Syst. Biol.

46(3)537-553, 1997

TREES WITHIN TREES: GENES AND SPECIES,

MOLECULES AND MORPHOLOGY
JEFF J. DOYLE
L. H. Bailey Hortorium, 462 Mann Library Building, Cornell University, Ithaca, New York 14853, USA;
E-mail: jjd5@cornell.edu

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

Abstract.—The construction and interpretation of gene trees is fundamental in molecular system-
atics. If the gene is defined in a historical (coalescent) sense, there can be multiple gene trees
within the single contiguous set of nucleotides, and attempts to construct a single tree for such a
sequence must deal with homoplasy created by conflict among divergent histories. On a larger
scale, incongruence is expected among gene tree topologies at different loci of individuals within
sexually reproducing species, and it has been suggested that this discordance can be used to
delimit species. A practical concern for such topological methods is that polymorphisms may be
maintained through numerous cladogenic events; this polymorphism problem is less of a concern
for nontopological approaches to species delimitation using molecular data. Although a central
theoretical concern in molecular systematics is discordance between a given gene tree and the
true "species tree," the primary empirical problem faced in reconstructing taxic phylogeny is
incongruence among the trees inferred from different sequences. Linkage relationships limit char-
acter independence and thus have important implications for handling multiple data sets in phy-
logenetic analysis, particularly at the species level, where incongruence among different histori-
cally associated loci is expected. Gene trees can also be reconstructed for loci that influence
phenotypic characters, but there is at best a tenuous relationship between phenotypic homoplasy
and homoplasy in such gene trees. Nevertheless, expression patterns and orthology relationships
of genes involved in the expression of phenotypes can in theory provide criteria for homology
assessment of morphological characters. [Gene trees; homoplasy; linkage; morphology; orthology;
phylogeny; species delimitation.]

Page (1993, 1994) summarized several elucidated and remain to be incorporated

ways in which trees from one hierarchical
level may "include" historically associated
trees from a lower level. The trees of the
lowest hierarchical and organizational lev-
el, the gene, are discussed here. Gene trees
are fundamental to molecular systematics,
i.e., to systematics generally. The primary
use of such trees, constructed from DNA
sequence variation at individual genetic
loci, is to infer phylogenies of the taxa
bearing those genes. As such, genes are lit-
tle more than collections of individual nu-
cleotides, each assumed to be an indepen-
dent and representative character for
organismal phylogeny reconstruction. The
fact that systematists are speaking of
"gene trees" at all and distinguishing
them from trees of taxic relationships rep-
resents a significant advance over earlier—
not long past—ways of thinking about mo-
lecular systematics. The full implications of
that distinction, necessitated at each taxo-
nomic level by different but often overlap-
ping biological processes, are still being
537
into the day-to-day work of systematics.
Two aspects of "inclusion" that involve
gene trees are explored here. The first is
gene flow, which in sexually reproducing
organisms brings together lineages with
different histories. Gene trees become his-
torically associated as genes are united in
different combinations within the genomes
of individual organisms. Two hierarchical
levels are involved: the lower level is that
of the gene, and the higher level is that of
the organism. The presence of different al-
leles at a locus provides the substrate for
effective recombination. Within a contigu-
ous set of nucleotides composing a struc-
tural gene, recombination may produce
new alleles that are phylogenetic mosaics,
with significant implications for gene tree
construction. Above the organizational lev-
el of single loci, recombinational mecha-
nisms (crossing over, gene conversion, in-
dependent assortment) permit individual
loci to evolve independently and poten-
tially to track different histories. An im-
538 SYSTEMATIC BIOLOGY VOL. 46

portant consequence is the classic "gene lar systematics (e.g., Rieger, 1991) is a con-
trees versus species trees" problem, whose tiguous set of nucleotides, usually with a
empirical effect is incongruence between particular function: coxl, adh, rbcL, etc.
different loci. Although these issues have However, as has been recognized for some
received considerable attention (e.g., Mad- time (e.g., Grant, 1957) and discussed ex-
dison, 1995), their implications for species plicitly more recently (e.g., Hudson, 1990),
delimitation have been less well explored the historically relevant unit (the coales-
(but see Baum and Shaw, 1995; Doyle, cent gene [c-gene]) may be either larger or

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

1995) and are discussed here. The exis- smaller than this standard definition indi-
tence of separate trees for different parts cates. With no effective recombination, the
of a gene or for different genes within spe- c-gene can be an entire genome, as in the
cies raises the issue of how to accommo- case of organellar genomes; with recom-
date potentially conflicting data sets, an bination, it can be as small as a single nu-
area of considerable controversy (e.g., de cleotide, and multiple c-genes can there-
Queiroz et al., 1995; Miyamoto and Fitch, fore exist within a single conventional
1995). gene.
A second type of "inclusion" also takes Hudson (1990) pointed out that the ge-
place within the genomes of individual or- nealogy of a single contiguous set of nu-
ganisms but involves different organiza- cleotides may be impossible to reconstruct
with any confidence if recombination lev-
tional levels rather than different hierarchi-
cal levels. Genes underlie the morphological els are sufficiently high. In discussing cla-
attributes that are the basis for most taxo- distic analyses, Nixon and Wheeler (1992)
nomic treatments and, until the advent of noted that homoplasy caused by recombi-
macromolecular data, were the characters nation does not result from incorrect ho-
of most phylogenetic analyses. Gene trees mology assessments in the usual sense but
are "included" in morphologies, and the re- rather from conflict among the different
lationship between their topologies and trees supported by characters belonging to
morphological character state trees is ex- regions differing in their historical patterns
plored here. (see example in Fig. 1). Thus, recombina-
GENE TREES AND SPECIES TREES
tion is one case (others are hybridization
Asexually reproducing species represent and gene conversion among paralogous
a simplified case for species delimitation loci) of mixed-signal homoplasy (e.g.,
and phylogeny reconstruction (e.g., Liden, Doyle, 1996).
1990) because there is only phylogeny, not Thus the "gene" whose assumed unique
tokogeny—Pfennig's (1966) term for retic- history molecular systematists attempt to
ulate relationships among sexually repro- reconstruct may itself be historically het-
ducing individuals. In such taxa, the entire erogeneous, having several different "in-
genome is a single historical unit and is cluded" c-genes. Because history is the pri-
thus a single "gene" in terms of coales- mary concern of those reconstructing
cence theory (e.g., Hudson, 1990). This is phylogenies, it would seem that this
not the case with sexually reproducing should be the first and most significant
species. For such taxa, consequently, every process partition (Bull et al., 1993) treated
stage of molecular systematics is compli- by those concerned with data partitions
cated, from the construction of a tree for and should be more important, for exam-
alleles at a given locus, to delimiting spe- ple, than the kinds of functional criteria
cies, to inferring relationships among or- outlined by Miyamoto and Fitch (1995).
ganismal taxa. However, there seems to have been little
thought given to this issue in the debate
Gene Trees within Genes: Recombination, over methods of combining data. It falls
Incongruence, and Homoplasy under the broader heading of historical in-
A gene as defined in classical genetics, congruence, a topic that, as noted recently
molecular biology, and empirical molecu- by de Queiroz et al. (1995), has received
1997 DOYLE—TREES WITHIN TREES 539

less attention in this debate than it de- tremes lies the species. Despite the contro-
serves. The practical implications are ob- versy surrounding this hierarchical level,
vious. While advocating strict separation most concepts agree that actual, potential,
of data sets, Miyamoto and Fitch (1995:66) past, or present gene flow is of critical im-
commented that "a lower limit on the sizes portance in defining species. Thus, in at
of process partitions is effectively set by least a broad sense (and strictly speaking,
the need for sufficient numbers of charac- in the case of some species definitions), re-
ters to test statistically for significant het-lationships within species are primarily to-

