from DIVA. Knowledge of congruences and
discordance between the results from DIVA and the
XIANG & THOMAS: Evolution & biogeography of Cornaceae

351
ML analysis is particularly useful for evaluating our
previous understanding of lineage biogeographic
histories and synthesis on global biogeography based
on DIVA (Donoghue & Smith, 2004; Sanmartín et al.,
2001; Sanmartín & Ronquist, 2004).
The dogwood family Cornaceae is the key mem-
ber of Cornales, a lineage occupying a special position
on the tree of life of angiosperms for being basal in
the Asteridae clade (APG, 2003; Soltis et al., 2005).
The family has been circumscribed differently among
authors (Eyde, 1988; Takhtajan, 1987; see Xiang et
al., 1993; APG, 2003; Fan & Xiang, 2003; Soltis et
al., 2005). Here we take the narrow concept proposed
by Takhtajan (1987) that Cornaceae contains only
Cornus L. s. l., based on the following reasons. First,
the closest genus to Cornus is Alangium Craib,
based on results of molecular phylogenetic studies
(Chase et al., 1993; Xiang et al., 1998, 2002; Fan &
Xiang, 2003). Second, Alangium has long been recog-
nized as a family, Alangiaceae DC. (de Candolle,
1828). Third, the name of Alangiaceae has been
proposed to be conserved (Bullock, 1959; APG, 2003).
Cornaceae s.s. consists of approximately 50–55
species that are morphologically diverse. Species of
the genus exhibit striking variation in inflorescence
architecture and display, fruit type, structure, and
color, among other attributes (see Xiang et al., 2006).
As a consequence of the morphological heterogeneity,
the species have been grouped in various ways, in
multiple genera, subgenera, or sections during the
past century (see Eyde, 1988; Murrell, 1993, 1996;
Xiang et al., 1996; Fan & Xiang, 2001; Xiang &
Boufford, 2005). Recent molecular phylogenetic
studies reconstructed a well resolved, robust species
phylogeny for Cornus and resolved the species into
four major clades (Xiang et al., 2006; Xiang et al., in
review, available upon request). These are: (1) the
blue-or-white fruited group (BW); (2) the cornelian
cherries (CC), (3) the big-bracted group (BB), and (4)
the dwarf dogwoods (DW). Relationships within and
among each group are also well resolved, and mostly
strongly supported (Xiang et al., 2006; Xiang et al., in
review). This provides a solid basis to track the evolu-
tionary histories of morphological characters in the
genus and help to understand the tempo and pattern of
key morphological changes associated with the dog-
woods radiation, such as changes in inflorescence and
fruit architecture and morphology that must have
played important roles in the genus diversification.
With its highest species diversity in eastern Asia
and eastern North America, the dogwood genus also
exhibits several historically important biogeographic
patterns involving intercontinental disjunctions in
eastern Asia, North America, Europe, tropical eastern
Africa, and South America (Fig. 1; Thorne, 1972;
Boufford & Spongberg, 1983; Wu, 1983; Milne &
Abbott, 2002; Donoghue & Smith, 2004; also see
Xiang et al., 2006; Xiang et al., in review). With an
excellent fossil record that includes fruit stones that
can be reliably identified into subgroups (Eyde, 1988;
Manchester, 1994; Crane et al., 1990; Manchester et
al., in press) and a well-resolved, robust phylogeny,
the genus is ideal for a comprehensive biogeographic
analysis and provides a good model for exploring
the impacts of different analytical methods and fossils
in phylogenetic biogeography. Xiang et al. (2006)
proposed, based on results of DIVA and fossils, that
the genus evolved and early diversified into the four
major clades in Europe followed by mutiple intercon-
tinental migrations mostly across the North Atlantic
Land Bridge. The migration routes in three of the four
disjunct lineages however remained to be evaluated
with divergence time information not available in
Xiang et al. (2006). In this study, the biogeographic
history of Cornus is reconstructed using the likeli-
hood-based method of Ree et al. (2005) and Ree &
Smith (2008). The ancestral areas at deep nodes are
compared to those inferred from DIVA (Ronquist,
1997). The impacts of fossils and different constraints
on maximum areas are also explored for each method.
Furthermore, the results from DIVA and ML are also
compared with an analysis treating distributions as a
single, multistate character in BAYESTRAITS.
The major goal of this study is to understand the
morphological evolution of Cornus in space and time
using the most recent and best method available. A
second goal of the study is to gain new insights on the
influences of method choice in reconstructing the
ancestral state of characters and ancestral areas of
biogeographic distributions.
1 Material and methods
1.1 Estimation of ancestral character states
Sampling of characters and phylogeny: Six char-
acters mostly from inflorescence architecture and fruit
morphology were chosen for the analyses (Table 1),
including (1) chromosome numbers, (2) inflorescence
bracts, (3) inflorescence type, (4) development of
inflorescence bud, (5) fruit type, and (6) fruit color.
Five of the six characters are multistate and one is
binary (character 5). For all of these characters, all
taxa have a single state except for character 6, which
is polymorphic within one terminal taxon (represented
Journal of Systematics and Evolution Vol. 46 No. 3 2008 352
























Fig. 1. Approximate geographic distributions of Cornus major clades. BB, big-bracted dogwoods; CC, cornelian cherries; DW, dwarf dogwoods;
BW, blue- or white-fruited dogwoods; BW1, C. oblonga and C. peruviana; BW2, C. alternifolia and C. controversa.


by C. racemosa Lam.), of the blue-fruited or white-
fruited clade (BW) and the outgroups. The six charac-
ters represent a range of character conditions that are
commonly encountered by investigators, e.g., charac-
ters without polymorphism and homoplasy (character
5), characters with missing data and homoplasy
(character 1), characters with homoplasy, but no
missing data (characters 2, 3, 4), and characters with
polymorphism and homoplasy (character 6). The
well-resolved molecular phylogenetic trees of Cornus
derived from six gene regions (rbcL, matK, ndhF,
atpB, ITS, and 26S rDNA; Xiang et al., in review)
were used as the basis for reconstruction of ancestral
states. These trees contain 22 Cornus species and two
outgroup taxa, Alangium and Diplopanax Hand.
-Mazz., sister and close relative of Cornus, respec-
tively (Xiang et al., 1998, 2002; Fan & Xiang, 2003).
The 22 Cornus species represent all the subgenera
recognized so far by different authors at different
times and all the species from each subgenus (see
Xiang et al., 2006), with the exception for subgen.
Syncarpea from the Big-bracted group (BB) and
subgen. Kraniopsis from the BW group. The former is
missing C. multinervosa, closely related to C. kousa
Hance, and C. elliptica (Chun) Q. Y. Xiang & Bouf-
ford, closely related to C. capitata Wall. Subgenus
Kraniopsis consists of ~30 species and was repre-
sented by three species, each representing one of the
three subclades indentified by the ITS-matK species
phylogeny that included all of the species (Xiang et
al., 2006). Species within each of the subclades are
morphologically highly similar (not variable for the
morphological characters under study), thus missing
these species will not affect tracking the evolutionary
histories of these characters in the genus.
1.2 Optimization of ancestral character states
Among the various methods, the Bayesian ap-
proach implemented in BAYESTRAITS 1.0 (Pagel &
Meade, 2006) stands out by not only taking into
account both branch length and phylogenetic uncer-
tainty, but also allowing the user to explore a variety
of models for character transition, defining nodes of
interest, and a reversible-jump Markov Chain Monte
Carlo (Pagel & Meade, 2006). We used 1,000 trees
with branch lengths randomly selected from the tree
pool after burn-in from the Bayesian analysis of the
six-gene data of Cornus (Xiang et al., in review). The
50% majority rule tree of these 1000 trees is identical
to that from all the Bayesian trees after burn-in,
suggesting that these 1000 trees are likely representative
XIANG & THOMAS: Evolution & biogeography of Cornaceae

353
Table 1 Character state of six morphological characters in Cornus
(character 1–6) and distribution of species (character 7)
Characters 1 2 3 4 5 6 7
C. nuttallii Audubon 0 2 2 1 0 4 3
C. florida L. 0 2 2 2 0 4 2
C. kousa Hance 0 2 2 3 1 4 1
C. disciflora Moc. & Sesse ex
DC.
0 1 2 2 0 2 34
C. capitata Wall. 0 2 2 1 1 4 1
C. oligophlebia Merr. 0 0 0 0 0 1 1235
C. hongkongensis Hemsley 0 2 2 1 1 4 1
C. alternifolia L. f. 1 0 0 3 0 1 2
C. controversa Hemsley 1 0 0 3 0 1 1
C. peruviana J. F. Macbr. 0 0 0 0 0 1 4
C. walteri Wangerin 0 0 0 0 0 1 1235
C. racemosa Lam. 0 0 0 0 0 01 23
C. oblonga Wall. 0 0 0 0 0 1 1
C. chinensis Wangerin 2 1 1 2 0 2 1
C. sessilis Torr. ex Durand 1 1 1 3 0 2 3
C. eydeana QY Xiang & YM
Shui
2 1 1 2 0 3 1
C. mas L. 2 1 1 2 0 4 5
C. officinalis Seib. & Zucc. 2 1 1 2 0 4 1
C. volkensii Harms ? 1 1 2 0 2 7
C. suecica L. 0 2 0 0 0 4 1235
C. canadensis L. f. 0 2 0 0 0 4 123
C. unalaschkensis Ledeb. 0 2 0 0 0 4 3
Alangium 0 0 0 0 0 01234 1 or
123567
Diplopanax 0 0 0 0 0 124 12345
Character: character state, score:
1. Chromosome number: n=11, 0; n=10, 1; n=9, 2.
2. Bracts: rudimentary (minute, early deciduous), distal to inflorescence
branches, 0; 4, nonpetaloid at the base of inflorescence, 1; 4 or 6,
petaloid at the base of inflorescence, 2.
3. Inflorescence type: branched compound cymes, 0; umbel (cymose),
1; glomerule (capitate cyme), 2.
4. Developmental state of winter inflorescence bud: floral buds
undeveloped, 0; floral buds developed and unprotected, 1; floral buds
developed and protected by inflorescence bracts, 2; floral buds
developed and protected by scales of a mixed bud, 3.
5. Fruit type: simple, 0; compound, 1.
6. Fruit color: white, 0, blue or black, 1; red then black, 2; dark purple
3; red, 4.
7. Distributions: eastern Asia, 1; eastern North America, 2; western
North America, 3; Central and South America, 4; Europe, 5; Austra-
lia, 6; Africa, 7.
Cornus racemosa represents a subclade in areas 2 and 3, and C.
oligophlebia and C. walteri represent a subclade occurring in areas of
1, 2, 3, and 5 (Xiang et al., 2006). These areas were coded for these
three species accordingly in the analysis. Diplopanax was coded as 1,
2, 3, 4, and 5 to incorporate distributions of taxa in the clade it repre-
sents (nysssoids, mastixioids, Grubbia and Curtisia) in the Cornales.


