Cladistics and taxonomy

Taxonomy — the science of naming, describing, and classifying organisms — is one of the oldest disciplines in biology. For more than two centuries, the framework erected by Carl Linnaeus in his Systema Naturae (1735, with the pivotal tenth edition in 1758) provided the standard architecture: a nested hierarchy of kingdoms, classes, orders, genera, and species, each organism assigned a binomial name composed of genus and species.¹ Linnaeus intended this hierarchy to reflect the order of divine creation, but with the publication of Darwin's On the Origin of Species in 1859 and the subsequent development of evolutionary theory, biologists gradually reinterpreted the Linnaean hierarchy as a reflection of genealogical descent. The enterprise of classifying organisms became inseparable from the enterprise of reconstructing their evolutionary history. It was not until the mid-twentieth century, however, that a rigorous methodology for making classification explicitly genealogical was formalized — a methodology now known as cladistics.

The Linnaean system and its legacy

Linnaeus's contribution was not the invention of biological classification, which had existed since Aristotle, but the standardization of a hierarchical system of ranks and the consistent use of binomial nomenclature. Every described species received a two-part Latin name — Homo sapiens, Canis lupus, Quercus robur — and was nested within a genus, which was grouped into an order, which was grouped into a class, and so on. The tenth edition of Systema Naturae (1758) is the starting point for zoological nomenclature by international convention, and botanists similarly trace formal nomenclature to Linnaeus's Species Plantarum (1753).¹ This system proved remarkably durable; its basic structure still organizes the International Code of Zoological Nomenclature and the International Code of Nomenclature for algae, fungi, and plants.

Yet the Linnaean system was designed to catalog perceived natural kinds, not evolutionary lineages. Linnaeus grouped organisms by overall morphological similarity — organisms that looked alike were placed together, and those that looked different were placed apart. This approach worked well enough for many purposes, but it had no principled way of distinguishing between similarity due to shared ancestry and similarity due to convergent evolution. Dolphins and sharks share a streamlined body form, but no post-Darwinian biologist would classify them together. As evolutionary thinking took hold in the late nineteenth century, taxonomists increasingly sought to make their classifications "natural" in the sense of reflecting genealogy, but the methods for doing so remained informal and often contradictory.¹⁷

The problem was compounded by the arbitrariness of ranks. There is no objective criterion for deciding what constitutes a family versus an order; the rank assignments in one group of organisms are not comparable to those in another. A family of beetles and a family of birds are both called "families," but they differ enormously in age, species richness, and degree of morphological divergence. Ernst Mayr's biological species concept, articulated in 1942 in Systematics and the Origin of Species, provided a rigorous definition for the species rank by defining species as groups of interbreeding populations reproductively isolated from other such groups.^{17, 25} But no comparable criterion existed for genera, families, or orders. The result was a classification system whose lowest unit had a biological definition but whose higher structure was largely a matter of convention and tradition.

Willi Hennig and the cladistic revolution

The decisive break came with the work of the German entomologist Willi Hennig. In his 1950 monograph Grundzüge einer Theorie der phylogenetischen Systematik (Foundations of a Theory of Phylogenetic Systematics), Hennig articulated a set of principles that, for the first time, provided an explicit and repeatable methodology for inferring evolutionary relationships from observed characteristics.² His ideas remained largely unknown outside the German-speaking world until the publication of an English translation, Phylogenetic Systematics, in 1966.³ The impact of the English edition was transformative: within two decades, Hennigian cladistics had become the dominant approach to systematics in both zoology and botany.

Hennig's central insight was that not all shared characters are equally informative about evolutionary relationships. He distinguished three types of character similarity. Symplesiomorphies are shared ancestral (primitive) characters — features inherited unchanged from a distant ancestor and therefore shared by a broad range of descendants. Having a vertebral column, for example, is a symplesiomorphy for all vertebrates; it tells us nothing about which vertebrates are most closely related to one another, because it was present in their common ancestor. Synapomorphies are shared derived characters — features that evolved in the common ancestor of a particular group and were inherited by its descendants but are absent in more distantly related lineages. The possession of feathers is a synapomorphy uniting birds with their closest theropod dinosaur relatives. Autapomorphies are derived characters unique to a single lineage and therefore uninformative about relationships with other groups — the trunk of an elephant, for instance, is an autapomorphy of elephants.^{2, 3}