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

erogeneities among these partitions." A se- kogenetic, and relationships between spe-
quence composed of c-genes that can be as cies are assumed to be phylogenetic and
small as a single nucleotide clearly poses thus represent a historical sequence of
significant challenges to those advocating cladogenic events. Whether any two indi-
data partitioning. Although few sequences viduals of species that diverged from one
within species may show this extreme a another most recently will be more closely
level of recombination, cases of recombi- related to one another than to an individ-
nation extensive enough to prohibit con- ual of a third species that diverged earlier
struction of well-resolved allele trees are is difficult to predict. This difficulty re-
known (e.g v adh in Drosophila pseudobscura: mains whether the criterion for "relation-
Schaeffer and Miller, 1992; see Doyle, ship" is overall similarity of their ge-
1995). nomes, the gene tree at any particular
locus from which species relationships
Genes and Their Trees in Species could be inferred, or the overall number of
Delimitation loci at which gene trees correctly predict
Reconstruction of a gene tree terminates the sequence of cladogenic events.
the initial stage of the molecular system- Individual sexually reproducing organ-
atics research program, the part that in- isms are not themselves species; their re-
volves objective phylogenetic analysis lationships are not phylogenetic. Consider
(Doyle, 1995). The second stage of the pro- the technical problem of reconstructing a
gram involves what I have elsewhere re- gene tree for alleles at a single locus (an
ferred to as a leap of faith (Doyle, 1992), in allele tree) for which heterozygous indi-
which the systematist substitutes the viduals exist. It is (theoretically) straight-
names of organismal taxa (species, genera, forward to construct a tree in which the
families) for the alleles or haplotypes, sam- operational taxonomic units are alleles
pled from individuals, that are the true ter- sampled from conspecific individuals that
minals of the objective analysis. For the fol- are linked by gene flow. The stage at which
lowing discussion, I assume that the many organismal taxa are substituted for genes
technical and theoretical difficulties inher- is not controversial when organellar hap-
ent in this first stage of the analysis have lotypes are sampled from a homoplasmic
been surmounted and that accurate gene individual or when alleles at a nuclear lo-
trees have been reconstructed. cus are sampled from a homozygote. But
The next question that must be ad- an individual that is heterozygous for a
dressed is whether there is a phylogenetic given locus possesses two alleles at that lo-
history of organisms and organismal taxa cus, each with its own unique historical
that can be recovered. This question seems position in the gene tree. It is therefore im-
uncontroversial with regard to taxa at, for possible to make a unique substitution of
example, the generic level and higher. "individual" for "allele" in the gene tree;
However, it is demonstrably untrue for in- such an individual must appear twice and
dividuals within sexually reproducing or- at different positions on the tree. The in-
ganisms connected by gene flow, the rela- terpretation of this situation in terms of ei-
tionships among which are tokogenetic ther individual or species relationships is
(reticulate) rather than phylogenetic (di- not apparent. Heterozygosity is a case of
vergent). Between these relatively clear ex- association among historical lineages, in
540 SYSTEMATIC BIOLOGY VOL. 46
(a) Character (b)
Allele 1 2 3 4 5 6 7 8 9 10 11 12 A B CD
m -7 8- '11
out
m-9 m -3
A 1 1 1 0 1 1 0 0 0 1 0 o i 1 m-6 -fi
m• 2 10
B 1 1 1 0 0 1 0 1 0 1 o o 1 |
"T12

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

L = 12
Cl = 1.0
Rl = 1.0

(c)
Allele 1 2 3 4 5 6 7 8 9 10 11 12

A I 1 1 0 1 1 0 0 0 1 o o T ~

1 1 0 1 1 0 0 0

(d) M 2 3 4 5 6 7 8 h
-| 9 10 11 12

A R B CD A B C R_c D
1 1-

-9
-10

L=8
r •12

4
Cl = 1.0 Cl = 1.0
Rl = 1.0 Rl = 1 0

(e)
A RacB CD A B Rac C D A RQC B CD
too

4-7 8- • •1 1 44= 7 4 - 3=|=4

1 0 8> m • -11
I I4
\

\ -3
-5 "•6 -3 6
• -2 = =10 1
-9
-1
•12 -12
L=14
Cl = 0.84 strict consensus
Rl = 0.80
1997 DOYLE—TREES WITHIN TREES 541

this case involving alleles with different the point where groups of individuals do
histories that are brought together in a sin- not share related alleles. This point of mul-
gle diploid genome. Heterozygosity is a tilocus coalescence defines the genealogi-
product of gene flow and thus is evidence cal species.
that relationships among individuals are An alternative approach makes use of
tokogenetic rather than phylogenetic. DNA sequence data in a different way, by
The use of sexually reproducing individ- using the distribution of alleles in individ-
uals as terminal taxa in a molecular phy- uals regardless of their positions in an al-

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

logenetic analysis therefore does not pro- lele tree. Individuals sharing alleles are ex-
vide even a technical solution to the pected to be close relatives if alleles are
species problem, as has been suggested defined as DNA sequences that are iden-
(Vrana and Wheeler, 1992). Some unit that tical at both coding and noncoding nucle-
includes many individuals is required, otides. But how can alleles, and by exten-
meaning that some criterion for recogniz- sion the individuals bearing them, be
ing "species" is necessary. It is reasonable grouped without reference to an allele
to ask whether gene sequences could be tree? The method of Doyle (1995) takes ad-
used to provide such a criterion, given the vantage of the fact that a heterozygous in-
current reliance of systematists on DNA dividual provides evidence of gene flow
data. between individuals that are homozygous
The most obvious approach is to some- for each of its two alleles. Alleles that co-
how utilize gene trees, the stock-in-trade exist in heterozygous form are part of a
of molecular systematics. Baum and Shaw common "allele pool" at their locus (Doyle,
(1995) proposed a method for delimiting 1995), the critical property of which is that
"genealogical" species on the basis of co- members of the same allele pool are po-
alescence of alleles or haplotypes at mul- tentially able to recombine. The group of
tiple loci. Their method operates on the as- individuals that possess alleles belonging
sumption that relationships among to a single allele pool thus are a field for
individuals within such species are toko- recombination (Carson, 1957; FFR of Doyle,
genetic and that individuals within a spe- 1995). The allele pool may include alleles
cies will therefore possess alleles having that never occur together in heterozygous
different relationships at each of many loci. form: for example, there may be no fl//het-
In the allele tree for one locus, the alleles erozygotes, but if the a/b and b/fhetero-
of individuals 1 and 2 may be sisters, zygotes join alleles a, b, and / in the same
whereas in the tree for an unlinked second allele pool, then all individuals bearing
locus, the allele possessed by individual 1 any of these alleles will belong in the same
may be more closely related to that of in- FFR. Moreover, these three alleles need not
dividual 5. Substituting individuals for al- bear any particular relationship to one an-
leles in the tree for each locus permits the other in the allele tree at this locus. Allele
construction of a strict consensus tree of / may be only distantly related to alleles a
individuals over all loci. This consensus and b, and other individuals may possess
tree will lack resolution among individuals alleles more closely related to alleles a or
that are linked by gene flow, collapsing to b. But if none of these exist in heterozy-

FIGURE 1. The effect of recombination on phylogeny reconstruction. L = length; CI = consistency index; RI