of the entire tree pool. The model of “multistates”,
MCMC mode, an exponential prior seeded from a
uniform on the interval 0 to 10 (rjhp exp 0.010), a
burn-in of 10,000 and sampling of the chain every 300
generations were applied. These parameters were
recommended by the program manual. Two separate
analyses with a rate parameter of 2 and 5, respectively
were performed. Both analyses were commanded to
estimate the ancestral state for each character at each
node of the phylogeny. The nodes were specified
using the command “addMRCA”. A total of
10,000,000 iterations were run for each analysis.
The evolutionary history of each of the six char-
acters was also traced over the 50% majority rule tree
from Bayesian analysis of Xiang et al. (in review)
using parsimony, maximum likelihood (ML), and
Stochastic character mapping (SCM) available in
MESQUITE 2.01 (Maddison & Maddison, 2007). For
ML and SCM analyses, both the “best” tree topology
and a chronogram (the “best” tree with branch length
in time) were used for a comparison. The chronogram
was derived from the divergence time analysis using
the program Multidivtime (Thorne et al., 1998; Ki-
shino et al., 2001) from Xiang et al. (in review). In
parsimony reconstruction, the character states were
treated as “unordered (i.e., allow free transformation
of a character state to any other states). The ML
reconstructions were conducted using the MK1 model
of evolution (Schluter et al., 1997; Pagel, 1999). The
Mk1 (Markov k-state 1 parameter model) is a k-state
generalization of the Jukes-Cantor model, corre-
sponding to Lewis’s (2001) Mk model, which gives
equal probability (or rate) for changes between any
two character states. The rate of change can be esti-
mated for each individual character based on the data
and branch length by MESQUITE. The stochastic
character mapping (Nielsen, 2002; Huelsenbeck et al.,
2003) in MESQUITE uses continuous-time Markov
models to derive posterior distributions of ancestral
state. It simulates realizations of precise histories of
character evolution in a way consistent with Mk1
model and ancestral states (Maddison & Maddison,
2007). Reconstructions of ancestral states for charac-
ter 6, which has polymorphism for some taxa, were
not allowed for ML and SCM on MESQUITE 2.01.
Therefore, only analyses with MP and BAYESTRAIT
were conducted for this character.
1.3 Optimization of ancestral distributions
The estimation of ancestral distributions was
performed using the same phylogenetic trees as for
character ancestral state reconstructions. Three meth-
ods, the parsimony-based dispersal-vicariance method
(DIVA) (Ronquist, 1997), the ML-based methods of
Ree et al. (2005) and Ree & Smith (2008), and the
Bayesian method using a character model (Pagel et al.,
2004) were employed. Seven geographic areas were
Journal of Systematics and Evolution Vol. 46 No. 3 2008 354
defined, eastern Asia (EAS), eastern North America
(ENA), western North America (WNA), Central and
South America (CSAM), Europe (EUR), Australia
(AUS), and Africa (AFR) to cover all the endemic
distributional areas of Cornus and its outgroups. The
method of DIVA finds the best biogeographic
pathways given the tree topology and distributions of
taxa by minimizing dispersal and extinction events. In
the analysis, each taxon was scored for presence (1) or
absence (0) for each of these areas. The distributions
of Cornus racemosa, C. oligophlebia Merr., and C.
walteri Wangerin, each representing a subclade in the
BW group, were scored according to the occurrence of
all species in the represented subclade. To explore
influences of maximum number of areas constrained
for each node, analyses with constraints of maximum
areas of 4 and 2, respectively, were performed for
comparison. Furthermore, analyses using two different
models of area coding for the outgroup Alangium were
also conducted under these two different maximum
area constraints. One model coded Alangium for all
areas of its occurrence including fossils (all of the
seven areas except Central and South America). The
other model coded Alangium for only the root area of
the genus (eastern Asia) that was inferred from bio-
geographic analysis of the genus based on the phy-
logeny using DIVA (Feng et al., in review). For the
other outgroup Diplopanax, the distribution was
scored for occurrence of all taxa of the clade it repre-
sents in the Cornales (nyssoids, mastixioids, Grubbia
P. Bergius and Curtisia (Burm.) G. A. Sm.; Fan &
Xiang, 2003; Xiang et al., 2002; Xiang et al., 2007).
Optimizations were conducted using the same tree
topology as for character ancestral state reconstruc-
tions described above.
ML inferences of geographic range evolution
were conducted for the same distribution matrix under
the constraints of maximum areas 4 and 2, respec-
tively, on a chronogram from the Bayesian analysis of
six gene regions, the same chronogram as the one for
ancestral character state reconstruction. The program
LAGRANGE (Ree & Smith, 2008) was employed to
run the analysis with a simple model of one rate of
dispersal and extinction constant over time and among
lineages. This program not only finds the most likely
ancestral areas at a node and the split of the areas in
the two descendant lineages, it also calculates the
probabilities of these most-likely areas (Ree & Smith,
2008) at each node, a feature lacking in the program
AReA that was based on Ree et al. (2005). The values
of rate of dispersal and extinction implemented in the
analysis with LAGRANGE were estimated based on the
tree using a maximum likelihood method by the
program. To reduce computation time, the following
areas were disallowed at all nodes due to their wide
disjunction, which requires prior extinction in their
intervening areas or long distance dispersal
(EAS-CSAM, EAS-AFR, ENA-AUS, ENA-AFR,
WNA-AUS, WNA-AFR, CSAM-EUR, EUR-AUS,
EAS-ENA-FR, EAS-WNA-FR, ENA-WNA-AUS,
ENA-WNA-FR, AM-EUR-AUS, EAS-ENA-WNA-
FR, ENA-NA-US-AFR). To explore the impacts of
excluding these areas as a prior and influences of rates
of extinction and dispersal, changes of physical con-
nection between geographic areas through time,
analyses were ran using AReA v. 2.1 (Smith, 2006;
based on Ree et al., 2005) without excluding areas as a
prior using a range of rates of V (dispersal) and L
(extinction) including those estimated from LA-
GRANGE as well as the following: V=L=0.005; V=
0.09, L=0.01; V=0.01, L=0.09; V=0.09, L=0.01; V=
0.02, L=0.18; V=0.18, L=0.02; V=0.0047, L=0.0033.
The analyses with AReA and LAGRANGE were both
conducted for the chronogram with and without
fossils. For analysis with fossils, the branch length in
time of the fossil lineages in the chronogram was
arbitrarily estimated as the length of time between the
earliest and latest appearances of the fossil taxa. If
there was only one report for a fossil taxon, the branch
length in time of that fossil was arbitrarily estimated
as the time range of the geological period of its occur-
rence. For example, the fossil species of BB was
found in the mid Oligocene bed. The branch length in
time of the BB fossil lineage was arbitrarily given as
5.1 million years (my), approximately the length of
time of the mid Oligocene. The fossil lineages were
connected to the stem lineage of the represented
crown group at the time point corresponding to the
age of the fossils. This analysis was done mainly to
see the impact of new geographic areas represented by
fossils, but not occupied by extant species of a line-
age, on optimization of ancestral areas of the perspec-
tive lineages. Information of age, occurrence, and
phylogenetic placements of fossils is from Xiang et al.
(2006). The analyses were run for 1,000,000 replicates
as suggested in the manual. For analyses with AReA,
the Mesophytic model for probabilities of dispersal
between geographic areas at different geological time
periods was implemented (Smith, 2006). The different
models have not been coded into LAGRANGE (Ree,
pers. comm.). Optimization of ancestral areas using
the Bayesian method in BAYESTRAITS v.1.0 was done
as described above for character mapping, but using
“rjhp exp 0.0 30 and “ratedev 15”.
XIANG & THOMAS: Evolution & biogeography of Cornaceae

355
1.4 Estimation of divergence time using BEAST
Phylogenetic dating using five-gene data (rbcL,
matK, ndhF, atpB, and 26S rDNA) was conducted
using r8s (Sanderson, 2002) and Multidivtime (Thorne
et al., 1998; Kishino et al., 2001) in a separate study
(Xiang et al., in review). These methods relied on only
a single phylogeny. Here we estimated the divergence
time using the same data set again using BEAST 1.4.3
(Drummond & Rambaut, 2006) for a comparison.
BEAST applies Bayesian methods and searches for the
optimal phylogeny and estimates divergence time
simultaneously. The divergence time estimated from
BEAST was a summary of estimations based on all the
optimal Bayesian trees found in the analysis, thus
taking the phylogenetic uncertainty into account. The
analyses with BEAST were performed using the model
suggested by Modeltest 3.7 (Posada & Crandall, 1998)
with an uncorrelated exponential relaxed molecular
clock model for 30 million generations. Multiple runs
were done to determine the number of generations
required to allow all node time estimates to reach a
stationary distribution. Four nodes were constrained
by fossils, as done for Multidivtime (Xiang et al.,
2005; Xiang et al., in review).
2 Results
2.1 Optimization of ancestral states
Results from ancestral character state optimiza-
tions from MP, ML, and SCM without temporal
constraints are presented in Figs. 2 and 3 for all nodes.
Results from BAYESTRAITS are provided in Table 2 as
well as Figs. 2 and 3. These results showed that for
binary characters without homoplasy, e.g., character 5,
the results at all nodes are identical and certain among
MP, ML, SCM, and Bayesian methods (Fig. 2). For
characters with homoplasy, there are substantial
differences in the relative certainty of optimal states at
some nodes (e.g., nodes m & l for character 1, node r
for character 2, nodes g & r for character 3, nodes f, g,
m, & r for character 4 among the four methods, as
well as nodes l, p, r, & s for character 6 between MP
and Bayesian inference). There are only a few nodes
among all characters showing striking difference in
the optimal states inferred from different methods.
These are nodes g and r for character 3 (inflorescence
type) and nodes a, b, c, f, g, r for character 4 (winter
condition of inflorescence bud) (Fig. 2B). For charac-
ter 3, nodes g and r were convincingly resolved as
state 0 by MP, ML, and SCM, but BAYESTRAITS
resulted in extremely low posterior probabilities for
state 0 at these two nodes and suggested the most
probable state to be 2 for node g (pp=0.38) and 1 for
node r (pp=0.5), which implies some degree of uncer-
tainty. For character 4, the MP method did not resolve
the ancestral states at g and r nodes (equivocal).
However, the ML and Stochastic character mapping
analyses show that the ancestral states of these two
nodes are state 0 (ML, most likely; SCM, 1.0 pp),
while BAYESTRAITS shows that the ancestral state at
this node is most probable state 2 (Fig. 2B; pp: 0.65).
For nodes a & b of character 4, ML and SCM sug-
gested alternative states (2 vs. 1), while BAYESTRAITS
suggested the two states are nearly equal probable,
similar to MP (Fig. 2B). For nodes c and f of Charac-
ter 4, the ancestral states are equivocal in MP and ML
analyses, but SCM and BAYESTRAITS resolved it as
state 1 (Fig. 2B). In general, the optimal states at the
lower nodes of the phylogeny estimated from BAYE-
STRAITS using 1,000 trees are less certain than estima-
tions from MP, ML, and SCM based on a single,
“best” phylogeny for all characters analyzed.
When temporal information was taken into ac-
count by using a chronogram for estimation with MP,
ML, and SCM, the results showed a few differences
for Characters 1, 2 and 4 under the same method. For
Character 1, node m (Fig. 2A) was reconstructed as
state 1 (green dots) by SCM and as equivocal by ML
based on the tree topology; whereas in the reconstruc-
tions based on a chronogram, this node was inferred
as state 2 (black dots) by both SCM and ML methods
(Fig. 3). For Character 2, node r (Fig. 2A) was in-
ferred to be equivocal by ML method in the topol-
ogy-based analysis, but in the chronogram-based
analysis, this node changed to be most likely state 0
(open circle) (Fig. 3). For character 4, the topol-
ogy-based analysis with the ML method optimized
nodes a and b to be mostly likely state 2 (green dots)
(Fig. 2), but in the chronogram-based ML analysis,
state 2 and state 1 (blue dots) at this node were equally
probable (Fig. 3). In the topology-bsed SCM analysis,
nodes a and b were both certain to be state 1 (blue
dots) (Fig. 2) while in the chronogram-based SCM
analysis, both nodes were reconstructed as state 2
(green dots) (Fig. 3). In general, discrepancies be-
tween ML and SCM in topology-based analyses
disappeared in the chronogram-based analyses (com-
pare node m for character 1 and 4 and node r for
character 2 in Figs. 2 & 3).
2.2 Ancestral distributions
2.2.1 Between coding models for Alangium The
results from analyses using AReA and DIVA with
Alangium coded for its root (indicated by a “*” in
Table 3) or for all areas of its occurrence (column
Journal of Systematics and Evolution Vol. 46 No. 3 2008 356




















