The methodological consequence of this distinction is profound. Hennig argued that only synapomorphies — shared derived characters — provide evidence for grouping organisms into clades (monophyletic groups). Shared primitive characters, no matter how conspicuous, are phylogenetically uninformative for resolving relationships within a group because they were already present before the group diversified. This principle gave systematists, for the first time, a clear rule for deciding which similarities count as evidence of close relationship and which do not.³

Monophyly, paraphyly, and polyphyly

Hennig imposed a strict requirement on taxonomic groups: every valid group must be monophyletic, meaning it includes an ancestor and all of its descendants. A monophyletic group, or clade, is defined by at least one synapomorphy — a shared derived character inherited from the group's most recent common ancestor. Mammals, for instance, form a monophyletic group because they descend from a single common ancestor and share derived features such as mammary glands, three middle ear bones, and a neocortex.^{3, 19}

A paraphyletic group, by contrast, includes an ancestor and some but not all of its descendants. The classic example is "Reptilia" as traditionally defined: it included turtles, lizards, snakes, tuataras, and crocodilians but excluded birds, even though birds evolved from within the archosaur lineage that also produced crocodilians. Cladistically, birds are nested within Reptilia, and excluding them renders the group paraphyletic — defined not by what its members share but by what they lack (the derived features of birds). Hennig argued that paraphyletic groups are artifacts of classification that obscure rather than reveal evolutionary history.^{3, 19}

A polyphyletic group is one whose members do not share an exclusive common ancestor — the group is assembled by convergence rather than genealogy. The old grouping "Pachydermata," which lumped elephants, rhinoceroses, and hippopotamuses together on the basis of thick skin, is polyphyletic because these animals descend from separate mammalian lineages and their thick skin evolved independently. No modern systematist defends polyphyletic groups; they represent a failure to distinguish homologous structures from analogous ones.^{3, 10}

The insistence on monophyly has had far-reaching consequences for taxonomy. "Fish" (Pisces) as traditionally conceived is paraphyletic because it excludes the tetrapods (amphibians, reptiles, mammals, and birds) that evolved from within the lobe-finned fish lineage. "Invertebrates" is paraphyletic for the same reason: it defines a group by the absence of a vertebral column rather than by shared derived features. Many traditional groupings that seemed natural to generations of biologists have been dismantled or redefined under the strict cladistic criterion of monophyly.^{19, 20}

The phenetic challenge and its decline

Cladistics was not the only alternative to traditional taxonomy that emerged in the mid-twentieth century. In the 1960s and 1970s, a school known as numerical taxonomy or phenetics sought to classify organisms on the basis of overall similarity, measured quantitatively across as many characters as possible, without attempting to distinguish ancestral from derived states. The approach was championed by Peter Sneath and Robert Sokal, whose 1973 textbook Numerical Taxonomy systematized the methods.⁴ Pheneticists argued that the distinction between ancestral and derived characters was subjective and unreliable, and that the most objective classification would emerge from clustering organisms by quantitative similarity scores computed from large character matrices.

Phenetics had certain advantages: it was explicit, repeatable, and could handle large datasets with emerging computer technology. But it suffered from a fundamental limitation. Because it did not distinguish shared derived characters from shared ancestral characters, it could not reliably distinguish similarity due to common ancestry from similarity due to convergent evolution. Two lineages that independently evolved similar features would be grouped together, producing a misleading classification. Furthermore, phenetic classifications proved to be highly sensitive to the choice and weighting of characters: different character sets often produced different groupings, undermining the objectivity that was phenetics' principal claim.^{4, 10} By the 1980s, cladistics had largely supplanted phenetics as the preferred methodology in systematic biology, though some phenetic techniques (particularly cluster analysis and ordination methods) remain in use for ecological and population-level studies.

Maximum parsimony

The earliest computational method for implementing cladistic analysis was maximum parsimony, developed in the 1960s and 1970s. The parsimony criterion selects the phylogenetic tree that requires the fewest evolutionary changes (character-state transformations) to explain the observed distribution of characters across taxa. If species A, B, and C share a particular derived character state and species D lacks it, the most parsimonious explanation is that the character evolved once in the common ancestor of A, B, and C, rather than evolving independently in each. Walter Fitch formalized the algorithmic procedure for counting the minimum number of changes on a tree in 1971.⁵

Parsimony has an appealing philosophical simplicity: it invokes the minimum amount of evolutionary change necessary to explain the data, following the principle of Occam's razor. In many cases, parsimony analysis recovers well-supported and biologically sensible phylogenies. However, Joseph Felsenstein demonstrated in 1978 that parsimony can be "positively misleading" under certain conditions — specifically, when two long branches in a tree are separated by a short internal branch. In this situation, known as long-branch attraction, convergent changes along the two long branches accumulate faster than synapomorphic changes along the short connecting branch, causing parsimony to group the long-branched taxa together even though they are not closely related.⁷ This vulnerability arises because parsimony does not explicitly model the process of character evolution; it simply counts changes without accounting for the probability that multiple changes could occur at the same site along a long branch.

