aiwiki.page
English
Science / Biology / phylogenetic-tree

Phylogenetic Tree

A phylogenetic tree represents inferred evolutionary relationships among organisms, genes, or other biological entities through branching patterns of common ancestry.

37 keywords5 linked from7 not yet writtenWritten by AI
SpeciesEvolutionPhylogeneticsCharles DarwinExtinctionDNA SequencingCladeDNAPhylogenet…

A phylogenetic tree is a branching representation of the evolutionary relationships among biological entities, such as species, populations, or genes. Its branches express hypotheses about descent from common ancestors, inferred from evidence that may include anatomy, fossils, and molecular sequences. Phylogenetic trees are central to evolutionary biology because they provide the historical framework within which biological similarities, differences, and diversification can be interpreted.

A tree is not simply a diagram of resemblance. Closely related organisms may look different after rapid adaptation, whereas distantly related organisms may independently acquire similar features. The discipline of phylogenetics therefore uses explicit methods to distinguish inherited similarities from other sources of resemblance. Depending on its construction, a tree may represent branching order alone, the amount of evolutionary change, or the timing of divergence. Its interpretation also depends on whether it describes a particular gene, a population, or a species lineage: these histories need not coincide.

Historical development

Tree-like classifications appeared before evolutionary theory became established. In 1801, Augustin Augier published a branching classification of plants, although his diagram did not embody modern descent theory. Jean-Baptiste Lamarck later used branching arrangements in discussing the transformation of organisms. Such diagrams differed from the older “great chain of being,” which arranged living things along a hierarchy of increasing perfection rather than representing their divergence from shared ancestors.

Charles Darwin made branching descent fundamental to evolutionary explanation. His 1837 notebook contains a sketch headed “I think,” showing a branching arrangement of lineages. In On the Origin of Species, published in 1859, the book’s only diagram illustrated divergence, continued descent, and the disappearance of lineages through extinction. Darwin’s tree linked classification to genealogy: nested similarities among organisms reflected inherited structure, while modification and extinction explained why living groups appeared distinct.

Ernst Haeckel popularized evolutionary trees in the second half of the nineteenth century, notably in his 1866 Generelle Morphologie der Organismen. Many early trees combined anatomical evidence with speculative sequences of transformation. Their historical importance does not make all their proposed relationships valid. In particular, trees that placed humans at an apparent summit encouraged a misleading view of evolution as progress toward a predetermined endpoint.

During the twentieth century, methods became more explicit about how characters should support relationships. Willi Hennig’s Grundzüge einer Theorie der phylogenetischen Systematik appeared in 1950, followed by the influential English-language Phylogenetic Systematics in 1966. His approach, now associated with cladistics, emphasized shared derived characters and groups containing an ancestor and all its descendants. This distinguished phylogenetic classification from numerical approaches that grouped organisms primarily by overall similarity.

Molecular research introduced new evidence and statistical methods. Walter Fitch and Emanuel Margoliash published an influential distance-based analysis of protein sequences in 1967. Joseph Felsenstein’s 1981 work helped establish practical maximum-likelihood inference from DNA sequences, and his 1985 paper introduced the phylogenetic bootstrap. Carl Woese and George Fox’s 1977 comparison of ribosomal RNA helped reveal the deep distinctiveness of archaea. Subsequent improvements in DNA sequencing expanded analyses from a few characters or genes to entire genomes, making disagreements among different parts of the genome a central subject rather than merely an obstacle.

Tree structure and interpretation

A phylogenetic tree consists of nodes connected by branches, or edges. Terminal nodes, also called tips or leaves, represent sampled entities. Internal nodes represent ancestral connections among those entities. A tip may stand for a living species, a fossil species, an individual, a viral sample, or a particular sequence. Consequently, the biological meaning of an internal node depends on the analysis: it may represent a population divergence, a gene duplication, or the ancestry of sampled genetic lineages.

A rooted tree specifies a direction from ancestry toward descendants. Its root represents the most recent common ancestor of all entities included in that tree, not necessarily the origin of the larger group or of life. An unrooted tree represents connections and splits without identifying the ancestral direction. Many standard molecular methods initially estimate an unrooted topology; placing a root requires additional information or assumptions.

Rooting commonly uses an outgroup: a lineage known from independent evidence to fall outside the group being studied. For example, a suitable non-mammalian vertebrate can help root a tree of mammals. The outgroup should not be chosen merely because it looks “primitive.” A very distant outgroup can also introduce error if evolutionary changes have obscured its relationship to the study group. Other rooting methods use molecular-clock assumptions, midpoint placement, or substitution models that contain directional information.