XIANG & THOMAS: Evolution & biogeography of Cornaceae

357




















































Journal of Systematics and Evolution Vol. 46 No. 3 2008 358













































Fig. 2. Evolutionary histories of six key morphological characters on inflorescence and fruit morphology reconstructed using four different methods
(MP, maximum parsimony; ML, maximum likelihood; SCM, stochastic character mapping in MESQUITE 2.01, and BAYESTRAITS 1.0). Results for
the same character from different methods are presented in the same column. Posterior probabilities of <100% from BAYESTRAITS are presented as
numbers. Character state and posterior probability at each node is presented in Table 2. Characters are referred to in Table 1. Species from left to right
on each tree: Cornus nuttallii, C. florida, C. disciflora, C. capitata, C. hongkongensis, C. kousa, C. suecica, C. canadensis, C. unalaschkensis, C.
sessilis, C. volkensii, C. eydeana, C. chinensis, C. mas, C. officinalis, C. oblonga, C. peruviana, C. racemosa, C. walteri, C. oligophlebia, C. alterni-
folia, C. controversa, Alangium, and Diplopanax. A, For characters 1 & 2. B, For characters 3 & 4. C, For characters 5 & 6.


XIANG & THOMAS: Evolution & biogeography of Cornaceae

359





















































Fig. 3. Evolutionary histories of characters inferred from the chronogram of the “best” tree. Only those different from inferences based on the “best”
tree topology are presented.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 360
Table 2 Results of reconstruction of ancestral states for six characters (explained in Table 1) from BAYESTRAITS 1.0 based on 1,000 Bayesian trees
Character
node 1 2 3 4 5 6
a 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 0.50+0.27 (1)
0.47+0.27 (2)
1.00+0.00 (0) 1.00+0.00 (4)
b 1.00+0.00 (0) 0.97+0.10 (2) 1.00+0.00 (2) 0.52+0.35 (1)
0.48+0.35 (2)
0.99+0.02 (0) 1.00+0.00 (4)
c 1.00+0.00 (0) 1.00+0.00 (2) 1.00+0.00 (0) 0.96+0.09 (1) 1.00+0.00 (1) 1.00+0.00 (4)
d 1.00+0.00 (0) 1.00+0.00 (2) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (4)
e 1.00+0.00 (0) 1.00+0.00 (2) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (4)
f 1.00+0.00 (0) 1.00+0.00 (2) 1.00 +0.00 (2) 0.42+0.33 (2)
0.57+0.34 (1)
0.98+0.05 (0) 1.00+0.00 (4)
g 0.96+0.04 (0) .094+0.07 (2) 0.38+0.19 (2) 0.22+0.48 (1)
0.48+0.19 (0)
0.96+0.04 (0) 0.95+0.05 (4)
h 0.99+0.01 (0) 0.99+0.01 (0) 0.99+0.01 (0) 0.99+0.01 (0) 0.99+0.01 (0) 0.99+0.01 (1)
i 0.99+0.02 (0) 1.00+0.00 (0) 1.00+0.00 (0) 0.94+0.12 (0) 1.00+0.00 (0) 0.99+0.01 (1)
j 1.00+0.00 (2) 1.00+0.00 (1) 1.00+0.00 (1) 1.00+0.00 (2) 1.00+0.00 (0) 0.99+0.01 (4)
k 1.00+0.00 (2) 1.00+0.00 (1) 1.00+0.00 (1) 1.00+0.00 (2) 1.00+0.00 (0) 0.50+ 0.01 (2)
0.35+0.29 (4)
0.13+0.18 (3)
l 0.70+0.26 (2) 1.00+0.00 (1) 1.00+0.00 (1) 0.99+0.03 (2) 1.00+0.00 (0) 0.87+0.20 (2)
m 0.64+0.30 (2) 1.00+0.00 (1) 1.00+0.00 (1) 0.94+0.10 (2) 1.00+0.00 (0) 0.95+0.14 (2)
n 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 0.99+0.01 (1)
o 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (0) 0.99+0.01 (1)
p 0.83+0.13 (0) 1.00+0.00 (0) 1.00+0.00 (0) 0.59+0.28 (0)
0.33+0.28 (3)
1.00+0.00 (0) 1.00+0.00 (1)
q 0.99+0.01 (1) 1.00+0.00 (0) 1.00+0.00 (0) 1.00+0.00 (3) 1.00+0.00 (0) 0.99+0.01 (1)
r 0.65+0.21 (0) 0.94+0.07 (2) 0.38+0.19 (2) 0.48+0.19 (0)
0.22+0.18 (1)
0.96+0.05 (0) 0.95+0.05 (4)
s 0.98+0.05 (0) 0.73+0.12 (0) 0.93+0.10 (0) 0.81+0.21 (0) 1.00+0.00 (0) 0.53+0.30 (1)
0.26+0.25 (2)
First number in each cell is the posterior probability and the second number is the standard deviation. Numbers in parenthesis indicate the character
state (explained in Table 1). Nodes a–s are marked on the phylogeny in Figs. 2 and 3.


without indicated by a star in Table 3) showed little
difference for possible optimal ranges at each node,
although there were more alternative solutions at some
nodes from one of the coding models in DIVA and
AReA (Table 3). There was also little difference
between optimizations with and without the outgroups
in a rooted tree.
2.2.2 Constraints of maximum areas at a node
For DIVA, optimal solutions at each node was usually
more numerous when maximum areas were con-
strained as four and contained the solutions from
maximum area constrained as 2 (Table 3). An excep-
tion was found in the analysis without fossils and
coded Alangium for its root area only. In this analysis,
solutions at most of the eight compared nodes were
either identical between the two constraints or more
inclusive from analyses with constraint of maximum
areas equal to 2. For ML analyses from LAGRANGE,
the most probable areas with the highest likelihood at
each node were highly consistent among the two
different constraints (Table 3). These results were also
quite consistent, especially at the four lowest nodes,
with the analyses from AReA without restriction on
maximum areas if fossils were included.
2.2.3 With and without fossils The reconstruction
of ancestral areas at some nodes were strikingly
different between including fossils and excluding
fossils (Table 3). With fossils, the reconstructions at
deepest nodes (nodes 30, 41, 28, 42) favored Europe.
This area was not included in any of the solutions at
these nodes from analyses without fossils except at
node 42 from the ML analysis with LAGRANGE with a
constraint of maximum area of 4.
2.2.4 Among DIVA, ML, and Bayesian methods
When all conditions were identical in the analyses,
results from DIVA and ML showed differences at
most nodes, and most of the time, DIVA suggested
more alternatives while ML was able to narrow down
to one or two most probable areas with the highest
likelihood (Table 3). ML and DIVA remarkably
XIANG & THOMAS: Evolution & biogeography of Cornaceae

361

Journal of Systematics and Evolution Vol. 46 No. 3 2008 362






















Fig. 4. Comparison of divergence times estimated from BEAST with those from Multidivtime and r8s in Xiang et al. (in review). Node numbers
correspond to those marked in Fig. 5 (22–42). Estimations from BEAST at nodes 22, 29, 31, 40, and 41 (shaded in green) are significantly younger
than estimates from Multidivtime and r8s.


differed in the estimations at nodes 30, 40, and 41 in
analyses without fossils and with maximum areas
constrained as 2 (Table 3, comparing MLa in Column
G and DIVAa in Column H) and differed at the nodes
34, 27, 40, 41, 42 in analyses without fossils and with
maximum areas constrained as 4 (Table 3, comparing
MLa in Column L and DIVAa in Column M). The
differences between ML and BayesTraits were much
less, however, only at nodes 37 and 40 (Table 3,
comparing MLb in column J and BAYESTRAITSb in
column K). In analyses with fossils, ML and DIVA
differed at all of the eight nodes (Table 3, comparing
MLa in column A and DIVAa in column B). At
nodes 30 and 28, DIVA favored western North Amer-
ica (area 3) while ML favored Europe (area 5). At
node 40, DIVA consistently optimized a disjunct
ancestral distribution in eastern Asia and Central and
South America (areas 14), but ML suggested distribu-
tions involving eastern Asia and eastern North Amer-
ica (areas 12) and a number of alternatives all with
low posterior probabilities (Table 3, column A). In
general, reconstruction of ancestral distributions at the
lower nodes of the Cornus phylogeny using different
methods showed substantial differences and uncer-
tainty. Probabilities of the most likely ancestral area
estimated from ML methods at each node were mostly
below 50% (Table 3).
2.3 Divergence time
Estimations of divergence time using BEAST
were congruent at most nodes with those from previ-
ous studies using Multidivtime and r8s (Xiang et al.,
in review) (Fig. 4). Exceptions were found for five
nodes (shaded in green in Fig. 5). The estimations
from BEAST were much younger for nodes 22, 31, and
40, 41, but older for node 29 from BEAST. Divergence
time estimations indicate that Cornus split from its
sister Alangium in the late Cretaceous and early
radiation of the genus into several clades occurred also
during the late Cretaceous and the earliest Tertiary
(Fig. 5).
3 Discussion
3.1 Does choice of methods matter?
The comparison of ancestral state reconstruction
for six characters shows that the impact of method
choice depends on the nature of characters. For char-
acters with no homoplasy, no polymorphism, and no
missing data (e.g., character 5), the reconstruction of
the ancestral state was consistent among all methods
compared and between topology-based and chrono-
gram-based analyses (see results & Fig. 2C). This
suggests that for perfect and clean data meeting these
conditions, choice of methods does not affect the
inference of evolutionary trend, although analysis
using a chronogram is better in that it provides tem-
poral information on the events of character state
transitions (see Fig. 3). However, for characters with
homoplasy, but no missing data and no polymorphism
(characters 2, 3, 4), character state reconstructed at
nodes including the taxa carrying the homoplasy (e.g.,
XIANG & THOMAS: Evolution & biogeography of Cornaceae

