The Delimitation of Phylogenetic Characters

Eric S. J. Harris Using information derived from an examination of the

Osher Research Center characteristics of objects or organisms under study, all biolo-
Harvard Medical School gists, indeed all scientists, need to develop classifications of
Cambridge, MA, USA the things they study. There are four desirable criteria for clas-
eric [email protected] sifications: (1) practicality: names that are easy to apply and
stable; (2) information content: names that index an optimal
Brent D. Mishler
summarization of what is known; (3) predictivity: names that
Department of Integrative Biology
University and Jepson Herbaria
maximally predict unknown features; (4) function in theories:
University of California names that capture entities acting in, or resulting from, nat-
Berkeley, CA, USA ural processes. These criteria sometimes seem contradictory,
[email protected] in which case debates erupt between pragmatists emphasiz-
ing criterion 1 and theoreticians emphasizing criterion 4 (e.g.,
debates over chemical classification in the 1960s and 1970s;
The bud disappears in the bursting-forth of the blossom, and one debates over biological classification in the 1970s and 1980s).
might say that the former is refuted by the latter; similarly, when Ultimately, however, these criteria should not be contradictory,
the fruit appears, the blossom is shown up in its turn as a false and should flow from criterion 4 to criterion 1, in the sense
manifestation of the plant, and the fruit now emerges as the truth of it
that representing an important natural process in a classifica-
instead. These forms are not just distinguished from one another, they
tion will lead to high predictivity, information content, and true
also supplant one another as mutually incompatible. Yet at the same
time their fluid nature makes them moments of an organic unity in
practicality for users of the classification (Mishler 2009). The
which they not only do not conflict, but in which each is as necessary key to “carving nature at its joints” is to find the joints first.
as the other; and this mutual necessity alone constitutes the life of Molecular biologists need to recognize and name genes, func-
the whole. tional regions of genes, and gene products. Ecologists need to
— Hegel, Preface to Phenomenology of Spirit characterize elements of food chains or nutrient cycles. Phy-
logenetic systematists need to name monophyletic groups.
Selecting a character requires denying the continuity of
form of the organism. But there are several ways that this con-
Characters and Classification in Biology
tinuity can be denied and the selection process depends on
The main difficulty of character delimitation in biology comes what purpose the characters will be used for. These different
from the fact that it is difficult to find objective suture lines purposes can have important consequences for the resulting
along which the analytical gaze of a biologist can cut and inferences or analyses that use the characters as their basis. In
define a character (Lewontin 2001). Yet biologists need to order to make the process of character delimitation repeatable
find such sutures in order to progress with their research. The and explicit, the underlying biological and evolutionary pro-
bud is isolated from the unity of the plant and distinguishing cesses must be examined carefully, and some general proce-
features noted, the blossom is examined, and so on at various dures must be adopted based on these processes. The purpose
levels of biological organization. Character delimitation plays of this article is to address the problem of character analysis
a fundamental role in any biological research, thus the process as it is manifested in the field of phylogenetic systematics.
of “carving nature at its joints” (Plato 1997, Phaedrus 265d– Any resolution to the problem of character delimitation, not
266a) needs careful examination if it is to be part of a rational only in phylogenetics but also biology more broadly, includes
scientific inference process. both a theoretical and a practical component. The theoretical
February 17, 2009; accepted October 10, 2009
230 c 2010 Konrad Lorenz Institute for Evolution and Cognition Research
Biological Theory 4(3) 2009, 230–234. !
 Eric S. J. Harris and Brent D. Mishler