= retention index, (a) Data matrix of four ingroup (A-D) alleles and one outgroup allele prior to recombination,
with 12 binary characters numbered sequentially, (b) Allele phylogeny reconstructed from the data matrix
shows no homoplasy. (c) Recombination of alleles A and C to produce a recombinant, R^; the corresponding
RM recombinant is not followed, (d) Phylogeny reconstruction of two coalescent genes whose boundary is the
site of the recombination event between alleles A and C; neither of the two c-gene trees shows homoplasy. (e)
Parsimony analysis of the whole data set, ignoring the presence of two historical units (c-genes), results in two
trees and homoplasy. Homoplasy is due to the mixed signals of the two c-genes.
542 SYSTEMATIC BIOLOGY
gous combinations with alleles a, b, or f,
then such individuals will not belong to
the same FFR. Once allele pools are de-
fined, gene pools can be defined in much
the same way, by considering the intersec-
tions of multiple allele pools: a multilocus
genotype with allele b at locus 1 and allele
x at locus 2 links alleles a, b, and/with all
alleles belonging to the same allele pool at
locus 2 as does allele x. From the gene
pool, a multilocus FFR can then be defined
(Doyle, 1995).
These topological and nontopological
methods both rely on the fact that individ-
uals within larger entities (genealogical
species or multilocus FFR) have different
sets of reticulate (tokogenetic) relation-
ships. Each suffers from different practical
problems. Topology-based approaches
must contend with loci at which ancient
polymorphisms persist. The coalescence
points of alleles at such a locus predate
many cladogenic events leading to taxa
among which gene flow has since ceased
and that may be well differentiated mor-
phologically, ethologically, etc. Classic ex-
amples are the major histocompatibility
complex (MHC) of animals (e.g., Gaur et
al., 1992) and self-incompatibility loci of
plants (e.g., Ioerger et al., 1990; Clark and
Kao, 1991; Richman et al., 1996). Using the
MHC as an example, the ancestor in which
all human DQB1 alleles coalesce predates
the divergence of humans, chimpanzees,
gorillas, macaques, and a number of other
primates (Gaur et al., 1992). For a topolog-
ical species definition, either this large and
diverse group of taxa must be accepted as
a single "species" or the locus must be ex-
cluded from consideration in species de-
limitation. If a goal of species delimitation
is objectivity, then it would seem reason-
able to justify either decision based on
some set of criteria. Recognizing apes and
humans as a single species may be unpal-
atable from a practical point of view, but
the decision to do so does use an objective
and repeatable approach to define a natu-
ral group with particular biological prop-
erties. It is more difficult to envision prac-
tical objective criteria for excluding a locus
from consideration. One might decide to
VOL. 46
exclude MHC loci on the basis that these
loci represent a classic case of selectively
maintained polymorphism (e.g., Hughes
and Nei, 1988), for which it is expected
that polymorphisms will transcend accept-
ed specific (or even generic) boundaries.
However, it is difficult to see how this so-
lution can be applied objectively for most
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
loci, where often little or nothing is known
about mechanisms maintaining polymor-
phism. Moreover, some of the best meth-
ods for detecting selection at a locus (e.g.,
Hudson et al., 1987) rely on comparisons
of polymorphism within species versus di-
vergence between species; clearly a species
concept must be invoked and applied prior
to using such tests. It does not appear that
a practical criterion has yet been proposed
for excluding loci that does not have as a
prerequisite agreement with existing or
"reasonable" taxonomic boundaries.
In contrast, trans-specific polymorphism
is theoretically not a problem for a nonto-
pological method because topological re-
lationships are expected to persist far lon-
ger than will allele identities. Consider two
extant DQB1 alleles, one sampled from a
human and the other from a chimpanzee,
that were identical immediately prior to
the cladogenic event that gave rise to the
two taxa. These sister alleles have persisted
to the present in both taxa, having been
prevented from either becoming extinct or
going to fixation by the selectional regime
at this locus. These alleles share a more re-
cent common ancestor with each other
than either does with any other DQB1 al-
lele (except for those derived from each
subsequent to the divergence of humans
and chimpanzees), and so they will be sis-
ters on the DQB1 gene tree, more closely
related than either is to other alleles found
in humans or chimpanzees. Yet these al-
leles have been diverged for some 6 mil-
lion years, and so it is likely that they will
differ from one another by at least one sub-
stitution, perhaps in a noncoding region or
at a synonymous codon position. Despite
being sisters on the allele tree, they are not
the same allele (e.g., Rieger, 1991) and so
will not provide a link between humans
and chimpanzees. There will not be any
1997 DOYLE—TREES WITHIN TREES
heterozygous individuals having these two
alleles, because there has been no gene
flow between chimpanzees and humans
since their divergence from their common
ancestor. Theoretically, therefore, even a lo-
cus like DQB1 can be accommodated in
the nontopological method of Doyle
(1995), although low levels of allele diver-
gence coupled with parallelism could pro-
duce "shared" alleles that were not iden-
tical by descent and so would provide
misleading estimates of relationship.
The fine resolution provided by DNA se-
quences on which the nontopological ap-
proach relies may be a source of "prob-
lems" in the form of unpalatable results,
although in a direction opposite that of the
topological method. The finer the resolu-
tion, die less likely that identical alleles
will be shared among individuals. For
many taxa, the units suggested by such a
procedure may be very small, and it may
then be desirable to identify some objec-
tive means of grouping FFRs together
(Doyle, 1995).
Gene Trees and Species Phytogeny
Once units of phylogeny have been de-
limited, the topology relating them can be
inferred by the application of the nonto-
pological molecular systematics program
(Doyle, 1995). However, some major prob-
lems remain. First, not all loci are suitable
for this purpose. For example, even if the
DQB1 locus can be used to delimit human,
chimpanzee, and gorilla species, the rela-
tionships of DQB1 alleles are tangled
enough that no historical inference could
be made from them for these species. For
a locus to be suitable for phylogenetic in-
ference, a minimum requirement is that its
alleles be monophyletic within terminal
taxa (Doyle, 1995). This phylogenetic "suit-
ability" requirement can be applied objec-
tively and uniformly to any locus once
species have been delimited. Although it
involves exclusion of some loci, it is not
subject to the criticism of topological meth-
ods for species delimitation that they may
require subjective decisions concerning
which loci must be excluded to give a de-
sired result.
543
More important than this technical con-
cern is the fact that even if all alleles are
monophyletic at a locus within the species
being studied, there is no guarantee that
the overall topology at that locus will re-
flect the cladogenic history of the species
themselves. This problem is simply the
now-familiar one of gene tree versus spe-
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
cies tree. However, although the problem
is usually phrased as one of incongnience
between a particular reconstructed gene
tree and "the" species tree, the true taxon
tree is unknown and unknowable, with
few exceptions (e.g., experimental phylog-
enies; Hillis et al., 1992). Therefore, to dis-
cuss incongruence in terms of the un-
known species tree is irrelevant from an
empirical perspective; although there is
presumably a unique species history to be
recovered by constructing gene trees for
different loci, only the individual gene
trees actually can be observed. If some sig-
nificant percentage of gene trees within a
set of species is expected to disagree with
the species history (e.g., Pamilo and Nei,
1988), then it is incongruence among gene
trees that will be observed when multiple
loci are studied. This conflict must some-
how be reconciled in arriving at a species-
level phylogenetic hypothesis. The fact that
incongruence among gene trees is expect-
ed is an important aspect of the gene trees
versus species trees issue that should per-
haps be more strongly stated than has of-
ten been the case.
In their review of issues concerning
combining data, de Queiroz et al. (1995)
distinguished three approaches to the
problem of incongruent data sets. The first
approach is to produce a collection of in-
dividual histories. However, species are
more than merely collections of indepen-
dent gene lineages, and usually they do
have a singular history. Systematists are
interested in historical associations of or-
ganismal lineages, not just in gene trees
for each of the many thousands of sequenc-
es that comprise the genome. De Queiroz
et al.'s (1995) second option is to depict re-
lationships as a reticulating tree. However,
lineage sorting, which is the primary sto-
chastic source of incongruence, is not retic-
544 SYSTEMATIC BIOLOGY VOL. 46

ulation per se, unlike secondary processes is, paradoxically, a primary source of char-
such as hybridization. The final alternative acter nonindependence when multiple
is the production of a hierarchical tree data sets are used to reconstruct species
"representing the single dominant pattern histories because the unit of incongruence
among data sets" (de Queiroz et al., 1995: among loci is not the single nucleotide but
674), i.e., combining data sets or topolo- rather the entire locus or even larger
gies; whether some taxa can or should be regions of DNA that assort independently
excised (de Queiroz et al., 1995) is a sepa- (e.g., nuclear or organellar chromosomes,