363































Fig. 5. Chronogram of Cornus from showing major variation and evolutionary trends of inflorescence and fruit characters, as well as ancestral areas
estimated from the maximum likelihood method. The chronogram was derived from Bayesian inference of the phylogeny of the six gene regions
rbcL,matK, ndhF, atpB, ITS, and 26S rDNA and divergence time estimation based on rbcL,matK, ndhF, atpB, and 26S rDNA using Multidivtime in
Xiang et al. (in review). This topology and chronogram was used for analyses of character evolution and biogeographic history. BB, big-bracted
dogwoods; BW, blue- or white-fruited dogwoods; CC, cornelian cherries; DW, dwarf dogwoods. All nodes are strongly supported with 1.00 posterior
probability and bootstrap values >90%, except a few terminal nodes, 32, 39, 22, 31 and one deep node, 28. Nodes 42, 34, 30, and 28 were constrained
by fossils in all divergence time analyses. Numbers connected by a dash at each node indicate geographic area splitting in the two descendant
lineages; the number before the dash represents areas inherited down to the upper daughter lineage while the number after the dash represents areas
inherited by the lower daughter. Alternative area combinations at a node are separated by forward slash. Areas in italics have lower likelihood and
lower probability (see Table 3). Areas in blue at eight deep nodes are inferred from ML analyses without fossils by LAGRANGE and areas in red are
inferred from ML analyses including fossils using LAGRANGE and AReA. 1=eastern Asia, 2=eastern North America, 3=western North America,
4=Central and South America, 5=Europe, 6=Australia, and 7=Africa. The mean divergence time estimation from Multidivtime for each node is
indicated by number in parentheses in million years ago (mya). The horizontal axis shows the time scale in million years ago. Nodes numbered from
22–42 correspond to those in Fig. 4. Nodes with divergence times estimated from BEAST that are significantly different from estimations from
Multidivtime and r8s (nodes 22, 29, 31, 40, and 41) are shaded in green.


nodes g and r for characters 2, 3, nodes c, f, g, m, and
r for character 4) is method-dependent (Figs. 1, 2). In
such cases, the certainty on the reconstructed state
increases from parsimony, to ML, and then to SCM
methods, while the reconstructions by BAYESTRAITS
often suggested a different character state as being
most probable. Furthermore, analyses based on the
best tree topology only or chronogram can result in
different reconstructions at some of these nodes (see
Figs. 2 & 3 for characters 2, 3, 4). For characters with
missing data and homoplasy, but no polymorphism
(character 1), reconstruction of the ancestral state
based on a chronogram at the node including the
lineage with missing data was strikingly different
from that based on the topology only (Figs. 2A, 3),
suggesting that reconstruction of the ancestral state
with missing data is sensitive to branch length (or time
available for evolution). Finally, for characters with
polymorphism and homoplasy, but no missing data,
the inference from parsimony and BAYESTRAITS was
quite congruent (character 6; Figs. 2C, 3). Comparis-
ons of results from ML and SCM for this type of
character cannot be made because the ML and SCM in
MESQUITE cannot perform estimation for data with
polymorphism within a taxon at the present. Devel-
opment of algorithms that can cope with this type of
Journal of Systematics and Evolution Vol. 46 No. 3 2008 364
character is desirable since polymorphism is not
uncommon. Nonetheless, these results indicate that
the incongruence among analyses mostly involves
homoplasious characters, implying difficulty and
uncertainty in reconstructing the ancestral state of
characters with homoplasy. This finding, although
may not be surprising, is particularly important to bear
in mind in studies of comparative biology, e.g., when
attempting to demonstrate adaptive evolution of
species or genes with repeated gain or loss of a certain
ecological trait or gene function in different species by
reconstructing the ancestral state on a phylogeny. It is
recommended that when reconstructing the ancestral
state for homoplasious characters, sensitivity to
methods and models of character evolution should be
explored. For example, one can use BAYESTRAITS to
explore the influence of different rates of character
transition and prior distributions on the reconstruction.
Careful analyses can help to avoid over confident or
incorrect conclusions on important biological prob-
lems derived by simple estimations using maximum
parsimony on a single tree topology.
The results from this study further indicate that,
for the nodes with phylogenetic uncertainty (e.g.,
relatively lower bootstrap support), the ML and SCM
analyses with the simple Mk1 model and a single best
tree (no matter whether including or excluding tem-
poral information) can result in confident reconstruc-
tions that are not favored by BAYESTRAITS (e.g., node
r for characters 2, 3, 4; corresponding to node 28 in
Fig. 5). Furthermore, the reconstructions with ML and
SCM at nodes with strong support (e.g., node g, for
characters 3, 4; Figs. 2B, 3) can also be overconfident
compared to BAYESTRAITS when some level of
uncertainty exists at other nodes of the phylogeny.
The discrepancy between ML, SCM, and BAYE-
STRAITS found in this study reflects a combination of
uncertainties in both ancestral state and phylogenetic
reconstructions. If the Mk1 model implemented in the
ML and SCM analyses is not the correct model for the
characters under investigation, the use of this model in
the ML and SCM analyses could have been responsi-
ble for the incongruence. Unfortunately, there are few
alternative models available for comparison in MES-
QUITE 2.01. At present, the Asymmetrical 2-parameter
Markov-K model is the only alternative model in
MESQUITE under the ML and SCM methods. Yet this
model is limited to binary data for defining differential
rates for forward and backward changes. There is no
option in MESQUITE for applying differential rates of
changes between states of multistate morphological
characters, which are probably most common and
interesting. User-defined step matrix for cost of
change for multistate characters can only be applied to
parsimony analysis. An experiment with the ML
analysis using the Asymmetrical 2-parameter
Markov-k model for the only binary character in this
study (character 5) with a cost of 0.5 for state 0 (clus-
tered fruits) to state 1 (fused fruits) and cost of 1 for
state 1 to state 0 did not change the result. Experi-
ments with MP analyses using step matrices for
multistate characters also showed little difference.
Only at node m (common ancestor of CC) of character
1, the reconstruction changed from equivocal to state
1 (x=10), a state inconsistent with the inferences from
BAYESTRAITS, ML, and SCM (Figs. 2A, 3).
Compared to the ML and SCM methods in
MESQUITE, the full Bayesian influences approach
implemented in BAYESTRAITS estimates the parame-
ters of a model of discrete trait evolution and infer
ancestral state by combining information from uncer-
tainty of phylogeny and uncertainty in the estimate of
ancestral state (Pagel et al., 2004). It uses the con-
tinuous-time Markov model (allowing more than one
change along a branch) to estimate the maximum
likelihood values and their posterior probability
distribution of rate parameters based on the phylogeny
and observations on the value of the traits in each
species. The method uses the most recent common
ancestor approach to overcome the limitation posed
from variation in tree topologies (e.g., not all of the
trees necessarily contain the internal node or nodes of
interest) (Pagel et al., 2004). In addition, the method
takes into account branch length (or temporal infor-
mation) in the estimation of ancestral state. These
features probably make BAYESTRAITS a better choice
for studying homoplasious character evolution. The
program is quite user-friendly and allows authors to
explore various parameters for prior distributions and
rate of character transitions.
3.2 Character evolution in Cornus
The following discussion on character evolution
in Cornus is based on results from BAYESTRAITS and
chronogram-based ML and SCM analyses.
3.2.1 Chromosome numbers In Cornus, most
species have x=11, three species have x=10 and three
have x=9 (see Table 1). In species with x=9 there are
two large metacentric chromosomes while in species
with x=10 one pair of the large metacentric chro-
mosomes are replaced by two telocentric pairs and in
species with x=11, the two pairs of large metacentric
chromosomes are replaced by four pairs of telocentric
chromosomes (Dermen, 1932). Based on this evi-
dence, Dermen (1932) proposed chromosomal
XIANG & THOMAS: Evolution & biogeography of Cornaceae

365
































Fig. 6. Biogeographic histories of the four intercontinental disjunct lineages of Cornus favored by the ML analysis including fruit fossils. BB,
big-bracted dogwoods; CC, cornelian cherries. Numerical numbers represent geographic areas corresponding to those defined in Table 1, Table 3, and
Fig. 5.


Robinsonian fission as the mechanism for chromo-
some number evolution in Cornus. This scenario was
not supported by the chloroplast DNA phylogeny
(Xiang et al., 1996). Based on the parsimony principle
and the chloroplast DNA phylogeny, Xiang et al.
(1996) and Xiang & Eyde (1995) suggested that the
chromosome number in Cornus evolved from x=11 to
10 and 9; with two independent reductions to x=10,
once in the common ancestor of the CC lineage (node
m) and once in the common ancestor of pagoda
dogwoods (node q), and a further reduction from x=10
to x=9 later within the CC lineage. This implies three
independent chromosomal fusion events. The hy-
pothesis partially agrees with the reconstruction of
ancestral state in the present study. The ML, SCM,
and BAYESTRAITS all suggested x=11 being ancestral,
but the evolutionary pathway reconstructed using
these model-based methods is more dynamic. These
methods inferred a change from x=11 to x=9 (rather
than 10 as in parsimony) in the common ancestor of
CC (node m) and x=9 to x=10 in C. sessilis Torr. ex
Durand (Fig. 3). This implies two chromosomal
fusions (at node m) and one fission (in C. sessilis) in
the CC lineage and one chromosomal fusion in the
pagoda dogwoods (node q) occurred during the evolu-
tion of Cornus. Furthermore, divergence time dating
and the reconstructions from ML, SCM and BAYE-
STRAITS suggest that the fusion events in the cornelian
cherries occurred back in the Eocene, while the fusion
event in the pagoda dogwoods occurred not earlier
than the Miocene (Figs. 3, 4).
Chromosome fission has been considered a major
mechanism leading to speciation and increases chro-
mosome numbers in closely related species that are
Journal of Systematics and Evolution Vol. 46 No. 3 2008 366
derived rapidly (see Godfrey & Masters, 2000; Kol-
nicki, 2000), while chromosome fusion is considered
the major mechanism causing a decrease in chromo-
some numbers (Murnane, 2006). Both, especially
fusion, have been reported from plants and animals,
e.g., fission in flies (Rousselet et al., 2000) and beans
(Schubert & Rieger, 1990), and fusion in human (Ijdo
et al., 1991 ), moths (Traut & Clarke, 1997), Droso-
phila (Yu et al., 1999), and Arabidopsis (Heacock et
al., 2004). Loss of telomeres, occurring spontaneously
or via exogenous DNA damage, (Murnane, 2006) has
been considered the major cause for genome instabil-
ity which provide “sticky ends” for chromosome
fusion (Murnane, 2006). In Cornus, it is possible that
spontaneous loss of telomeres in acrocentric chromo-
somes had led to chromosome fusion and resulted in a
decrease of chromosome numbers in three lineages at
different geological times.
3.2.2 Inflorescence architecture and morphology
The different major lineages of Cornus exhibit strik-
ingly different inflorescence architectures (Fig. 5). All
species of the big-bracted (BB) lineage bear cymose
heads (glomerule) subtended by four petaloid bracts
except in two species. The pacific flowering dogwood
C. nuttallii Audubon usually has six bracts and the
Mexican flowering dogwood C. disciflora Moc. &
Sesse ex DC. has four, small, broadly ovate, scale-like
bracts that are deciduous prior to anthesis and before
they expand to become petaloid. The conditions of
inflorescence buds among species of the BB lineage
also vary. In some species the bud is preformed in the
summer or fall and are covered completely by the
scale-like bracts (e.g., in C. florida L. and C. dis-
ciflora). These bracts will expand and develop into the
large petaloid structure during the next growing
season in C. florida, but will fall off before their
expansion in C. disciflora. In two species (e.g., in C.
kousa and C. multinervosa (Pojark.) Q. Y. Xiang), the
inflorescence bud is covered by the expanded leaf bud
scales beneath the inflorescence bud (thus the whole
bud is mix bud; Xiang, 1987). In a few species (e.g.,
C. nuttallii, C. hongkongensis Hemsley, C. capitata,
& C. elliptica), the inflorescence bud is not protected.
The young bracts are non-scaly and only partially
cover the bud. In the cornelian cherry group (CC),
species bear cymose umbels subtended by four
broadly ovate, scale-like bracts (like those in C.
disciflora) (see Fig. 5), which are persistent through
the anthesis and young fruit stage. The inflorescence
buds are also preformed in the fall and covered by the
scale-like bracts, but the bracts never expand to
become petaloid during the growing season of the
following year. The blue- or white-fruited lineage
(BW) produce large and branched compound cymes
with one rudimentary, early deciduous (before anthe-
sis) and minute bract at the branching point of the
lower part of the inflorescence (see Fig. 5). The flower
buds are not preformed in this group, except probably
in the pagoda dogwoods (C. alternifolia L.f. and C.
controversa Hemsley). In these two species, the
inflorescence is preformed in the fall and covered by
several alternatively arranged and imbricate scales that
also protect the leaf buds. In the fourth lineage, the
dwarf dogwood group (DW), the inflorescence archi-
tecture is an intermediate between the BB and BW
(see Fig. 5). These species have a highly reduced
branched compound cyme that is not preformed, but
subtended by four large petaloid bracts.
The ancestral form of inflorescence architecture
in Cornus was inferred to be branched compound
cymes with rudimentary bracts, like those found in the
BW lineage (Fig. 5). Two alternative pathways of
subsequent evolution of inflorescences were sug-
gested, one revealed by BAYESTRAITS and the other
by the ML and SCM analyses. The two pathways
mainly differ in the reconstruction of character state at
the two deep nodes (g and r in Figs. 2, 3, correspond-
ing to nodes 28 and 30 in Fig. 5). The BAYESTRAITS
reconstructions at these two nodes are illustrated with
images at the right side of the nodes in Fig. 5, while
those by ML and SCM are illustrated with images at
the left of the nodes in Fig. 5. The Bayesian pathway
implies a series of dramatic changes of inflorescence
architecture during the early diversification of the
genus in the late Cretaceous or early Tertiary. Sup-
pression of inflorescence branching, preformation of
inflorescence bud, development of four base bracts
into broadly ovate scales, and enlargement and peta-
loidy of the scales all occurred together during the
intial divergence of the genus, leading to umbels
subtended by four large petaloid bracts for the com-
mon ancestor of BB , CC, and DW (image at right of
node 28 in Fig. 5; node r in Figs. 2, 3). During the
divergence between CC and BB+DW lineages, bract
expansion and petaloidy were lost in the CC stem
lineage (node 27 in Fig. 5; node m in Figs. 2, 3), while
umbels were replaced by a head via suppression of
pedicel development along with a reversal to a
non-preformed inflorescence bud in the BB+DW stem
lineage, resulting in glomerules subtended by four
large petaloid bracts in the common ancestor of BB
and DW (image at right of node 30 in Fig. 5; node f in
Figs. 2, 3). When DW diverged from BB, a reversal to
the branched compound cyme occurred in the DW
XIANG & THOMAS: Evolution & biogeography of Cornaceae