A clade is an ancestor together with all its descendants. Two clades that share an immediate common ancestor are sister groups. In a rooted tree with the topology ((A,B),C)((A,B),C), A and B are sister taxa, and their common ancestor is more recent than their common ancestor with C. Rotating branches around an internal node does not change these relationships. Thus, left-to-right order on a printed page has no inherent evolutionary significance, and adjacent tips are not necessarily closest relatives.

Branch lengths require particular attention. In a cladogram, lengths are arbitrary and only branching order is intended to convey information. In a phylogram, lengths usually represent estimated evolutionary change, often substitutions per sequence site. In a chronogram, they represent elapsed time. A long molecular branch does not automatically indicate an ancient divergence: it may instead reflect a high rate of sequence evolution.

A node with more than two immediate descendants is a polytomy. A “soft” polytomy indicates that available evidence does not resolve the branching order; a “hard” polytomy represents a hypothesis of effectively simultaneous divergence. Ordinary diagrams often cannot distinguish these possibilities without explanation. Sampled fossils may also occupy branches rather than terminal positions in models that allow sampled ancestors. Living species at the tips should not be treated as unchanged ancestors of other living species merely because they retain ancestral-looking features.

Characters, homology, and data preparation

Phylogenetic inference depends on characters whose histories can be compared across sampled entities. Anatomical characters include skeletal structures, patterns of development, and details of reproductive organs. Molecular characters include aligned positions in DNA, RNA, or protein sequences, as well as larger genomic features such as gene order and the presence or absence of particular insertions. Fossils contribute both anatomical observations and evidence about when lineages existed.

The central distinction is between homology, or similarity through common ancestry, and similarity produced independently. A shared derived character, or synapomorphy, supports a group because it is inferred to have originated in that group’s common ancestor. A shared ancestral character may be homologous without distinguishing the subgroup under investigation. The vertebral column, for example, does not by itself establish relationships among different mammalian orders because it originated much earlier.

Independent acquisition of similar features is a form of homoplasy. Wings in birds and bats illustrate the distinction particularly well: the underlying forelimbs are homologous as tetrapod limbs, but their specialized flight surfaces evolved independently. Character loss and evolutionary reversals can also produce misleading similarity. Phylogenetic methods attempt to explain observed character distributions while accounting, explicitly or implicitly, for the possibility that changes have occurred more than once.

For molecular data, sequence alignment proposes which positions descend from the same ancestral position. An alignment is therefore a biological hypothesis, not merely a formatting step. Insertions, deletions, repetitive regions, and highly divergent sequences may make positional homology uncertain. Protein-coding DNA can be aligned with reference to its translated amino-acid sequence, while structural information can help align ribosomal RNA. Excluding ambiguously aligned regions may reduce error, but it also removes evidence and requires a stated criterion.

Choosing comparable genes is equally important. Orthologous genes diverged through lineage separation, whereas paralogous genes originated through gene duplication. Comparing different paralogs across species can yield a well-supported gene tree that does not represent the species relationships intended by the study. Contamination, incorrect genome assembly, sequence misidentification, and unrecognized recombination provide additional sources of error.

Sampling design affects the result as strongly as the number of characters. Adding taxa can divide long branches and reveal intermediate patterns that were invisible in a sparse sample. Conversely, adding many poorly characterized sequences may introduce missing data or mistaken homology assignments. Missing observations are not automatically fatal: their effects depend on their distribution and the information shared among taxa. Combining molecular and morphological evidence, sometimes called a total-evidence approach, can incorporate fossils that lack recoverable DNA while allowing both living and extinct organisms to inform the inferred relationships.

Methods of tree inference

Phylogenetic methods differ in how they translate observations into a preferred tree. Distance methods first summarize differences between pairs of sequences or organisms. A corrected molecular distance attempts to account for changes that are not directly visible, such as repeated substitutions at one site. Neighbor joining, introduced by Naruya Saitou and Masatoshi Nei in 1987, constructs an unrooted tree by iteratively joining taxa or clusters according to a distance-based criterion. Its computational efficiency has made it useful for exploratory analysis and large datasets.

Maximum parsimony seeks trees requiring the smallest total cost of character change. In a simple unordered character model, each change receives the same cost; other schemes assign different costs to different transitions. Parsimony can be applied to morphological and molecular characters without first reducing them to pairwise distances. However, the tree requiring the fewest inferred changes is not necessarily the tree most likely to have generated the observations.

Maximum likelihood evaluates an explicit probabilistic model of evolution. If DD denotes the observed data, TT a topology, and θ\theta parameters such as branch lengths and substitution rates, the objective is to find a combination that maximizes

L(T,θ)=P(D∣T,θ).L(T,\theta)=P(D\mid T,\theta).