aspects of phylogenetic character coding are addressed in this on the topic (e.g., Ereshefsky 2007). Dupré (1981) echoes
article. The practical issues of phylogenetic character delimi- the sentiment expressed here in emphasizing that characters
tation are addressed in a separate paper (Harris and Mishler in can never be a full definition of a taxon, since a taxon can
preparation). never be circumscribed by necessary and sufficient conditions.
The reason for this is that taxa are evolutionary lineages. As
such, a taxon is defined as the assemblage of all things that
“Individuals” Versus “Classes”
arise from a common ancestor, even though none of those
The problem of character delimitation is not unique to biology, things necessarily share any one character in common (Hull
of course. In fact, it can be thought of as a particular instan- 1978; Grant and Kluge 2004). Hennig (1966) realized this too,
tiation of the very general problem of assigning predicates to acknowledging that “holomorphology” could only be used as
objects, of defining and describing things. The relation be- a proxy for evolutionary descent.
tween an object and its predicates remains a problematic issue However, it is very clear from the vast literature describing
in the philosophy of language. The notion of “character” has the world’s biodiversity that it is possible to use characteristics
been under discussion for many years (e.g., Wilkins 1668). to describe biological taxa and, through the use of identifica-
Traditionally, objects have been considered to be definable by tion keys, identify an unknown organism. It is important to
a set of necessary and sufficient conditions (Schwartz 1977). note that characters used for identification are not being used
The traditional view has been that the definition of a term is for the definition of a taxon. As indicated by the arguments
made through reference to the essential properties of an object, above, identifiable characteristics of an organism are inciden-
and if something exhibits those essential properties then it is tal to it as a result of its status as an “individual.” A character of
ipso facto the object specified. any nature is never an abstract, essential, defining property of
However, some philosophers have argued that even gen- an organism in the way that the atomic number of an element
eral terms are defined demonstratively through specific ref- is. The most they can be are convenient “handles” by which
erence to a particular object or group of objects, that is, by one can recognize and refer to an organism or taxon. Based on
ostension (e.g., Kripke 1980). These philosophers of language these arguments, we recognize that characters in an identifica-
have emphasized the importance of concrete spatiotemporal tion key or taxonomic description are technical instruments of
recognition of an object. In this view, objects are initially vocabulary selected for their ability to describe and differen-
referred to through demonstration. The connection between tiate biological taxa (cf. “description” and “diagnosis”; Hull
word and object is then perpetuated through a community of 1978). Phylogenetic characters serve a very different purpose,
speakers that ultimately reaches back to the object, person, or yet they are also technical instruments, selected in this case for
thing itself (Kripke). Consequently, a word is defined histor- their ability to discover phylogenetic relationships. The nature
ically through a community of speakers reaching back to the of characters, then, depends on pragmatic and utilitarian con-
original referent(s), rather than being defined by a certain set siderations as well as on theory. Their definition must always
of properties. relate to their function.
This debate relates to the distinction between individuals A confusion may arise in discerning between characters
and classes (Ghiselin 1975; Hull 1978). An individual is a used for various purposes, such as for phylogenetic analyses,
historically bounded entity, with no particular properties that identification keys, classification of ecological communities,
can define it. An individual in this sense must ultimately be and so on. The confusion results because the use of characters
referred to through ostensive definition. Type specimens take in one context, say phylogenetic analysis, may often make use
the function of ostensive definition in scientific taxonomy, and of similar characteristics as other contexts (e.g., description,
as such a taxonomic name is ultimately defined by reference to diagnosis, ecology). In general, characters of organisms can
a type specimen, not by any particular property (Hull). Ghiselin function in different ways and serve many purposes in sci-
and others (e.g., Hull) have successfully argued that biological entific research. Ultimately, the selection of various types of
species are “individuals” in this sense. By contrast, a “class” is characters for different purposes operates by different criteria
any object that conforms to a certain set of essential properties. and thus characters are not really synonymous when similar
Any one member of a class is interchangeable with any other characters are used for different purposes. For example, phylo-
member as long as they share those properties. Elements of genetic characters represent a particular subset of all possible
the periodic table are examples of a class, since to be a specific attributes or features of organisms, while traits used in an iden-
atomic element the object must always have a certain atomic tification key represent a different, though sometimes overlap-
number. ping, subset of all possible attributes or features of organisms.
A biological taxon does not have essential properties, and A common misconception about the delimitation of phy-
ultimately must be defined through ostension. This view is logenetic characters—that they represent any distinguishing
the one adopted in this article, although there remains debate feature or attribute of an organism—can thus be laid to rest.