367
stem lineage through slight elongation of inflores-
cence branches, leading to the mini-form of branched
compound cymes with 4, large petaloid bracts ob-
served in the modern dwarf dogwood species (C.
canadensis L. f., C. suecica L., and C. unalaschkensis
Ledeb.) (image on the branch of DW in Fig. 5). In the
BB group, later losses of bract enlargement and
petaloidy occurred in C. disciflora in the mid Tertiary,
associated with a southward migration in North
America into moister mountains of Mexico and
further to Central America.
This scenario, although favored (has the highest
posterior probability) by BAYESTRAITS among the
alternatives, has much uncertainty as seen from the
variable posterior probability values at the two deep
nodes (all pp < 0.95) (see characters 2, 3, 4 in Figs. 2,
3). The uncertainty does not seem too closely corre-
lated with the phylogenetic uncertainty. First, the
posterior probabilities at nodes with uncertain recon-
structions of ancestral states are not similar to the
posterior probabilities in the phylogenetic trees (Figs.
2, 3, 5). Nodes with high pp values in the phylogenetic
tree (e.g., nodes f and g) can have low pp values for
the favored state or two alternative favored states of
the inflorescence characters (Figs. 2, 3). Thus, the
uncertainty must come from a combination of phy-
logenetic uncertainty and ambiguity in ancestral state
reconstruction. However, it is noteworthy that this
scenario suggested by BAYESTRAITS on inflorescence
evolution is novel. An ancestral inflorescence of
umbels subtended by petaloid bracts has never been
proposed and is not found in extant species or fossils.
If it is true, an umbel subtended by petaloid bracts
must have become extinct (due to unknown disadvan-
tages). The early Tertiary cornelian cherry fossils (all
represented by fruits; see Eyde, 1988; Crane et al.,
1990; Xiang et al., 2003, 2005) might have been the
lineage bearing this type of inflorescence that has
become extinct. It would be interesting to find out if
there are any fossils with such morphology from the
sites where the fossil fruit stones were discovered. If
this evolutionary pathway is true, the umbel with four
scale-like bracts in the modern cornelian cherries (CC)
represents a retention of the ancestral inflorescence
from the common ancestor of CC and BB+DW, while
the compound cyme in the dwarf dogwoods (DW)
represents a reversal and the four showy bracts in the
DW species are plesiomorphic.
In the alternative pathway suggested by
ML-SCM (“best”-phylogeny based) the early ances-
tors of the four major lineages of Cornus (nodes g, r, s
in Figs. 2, 3) all retained the ancestral inflorescence
form (compound cymes) (images at left of nodes 42,
28, 30 in Fig. 5). Major changes of inflorescence
architecture occurred more recently, after the diversi-
fication into the four major clades. The suppression of
inflorescence branching occurred in CC and BB
lineages separately. The development of 4-fold bracts
and their petaloidy occurred before the condensation
of the inflorescence branches in the BB lineage. This
pathway implies that the umbels with four scale-like
bracts in the cornelian cherries (CC) are an apomor-
phy and evolved from the ancestral inflorescence type
(compound cymes), while the minicompound cymes
with four showy bracts in the dwarf dogwoods
(DW) represent a plesiomorphy and a retention of the
ancestral inflorescence (Fig. 5), in contrast to the
inferences from BAYESTRAITS. It is difficult to dis-
tinguish the two hypotheses without additional evi-
dence, e.g., developmental and genetic data, to deter-
mine which pathway is more likely. Given that the
ML and SCM analysis used the “best” phylogeny
based on six gene regions (ITS, 26S rDNA, matK,
rbcL, atpB, and ndhF) (Xiang et al., in review), which
is also supported by four other low copy nuclear genes
(PI & AP3 genes: Zhang, 2006; Zhang et al., 2008.
antR-Cor: Fan et al., 2004, 2007. waxy: Xiang, unpub-
lished), this “best” phylogeny may well represent the
species phylogeny, adding some support to the ML
and SCM pathway. In other words, the ancestor of
BB-DW-CC was more likely bearing branched com-
pound cymes with early deciduous rudimentary bracts
and the inflorescence was not preformed, and the
ancestor of BB-DW was more likely to bear a con-
densed compound cyme with four petaloid bracts, like
the DW species (Fig. 5). Nonetheless, the ancestral
state of inflorescence architecture at the two deep
nodes above the root of the genus remains uncertain to
some extent. Histological studies on the morphogene-
sis of different inflorescence types will be helpful to
evaluate which ancestral state at these nodes may be
more likely from a developmental perspective. Cou-
pling with a molecular genetic approach, such studies
would further unravel the molecular basis underlying
the evolution of inflorescence architectures, a project
now is being on-going in Xiang’s lab.
3.3 Evolution of fruits and potential ecological
causes
Fruits of the dogwoods are drupaceous develop-
ing from a 2-locular inferior ovary, with a fleshy layer
of pulp and a stony endocarp. In most species, the
fruits are simple, but in the Asian species of the
big-bracted (BB) lineage, fruits are compound (fused,
appearing like a strawberry with stones) (see Fig. 5).
Journal of Systematics and Evolution Vol. 46 No. 3 2008 368
Character ancestral state reconstruction and diver-
gence time dating indicate that this dramatic change in
fruit structure of Cornus, fusion of clustered fruits,
occurred in the mid or late Miocene, associated with
an isolation of the BB lineage in eastern Asia, and was
proposed to be a result of selections by monkeys (Fig.
5; Eyde, 1985).
The fleshy fruits exhibit various colors at the
mature stage among species, including blue, black,
white, red, red-black and dark purple (Fig. 5). The
blue and black fruits are found in most species of the
BW lineage and a few species of the group have white
fruits. Red fruits characterize the BB and DW line-
ages, except for C. disciflora which has red fruits
maturing into black color. The fruit colors in the
cornelian cherries (CC lineage) vary among species
from red, red-black, to dark purple. Ancestral state
reconstructions using parsimony and BayesTraits
(analyses with ML and SCM methods in MESQUITE
were not permitted due to polymorphism) suggested
that the fruit color at the root of Cornus is more likely
to be blue or black, and several evolutionary changes
in fruit color occurred during the genus radiation in
different geological time periods. The early changes
appeared to be accompanied with the major alterations
in inflorescence architecture (Fig. 2C, character 6; Fig.
5). A red color stage in fruits evolved from the blue
and black fruits in the early Tertiary (node r; Fig.
2C). Red fruits evolved in lineages with petaloid
bracts (node g) in the early Tertiary and also in the
ancestor of C. mas L. and C. officinalis Seib. & Zucc.
of the cornelian cherry group (CC) in the mid to late
Tertiary (node j in Figs. 2C, 5). Red maturing into
black fruits also evolved twice, once in C. disciflora in
the mid Tertiary and once in ancestor of CC lineage
(node m) or in the ancestor of CC+BB+DW clade
(node r) in the very early Tertiary (Fig. 2C). Further-
more, dark purple fruit evolved in C. eydeana QY
Xiang & YM Shui from red-black fruits in the CC
lineage, and white fruits evolved from blue-black
fruits in the BW clade (Fig. 2C).
Evolution of fruit color in fleshy fruited plants is
an old topic in evolutionary ecology, but our under-
standing of the underlying ecological causes has been
very limited (Willson et al., 1989; Willson & Whelan,
1990). Most fleshy fruits consumed by vertebrates in
the temperate zones are red and/or black which are
also common in tropical and subtropical fruits (Will-
son et al., 1989; Willson & Whelan, 1990). Willson
and Whelan (1990) invoked a wide array of ecological
and evolutionary causes to explain the globally most
frequent occurrence of red and black fruits along with
the occurrence of other colored fruits at lower fre-
quency. A partial list includes effects on avian forag-
ing and against natural enemies, physiological adapta-
tions, results of selection acting on correlated charac-
ters, constraint by metabolic costs, competition for
dispersal agents, as well as constraint by phylogenetic
history (Willson & Whelan, 1990). However, as
indicated by the authors, there appears to be no known
data to strongly support any of these hypotheses.
In Cornus, given that the differentiation in fruit
color seems to correlate with phylogenetic lineages
and with changes in inflorescence architecture and
morphology, it is possible that phylogenetic constraint
coupled with selection acting on the inflorescences
could have contributed to the fruit color divergence
among the major lineages of Cornus. However, other
factors, such as avian foraging, defense against
pathogens, and differences in taste or nutrients might
have also played important roles in the fruit color
evolution in the genus as discussed below.
The fruits of dogwoods are eaten by mostly birds,
but also by rodents and other mammals (Eyde, 1988).
There is no clear evidence for color preference (e.g.,
white fruits of C. alba Linn. Mant. and black fruits of
C. sanguinea L. are both eaten by European redstarts),
but there seems to be a preference in fruit size and
display positions (on branches vs. on ground). For
example, small birds do not swallow and disperse the
relatively large cornelian cherry fruits (CC) and some
species have fruits persistent on the branch (e.g., C.
sanguinea with black fruit) while other species have
fruits fallen to the ground (e.g., C. florida with red
fruit) (Baird, 1980; Eyde, 1988). In the later case,
color may make a difference. Red color is easier for
birds to find on the ground than blue or black fruits.
The red fruits of the rhizomatous little dwarf dog-
woods (DW) are dispersed mostly by ground-feeding
birds and mammals (see Eyde, 1988). The fused red
fruits in the Asian big-bracted dogwoods (like the
fruits in C. kousa) are favored by monkeys that take
the sweet pulp and discard the seeds, and were con-
sidered as the mechanism for the evolution of the
fused large fruits (Eyde, 1985). Furthermore, phyto-
chemistry studies showed that the different colors in
fruits are a result of production of different kinds of
anthocyanins (Vareed et al., 2006; Bjorøy et al., 2007)
and their relative quantities. These studies reported
that fresh fruits of C. alternifolia and C. controversa
(BW lineages, black fruits) contain mostly delphinidin
3-O-glucoside and delphinidin 3-O-rutinoside, as well
as a very small amount of cyanidin 3-O-glucoside
(Vareed et al., 2006). Fresh fruits of C. florida and C.
XIANG & THOMAS: Evolution & biogeography of Cornaceae