Despite this limitation, parsimony remains in use, particularly in morphological systematics where explicit probabilistic models of character evolution are harder to specify. For molecular data, however, model-based methods have become standard because they can account for the complexities of nucleotide substitution that parsimony ignores.¹¹

Maximum likelihood and Bayesian inference

Maximum likelihood phylogenetics, introduced by Felsenstein in 1981, evaluates phylogenetic trees by calculating the probability of the observed sequence data given the tree topology, branch lengths, and an explicit model of nucleotide or amino acid substitution.⁶ The tree that makes the observed data most probable is selected as the best estimate of phylogeny. This approach requires specifying a substitution model — for example, the Kimura two-parameter model, which distinguishes transitions from transversions and allows them to occur at different rates, or the more complex general time-reversible (GTR) model, which allows each of the six possible nucleotide exchanges to have its own rate.¹⁶ Model selection criteria such as the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) guide the choice of substitution model, balancing fit to the data against model complexity.²³

The chief advantage of maximum likelihood over parsimony is its ability to account for multiple substitutions at the same site, rate variation among sites, and differences in substitution rates among lineages. By modeling the evolutionary process explicitly, likelihood methods are less susceptible to long-branch attraction and generally perform better when rates of evolution are heterogeneous across the tree.^{6, 7} The principal disadvantage is computational cost: evaluating the likelihood of a tree requires integrating over all possible ancestral states at every internal node, and searching the space of possible tree topologies is an NP-hard problem. Modern software packages such as RAxML and IQ-TREE have made maximum likelihood analysis feasible for datasets with thousands of taxa and millions of aligned sites, but the computational demands remain substantial.¹¹

Bayesian inference, implemented in programs such as MrBayes, extends the likelihood framework by incorporating prior probability distributions on tree topologies, branch lengths, and model parameters.⁹ Rather than finding a single best tree, Bayesian methods use Markov chain Monte Carlo (MCMC) algorithms to sample trees in proportion to their posterior probability — the probability of the tree given the data and the prior. The result is a posterior distribution of trees from which clade support can be assessed as posterior probabilities: if a particular clade appears in 95% of the sampled trees, it receives a posterior probability of 0.95. Bayesian posterior probabilities have been shown to provide a more intuitive and often more accurate measure of clade support than the bootstrap, though they can be sensitive to prior specification and may overestimate support under certain model misspecifications.^{8, 9}

Morphological versus molecular data

For most of the history of taxonomy, classification was based entirely on morphological characters — the observable anatomical features of organisms. Bones, teeth, shells, leaf shapes, flower structures, and developmental patterns provided the data from which systematists inferred relationships. Morphological data have the advantage of being applicable to fossils, which lack preserved DNA, and they directly reflect the phenotypic diversity that taxonomy seeks to organize.¹⁰

The molecular revolution, beginning with protein electrophoresis in the 1960s and accelerating with DNA sequencing from the 1980s onward, transformed systematics by providing an independent and vastly richer source of phylogenetic information. A single gene can yield hundreds or thousands of aligned nucleotide positions, each a potential phylogenetic character, and whole-genome sequencing now provides millions of characters for analysis. Molecular data have several intrinsic advantages over morphological data: characters are discrete and unambiguous (the four nucleotide bases), the same genes can be compared across enormously divergent organisms, and explicit models of sequence evolution allow for sophisticated statistical analysis. Early molecular phylogenies were often constructed using distance-based algorithms such as neighbor-joining, which clusters taxa by pairwise sequence divergence and can produce reasonable trees rapidly even for large datasets.¹⁸ As computational power grew, model-based methods largely supplanted distance approaches for inferring deep evolutionary relationships.^{11, 12}

Molecular phylogenetics has resolved many longstanding taxonomic disputes that morphology alone could not settle. The relationship of turtles to other reptiles, debated for over a century on morphological grounds, was resolved by molecular data showing that turtles are archosaurs — more closely related to birds and crocodilians than to lizards and snakes — contrary to the traditional placement based on skull morphology.²⁴ Similarly, molecular studies have demonstrated that the traditional mammalian order Insectivora was polyphyletic, with hedgehogs, shrews, tenrecs, and golden moles belonging to separate evolutionary lineages that converged on a small, insect-eating body plan.¹² These results vindicate the cladistic insistence on distinguishing shared derived from convergent characters, with molecular phylogenetics methods providing the tools to do so at a scale morphology could not match.