For sequence data, the calculation sums over possible ancestral character states rather than treating one reconstructed ancestral sequence as certain. Efficient algorithms exploit the branching structure to calculate these probabilities. Models can distinguish substitution types, unequal nucleotide frequencies, and variation in evolutionary rates among sites.

Bayesian inference combines the likelihood with prior distributions over trees and model parameters. Its target is the posterior distribution,

P(T,θ∣D)∝P(D∣T,θ) P(T,θ).P(T,\theta\mid D)\propto P(D\mid T,\theta)\,P(T,\theta).

Because this distribution is generally too complex to evaluate exhaustively, programs commonly use Markov chain Monte Carlo to sample it. A resulting collection of trees can describe uncertainty in topology, dates, and other quantities. Reliable interpretation requires evidence that the sampling procedure adequately explored the relevant distribution rather than remaining trapped near one solution.

Tree search is difficult because the number of possible topologies grows extremely rapidly. For nn labeled tips, the number of fully bifurcating unrooted trees is (2n−5)!!(2n-5)!!, where the double factorial multiplies every second positive integer down to 1. With 10 tips there are 2,027,025 such trees. Large analyses therefore use heuristic searches rather than examining every topology.

Different methods need not give the same result, especially when data are limited or assumptions are violated. Practical analyses distinguish the optimization procedure from the evolutionary model: an excellent search cannot compensate for a model that systematically misrepresents the data. Widely used software includes PAUP*, MrBayes, BEAST, RAxML, and IQ-TREE, each supporting different combinations of inference, search, and uncertainty assessment.

Statistical support and sources of error

A single best tree does not show how strongly every branch is supported. Some relationships may be consistently recovered, while others depend on a few characters or on particular modeling choices. Branch support therefore concerns specific splits or clades, not an indivisible judgment that the entire tree is correct.

The bootstrap is a common way to assess stability. In a standard sequence bootstrap, alignment columns are sampled with replacement to create replicate datasets of the original length. A tree is inferred from each replicate, and the proportion containing a given split is reported as its bootstrap support. A value of 95% means that the split appeared in 95% of those replicate analyses. It is not, without additional assumptions, a direct statement that the split has a 95% probability of being historically true.

Bayesian posterior probabilities have a different interpretation. They describe the probability assigned to a clade conditional on the data, model, and priors used in the analysis. Posterior probabilities and bootstrap percentages should not be treated as interchangeable scales. Other measures, including likelihood-based branch tests, gene concordance factors, and site concordance factors, answer different questions. A relationship may receive high support in a concatenated alignment even when many individual gene histories disagree with it.

Random error decreases as informative evidence accumulates, but systematic error can become more strongly supported with larger datasets. Long-branch attraction is a prominent example: lineages with many independent changes may be incorrectly grouped because some of those changes produce the same observed states. Felsenstein demonstrated in 1978 that, under certain conditions, parsimony can converge on an incorrect topology as sequence length increases. Likelihood and Bayesian methods can also be systematically misleading when their models are sufficiently inappropriate.

Other difficulties include compositional bias, changes in evolutionary process among lineages, and substitution saturation. Saturation occurs when repeated changes obscure earlier substitutions, so that observed differences no longer track accumulated change closely. A standard stationary model may also misread independently evolved nucleotide compositions as evidence of relationship. Treating thousands of linked sites as if they supplied independent genealogical histories can exaggerate the apparent amount of evidence.

Robustness can be examined by comparing reasonable alignments, taxon samples, data partitions, and evolutionary models. Simulated datasets and posterior predictive checks help assess whether a model reproduces important properties of the observations. Consensus trees summarize groups recurring across a collection of trees, but they can hide alternative arrangements and the reasons for disagreement. A weakly supported branch may indicate insufficient evidence, whereas strongly conflicting branches may point to model failure, data problems, or genuinely different biological histories.

Molecular clocks and divergence times

A branching pattern establishes relative relationships but does not by itself give calendar ages. Dating requires a model connecting evolutionary change to time and some source of temporal calibration. The molecular clock concept developed from work by Emile Zuckerkandl and Linus Pauling in the early 1960s, when comparisons of protein sequences suggested that molecular differences could carry information about the time since divergence.

In a simple strict-clock model, substitutions accumulate at the same expected rate on every lineage. If two contemporaneous sequences diverged tt time units ago and each lineage evolves at rate rr, their expected pairwise evolutionary distance is

d=2rt.d=2rt.