Biological Theory 4(3) 2009 231

The Delimitation of Phylogenetic Characters

There is no theory-neutral way to individuate characters

at any level from DNA sequence data to morphology. It is
true that any particular aspect of an organism’s morphology,
anatomy, chemistry, distribution, genetics, or any of its other
peculiarities may be examined as a potential phylogenetic
character. However, many characteristics of organisms, such
as their morphological features, distribution, trophic level,
and so on, might be useful in examining the organism in
the context of its ecology or behavior, yet not be useful as
phylogenetic characters. The point is that the determination
of a phylogenetic character must always relate to the purpose
of the reconstruction of phylogeny. That is, phylogenetic
Figure 1.
characters represent a subset of attributes, features, properties
As first explained clearly by Hennig (1966), evolution occurs through descent
of an organism that are chosen for the specific purpose of with modification along lineages that split occasionally. These evolutionary
revealing the evolutionary history of organisms. Thus, in modifications can serve as evidence for the existence of a lineage in the future.
order to clarify the problem of character delimitation for As in this figure, the evolutionary modification is represented by the change
from a light color to a dark one. The purpose of delimiting phylogenetic
phylogenetics, we must first clearly understand what it is we characters is to discover modifications such as this that can serve as evidence
want these particular types of characters to do. of evolutionary relatedness.

literature remains fragmentary on the topic, addressing only

What Do We Use Phylogenetic Characters For?
parts of the problem. Frequently only the theory and philo-
Much of the literature on cladistic methodology places con- sophical basis of phylogenetic characters (e.g., Kluge 2003;
siderably more emphasis on analysis than exploration, taking Richards 2003; Grant and Kluge 2004; Fitzhugh 2006), or the
as the starting point that one has a data matrix, and discussing actual method of coding the characters (e.g., Wiens 1995) are
how to analyze it (Thiele 1993). Character coding represents covered in any one paper that deals with this topic. Obviously,
the link between observation and analysis and greatly influ- both of these two complementary aspects (theory/method) of
ences the results, but has nevertheless received little attention character coding are required for a full resolution of this prob-
(Pleijel 1995). lem. We need to understand the relation that characters have
In general, phylogenetic analysis consists of two main to phylogenetic theory and we need to have some practical
phases: character analysis and phylogenetic tree building (Neff method so that character delimitation is a repeatable process.
1986). There appears to be a common view among phyloge- The first criterion reflects the need for a clarification of the
neticists that the character analysis phase has received much theoretical component to the character definition; it is an on-
less attention as compared to the tree building step (Thiele tological criterion. The second aspect represents a procedural
1993; Pleijel 1995; Mishler 2005). Additionally, in the litera- requirement to maintain objectivity; it is an epistemological
ture on phylogenetic character analysis there is little consen- criterion. The central problem for character coding comes from
sus on methods and approaches to code phylogenetic charac- the tension between the procedural requirement of repeata-
ters. Even the definition of the word “character” remains an bility and the theoretical requirement of biological meaning-
open question (Colless 1985; Grant and Kluge 2004), despite fulness. Of course, the tension between theory and practice
many attempts to clearly define the word. By contrast, some underlies any scientific endeavor.
degree of consensus has been reached on a limited number The four desired criteria for classifications discussed
of methods of phylogenetic inference (e.g., parsimony, max- above can help relieve this tension, particularly the empha-
imum likelihood, Bayesian methods). Phylogenetic inference sis placed above on the fourth one (function in theories about
almost always uses computer algorithms, and there are only process). What theory needs to guide the specific problem of
a handful of different computer algorithms available for this delimiting characters and character states in phylogenetic sys-
purpose—e.g., PHYLIP (Felsenstein 1989), PAUP (Swofford tematics? What is the fundamental underlying process model
2003), TNT (Goloboff et al. 2000). However, no such com- that underlies phylogenetic reconstruction? As first explained
parable homogeneity of methods yet exists for the process clearly by Willi Hennig (1966), it is descent with modifica-
of character coding, since characters come in a vast array of tion along lineages that split occasionally, allowing the mod-
forms, whether they are morphological, molecular, or of any ifications to serve as evidence for the existence of a lineage
other type. in the future (Figure 1). Evolution occurs via changes along
Because of the lack of consensus regarding character branches, where a prior condition in a feature changes to a
coding in phylogenetic systematics, much of the published posterior condition. After lineage splitting, the presence of the