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

rate issue. The entire spectrum of alterna- c-genes). Organellar genomes represent a
tive ways of dealing with multiple data familiar and relatively simple example be-
sets thus can be discussed under this third cause all of their genes are irrevocably
option. A consensus tree of gene trees linked and travel as single historical units.
from several loci in a taxonomic congru- When the organellar genome has a history
ence study is clearly a means to this end, different from that of the nuclear genome,
and if data (instead of topologies) are to be every nucleotide in every one of its genes
combined, the choice between direct com- will give the "wrong" phylogenetic pat-
bination and recoding methods (e.g., tern for those taxa affected (e.g., in lineage
Doyle, 1992) boils down to a question of sorting or introgression). Under such cir-
whether one should trust the species tree cumstances, combining organellar and nu-
supported by the most nucleotides or the clear sequences to produce a single tree in
species tree that is supported by the most a character congruence approach will pro-
loci. Both are total evidence trees, and both duce mixed-signal homoplasy, as does re-
are character congruence methods, but combination at a single locus. In the case
with different definitions of the relevant of multiple loci, larger regions of the ge-
character. nome—loci, blocks of loci, or whole chro-
Character definition is an important mosomes—rather than different regions
problem in any phylogenetic analysis, and along the same gene have different "in-
a fundamental attribute of a useful char- cluded" histories. Because of their historical
acter is independence (e.g., Davis and Hey- linkage to one another as a single c-gene,
wood, 1973). Independence normally the characters involved are not indepen-
means that the state of one character does dent; homoplasy at one predicts homopla-
not predict the state of another. Systema- sy at all others (Doyle, 1992). Numerous
tists seek to explain observed character co- cases of incongruence between organellar
variation in terms of phylogeny; other forc- and nuclear genomes are now known (re-
es or phenomena that cause the states of viewed by Wendel and Doyle, in press).
several characters to vary together pose However, incongruence among nuclear loci
problems for phylogeny reconstruction (al- that are in linkage equilibrium with one
though they are of obvious interest evolu- another should be just as much a problem
tionarily; e.g., Sanderson and Hufford, unless there is some general propensity for
1996). Natural selection is the primary organellar and nuclear sequences to be-
force leading to nonindependence, e.g., se- have differently evolutionarily (see Moore,
lection for particular amino acids in a pro- 1995; Hoelzer, 1997).
tein such as lysozyme (Messier and Stew- Some significant fraction of all neutrally
art, 1997) or selection to maintain evolving loci are expected to show patterns
secondary structure in ribosomal RNAs other than the true phylogeny and will
(Wheeler and Honeycutt, 1988). Con- therefore disagree with one another, which
straints involving function, secondary makes this topic important in the debate
structure, nearest-neighbor effects, base over data combination. This issue has re-
compositional biases, etc., are all concerns ceived much less attention than it deserves
for the construction of individual gene (as noted by de Queiroz et al., 1995) per-
trees. haps because fundamental historical in-
The historical independence of c-genes congruence of different loci is a problem
1997 DOYLE—TREES WITHIN TREES
15 2 3 4
1 2 3 4 5
Locus A allele: a b b b a "a" (open bars in Fig. 2) or type " b " (solid
Locus X allele: x x x x x bars). Suppose that all three loci yield gene
Locus V allele: y y y y y 12 3 4 5
(a)
12 3 4 5 15 2 3 4 12 3 4 5
lljll
(b) (c) (d)
FIGURE 2. The effect of incongruence due to lin-
eage sorting on different approaches to combining
data sets, (a) Species tree for five species (1-5), show-
ing the allelic composition at three loci (A, X, Y). The
common ancestor of all five species was polymorphic
for alleles a and b at locus A but fixed at loci X and
Y. Lineage sorting leads to inference of a species tree
from locus A that does not agree with the true species
tree; loci X and Y in contrast track the true species
history, (b) The strict consensus topology resulting
from the combination of topologies from the three loci
is completely unresolved, (c) Because locus A provides
three times as many phylogenetically informative
characters as do either locus X or locus Y (see text),
simultaneous unweighted parsimony analysis of se-
quences from all three gene trees results in a single
tree (consistency index [CI] = 0.71; retention index
[RI] = 0.70) that has the same topology as the locus
A tree, (d) Recoding of the three trees as ordered mul-
tistate characters, followed by parsimony analysis of
the combined characters, produces two trees (CI =
0.69; RI = 0.66) whose strict consensus topology cap-
tures the 1 + 2 versus 3-5 split of the true gene tree
but is unresolved for species 3-5.
for any method that deals with data from
multiple genes and still has as its goal the
construction of a single species tree. Each
approach makes different compromises in
dealing with this problem.
Consider three loci (A, X, Y) sampled
from a group of five species related by a
nonreticulating tree (Fig. 2). Loci X and Y
are genes of 500 nucleotides each, and lo-
cus A is a gene of 1,500 nucleotides. The
common ancestor of all five species was
fixed for allele x at locus X and for allele y
at locus Y but was polymorphic for alleles
a and b at locus A. This polymorphism was
carried through other ancestors, until fix-
545
ation occurred for either an allele of type
trees with no homoplasy and robust sup-
port for all nodes, that the number of syn-
apomorphies provided by each gene is in
direct proportion to its length, and that an
identical proportion of the total number of
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
synapomorphies supports each clade in
each gene tree. This classic example of lin-
eage sorting leads to the inference of a spe-
cies tree from locus A that does not agree
with the true species tree; it is also incon-
gruent with trees reconstructed for loci X
and Y, which track the true species history.
In this situation, a taxonomic congru-
ence approach using currently available
methods (but see Miyamoto and Fitch,
1995) would combine the topologies of all
three trees to produce a strict consensus
tree, which in this case fails to resolve a
dominant pattern among the data sets (Fig.
2b). In a character congruence approach
(e.g., Kluge, 1989), all three loci are com-
bined to form a single data set. In this case,
parsimony analysis recovers the topology
of locus A because this locus provides
more nucleotide characters than the com-
bined X and Y data sets (Fig. 2c). A dom-
inant pattern is resolved, but it is not the
pattern of the true species tree. A third al-
ternative is to treat each locus as a single
ordered multistate character whose states
are the alleles and whose transformational
order is given by the gene tree (Doyle,
1992). This approach is analogous to
Brooks parsimony analysis (e.g., Brooks,
1990) as used in biogeography and host/
parasite coevolution studies. In an analysis
of these three "characters," two trees are
identified, one with a topology identical to
that of the tree from loci X and Y and the
second with a topology found in no other
analysis (Fig. 2d). The three approaches
differ markedly in their sensitivities to dif-
ferent factors. Taxonomic congruence em-
phasizes topology, and strict consensus is
sensitive to any single locus that shows to-
pological differences, regardless of how
many loci are studied. Perhaps other con-
sensus methods (e.g., majority rule consen-
sus of the strict consensus trees from each
546 SYSTEMATIC BIOLOGY VOL. 46

locus?) would be preferable as an ap- ing separately with individual loci,

proach to taxonomic congruence, once regardless of whether a consensus or re-
enough loci were sampled. Character con- coding approach were taken.
gruence is sensitive to the number of char- Unfortunately, detailed knowledge of
acters supporting contrasting patterns, re- nuclear gene linkage relationships is gen-
gardless of the number of historically erally not available to systematists. Even
distinct linkage groups (c-genes) involved. were such information available, however,
The recoding method is intermediate, em- recombination within chromosomes would

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

phasizing individual c-genes as the units severely limit the utility of any method
of analysis but at the cost of ignoring the that assumed strict correlation between
relative amounts of the genome represent- physical and historical linkage groups.
ed by each and the relative support for in- Only in organellar genomes does it seem
dividual clades in each tree. possible to make assumptions about link-
To the extent that linkage (historical as- age of different genes or about the likeli-
sociation) is a key consideration, then none hood of their historical independence from
of these options is adequate by itself. Con- nuclear genes (e.g., Moore, 1995). No so-
sider the following simple example. A lutions are proposed here, but it is clear
group of sexually reproducing eukaryotic that any proposals about data combination
species each has a strongly bimodal karyo- should consider the fundamental issue of
type composed of four chromosomes in historical lineage association of linked
the haploid set. The single large chromo- genes.
some is considerably longer than the sum
of the smaller chromosomes, and each The Sufficiency of Gene Trees
chromosome bears a number of genetic The case has been made (e.g., Doyle,
loci proportional to its length. There is in- 1995) that gene trees are insufficient for
dependent assortment but no recombina- some purposes, such as delimiting species,
tion within chromosomes; therefore, the because they do not adequately track the
units of lineage sorting are whole chro- relationships of the organismal units with-
mosomes, each of which behaves as a sin- in which they are "included." There are
gle c-gene. Lineage sorting produces a sit- other circumstances, however, when spe-
uation in which all but one linkage group cies relationships may be in some sense ir-
tracks the correct species tree, but that relevant and gene trees are quite adequate;
linkage group, by chance, is the largest for example, in studies of organellar
chromosome. The entire genomes of the genome evolution, knowledge of organis-
species are sequenced, and trees are pro- mal phylogenies may not be necessary. Mi-
duced for each of their thousands of loci. yamoto et al. (1994) made effective use of
Because all loci on each chromosome have this situation in testing the utility of dif-
the same history, each linked locus should ferent portions of the mitochondrial ge-
potentially capture that history. Given the nome for phylogeny reconstruction. Simi-
pattern of lineage sorting described and larly, the question of whether losses of
ignoring conventional sources of homopla- sequences from plant chloroplast genomes
sy, a direct combination of all loci across represent independent molecular events
all chromosomes is likely to result in a tree (e.g., Doyle et al., 1995) can be addressed
with the topology of the tree based on the adequately with chloroplast gene trees
large chromosome, the single linkage whether or not those same trees are faith-
group that does not agree with the re- ful as species trees.
maining chromosomes or with the species Biogeographic studies use taxon trees to
tree. This example is a restatement of the reconstruct area relationships (e.g., Page,
example in Figure 2 using units larger 1993). For such studies, the problem of dis-
than the genetic locus. Treating the nucle- cordance between gene trees and taxon
otide as the unit character would cause in- trees may be irrelevant. Avise et al. (1987)
dependence problems but so would deal- coined the term intraspecific phylogeography
1997 DOYLE—TREES WITHIN TREES 547