369
kousa (BB lineage, red fruits) contain mostly cyanidin
3-O-galactoside and cyanidin 3-O-glucoside, with a
negligible amount of delphinidin 3-O-glucoside, while
major anthocyanins in C. mas and C. officinalis
fruits (CC lineage, red fruits) are delphinidin 3-O-
galactoside, cyanidin 3-O-galactoside and pelargo-
nidin 3-O-galactoside. In Cornus alba fruits (BW,
white), five anthocyanins were detected (delphinidin
3-O-β-galactopyranoside-3′,5′-di-O-β-glucopyranoside,
delphinidin 3-O-β-galactopyranoside-3′-O-β-glucopy-
ranoside cyanidin 3-O-β-galactopyranoside-3′-O-β-
glucopyranoside, the 3-O-β-galactopyranosides of
delphinidin and cyanidin), with delphinidin 3-O-β-
galactopyranoside-3′,5′-di-O-β-glucopyranoside being
the most abundant (Bjorøy et al., 2007). The relative
amounts of the major anthocyanins within each spe-
cies also vary. Therefore, the alternations in fruit color
of Cornus must be related to changes in regulation of
the anthocyanin pathway. These data also indicate that
the red fruits in cornelian cherries and big-bracted
dogwoods do not contain the same kinds of antho-
cyanins, suggesting the origins of red fruits in the two
lineages were mostly likely via different developmen-
tal pathways.
Although functions of anthocyanins in plants are
still debated (Willson & Whelan, 1990; Gould &
Lister, 2006), masking of chlorophylls by antho-
cyanins was found to reduce risk of photo-oxidative
damage to leaf cells in some plants as they senesce,
including the white-fruited red osier dogwood (Cor-
nus sericea L.) and black-fruited Pagoda dogwood
(Cornus alternifolia) (Feild et al., 2001; Hoch et al.,
2003; Lee et al., 2003). There also have been reports
on anti-cancer, anti-inflammatory and antioxidant
effects of Cornus anthocyanins (Seeram et al., 2002;
Vareed et al., 2006) as well as their uses in treatment
of diabetes mellitus-related disorders (Jayaprakasam
et al., 2006; Nair et al., 2006). The pigments del-
phinidin 3-O-β-galactopyranoside-3′,5′-di-O-β-glu-
copyranoside and delphinidin 3-O-β-galactopyrano-
side-3′-O-β-glucopyranoside rich in the black fruits of
C. controversa and C. alternifolia were confirmed to
show growth inhibitory activity towards various
human cancer cell lines (Vareed et al., 2006). The
evidence suggests that differences in avian foraging,
alteration in nutrient and taste, and defense against
pathogens could have played a role in fruit color
evolution in Cornus.
3.4 Biogeographic history
Ancestral area estimation using parsimony and
maximum likelihood methods under different condi-
tions show substantial differences at some nodes and
indicate uncertainty in ancestral area reconstructions
at lower phylogenetic nodes (Table 3). Inclusion and
exclusion of fossils have a striking effect on the
uncertainty (see Table 3 and Fig. 5). Other factors,
such as coding models for outgroup distributions,
maximum area constraints, and including or excluding
outgroups in the rooted tree influenced the ancestral
area estimation to a lesser extent (Table 3). The
differences among the analyses clearly suggest chal-
lenge and limitation in estimating lineage bio-
geographic histories. At one hand, adding fossils in
the analyses could have biased the results if the fossils
were not correctly placed on the phylogeny and the
branch length leading to the fossil taxa were errone-
ously constrained in the ML analyses. Thus caution
must be taken to evaluate the fossil data (e.g., taxo-
nomic and age identifications) and to determine the
phylogenetic affinities of the fossil taxa to lessen this
problem. On the other hand, one might take an ex-
treme approach by excluding fossils in the analysis to
avoid the problems associated with fossils. The short-
coming of this approach is obvious. The bio-
geographic history inferred with incomplete informa-
tion (e.g., without considering the fossils) could also
be heavily biased. A cautious and feasible approach
would be to compare results from analyses including
and excluding fossils.
Results from this study and those from Xiang et
al. (2006) have also shown that DIVA is sensitive to
approaches of geographic area coding for terminal
taxa of the ingroup. Coding only the ancestral area of
a terminal taxon (recommended by DIVA) vs. coding
distributional areas of all of the constitutent species of
the terminal taxon could lead to different conclusions.
For example, when the fossil lineage of cornelian
cherries was coded for occurrence only in Europe (the
ancestral area of the lineage estimated from Xiang et
al., 2005), the root area of Cornus was optimized as
Europe under the maximum area constraint of 2 using
DIVA (Xiang et al., 2006). However, in the present
study, the fossil lineage of cornelian cherries was
coded for all distribution areas of its occurrrence
(Europe and western North America), a case probably
most common since the root areas of most lineages
included in an analysis are unknown, the same analy-
sis with DIVA under maxiareas of 2 optimized the
root area of Cornus to be quite uncertain, most in-
cluding western North America (Table 3, Column E).
The ancestral areas optimized in Xiang et al. (2006)
for these lower phylogenetic nodes with the coding
approach favored by DIVA (coding for ancestral areas
for higher hierarchical terminal taxa) are largely
Journal of Systematics and Evolution Vol. 46 No. 3 2008 370
congruent with the results inferred from ML analyses
in the present study (see Table 3, column A and D),
which mostly include Europe. However, without
divergence time information in Xiang et al. (2006),
evaluation of alternative intercontinental dispersal
routes for the genus was limited.
The evolutionary pathway of Cornus in space
and time favored by ML analyses using LAGRANGE
including fossils is illustrated in Figures 5 and 6. The
origin and early diversification of Cornus was sug-
gested to be in the Eurasian continent in the late
Cretaceous and early Tertiary (Fig. 5). Three intercon-
tinental dispersals out of Europe into eastern North
America occurred in the BW lineage in different
geological times, twice in the early Tertiary at ~52
mya and ~46 mya and once in the mid ~22 mya (Fig.
5). These migrations must have crossed the North
Atlantic Land Bridge (NALB). Although NALB was
believed to have been broken by the early Eocene,
migration of temperate plants up to the early Miocene
by hopping across the island chains was considered
possible (Tiffney, 1985). This result is congruent with
post-Eocene Trans-NALB migration of plants (see
Xiang et al., 2005). The migrations of the BB lineage
from Europe into eastern and western North America,
as well as Central America at ~27 mya could have,
however, occurred via either NALB and the Bering
Land Bridge (BLB) which was believed to be avail-
able throughout most of the Tertiary (Tiffney &
Manchester, 2001). The early dispersal of the CC
lineage from Europe into western North America at
~42 mya was more likely to have taken the NALB
route because in the Eocene, Tran-BLB dispersal
through Asia was blocked by the Turgai Strait (Tiff-
ney & Manchester, 2001). However, the recent dis-
covery of dogwood leaf fossils from the Paleocene of
western North America and northeastern Asia where
fruit fossils of the CC lineage of the same age were
also found (Manchester et al., in press) may lend
support to early Tertiary Trans-BLB migration of the
CC lineage at high latitudes. The leaf fossils were not
used in the biogeographic analysis due to their phy-
logenetic uncertainty within the genus.
It must be noted that the biogeographic history
inferred here and shown in Fig. 5 and Fig. 6 is con-
tingent upon the fruit fossil evidence available at the
present and only the ancestral ranges with the highest
likelihood scores and greatest probabilities. Given that
there are alternative ranges with slightly lower likeli-
hood scores and lower probabilities at many nodes,
uncertainty clearly remains with this biogeographic
history. Furthermore, long distance dispersal can
never be ruled out. This again demonstrates the limi-
tation and difficulty in biogeographic analyses. Con-
sidering all of the fossil evidence and results from
different analyses for Cornus, it is likely that the
genus diversified into major lineages and spread
around the globe via NALB and BLB quickly in the
early Tertiary, soon after its origin in the late Creta-
ceous. Taken at face value, the results support
post-Eocene migration of plants across the NALB in
the BW lineage (from node 37 to 36) (Xiang et al.,
2005), suggesting that the NALB may have been more
important for plant migration during the later Tertiary
time than previously thought (Tiffney and Manches-
ter, 2001; see Xiang et al., 2005). The results are in
agreement with the role of the NALB in connecting
the floras in Tropical Asia, America and Africa (Fig.
6, CC lineage, C. oblonga-C. peruviana J. F. Macbr.),
as suggested by Xiang et al. (2005, 2006), Lavin et al.
(2000), and Davis et al. (2002).
The divergence time between Cornus oblonga
Wall. and C. peruviana J. F. Macbr. was estimated to
be in the Eocene by Multidivtime and r8s and in the
Miocene by BEAST (node 40 in Fig. 4; Figs. 5, 6).
Based on the ancestral areas reconstructed by LA-
GRANGE and AReA (Table 3), both of these diver-
gence times agree with migration from Europe east-
ward to Asia and westward to eastern North America
via NALB and further southward via long distance
dispersal into South America by birds or by water (see
discussion in Xiang et al., 2006 and references
therein). However, a divergence time in the Eocene
also agrees with migration along the Tethys seaway
margins, which was considered to be a viable migra-
tion route for the thermophilic boreotropical flora in
the early Eocene (Tiffney, 1985) and was proposed to
be used for spread of the tropical family Alangiaceae
(Feng et al., in review).
In summary, this study showed that reconstruc-
tions of the ancestral state of characters and ancestral
area distributions are affected by both analytical
methods and phylogenetic uncertainty. Researchers
are encouraged to apply the most sophisticated
method possible available and conduct vigorous
analyses to explore alternative hypotheses before
reaching a conclusion for evolution of characters that
are homoplasious. The substantial differences between
DIVA and ML analyses that included and excluded
fossils suggest the need of a reevaluation of the global
biogeographic patterns previously assembled using
DIVA and raise special attentions to the use of fossil
data in biogeographic analyses. Researchers
reconstructing ancestral areas with a phylogeny
XIANG & THOMAS: Evolution & biogeography of Cornaceae