232 Biological Theory 4(3) 2009

 Eric S. J. Harris and Brent D. Mishler

posterior (derived, or apomorphic) condition can serve as a evolutionary lineage: phylogenetic characters are used as ev-
marker for the past lineage in which it arose. It is these mark- idence of common evolutionary history at any level of inclu-
ers that the phylogeneticist seeks to infer from the relative siveness in the tree of life. Phylogenetic characters are taken
recency of common ancestry of the members of a study group. to serve as putative evidence for the existence of monophyletic
Ontologically, these changes along lineages underlie the groups, whether that monophyletic group includes all living
important concept of homology—the general definition of things or a single individual organism, a group of cells, or a
which is two or more features that have evolved via common gene family. To do this, the character in question must vary—
descent from an ancestor that had that feature (Roth 1988). there must be at least two states (Figure 1). Certain traits may
There are several subtypes of homology, however, some of vary at one level of phylogenetic analysis, but be invariant at
which involve features duplicated within one organism (par- another. Consequently, the delimitation of characters is always
alogy and serial homology). The subtype of homology we are relative to the breadth of the particular phylogenetic study be-
most concerned with in this article is phylogenetic homology, ing undertaken. For example, the “presence of chlorophyll a &
or the sharing of homologous traits between organisms. There b” would be a useless character for a phylogenetic study of the
are two kinds of phylogenetic homology: transformational ho- genus Quercus (= oak), but may be useful in a phylogenetic
mology and taxic homology. Transformational homology is study of major lineages of eukaryotic organisms. Furthermore,
the diachronic relationship that occurs along a lineage between a certain trait of an organism may appear to be evidence for
the prior and the posterior conditions of a feature (“events,” common evolutionary history at one level in the evolutionary
according to Kluge 2007). Taxic homology is a synchronic re- tree, but a result of convergent evolution at another (e.g., the
lationship among two or more lineages that share the same presence of wings would be considered the result of conver-
condition. gent evolution at the scale of all amniotes, but might serve as a
Epistemologically, these markers serve as models for phy- homology within a smaller group such as bats and their close
logenetic character definition. The process of character analy- relatives). At the molecular level, the best comparative align-
sis involves searching for sets of conditions that appear to be ment of a gene region depends on the phylogenetic breadth
transformational homologs of each other. Each independent set of the organisms being compared. Phylogenetic characters are
of postulated transformational homologs is a character; each always related to the phylogenetic question that they provide
condition within the set is a character state. There are extensive evidence for and vice versa. Character analysis cannot be ab-
rules of inference that govern initial hypotheses of homology, solute. Any attempt at automation of the character analysis
which we will cover in another paper (Harris and Mishler in step and standardization of character terms (e.g., Pullan et al.
preparation), but for now it is just important to establish that 2005) must take this fact into account.
each character and its states are evaluated by a host of empirical We refer to a terminal unit in the cladogram (usually rep-
criteria that do not involve any particular phylogeny. resented as a row in the character matrix) as an operational tax-
These individual character hypotheses are then combined onomic unit (OTU). This is a general term that can encompass
in the form of a table—a matrix that can subsequently be an- any level of inclusiveness. An OTU could represent anything
alyzed to infer patterns of phylogenetic relationship, using a from a broadly defined group of organisms (e.g., a hypothe-
parsimony, maximum likelihood, or other criteria, a full de- sized monophyletic group representing a taxonomic family)
scription of which is beyond the scope of this article (see Wiley to the organism at a snapshot of its life (i.e., a semaphoront
et al. 1991; Kitching et al. 1998; Felsenstein 2004; Mishler sensu Hennig 1966), to a particular gene within one genome.
2009). A joint solution of the matrix is sought—what phylo- An important corollary to the above point that characters need
genetic tree best fits all the separate hypotheses of homology? to have at least two states is that in general the character should
When a “best fit” tree is chosen, the character states that are be variable between OTUs but invariant within those OTUs.
congruent in their distribution with the tree are judged to be That is, to be useful in phylogenetic reconstruction, a charac-
homologies, those that are not are judged to be homoplasies. ter should not be polymorphic within an OTU. However, the
Thus a second test of homology (congruence) is applied to actual boundaries of the OTU are not known a priori, they
the initial hypotheses of homology that were made before the must be inferred. An OTU represents a hypothesis of a his-
matrix was assembled (Patterson 1982). torical individual, a spatiotemporally restricted entity. But the
spatiotemporal boundaries of the OTU need to be tested. The
phylogenetic characters are used for this purpose.
Taxonomic Breadth and the Mutual Constitution
Therefore, characters and OTUs are related by being mu-
of OTU and Character
tually tested by one another. In the process of assembling a data
As the above section illustrates, the function of the phyloge- set during character analysis, the OTUs and character states
netic character is to serve as evidence of common evolutionary are assembled iteratively. The discovery of a new character
history. This function does not rely on any particular scale of may well split what was considered one OTU into two, or