for the study of area relationships from many morphological features, compared
gene trees. If discoverable geographic re- with the small number of states of nucle-
lationships among alleles or haplotypes otides or amino acids, is an additional ad-
exist, a gene tree will be sufficient for in- vantage of morphology in homology as-
ference of area relationships. It should sessment, but that very advantage is
make no difference from which taxa the al- compromised by the ambiguities associat-
leles or haplotypes are sampled, and dis- ed with scoring morphological characters
agreements between gene trees and spe- (e.g., Stevens, 1991). Morphological com-

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

cies trees should not be important. Of plexity has roots in the network of gene
course, different samples (loci) from the interactions that underlie all but the most
same taxa are not likely to be historically simple phenotypic characters. Is there a
independent, being "included" in the way to combine the advantageous aspects
same set of genomes. Thus, one is still con- of molecular and morphological data? Hil-
fronted with the need to construct area hy- lis and Moritz (1990:515) suggested that
potheses based on multiple unrelated taxa "molecular biology may begin to provide
and not simply based on multiple loci more information on the molecular basis of
from the same set of organisms. development, so that a true synthesis of
molecular and morphological data can oc-
GENE TREES IN MORPHOLOGY cur." If that synthesis is to occur, it will
Species histories prune and shape gene require a shift of emphasis in "molecular"
trees, and therefore gene trees can be used systematics, which currently treats all
to make inferences about species histories. genes as essentially equivalent grab-bags
This use of "included" gene histories of nucleotides and ignores their functional
spans hierarchical levels. In contrast, genes roles.
and morphological characters belong to There has been a tremendous explosion
different organizational rather than hier- of knowledge about the molecular under-
archical levels: genotype and phenotype. pinnings of morphological features and
These organizational levels are connected ontogeny in both plants and animals, and
by the products of gene expression, pro- the evolution of developmental systems is
teins and structural RNAs, from which or- under active investigation (e.g., Akam et
ganisms are constructed. To this extent, al., 1994). Forging links between system-
genes are "included" as a lower level with- atics and molecular developmental genet-
in morphological characters, and gene tree ics requires, among other things, identify-
topologies are connected to morphological ing common areas in which the techniques
character distributions by spatial and tem- of one field can be applied to the other. I
poral patterns of gene expression and reg- have argued (Doyle, 1994a) that providing
ulation. objective (i.e., phylogenetic) determina-
Although systematists generally agree tions of paralogy and orthology is a poten-
that molecules and morphology are com- tial major contribution of systematics to
plementary sources of phylogenetic infor- molecular developmental genetics. Many
mation, molecular data now provide the workers studying homeoboxes make ex-
preponderance of characters for phylogeny plicit use of these phylogenetic concepts
reconstruction. Nonmolecular characters (e.g., Schubert et al., 1993), and laboratories
should not be forgotten, as suggested by studying floral development have recently
Lanyon's (1988:572) observation that sto- taken note of the information available in
chastically evolving molecular characters phylogenetic studies (Purugganan et al.,
may be inferior to selectively maintained 1995; Tandre et al., 1995). In return, molec-
morphological features in studies of an- ular developmental genetics offers system-
cient radiations: "morphological characters atists detailed information about the
may well retain synapomorphies under connection between genes and morpholo-
circumstances under which molecular gies. Such information can be useful in a
characters would not." The complexity of variety of ways for phylogenetic issues. For
548 SYSTEMATIC BIOLOGY
example, Whiting and Wheeler (1994)
drew on the detailed understanding of
dipteran development to formulate a hy-
pothesis of the origin of halteres in Strep-
siptera, combining this information with
their own studies of insect phylogeny us-
ing 18S ribosomal genes. For phylogeny re-
construction from morphological charac-
ters, molecular developmental genetic data
are of clear relevance for the issue of char-
acter independence, most simply in cases
of pleiotropy. Mutations in several of the
genes involved in floral development are
expressed at more than one stage of de-
velopment, and some have dramatic effects
on more than one organ. Mutations in the
SUPERMAN gene of Arabidopsis, for ex-
ample, affect the number of stamens as
well as the texture of the seed coat (Gaiser
et al, 1995).
Of interest here is the general connection
between molecules and morphology in
phylogeny reconstruction, particularly the
implications of the association between
morphological character patterns and the
trees of the genes "included" in them. Ex-
pectations concerning levels of homoplasy
are central to debates about the relative
utilities of different classes of characters,
whether the classes are both morphologi-
cal or both molecular or whether mor-
phology and molecular data are being
compared. Specifically, it is worth explor-
ing how homoplasy at one level is con-
nected to homoplasy at the other.
The connection is not simple nor is it ex-
pected to be, given the complexity of the
overall connection between genotype and
phenotype (Roth, 1988, 1991). After all,
genes do not so much control morphology
as participate in it (Nijhout, 1990), and the
participation is a complex one, better de-
scribed as a network than as a linear path-
way. But the fact that the connection be-
tween genotype and phenotype is not
simple does not mean that the disconnec-
tion in homoplasy between the two levels
cannot be illustrated by simple examples
(e.g., Doyle, 1996). Phenotypes are pro-
duced as an expression of genotype, and
morphology is only one of several mani-
festations of gene expression. Mobility of a
VOL. 46
native protein in an electric field is also a
phenotypic character, albeit a much more
simple one than morphology. Mobility dif-
ferences among proteins encoded by or-
thologous loci (allozymes) are due to dif-
ferences in net charge under particular
electrophoretic conditions. Charge in turn
is determined by the primary amino acid
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
sequences and thus ultimately by the nu-
cleotide sequences of the genes encoding
the proteins. The character "electrophoretic
mobility" could show homoplasy as a re-
sult of parallel changes in net charge. This
homoplasy could be due to a parallel non-
synonymous nucleotide substitution, in
which case there would be a one-to-one
correspondence in homoplasy at the nucle-
otide, amino acid, and phenotypic (mobil-
ity) levels. But the same false homology for
mobility could be caused by different non-
synonymous substitutions at the same co-
don, resulting in identical but nonho-
mologous amino acids, or even by
substitutions involving completely differ-
ent amino acids. In these cases (illustrated
by Doyle, 1996), homoplasy occurs in the
phenotype (mobility) but not at the level of
the gene. Conversely, because relatively
few nonsynonymous nucleotide substitu-
tions will result in net charge changes,
many homoplastic replacement substitu-
tions that would be detected at either the
nucleotide or amino acid levels will not ap-
pear as homoplasies for the phenotypic
character "mobility." Of course, the most
dramatic—and familiar—disconnection
between organizational levels occurs be-
tween the nucleotide and amino acid levels
because of the expectation that synony-
mous changes will generally far outnum-
ber nonsynonymous substitutions.
These considerations presumably apply
for any gene whose tree one might wish
to compare with the phylogenetic distri-
bution of a phenotypic character it under-
lies. With morphological characters, the
connection is still more tenuous, however.
Gene trees are usually constructed from
coding regions, yet sequences critical for
the spatial and temporal expression of the
gene are generally located outside this re-
gion, and even subtle changes in expres-
1997 DOYLE—TREES WITHIN TREES
sion patterns can produce profound effects
on the phenotype. AH of these factors rep-
resent disconnections between single
genes and morphological characters. Be-
yond this level lie the further sources of
disconnection caused by interactions
among several genes (for examples, see
Doyle, 1996). Weston (1987) suggested that
biosynthetic pathways can serve as useful
models for the more complex network of
ontogenetic interactions (cf. Alberch, 1985).
An example of the disconnection between
gene tree homoplasy and homoplasy in-
volving micromolecular characters is dis-
ruption of a pathway at two different steps
in two unrelated taxa, leading to a parallel
loss of a particular compound. There need
be no homoplasy at all in the trees for any
of the pathway's genes, yet homoplasy will
be present for the chemical character.
A consequence of the disconnection be-
tween homoplasy at the two organizational
levels is that it defies facile characterization
in the "molecules versus morphology" de-
bate One simplistic view that may subcon-
sciously pervade perceptions of the connec-
tion between genotype and phenotype in
phylogeny reconstruction is suggested by
the following statement (Doyle, 1990:6):
morphological characters . . . are likely to be largely
determined by genes undergoing the same complex
evolutionary processes as those from which we at-
tempt to construct our molecular phylogenies, and
in addition are likely to show further layers of com-
plication from . . . a host of other processes.
This statement implies a belief that mor-
phological characters must somehow sum
the homoplasy of the genes that influence
them. If four genes affect a particular mor-
phological feature, each gene is expected to
show some homoplasy, which may lead to
uncertainty in its gene tree and, conse-
quently, in the inference of taxic phylogeny
made from it. In this naive view, the mor-
phological character would be much more
(perhaps four times more) unreliable than
any of the genes that influence its expres-