371
containing widely distributed higher hierarchical
terminal taxa (e.g., genus, family, etc.) should be
particularly careful with the choice of methods.
Acknowledgements The authors thank R. Ree, S.
Smith, and B. Moore for assistance and guidance in
biogeographic analyses using LAGRANGE and AReA.
We also thank the Deep Time Research Coordination
Network supported by the National Science Founda-
tion (NSF) funded to D. E. Soltis (DEB-0090283), the
Phytogeography of the Northern Hemisphere working
group supported by NESCent funded to M. Donoghue
and P. Manos, and the ClockWork group supported by
NESCent funded to J. Clark and B. Wiegmann for the
workshops, symposia, and discussion with the par-
ticipants of each group. We also thank the two re-
viewers and the journal editors Y-L Qiu and Z-D Chen
for valuable comments to improve this manuscript.
This research was supported by the National Science
Foundation (DEB-0444125) funded to Xiang and
benefited from a Multidisciplinary grant for faculty
research and professional development grant at North
Carolina State University funded to Xiang, Franks and
Xie.
References
APG (Angiosperm Phylogeny Group). 2003. An update of the
Angiosperm Phylogeny Group classification for the orders
and families of flowering plants: APG II. Botanical
Journal of Linnean Society 141: 399–436.
Baird JW. 1980. The selection and use of fruit by birds in an
eastern USA forest. Wilson Bulletin 92: 63–73.
Bjorøy Ø, Fossen T, Andersen ØM. 2007. Anthocyanin
3-galactosides from Cornus alba “Sibirica” with
glucosidation of the B-ring. Phytochemistry 68: 640–645.
Blanchette M, Green ED, Miller W, Haussler D. 2004.
Reconstructing large regions of an ancestral mammalian
genome in silico. Genome Research 14: 2412–2423.
Bollback JP. 2006. SIMMAP: Stochastic character mapping of
discrete traits on phylogenies. BMC Bioinformatics 7: 88.
Boufford DE, Spongberg SA. 1983. Eastern Asian-eastern
North American phytogeographical relationships—a
history from the time of Linnaeus to the twentieth century.
Annals of the Missouri Botanical Garden 70: 423–439.
Bullock AA. 1959. Nomina familiarum conservanda proposita.
Taxon 8: 154–181.
Chase MC, Soltis DE, Olmstead RG, Morgan D, Les DH,
Mishler BD, Duvall MR, Price RA, Hills HG, Qiu Y-L,
Kron KA, Rettig JH, Conti E, Palmer JD, Manhart JR,
Sytsma KJ, Michaels HJ, Kress WJ, Karol KG, Clark WD,
Hedren M, Gaut BS, Jansen RK, Kim K-J, Wimpee CF,
Smith JF, Furnier GR, Strauss SH, Xiang Q-Y, Plunkett
GM, Soltis PS, Swensen S, Williams SE, Gadek PA,
Quinn CJ, Eguiarte LE, Golenberg E, Learn GH Jr,
Graham SW, Barrett SCH, Dayanandan S, Albert VA.
1993. Phylogenetics of seed plants: an analysis of
nucleotide sequences from the plastid gene rbcL. Annals
of the Missouri Botanical Garden 80: 528–580.
Crane PR, Manchester SR, Dilcher DL. 1990. A preliminary
survey of fossil leaves and well-preserved reproductive
structures from the Sentinel Butte Formation (Paleocene)
near Almont, North Dakota. Fieldiana Geol. 1418: 1–63.
Cunningham C. 1999. Some limitations of ancestral
character-state reconstruction when testing evolutionary
hypotheses. Systematic Biology 48: 665–674.
Davis CC, Bell CD, Mathews S, Donoghue MJ. 2002.
Laurasian migration explains Gondwanan disjuncts:
evidence from Malpighiaceae. Proceedings of the National
Academy Sciences USA 99: 6833–6837.
de Candolle AP. 1828. Prodromus Systematis Naturalis Regni
Vegetabilis 3: 203.
Dermen H. 1932. Cytological studies of Cornus. Journal of
Arnold Arboretum 13: 401–417.
Donoghue MJ, Smith SA. 2004. Patterns in the assembly of
temperate forests around the Northern Hemisphere.
Philosophical Transactions of the Royal Society, London B
359: 1633–1644.
Drummond AJ, Rambaut A. 2006. BEAST v1.4 [online].
Available from http://beast.bio.ed.ac.uk/.
Ekman S, Andersen LH, Wedin M. 2008. The limitations of
ancestral state reconstruction and the evolution of the
Ascus in the Lecanorales (Lichenized Ascomycota).
Systematic Biology 57: 141–156.
Eyde RH. 1985. The case for monkey-mediated evolution in
big-bracted dogwoods. Arnoldia (Jamaica Plain) 45: 2–9.
Eyde RH. 1988. Comprehending Cornus: Puzzles and progress
in the systematics of the dogwoods. Botanical Review
(Lancaster) 54: 233–351.
Fan CZ, Purugganan MD, Thomas DT, Wiegmann BM, Xiang
QY. 2004. Heterogeneous evolution of the Myc-like
Anthocyanin regulatory gene and its phylogenetic utility
in Cornus L. (Cornaceae). Molecular Phylogenetics and
Evolution 33: 580–594.
Fan CZ, Xiang QY. 2001. Phylogenetic relationships within
Cornus L. (Cornaceae) based on 26S rDNA sequences.
American Journal of Botany 88: 1131–1138.
Fan C, Xiang QY. 2003. Phylogenetic analyses of Cornales
based on 26S rRNA and combined 26S rDNA-matK-rbcL
sequence data. American Journal of Botany 90:
1357–1372.
Fan C, Xiang QY, Remington DL, Purugganan MD, Wiegmann
BM. 2007. Evolutionary patterns in the antR-Cor gene in
the dwarf dogwood complex (Cornus, Cornaceae).
Genetica 130: 19–34.
Feild TS, Lee DW, Holbrook NM. 2001. Why leaves turn red in
autumn. The role of anthocyanins I senescing leaves of
red-osier dogwood. Plant Physiology 127: 566–574.
Feng CM, Manchester SR, Xiang QY(J). Phylogeny and
biogeography of Alangiaceae (Cornales) inferred from
DNA sequences, morphology, and fossils. Molecular
Phylogenetics and Evolution. (in review, available from
the first author upon request)
Godfrey LR, Masters JC. 2000. Kinetochore reproduction
theory may explain rapid chromosome evolution
Proceedings of National Academy of Sciences USA 97:
9821–9823.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 372
Gould KS, Lister C. 2006. Flavonoid functions in plants. In:
Andersen OM, Markham KR eds. Flavonoids: Chemistry,
Biochemistry and Applications. Boca Raton: CRC Press.
397–441.
Harvey PH, Pagel MD. 1991. The Comparative Method in
Evolutionary Biology. Oxford: Oxford University Press.
Heacock M, Spangler E, Riha K, Puizina J, Shippen DE. 2004.
Molecular analysis of telomere fusions in Arabidopsis:
multiple pathways for chromosome end-joining. The
EMBO Journal 23: 2304–2313.
Hines HM. 2008. Historical biogeography, divergence times,
and diversification patterns of Bumble Bees
(Hymenoptera: Apidae: Bombus). Systematic Biology 57:
58–75.
Hoch WA, Singsaas EL, McCown BH. 2003. Resorption
protection. Anthocyanins facilitate nutrient recovery in
autumn by shielding leaves from potentially damaging
light levels. Plant Physiology 133: 1296–1305.
Huelsenbeck JP, Nielsen R, Bollback JP. 2003. Stochastic
mapping of morphological characters. Systematic Biology
52: 131–158.
Huelsenbeck JP, Rannala B, Masly JP. 2000. Accommodating
phylogenetic uncertainty in evolutionary studies. Science
288: 2349–2350.
Huelsenbeck JP, Ronquist F, Nielsen R, Bollback JP. 2001.
Bayesian inference of phylogeny and its impact on
evolutionary biology. Science 294: 2310–2314.
Ijdo JW, Baldini A, Ward DC, Reeders ST, Wells RA. 1991.
Origin of human chromosome 2: an ancestral
telomere-telomere fusion. Proceedings of the National
Academy of Sciences USA 88: 9051–9055.
Jayaprakasam B, Olson LK, Schutzki RE, Tai MH, Nair MG.
2006. Amelioration of obesity and glucose intolerance in
high-fat-fed C57BL/6 mice by anthocyanins and ursolic
acid in Cornelian cherry (Cornus mas). Journal of
Agriculture and Food Chemistry 54: 243–248.
Kishino H, Thorne JL, Bruno WJ. 2001. Performance of a
divergence time estimation method under a probabilistic
model of rate evolution. Molecular Biology and Evolution
18: 352–361.
Kolnicki RL. 2000. Kinetochore reproduction in animal
evolution: Cell biological explanation of karyotypic
fission theory. Proceedings of the National Academy of
Sciences USA 97: 9493–9497.
Kondo B, Omland KE. 2007. Ancestral state reconstruction of
migration: multistate analysis reveals rapid changes in
new world orioles (Icterus spp.) The Auk 124: 410–419.
Lavin M, Thulin M, Labat JN, Pennington RT. 2000. Africa, the
odd man out: molecular biogeography of Dalbergioid
legumes (Fabaceae) suggests otherwise. Systematic
Botany 25: 449–467.
Lee DW, O’Keefe J, Hobrook NM, Field TS. 2003. Pigment
dynamics and autumn leaf senescence in a New England
deciduous forest, eastern USA. Ecological Research 18:
677–694.
Lewis PO. 2001. A likelihood approach to estimating
phylogeny from discrete morphological character data.
Systematic Biology 50: 913–925.
Maddison DR, Maddison WP. 1992. MacClade: Analysis of
phylogeny and character evolution. Version 3.0.
Sunderland, MA: Sinauer Associates.
Maddison DR, Maddison WP. 2001. MacClade4: Analysis of
phylogeny and character evolution. Version 4.03.
Sunderland, MA: Sinauer Associates.
Maddison WP, Maddison DR. 2007. Mesquite: a modular
system for evolutionary analysis [online]. Version 2.01.
Available from http://mesquiteproject.org.
Manchester SR. 1994. Fruits and seeds of the Middle Eocene
Nut Beds flora, Clarno Formation, North Central Oregon.
Palaeontography of America 58: 1–205, 70 pls.
Manchester SR, Xiang QY, Kodrul TM, Akhmetiev M.
Leaves of Cornus (Cornaceae) from the paleocene of
North American and Asia confirmed by trichome
characters. International Journal of Plant Sciences. (in
press)
Milne RI, Abbott RJ. 2002. The origin and evolution of Tertiary
relict floras. Advances in Botanical Research 38: 282–314,
pl. 5–7.
Moore BR, Smith SA, Donoghue MJ. 2006. Increasing data
transparency and estimating phylogenetic uncertainty in
supertrees: Approaches using nonparametric
bootstrapping. Systematic Biology 55: 662–676.
Murnane JP. 2006. Telomeres and chromosome instability DNA
Repair. Vol. 5. No. 9-10. (8 September 2006). 1082–1092.
Murrell ZE. 1993. Phylogenetic Relationships in Cornus
(Cornaceae). Systematic Botany 18: 469–495.
Murrell ZE. 1996. A new section of Cornus in South and
Central America. Systematic Botany 21: 273–288.
Nair MG, Jayaprakasam B, Olson LK, Vareed SK. 2006. Insulin
secretion by anthocyanins and anthocyanidins. US Patent
and Trademark Office, United States Patent Application
No:20060025353.
Nepokroeff M, Sytsma KJ, Wagner WL, Zimmer EA. 2003.
Reconstructing ancestral patterns of colonization and
dispersal in the Hawaiian understory tree genus Psychotria
(Rubiaceae): a comparison of parsimony and likelihood
approaches. Systematic Biology 52: 820–838.
Nielsen R. 2002. Mapping mutations on phylogenies.
Systematic Biology 51: 729–739.
Ouzounis CA. 2005. Ancestral state reconstructions for
genomes. Current Opinion in Genetics & Development
15: 595–600.
Pagel M. 1999. The maximum likelihood approach to
reconstructing ancestral character states of discrete
characters on phylogenies. Systematic Biology 48:
612–622.
Pagel M, Meade A. 2006. Bayesian analysis of correlated
evolution of discrete characters by reversible-jump
Markov chain Monte Carlo. American Naturalist 167:
808–825.
Pagel M, Meade A, Barker D. 2004. Bayesian estimation of
ancestral character states on phylogenies. Systematic
Biology 53: 673–684.
Pedersen N, Holyoak DT, Newton AE. 2007. Systematics and
morphological evolution within the moss family Bryaceae:
A comparison between parsimony and Bayesian methods
for reconstruction of ancestral character states. Molecular
Phylogenetics and Evolution 43: 891–907.
Posada D, Crandall KA. 1998. Modeltest: testing the model of
DNA substitution. Bioinformatics 14: 817–818.
Ree RH, Moore BR, Webb CO, Donoghue MJ. 2005. A
likelihood framework for inferring the evolution of
XIANG & THOMAS: Evolution & biogeography of Cornaceae