Biological Theory 4(3) 2009 233

The Delimitation of Phylogenetic Characters

rejection of a character may lump two previously considered Ghiselin MT (1975) A radical solution to the species problem. Systematic
OTUs into one. Likewise consistency of variation of a po- Zoology 23: 536–544.
tential character within and between already well-supported Goloboff P, Farris S, Nixon K (2000) TNT (Tree Analysis Using New Tech-
nology). (BETA) Published by the authors, Tucumán, Argentina.
OTUs, as described above, is one criterion for accepting it
Grant T, Kluge AG (2004) Transformation series as an ideographic character
as a phylogenetic character. Epistemological details of this concept. Cladistics 20: 23–31.
process are given in Harris and Mishler (in preparation). For Harris ESJ, Mishler BD (in preparation) The practice of phylogenetic character
the purpose of this article it is enough to recognize that there delimitation.
is considerable reciprocal illumination between the concept Hegel GWF (1977) The Phenomenology of Spirit (Miller AV, trans). Oxford:
Oxford University Press.
of character and the concept of OTU. A character is chosen
Hennig W (1966) Phylogenetic Systematics. Chicago: University of Illinois
partly because it is invariant within an OTU and the OTUs are Press.
recognized because they are homogeneous for all the character Hull DL (1978) A matter of individuality. Philosophy of Science 45: 335–360.
states currently known (Mishler 2005). Kitching IJ, Forey PL, Humphries CJ, Williams DM (1998) Cladistics: The
Theory and Practice of Parsimony Analysis. Oxford: Oxford University
Summary Press.
Kluge AG (2003) The repugnant and the mature in phylogenetic inference:
The delimitation of phylogenetic characters is a process that Atemporal similarity and historical identity. Cladistics 19: 356–368.
fragments the continuity of the organism into comparable at- Kluge AG (2007) Completing the neo-Darwinian synthesis with an event
tributes, features, and properties that can function as potential criterion. Cladistics 23: 613–633.
indicators of evolutionary history (i.e., homologs) within the Kripke S (1980) Naming and Necessity. Cambridge, MA: Harvard University
context of a specified breadth of phylogenetic study. Taxa are Press.
Lewontin RC (2001) Foreword. In: The Character Concept in Evolutionary
individuals. Two main conclusions result from these points.
Biology (Wagner GP, ed), xvii–xxiii. San Diego: Academic Press.
First, when a phylogeneticist attempts to delimit phylogenetic Mishler BD (2005) The logic of the data matrix in phylogenetic analysis. In:
characters, those characters cannot function as essential Parsimony, Phylogeny, and Genomics (Albert VA, ed), 57–70. Oxford:
properties or elements of an exact definition of a lineage Oxford University Press.
or taxon. Rather, phylogenetic characters are specialized Mishler BD (2009) Three centuries of paradigm changes in biological classi-
fication: Is the end in sight? Taxon 58: 61–67.
technical words or other aspects (e.g., base pairs of DNA)