sion. The above examples show that this
will rarely, if ever, be the case.
Knowledge of the genotypic basis of
phenotypic characters should be helpful in
providing new criteria for homology as-
549
sessment. Genie similarity extends the
usual concept of structural similarity: are
the same genes involved in a character hy-
pothesized to be homologous? Moreover,
are these genes expressed in similar ways,
spatially and temporally, in different taxa?
It might not be apparent that application
of these criteria involves "included" gene
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
trees. Certainly, it may be trivial to deter-
mine that two genes are not the same (non-
homologous) if, for example, they show
virtually no nucleotide similarity above
random levels. But homology is certainly
(if not exclusively) a phylogenetic concept
(e.g., de Pinna, 1991), and relationships
among genes are often complex, in many
cases caused by successive waves of du-
plication and divergence that are best de-
picted and understood by hierarchical re-
lationships, i.e., gene trees. Ultimately,
what is meant by "the same" gene is
"identical alleles at orthologous loci," al-
though in studies at deeper levels it may
be sufficient to focus on orthology relation-
ships alone.
To illustrate the application of gene trees
to assessment of homology in a phenotypic
character, consider a unique morphological
structure, found in relatively few taxa of
some larger group. A phylogenetic analy-
sis of these taxa suggests that the structure
is not homologous in all of the taxa but
rather has arisen more than once in the
group. Obvious tests of this hypothesis
would include anatomical and ultrastruc-
tural studies. One could also ask how the
features that make this structure unique at
the morphological level develop at the mo-
lecular level. Are these unique features the
result of the action of orthologous genes
expressed in similar ways in the structures
of all taxa? The nodule of leguminous
plants provides an example of such a char-
acter. Symbiotic fixation of atmospheric ni-
trogen by leguminous plants and soil bac-
teria (nodulation) is most parsimoniously
hypothesized to have arisen independently
in at least two, and perhaps three, groups
of Leguminosae (Doyle, 1994b; Doyle et al.,
1997). The nodules that house the bacteria
are considered novel plant organs, and
their homologies with other organs are a
550 SYSTEMATIC BIOLOGY
matter of debate (Hirsch and LaRue, in
press). Within nodules, bacteria in their ni-
trogen-fixing stage (bacteroids) are sur-
rounded by a complex membrane of host
origin, forming what has been called a
novel organelle, the symbiosome (reviewed
by Mylona et al., 1995). Nodules show con-
siderable morphological, developmental,
and biochemical diversity in the Legumi-
nosae but, based on positional and struc-
tural criteria, appear to be homologous
throughout the family. The apparent ho-
mology of nodules in diverse legumes
combined with the complexity of the nod-
ulation syndrome make it perhaps more
plausible that nodulation arose only once
in the family but was lost independently
in many lineages.
Additional criteria for assessing nodule
homologies would therefore be useful as
tests of the competing hypotheses of single
versus multiple origins, and such criteria
could potentially be provided by genes en-
coding nodulins, proteins that perform a
variety of enzymatic and structural roles in
the nodule (Mylona et al., 1995). Nodulins
could be products of truly novel genes,
produced by exon shuffling or by recent
gene duplication from genes with other
roles in the plant; alternatively, they could
be recruited for function in the nodule
while retaining their normal functions
elsewhere in the plant. In principle, the ho-
mology tests proposed are simple: con-
struct gene trees for gene families that in-
clude nodule-expressed genes and
determine the orthology relationships of
nodulin genes across taxa (Fig. 3; Doyle,
1994b). Finding similar roles performed by
paralogous genes would constitute a fal-
sification of the hypothesis of nodule ho-
mology because it would suggest indepen-
dent gene origin or recruitment. However,
finding orthologous genes performing
similar roles in all taxa would be consis-
tent either with nodule homology (a single
gene origin or recruitment) or nonhomol-
ogy (e.g., independent recruitment of or-
thologous genes).
In practice, the situation is much more
complex. For one thing, some gene families
evolve concertedly, obscuring orthology
VOL. 46
taxon: 1 2 3 4 5 6 7
(a)
nodule expression
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
(b)
nodule expression
1 2 3 4 5 6 7
B B B B B B B
FIGURE 3. Using gene trees and expression pat-
terns of genes "included" in a morphological charac-
ter (nodules) to assess homologies of that character,
(a) Species tree for seven taxa; taxa 1, 6, and 7 produce
nodules, and the most-parsimonious optimization of
the nodule character is a parallel gain (solid bars), (b,
c) Gene trees for two loci, one of whose paralogous
genes encodes a product expressed in the nodule. In
(b), paralogous genes appear to have been recruited
for a role in nodulation in taxon 1 and in the ancestor
of taxa 6 and 7. This situation is consistent with a
separate recruitment of genes in the course of a par-
allel origin of nodulation in these taxa. In (c), orthol-
ogous genes are expressed in the nodules of these
taxa. This situation is consistent with either a single
origin of nodulation (with loss of the trait in taxa 2-
5) or parallel recruitments during separate origins.
relationships and making the test impos-
sible to perform; one of the most abundant
nodulins, leghemoglobin, is encoded by
such a family (e.gv Doyle, 1994b). Even in
gene families where orthology relation-
ships can be hypothesized, patterns of
gene expression may be very complex. In
the cytosolic glutamine synthetase family,
for example, some species include consti-
tutively expressed genes with low level
nodule expression, as well as paralogous
genes that are dramatically up-regulated
in the nodule but also are expressed in
some other tissues (e.g., Temple et al., 1995;
Doyle and Doyle, 1997). Shifts in nodulin
gene expression seem likely to have oc-
curred, given the structural and biochem-
1997 DOYLE—TREES WITHIN TREES
ical diversity that exists even among nod-
ules of taxa sharing recent common
ancestors. Such shifts could result in mis-
leading conclusions: paralogous genes
could be expressed identically in homolo-
gous nodules, or orthologous genes could
encode nodulins in nonhomologous nod-
ules.
Because the tests are not foolproof, re-
sults from any given nodulin gene family
may not be reliable, and therefore infor-
mation from several different gene families
will be required to produce a robust hy-
pothesis of nodule homology. This is, of
course, precisely the same situation that
exists in other studies that rely on "includ-
ed" trees. The area cladogram for any par-
ticular taxonomic group may be mislead-
ing for a variety of reasons; only after
several such cladograms are available are
conclusions considered well supported.
The gene tree for any particular gene may
not track organismal histories accurately;
several independent gene trees are re-
quired for a robust hypothesis of species
relationships.
COMBINING DATA FROM "INCLUDED"
TREES: SOME FINAL COMMENTS
The typical solution to all of these prob-
lems is to obtain data from additional
genes or taxa whose "included" trees are
being reconstructed. How many genes or
taxa and how to utilize that information
are open and controversial questions. The
gene/species tree issue can be treated as a
direct parallel to biogeography: the goal in
phylogenetic reconstruction is to infer spe-
cies relationships from "included" gene
trees, whereas in biogeographic studies the
goal is to infer area relationships from
trees of "included" taxa. Both types of
studies are conducted under the assump-
tion that a direct inference of relationships
at one level can be made from those at the
other, an assumption that in both cases is
expected to be invalid under a number of
circumstances. For molecular systematics,
problems of historical independence (bear-
ing on the very definition of the gene) be-
devil the issue of how to combine data at
the level of closely related species, where
551
the greatest potential for incongruence ap-
pears to lie. The definition of species is
critical in such studies because delimita-
tion of species to a considerable degree
sets the threshold for expected incongru-
ence among gene trees. Nontopological al-
ternatives exist at various stages in the
analysis of species using molecular data;
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
these alternatives do not seem to be avail-
able for biogeography.
The issue of combining data takes quite
a different form when the final class of "in-
cluded" trees discussed here is considered.
I have argued that topologies of "includ-
ed" gene trees may have little relationship
to gene expression and hence to phenotyp-
ic characters. In the discussion of testing
homologies of morphological characters
with gene trees, the question to be an-
swered requires a simple "yes" or "no":
structures are either homologous or they
are not. Correspondingly, the information
provided by each "included" gene tree is
also binary: are orthologous genes per-
forming a similar role in the taxa under
consideration? Combining data to address
the overall homology question takes the
form of a simple tally of "yes" and "no"
answers, one for each locus considered. Of
course, homologous morphological char-
acters exist as transformation series. Once
characters are shown to be homologous,
the topologies of "included" trees of an or-
thologous gene could perhaps be useful in
ordering such a series. The disconnection
between gene trees and phenotype sug-
gests that this ordering could often be dif-
ficult, but in at least some cases it may be
possible (e.g., Rogers et al., 1997); how
trees from the various genes involved in a
morphological character could be com-
bined for this purpose is not immediately
obvious.
If there is a single conclusion from all of
this, it is that the business of inferring re-
lationships at one hierarchical or organi-
zational level using trees from another lev-
el is a tricky one because of the
disconnections that exist between the lev-
els. Nevertheless, as the history of system-
atics over the past decade has shown, gene
trees represent a vast source of information
552 SYSTEMATIC BIOLOGY VOL. 46