Morphological data remain essential, however, for two reasons. First, fossils can only be placed in phylogenies using morphological characters, and the inclusion of fossil taxa is critical for understanding the timing and sequence of evolutionary events. Second, morphological and molecular data sometimes conflict, and these conflicts can be informative about processes such as convergent evolution, hybridization, or incomplete lineage sorting. The most robust phylogenetic analyses combine morphological and molecular data in a "total evidence" framework, allowing fossils and living organisms to be analyzed together.^{10, 20}

Reading phylogenetic trees

The primary output of a cladistic analysis is a phylogenetic tree, also called a cladogram when branch lengths are not specified or a phylogram when branches are proportional to the amount of evolutionary change. A phylogenetic tree is a branching diagram in which the tips (terminal nodes) represent the taxa under study — species, populations, or genes — and the internal nodes represent hypothesized common ancestors. Each branching point (node) represents a speciation or divergence event, and the pattern of branching reflects the inferred sequence of clade-splitting events.^{3, 21}

A common misconception about phylogenetic trees is that the order of taxa along the tips reflects a linear progression from "primitive" to "advanced." In reality, the branches of a tree can be rotated around any internal node without changing the evolutionary relationships depicted. What matters is the branching pattern — which taxa share more recent common ancestors — not the left-to-right ordering of the tips. Another misconception is that a node on a tree represents an extant species that is the ancestor of the taxa above it; in fact, internal nodes represent hypothetical ancestral populations, not any particular species that has been sampled.²¹

Branch support is assessed using statistical methods. The bootstrap, introduced by Felsenstein in 1985, resamples characters from the original data matrix with replacement, constructs a tree from each resampled dataset, and reports the percentage of resampled trees in which a given clade appears.⁸ Bootstrap values above 70% are generally considered moderate support, and values above 95% are considered strong. Bayesian posterior probabilities, as described above, offer an alternative measure of clade support. Both methods allow systematists to distinguish well-supported clades from uncertain ones, a crucial distinction when classifications are built upon the resulting trees.^{8, 9}

Phylogenetics and the molecular clock

An important extension of phylogenetic analysis is the use of a molecular clock to estimate the absolute timing of divergence events. The molecular clock hypothesis, first proposed by Emile Zuckerkandl and Linus Pauling in the 1960s, holds that mutations accumulate in DNA and the genetic code at a roughly constant rate over time, so that the number of sequence differences between two lineages is proportional to the time since they diverged from their most recent common ancestor. If the rate of molecular evolution can be calibrated using fossils or biogeographic events of known age, the molecular clock can be used to assign absolute dates to the nodes of a phylogenetic tree.^{11, 16}

In practice, the rate of molecular evolution varies among lineages, genes, and sites, violating the strict clock assumption. Modern "relaxed clock" methods accommodate this rate variation by allowing each branch of the tree to have its own rate, drawn from a statistical distribution, while still estimating divergence times. Bayesian frameworks are particularly suited to this task because they can simultaneously estimate the tree topology, branch-specific rates, and divergence times, integrating over the uncertainty in each parameter. The result is a time-calibrated phylogeny — a tree in which the internal nodes are dated with explicit confidence intervals, providing a timeline for the evolutionary history of the group under study.¹¹

Gene trees, species trees, and discordance

One of the most important conceptual advances in modern phylogenetics is the recognition that the evolutionary history of a gene is not necessarily identical to the evolutionary history of the species that carries it. A gene tree traces the ancestry of a particular locus through time; a species tree traces the branching history of populations and species. When lineages diverge rapidly or when ancestral populations are large, different genes can have different genealogical histories — a phenomenon known as incomplete lineage sorting. The result is gene tree discordance: different loci yield different phylogenies, and no single gene tree may match the true species tree.²²

Hybridization and horizontal gene transfer can produce additional sources of discordance. When two species interbreed and produce fertile offspring, genetic material from one species is incorporated into the genome of the other, creating a reticulate pattern of relationships that cannot be represented by a simple bifurcating tree. Horizontal gene transfer is pervasive in prokaryotes and occurs in eukaryotes as well, further complicating the inference of a single tree of life.²²