Neff NA (1986) A rational basis for a priori character weighting. Systematic
selected on the basis of their ability to reveal phylogenetic Zoology 35: 102–109.
relationships. Second, characters are always relative to the Patterson C (1982) Morphological characters and homology. In: Problems of
taxonomic breadth of the study being undertaken and the Phylogenetic Reconstruction (Joysey KA, Friday AE, eds), 21–74. New
terminal units in the analysis. Characters serve as putative York: Academic Press.
evidence of shared evolutionary history. Characters should Plato (1997) Complete Works (Cooper JM, Hutchinson DS, eds). Indianapolis:
Hackett.
ideally not vary within an OTU, yet vary between them. As
Pleijel F (1995) On character coding for phylogeny reconstruction. Cladistics
such, phylogenetic characters will be related to the total group 11: 309–315.
of organisms being considered in a given study as well as the Pullan MR, Armstrong KE, Paterson T, Cannon A, Kennedy JB, Watson MF,
terminal units (OTUs) whose evolutionary relationships are McDonald S, Raguenaud C (2005) The Prometheus description model:
in question. In contrast to the “real characters” sought after An examination of the taxonomic description-building process and its
representation. Taxon 54: 751–765.
by Wilkins (1668), phylogenetic characters are not absolute;
Richards R (2003) Character individuation in phylogenetic inference. Philos-
characters are always relative to their purpose, phylogenetic ophy of Science 70: 264–279.
breadth, and the operational taxonomic units in the phylogeny. Roth VL (1988) The biological basis of homology. In: Ontogeny and System-
atics (Humphries CJ, ed), 1–26. New York: Columbia University Press.
References Schwartz SP (1977) Introduction. In: Naming, Necessity, and Natural Kinds
(Schwartz SP, ed), 13–41. Ithaca: Cornell University Press.
Colless DH (1985) On “character” and related terms. Systematic Zoology 34: Swofford DL (2003) PAUP: Phylogenetic Analysis Using Parsimony and
229–233. Other Methods, Version 4. Sunderland: Sinauer.
Dupré J (1981) Natural kinds and biological taxa. Philosophical Review 90: Thiele K (1993) The holy grail of the perfect character: The cladistic treatment
66–90. of morphometric data. Cladistics 9: 275–304.
Ereshefsky M (2007) Foundational issues concerning taxa and taxon names. Wiens JJ (1995) Polymorphic characters in phylogenetic systematics. Sys-
Systematic Biology 56: 295–301. tematic Biology 44: 482–500.
Felsenstein J (1989) PHYLIP—Phylogeny inference package, Version 3.2. Wiley EO, Siegel-Causey D, Brooks DR, Funk VA (1991) The Compleat
Cladistics 5: 164–166. Cladist: A Primer of Phylogenetic Procedures. Lawrence: University of
Felsenstein J (2004) Inferring Phylogenies. Sunderland: Sinauer. Kansas Museum of Natural History.
Fitzhugh K (2006) The philosophical basis of character coding for the infer- Wilkins J (1668) An Essay Towards a Real Character, and a Philosophical
ence of phylogenetic hypotheses. Zoologica Scripta 35: 261–286. Language. London: Gellibrand and John Martyn.

234 Biological Theory 4(3) 2009