that can be exploited to address a host of DE QUEIROZ, A., M. J. DONOGHUE, AND J. KlM. 1995.
phylogenetic questions. The disconnec- Separate versus combined analysis of phylogenetic
evidence. Annu. Rev. Ecol. Syst. 26:657-681.
tions between genes and species are reason DOYLE, J. J. 1990. Plant systematics at the DNA level:
for caution, not despair. Similarly, despite Promises and pitfalls. Int. Organ. Plant Biosyst.
the disconnections between genes and Newsl. 8:3-7.
morphological characters, there is hope DOYLE, J. J. 1992. Gene trees and species trees: Mo-
that the marriage of molecular develop- lecular systematics as one-character taxonomy. Syst.
Bot. 17:144-163.
mental genetics and systematics can be a DOYLE, J. J. 1994a. Evolution of a plant homeotic mul-

Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021

productive one. tigene family: Toward connecting molecular sys-
ACKNOWLEDGMENTS
I thank Mark Siddall and Richard O'Grady for the
invitation to participate in the 1995 SSB symposium
that led to this article. I am grateful for many pro-
ductive discussions with my colleagues in the Bailey
Hortorium, especially Jerry Davis. I thank those who
read drafts of the manuscript when it was first writ-
ten, long ago, and also Dick Olmstead, Dave Canna-
tella, and two anonymous reviewers for comments
during its final stages. My research has been sup-
ported by NSE most recently by grant DEB 94-20215.
REFERENCES
AKAM, M., P. HOLLAND, P. INGHAM, AND G. WRAY
(eds.). 1994. The evolution of developmental mech-
anisms. Company of Biologists, Cambridge, En-
gland.
ALBERCH, P. 1985. Problems with the interpretation of
developmental sequences. Syst. Zool. 34:46-58.
AVISE, J. C, J. ARNOLD, R. M. BALL, E. BERMINGHAM,
T. LAMB, AND J. E. NEIGEL. 1987. Intraspecific phy-
logeography: The mitochondrial DNA bridge be-
tween population genetics and systematics. Annu.
Rev. Ecol. Syst. 18:489-522.
BAUM, D. A., AND K. L. SHAW. 1995. Genealogical per-
spectives on the species problem. Pages 289-303 in Ex-
perimental and molecular approaches to plant biosys-
tematics (P. C. Hoch, A. G. Stevenson, and B. A. Schaal,
eds.). Missouri Botanical Garden, St. Louis.
BROOKS, D. R. 1990. Parsimony analysis in historical
biogeography and coevolution: Methodological and
theoretical update. Syst. Zool. 39:14-30.
BULL, J. J., J. P. HUELSENBECK, C. W. CUNNINGHAM, D.
L. SWOFFORD, AND P. J. WADDELL. 1993. Partition-
ing and combining data in phylogenetic analysis.
Syst. Biol. 42:384-387.
CARSON, H. L. 1957. The species as a field for recom-
bination. Pages 23-38 in The species problem (E.
Mayr, ed.). American Association for the Advance-
ment of Science Press, Washington, D.C.
CLARK, A. G., AND T. H. KAO. 1991. Excess nonsyn-
onymous substitution at shared polymorphic sites
among self-incompatibility alleles of Solanaceae.
Proc. Natl. Acad. Sci. USA 88:9823-9827.
DAVIS, P. H., AND V. H. HEYWOOD. 1973. Principles of
angiosperm taxonomy. Krieger, Huntington, New
York.
DE PINNA, M. C. C. 1991. Concepts and tests of ho-
mology in the cladistic paradigm. Cladistics 7:367-
394.
tematics and molecular developmental genetics.
Syst. Biol. 43:307-328.
DOYLE, J. J. 1994b. Phytogeny of the legume family:
An approach to understanding the origins of nod-
ulation. Annu. Rev. Ecol. Syst. 25:325-349.
DOYLE, J. J. 1995. The irrelevance of allele tree topol-
ogies for species delimitation, and a non-topological
alternative. Syst. Bot. 20:574-588.
DOYLE, J. J. 1996. Homoplasy connections and discon-
nections: Genes and species, molecules and mor-
phology. Pages 37-66 in Homoplasy and the evo-
lutionary process (M. J. Sanderson and L. Hufford,
eds.). Academic Press, New York.
DOYLE, J. J., AND J. L. DOYLE. 1997. Phylogenetic per-
spectives on the origins and evolution of nodulation in
the legumes and allies. NATO ASI Ser. 39:307-312.
DOYLE, J. J., J. L. DOYLE, J. A. BALLENGER, E. E. DICK-
SON, T. KAIITA, AND H. OHASHI. 1997. A phytogeny
of the chloroplast gene rbcL in the Leguminosae:
Taxonomic correlations and insights into the evolu-
tion of nodulation. Am. J. Bot. 84:541-554.
DOYLE, J. J., J. L. DOYLE, AND J. D. PALMER. 1995. Multiple
independent losses of two genes and one intron from
legume chloroplast genomes. Syst. Bot. 20:272-294.
GAISER, J. C, K. ROBINSON-BEERS, AND C. S. GASSER.
1995. The Ambidopsis SUPERMAN gene mediates
asymmetric growth of the outer integument of
ovules. Plant Cell 7:333-345.
GAUR, L. K., A. L. HUGHES, E. R. HEISE, AND J. GUT-
KNECHT. 1992. Maintenance of DQB1 polymor-
phisms in primates. Mol. Biol. Evol. 9:599-609.
GRANT, V. 1957. The plant species in theory and prac-
tice. Pages 39-80 in The species problem (E. Mayr,
ed.). American Association for the Advancement of
Science Press, Washington, D.C.
HENNIG, W. 1966. Phylogenetic systematics. Univ. Il-
linois Press, Urbana.
HILLIS, D. M., J. J. BULL, M. E. WHITE, M. R. BADGETT,
AND I. J. MOLINEUX. 1992. Experimental phyloge-
netics: Generation of a known phytogeny. Science
255:589-592.
HiLLUS, D. M., AND C. MORrrz. 1990. An overview of
applications of molecular systematics. Pages 502-515
in Molecular systematics, 1st edition (D. M. Hillis and
C. Moritz, eds.). Sinauer, Sunderland, Massachusetts.
HIRSCH, A. M., AND T. A. LARUE. In press. Is the
legume nodule a modified root or stem or an organ
sui generis? Crit. Rev. Plant Sci.
HOELZER, G. A. 1997. Inferring phytogenies from
mtDNA variation: Mitochondrial gene-trees versus
nuclear gene-trees revisited. Evolution 51:622-626.