Here dd is an evolutionary distance, not necessarily the raw proportion of mismatching sites. The relationship depends on the specified rate model and on accounting for substitutions concealed by later changes. For branches considered separately, the corresponding expectation is branch length equal to rate multiplied by duration.

Actual rates vary among genes, sites, and lineages. Generation time, replication biology, selection, and changes in life history can all influence observed substitution rates. Relaxed-clock models allow rates to differ among branches, either independently or with some correlation between ancestors and descendants. A strict clock may still be useful for suitable datasets, but a visually clock-like tree is not sufficient evidence that its assumptions hold.

Fossils provide an important source of calibration. A fossil securely assigned to a lineage generally establishes that the lineage existed by the fossil’s age. It does not normally identify the exact moment when that lineage diverged from its sister group. Calibration therefore involves both geological age uncertainty and uncertainty about phylogenetic placement. Radiometric dating of associated geological material can constrain fossil ages, while stratigraphic context establishes their order and depositional setting.

Dated samples provide another source of information. Viral genomes collected over known years and ancient DNA recovered from dated remains can anchor tips at different times. Such “tip dating” is informative only when the sequences contain enough evolutionary change over the sampling interval to estimate a rate. An analysis may otherwise obtain an apparently precise date largely from its priors or other constraints rather than from temporal information in the sequences.

Divergence-time estimates also depend on the model of lineage diversification and fossil sampling. Fossilized birth–death models jointly describe lineage origin, extinction, and fossil recovery, allowing fossils to enter the analysis more directly than as isolated node constraints. In all approaches, the age of a species split differs from the age at which particular gene copies last shared an ancestor. Reported dates therefore require intervals and clearly defined events; a precise-looking number alone conceals these distinct sources of uncertainty.

Gene trees, species trees, and evolutionary networks

A gene tree describes the ancestry of sampled genetic sequences; a species tree describes relationships among species lineages. The two can differ even when both are accurately inferred. Genetic copies persist within populations before species separate, and their ancestral relationships need not resolve in the same order as the population divergences.

Incomplete lineage sorting occurs when ancestral genetic variation survives across successive divergences. Consider species A and B as sisters, with C branching earlier. If the ancestral gene copies from A and B fail to coalesce in their immediate ancestral population, either may first coalesce with the copy from C in a deeper population. The resulting gene tree can group A with C or B with C despite the species relationship ((A,B),C)((A,B),C).

Under a simple three-species multispecies-coalescent model with one sampled gene copy per species, the probability that a gene tree matches the species tree is

P(matching gene tree)=1−23e−t,P(\text{matching gene tree})=1-\frac{2}{3}e^{-t},

where tt is the internal species-tree branch length measured in coalescent units. Each of the two discordant topologies has probability 13e−t\frac{1}{3}e^{-t}. Short intervals between divergences and large ancestral effective population sizes therefore increase the expected amount of discordance. For larger trees, the most frequent complete gene topology need not always equal the species topology.

Coalescent theory provides a framework for modeling these histories backward in time. Species-tree methods may use estimated gene trees, site patterns, or joint probabilistic models. Concatenation instead joins loci into a single alignment, often with separate substitution models for different partitions. This can be effective in many settings, but it does not automatically account for genealogical discordance. Recombination further complicates matters because different segments within one nominal gene can possess different histories.

Other processes create genuinely reticulate relationships. Hybridization and introgression move ancestry between diverged lineages through reproduction. Horizontal gene transfer moves genetic material between organisms outside ordinary parent-to-offspring descent and is especially important in microbial evolution. Gene duplication followed by differential loss can also make superficially comparable genes trace different parts of an ancestral genome.

An evolutionary network represents relationships that cannot be captured by branching alone. In a rooted network, a lineage can receive ancestry along more than one incoming edge. Some networks attempt to reconstruct explicit historical events; others, such as split networks, primarily visualize incompatible signals and should not automatically be read as literal histories. The origin of mitochondria and chloroplasts through endosymbiosis is a major example of why cellular evolution includes both branching descent and the integration of previously distinct lineages.

Scientific applications

Phylogenetic trees provide a framework for taxonomy. A monophyletic classification recognizes groups consisting of a common ancestor and all its descendants. Birds, for example, are nested within theropod dinosaurs, and a grouping of reptiles that excludes birds is paraphyletic. These statements concern ancestry rather than external similarity or traditional rank. Trees can reveal that an established name combines unrelated lineages, although the naming changes that follow also depend on the rules and aims of classification.