373
geographic range on phylogenetic trees. Evolution 59:
2299–2311.
Ree RH, Smith SA. 2008. Maximum-likelihood inference of
geographic range evolution by dispersal, local extinction,
and cladogenesis. Systematic Biology 57: 4–414.
Renner SS, Beenken L, Grimm GW, Kocyan A, Ricklefs RE.
2007. The evolution of dioecy, heterodichogamy, and
labile sex expression in Acer. Evolution 61: 2701–2719.
Ronquist F. 1996. DIVA version 1.1 [online]. Computer
program and manual available by anonymous FTP from
Uppsala University (ftp.uu.se or ftp.systbot.uu.se).
Ronquist F. 1997. Dispersal-vicariance analysis: a new
approach to the quantification of historical biogeography.
Systematic Biology 46: 195–203.
Ronquist F. 2004. Bayesian inference of character evolution.
Trends in Ecology and Evolution 19: 475–481.
Rossnes R, Eidhammer I, Liberles DA. 2005. Phylogenetic
reconstruction of ancestral character states for gene
expression and mRNA splicing data. BMC
Bioinformatics, 6: 127.
Rousselet J, Monti L, Auger-Rozenberg MA, Parker JS,
Lemeunier F. 2000. Chromosome fission associated with
growth of ribosomal DNA in Neodiprion abietis
(Hymenoptera: Diprionidae). Proceedings of the Royal
Society: Biological Sciences 267: 1819–1823.
Sanderson MJ. 2002. Estimating absolute rates of molecular
evolution and divergence times: A penalized
likelihood approach. Molecular Biology and Evolution 19:
101–109.
Sanmartín I, Enghof H, Ronquist F. 2001. Patterns of animal
dispersal, vicariance and diversification in the Holarctic.
Biological Journal of Linnean Society 73: 345–390.
Sanmartín I, Ronquist F. 2004. Southern Hemisphere
biogeography analyzed with event-based models: Plant
versus animal patterns. Systematic Biology 53: 216–243.
Schluter D, Price T, Mooers AO, Ludwig D. 1997. Likelihood
of ancestor states in adaptive radiation. Evolution 51:
1699–1711.
Schubert I, Rieger R. 1990. Alteration by centric fission of the
diploid chromosome number in Vicia faba L. Genetica 81:
67–69.
Seeram NP, Schutzki R, Chandra A, Nair MG. 2002.
Characterization, quantification, and bioactivities of
anthocyanins in Cornus species. Journal of Agricultural
and Food Chemistry 50: 2519–2523.
Smith SA. 2006. AReA: Ancestral Range Analysis v.2.1
[online]. Available from http://blackrim.org/programs/
area.html.
Soltis DE, Soltis PS, Endress PK and Chase MW. 2005.
Phylogeny and Evolution of Angiosperms. Sunderland,
MA: Sinauer Associates.
Takhtajan AL. 1987. Sistema Magnoliofitov. Izdatelístvo
ìNauka,î Leningrad.
Thorne JL, Kishino H, Painter S. 1998. Estimating the rate of
evolution of the rate of molecular evolution. Molecular
Biology and Evolution 15: 1647–1657.
Thorne RF. 1972. Major disjunctions in the geographic ranges
of seed plants. Quaternary Review in Biology 47: 365–411.
Tiffney BH. 1985. The Eocene North Atlantic land bridge: its
importance in Tertiary and modern phytogeography of the
Northern Hemisphere. Journal of Arnold Arboretum 66:
243–273.
Tiffney BH, Manchester SR. 2001. The use of geological and
paleontological evidence in evaluating plant
phylogeographic hypotheses in the Northern Hemisphere
tertiary. International Journal of Plant Sciences 162:
S3–S17.
Traut W, Clarke CA. 1997. Karyotype evolution by
chromosome fusion in the moth genus Orgyia. Hereditas
126: 77–84.
Vareed SK, Reddy MK, Schutzki RE, Muraleedharan G. 2006.
Anthocyanins in Cornus alternifolia, Cornus controversa,
Cornus kousa. Life Sciences 78: 777–784.
Willson MF, Irvine AK, Walsh NG. 1989. Vertebrate dispersal
syndromes in some Australian and New Zealand plant
communities, with geographic comparisons. Biotropica
21: 133–147.
Willson MF, Whelan CJ. 1990. The evolution of fruit color in
fleshy-fruited plants. The American Naturalist 136:
790–809.
Wu ZY. 1983. On the significance of Pacific intercontinental
discontinuity. Annals of the Missouri Botanical Garden
70: 577–590.
Xiang QY(J). 1987. System and synopsis of Cornus subgen.
Syncarpea (Nakai) Q. Y. Xiang (Cornaceae). Bulletin of
Botanical Research (Harbin) 7 (2): 33–52.
Xiang QY(J), Boufford DE. 2005. Cornaceae, Mastixiaceae,
Toricelliaceae, Helwingiaceae, Aucubaceae. In: Wu ZY,
Raven PH eds. Flora of China. Beijing: Science Press; St.
Louis: Missouri Botanical Garden Press. 14: 206–234.
Xiang QY(J), Brunsfeld SJ, Soltis DE, Soltis PS. 1996.
Phylogenetic relationships in Cornus based on chloroplast
DNA restriction sites: implications for biogeography and
character evolution. Systematic Botany 21: 515–534.
Xiang QY(J), Eyde RH. 1995. Chromosome number of Cornus
sessilis (Cornaceae): phylogenetic affinity and evolution
of chromosome numbers in Cornus. Sida 16: 765–768.
Xiang QY(J), Manchester SR, Thomas D, Zhang WH, Fan CZ.
2005. Phylogeny, biogeography, and molecular dating of
cornelian cherries (Cornus, Cornaceae)—tracking Tertiary
plant migration. Evolution 59: 1685–1700.
Xiang QY(J), Moody ML, Soltis DE, Fan CZ, Soltis PS. 2002.
Relationships within Cornales and circumscription of
Cornaceae—matK and rbcL sequence data and effects of
outgroups and long branches. Molecular Phylogenetics
and Evolution 24: 35–57.
Xiang QY(J), Shui Y, Murrell Z. 2003. Cornus eydeana
(Cornaceae), A new cornelian cherry from China—Notes
on Systematics and Evolution. Systematic Botany 28:
757–764.
Xiang QY(J), Soltis DE. 2001. Dispersal-Vicariance analyses
of intercontinental disjuncts: historical biogeographical
implications for angiosperms in the Northern Hemisphere.
International Journal of Plant Sciences 162: S29–39.
Xiang QY(J), Soltis DE, Morgan DR Soltis PS. 1993.
Phylogenetic relationships of Cornus L. sensu lato and
putative relatives inferred from rbcL sequence data.
Annals of the Missouri Botanical Garden 80: 723–734.
Xiang QY(J), Soltis DE, Soltis PS. 1998. Phylogenetic
relationships of Cornaceae and close relatives inferred
from matK and rbcL sequences. American Journal of
Botany 85: 285–297.
Journal of Systematics and Evolution Vol. 46 No. 3 2008 374
Xiang QY(J), Thomas DT, Zhang W, Manchester SR, Murrell
Z. 2006. Species level phylogeny of the Dogwood genus
Cornus (Cornaceae) based on molecular and
morphological evidence—implication in taxonomy and
Tertiary intercontinental migration. Taxon 55: 9–30.
Xiang QY(J), Thomas DT, Manchester MR. 2007. Phylogeny
and Biogeography of Cornales—Increased Sampling in
Taxa and Characters, Placement of Hydrostachys, BEAST
vs. Multidivtime, and DIVA vs. AreA. Abstract presented
at meeting of Botany & Plant Biology. July, 2007, Chicago.
Xiang QY(J), Thone JL, Seo TK, Zhang WH, Thomas DT,
Ricklefs RE. Rate of nucleotide substitution in Cornus L.
(Cornaceae)—A remarkable increase in the circumboreal
lineage. Molecular Phylogenetics and Evolution. (in
review, available from the first author upon request)
Yu Y-C, Linà F-J, Chang H-Y. 1999. Stepwise chromosome
evolution in Drosophila albomicans. Heredity 83: 39–45.
Zhang WH. 2006. Molecular evolution of floral homeotic
B-class genes in the dogwood genus Cornus
(Cornaceae)—gene duplication, selection, and
coevolution. Ph.D. Dissertation. North Carolina State
University. Available from http://www.lib.ncsu.edu/
theses/available/etd-10142006-234225.
Zhang WH, Xiang QY, Thomas DT, Wiegmann BM, Frohlich
MW, Soltis DE. 2008. Molecular evolution of
PISTILLATA-like genes in the dogwood genus Cornus
(Cornaceae). Molecular Phylogeny and Evolution 47:
175–195.