Modern coalescent-based methods address gene tree discordance by explicitly modeling the process by which gene lineages sort within species lineages. Methods such as ASTRAL and *BEAST estimate the species tree that is most compatible with the observed distribution of gene trees, rather than concatenating all genes into a single alignment and treating them as if they share a single history. These approaches have revealed that many previously accepted phylogenies based on concatenation were artifacts of systematic bias, and they have provided more reliable estimates of species-level relationships in groups where rapid radiation produced short internal branches and extensive lineage sorting.²²

Phylogenetic nomenclature and the PhyloCode

The cladistic revolution in systematics has created a tension with the traditional Linnaean system of nomenclature. The Linnaean system is based on ranks — kingdom, phylum, class, order, family, genus, species — and every named taxon is assigned a rank. But cladistic analysis produces trees with many more branching levels than there are available ranks, and the assignment of ranks to clades is ultimately arbitrary. Two families might be of very different ages, species richnesses, or degrees of genetic divergence, yet they share the same rank.^{13, 14}

In response to these difficulties, Kevin de Queiroz and others have developed a rank-free system of phylogenetic nomenclature, codified in the PhyloCode, which was formally published in 2020.¹³ Under the PhyloCode, taxon names are defined not by rank but by reference to phylogenetic relationships. A clade name is defined by specifying the clade it refers to — for example, the most inclusive clade containing species A and species B but not species C. This approach ensures that names are anchored to phylogenetic hypotheses rather than to arbitrary rank assignments, and that they remain stable even as new species are discovered or trees are revised.¹⁴

The PhyloCode has not replaced the traditional codes of nomenclature and is used alongside them rather than as a substitute. Many systematists continue to use Linnaean ranks for practical convenience, particularly in applied contexts such as conservation biology and ecology where a standardized hierarchical system facilitates communication. The debate between rank-based and rank-free nomenclature reflects a deeper question about whether the purpose of taxonomy is to create a stable, general-purpose filing system or to provide a nomenclature that faithfully mirrors the current best estimate of evolutionary history.^{13, 14}

The modern synthesis of taxonomy and phylogenetics

The integration of cladistic principles, molecular data, and computational methods has produced a modern synthesis of taxonomy and phylogenetics that differs profoundly from the taxonomy practiced a century ago. Today, a proposed classification is expected to be consistent with a well-supported phylogenetic hypothesis, and a phylogenetic hypothesis is expected to be derived from an explicit analysis of character data using a specified method and model. Classifications that are inconsistent with phylogenetic evidence are revised; phylogenetic analyses that use inadequate data or methods are challenged. The result is a dynamic, self-correcting system in which taxonomy and phylogenetics are mutually reinforcing enterprises.^{20, 21}

This synthesis has produced major revisions to the tree of life at every scale. The three-domain system (Bacteria, Archaea, Eukarya), proposed by Carl Woese on the basis of ribosomal RNA sequences, fundamentally restructured the highest levels of biological classification. Within the animals, molecular phylogenetics revealed the division of bilaterian animals into three major clades — Deuterostomia, Ecdysozoa, and Lophotrochozoa — a grouping that was unanticipated by morphological taxonomy. Within vertebrates, the recognition that birds are dinosaurs and that whales are deeply nested within the artiodactyl mammals (the even-toed ungulates) required the revision of traditional orders and classes.^{12, 19, 21}

At the species level, molecular data have revealed cryptic species — morphologically indistinguishable populations that are genetically distinct and reproductively isolated — and have collapsed nominal species that prove to be variants of a single lineage. The species concept itself remains contentious, with different operational criteria (morphological, biological, phylogenetic, and genealogical) sometimes producing different species boundaries. Cladistic methods can identify the branching pattern of speciation events, but defining where one species ends and another begins remains, in many cases, a matter of judgment informed by multiple lines of evidence including morphology, genetics, ecology, and reproductive behavior.^{15, 25}

The ongoing assembly of the tree of life — a comprehensive phylogeny of all known species — represents one of the grand challenges of twenty-first-century biology. Initiatives such as the Open Tree of Life project aim to synthesize published phylogenies into a single, continuously updated supertree. The task is immense: there are an estimated 8.7 million eukaryotic species on Earth, of which perhaps 1.5 million have been formally described, and the phylogenetic position of many described species remains unknown. Yet the tools of modern cladistics and molecular phylogenetics methods have made the goal plausible in a way that would have been unimaginable in Linnaeus's era, and the resulting tree — when achieved — will be the single most comprehensive statement of the evolutionary relationships among all living things.^{12, 21}