1997 DOYLE—TREES WITHIN TREES
HUDSON, R. R. 1990. Gene genealogies and the co-
alescent process. Oxf. Surv. Evol. Biol. 7:1-44.
HUDSON, R. R., M. KREITMAN, AND M. AGUADE. 1987.
A test of neutral molecular evolution based on nu-
cleotide data. Genetics 116:153-160.
HUGHES, A. L., AND M. NEI. 1988. Pattern of nucleo-
tide substitution at major histocompatility complex
Class I loci reveals overdominant selection. Nature
335:167-170.
IOERGER, T. R., A. G. CLARK, AND T. H. KAO. 1990.
Polymorphism at the self-incompatibility locus of
Solanaceae predates speciation. Proc. Natl. Acad.
Sci. USA 87:9732-9735.
KLUGE, A. G. 1989. A concern for evidence and a phy-
logenetic hypothesis of relationships among Epicra-
tes (Boidae, Serpentes). Syst. Zool. 38:7-25.
LANYON, S. M. 1988. The stochastic mode of molec-
ular evolution: What consequences for systematic
investigations? Auk 105:563-573.
LIDEN, M. 1990. Replicators, hierarchy, and the spe-
cies problem. Cladistics 6:183-186.
MADDISON, W. P. 1995. Phylogenetic histories within
and among species. Pages 273-287 in Experimental
and molecular approaches to plant biosystematics
(P. C. Hoch, A. G. Stevenson, and B. A. Schaal, eds.).
Missouri Botanical Garden, St. Louis.
MESSIER, W, AND C.-B. STEWART. 1997. Episodic adap-
tive evolution of primate lysozymes. Nature 385:151-
154.
MIYAMOTO, M. M., M. W. ALLARD, R. M. ADKINS, L.
L. JANECEK, AND R. L. HONEYCUTT. 1994. A con-
gruence test of reliability using linked mitochon-
drial DNA sequences. Syst. Biol. 43:236-249.
MIYAMOTO, M. M., AND W. M. FITCH. 1995. Testing
species phylogenies and phylogenetic methods with
congruence. Syst. Biol. 44:64-76.
MOORE, W. S. 1995. Inferring phylogenies from
mtDNA variation: Mitochondrial-gene trees versus
nuclear-gene trees. Evolution 49:718-726.
MYLONA, P., K. PAWLOWSKI, AND T. BISSELING. 1995.
Symbiotic nitrogen fixation. Plant Cell 7:869-885.
NIJHOUT, H. F. 1990. Metaphors and the role of genes
in development. BioEssays 12:441-446.
NIXON, K. C , AND Q. D. WHEELER. 1992. Extinction
and the origin of species. Pages 119-143 in Extinc-
tion and phytogeny (M. J. Novacek and Q. D. Whee-
ler, eds.). Columbia Univ. Press, New York.
PAGE, R. D. M. 1993. Genes, organisms, and areas: The
problem of multiple lineages. Syst. Biol. 42:77-84
PAGE, R. D. M. 1994. Maps between trees and cladis-
tic analysis of historical associations among genes,
organisms, and areas. Syst. Biol. 43:58-77.
PAMILO, P., AND M. NEL 1988. Relationships between
gene trees and species trees. Mol. Biol. Evol. 5:568-583.
PURUGGANAN, M . D., S. D. ROUNSLEY, R. J. SCHMIDT, AND
M. F. YANOFSKY. 1995. Molecular evolution of flower
553
development: Diversification of the plant MADS-box
regulatory gene family. Genetics 140:345-356.
RICHMAN, A. D., M. K. UYENOYAMA, AND J. R. KOHN.
1996. Allelic diversity and gene genealogy at the
self-incompatibility locus in the Solanaceae. Science
273:1212-1216.
RIEGER, R. 1991. Glossary of genetics: Classical and
molecular. Springer-Verlag, New York.
ROGERS, B. T., M. D. PETERSON, AND T. C. KAUFMAN.
1997. Evolution of the insect body plan as revealed
Downloaded from https://academic.oup.com/sysbio/article/46/3/537/1651372 by guest on 16 September 2021
by the Sex combs reduced expression pattern. De-
velopment 124:149-157.
ROTH, V. L. 1988. The biological basis of homology.
Pages 1-26 in Ontogeny and systematics (C. J. Hum-
phries, ed.). Columbia Univ. Press, New York.
ROTH, V. L. 1991. Homology and hierarchies: Prob-
lems solved and unresolved. J. Evol. Biol. 4:167-194.
SANDERSON, M. J., AND L. HUFFORD (eds.). 1996. Ho-
moplasy and the evolutionary process. Academic
Press, New York.
SCHAEFFER, S. W, AND E. L. MILLER. 1992. Estimates
of gene flow in Drosophila psendoobscura determined
from nucleotide sequence analysis of the alcohol de-
hydrogenase region. Genetics 132:471-480.
SCHUBERT, F. R., K. NIELSELT-STRUWE, AND P. GRUSS.
1993. The Antennapaedia-type homeobox genes have
evolved from three precursors separated early in
metazoan evolution. Proc. Natl. Acad. Sci. USA 90:931-
936.
STEVENS, P. F. 1991. Character states, morphological
variation, and phylogenetic analysis: A review. Syst.
Bot. 16:553-583.
TANDRE, K., V. A. ALBERT, A. SUNDAS, AND P.
ENGSTROM. 1995. Conifer homologues to genes that
control floral development in angiosperms. Plant
Mol. Biol. 27:69-78.
TEMPLE, S. J., J. HEARD, G. GANTER, K. DUNN, AND C.
SENGUPTA-GOPALAN. 1995. Characterization of a nod-
ule-enhanced gluatmine synthetase from alfalfa: Nu-
cleotide sequence, in situ localization, and transcript
analysis. Mol. Plant-Microbe Interact. 8:218-227
VRANA, P., AND W. WHEELER. 1992. Individual organ-
isms as terminal entities: Laying the species prob-
lem to rest. Cladistics 8:67-72.
WENDEL, J. F, AND J. J. DOYLE. In press. Phylogenetic
incongruence: Window into genome history and
molecular evolution. In Molecular systematics of
plants, 2nd edition (D. E. Soltis, P. S. Soltis, and J. J.
Doyle, eds.). Chapman & Hall, New York.
WESTON, P. H. 1987. Indirect and direct methods in sys-
tematics. Pages 27-56 in Ontogeny and systematics (C.
J. Humphries, ed.). Columbia Univ. Press, New York.
WHEELER, W. C, AND R. L. HONEYCUTT. 1988. Paired
sequence difference in ribosomal RNAs: Evolutionary
and phylogenetic implication. Mol. Biol. Evol. 4:90-96.
WHITING, M. F., AND W. C. WHEELER. 1994. Insect
homeotic transformation. Nature 368:696.
Received 24 September 1996; accepted 3 January 1997
Associate Editor: Richard Olmstead