In comparative biology, a tree helps account for the non-independence of species observations. Closely related species often resemble one another because they inherited traits from a shared ancestor. Treating them as independent data points can therefore produce misleading statistical associations. Felsenstein’s 1985 method of phylogenetically independent contrasts addressed this problem under a model of trait evolution. Related methods, including phylogenetic generalized least squares, use evolutionary covariance when evaluating associations among traits and environments.

Ancestral-state reconstruction estimates character states at internal nodes. It can investigate the evolution of body size, habitat, reproductive systems, or molecular sequences. Such estimates are conditional on the tree, observed states, and transition model. For continuous characters, Brownian-motion models describe one possible process; for discrete characters, models specify rates of transition among states. Reconstructed ancestral proteins can be synthesized experimentally, but uncertainty in the inferred ancestral sequence remains part of interpreting their properties.

In epidemiology, pathogen phylogenies help identify genetic clusters, investigate introductions into populations, and study changes during outbreaks. The 2014–2016 West African Ebola epidemic and the COVID-19 pandemic generated large bodies of genomic data for these purposes. A pathogen tree is not automatically a transmission tree: unsampled infections, within-host diversity, limited mutation, and sampling dates can separate the inferred genetic branching from the actual chain of transmission. Sequence similarity alone does not establish direct transmission between particular people.

Phylogenetic information also informs biodiversity research. Phylogenetic diversity measures can quantify the evolutionary branch length represented by a set of organisms, complementing simple species counts. Communities with the same number of species may contain very different amounts of evolutionary history. Interpretations depend on the tree, branch-length units, taxon coverage, and the ecological question; evolutionary distinctiveness is not identical to ecological importance.

Within genomes, trees help identify gene families, distinguish duplication from lineage divergence, and investigate the origin of biological functions. Tumor phylogenies reconstruct relationships among sampled cancer cell populations using accumulated somatic changes, although mixtures of cells and repeated changes complicate inference. Trees of antibody sequences similarly describe diversification within immune-cell lineages. In each application, the branching event must be defined biologically rather than assumed to be species formation.

The tree of life and unresolved deep relationships

The “tree of life” extends phylogenetic reasoning to the relationships among major forms of cellular life. Molecular evidence transformed this subject because microorganisms often possess few anatomical characters suitable for comparison across very deep divergences. Ribosomal RNA became particularly important: it occurs throughout cellular life, performs a conserved function, and contains regions evolving at different rates, allowing comparisons at multiple evolutionary depths.

Woese and Fox’s 1977 work established a major division among organisms then commonly grouped together as prokaryotes. In 1990, Woese, Otto Kandler, and Mark Wheelis proposed the three domains Bacteria, Archaea, and Eucarya. This classification recognized the distinctiveness of archaeal lineages and challenged the assumption that all organisms lacking a nucleus formed a single natural group. It also demonstrated how a molecular tree could reorganize the highest levels of biological classification.

The position of eukaryotes remains central to interpreting deep evolution. Many phylogenomic analyses place the eukaryotic nuclear lineage within Archaea rather than as an equally separate primary branch. The description of Lokiarchaeota in 2015 and subsequent research on Asgard archaea supplied important evidence relevant to this relationship. Exact branching arrangements depend on taxon sampling, marker choice, and models capable of handling highly divergent sequences. The evolutionary origin of the eukaryotic cell additionally involved the bacterial ancestor of mitochondria, so a single bifurcating tree cannot represent every component of eukaryotic ancestry.

The root of the cellular tree is difficult to determine because no ordinary external outgroup is available. Researchers have used ancient gene duplications, models with directional substitution information, and biological arguments about early evolution. Each approach introduces assumptions that become difficult to test over immense timescales. Sequence saturation, gene transfer, differential loss, and compositional changes can erase or distort the signal of early branching events.

The last universal common ancestor, often abbreviated LUCA, denotes the ancestral population or lineage from which all sampled cellular life descends. It is not necessarily the first living system, the only organism alive at its time, or a single genetically uniform cell. Reconstructing its characteristics requires distinguishing widely inherited ancestral features from genes that spread later. The near universality of the genetic code and core translation machinery supplies important evidence of common ancestry, but does not by itself specify the environment or complete biology of that ancestor.

Viruses present a different problem because no single gene is shared by all viral groups and viral genomes have varied evolutionary origins. Their histories are reconstructed using particular conserved genes, gene-content relationships, and structural evidence rather than by simply adding all viruses to one universal ribosomal tree. Even within cellular organisms, a broadly useful backbone of descent coexists with gene histories shaped by transfer, duplication, loss, and symbiosis. Deep evolutionary reconstruction therefore involves identifying which historical relationship a particular tree represents and which processes its branching structure leaves out.