Phylogenetics

Phylogenetics is the discipline of evolutionary biology and bioinformatics dedicated to reconstructing the historical patterns of descent, lineage splitting (speciation), and genealogical relationships among biological entities—ranging from populations and species to individual genes, proteins, and viral genomes. These evolutionary relationships are represented mathematically and graphically as a phylogenetic tree (or evolutionary network), a directed acyclic or unrooted branching graph in which extant or fossil entities occupy the tips and ancestral divergence events occupy the internal vertices. First conceptualized visually by Charles Darwin's 1837 "I think" notebook sketch and the single branching diagram in On the Origin of Species (1859), phylogenetic systematics was transformed into an objective, testable analytical science in the mid-20th century through two breakthroughs: the formulation of cladistics by the German entomologist Willi Hennig (1950, 1966) and the introduction of molecular sequence comparison by Emile Zuckerkandl, Linus Pauling, Walter Fitch, and Allan Wilson.

Modern molecular phylogenetics analyzes aligned nucleotide (DNA/RNA) or amino-acid sequences of homologous loci across taxa to infer both the branching order (topology) and the evolutionary distance or time elapsed along each lineage (branch lengths, often calibrated against fossil constraints via a relaxed molecular clock). Beyond biological taxonomy, phylogenetic methods are indispensable for tracing the zoonotic spillover and real-time transmission of emerging viral pathogens, reconstructing ancestral protein sequences, detecting positive and purifying selection, and mapping the coevolution of hosts and parasites.

Anatomy of a Phylogenetic Tree

A phylogenetic tree consists of discrete topological and metric components:

• Terminal Nodes (Tips or Leaves): Represent the sampled operational taxonomic units (OTUs)—extant species, fossil taxa, viral isolates, or orthologous gene sequences.
• Internal Nodes (Vertices): Represent hypothetical most recent common ancestors (MRCAs) at the point of lineage bifurcation. A node that splits into two descendant lineages is dichotomous (bifurcating), whereas a node splitting into three or more lineages is a polytomy (indicating either statistical uncertainty—a "soft polytomy"—or rapid simultaneous radiation—a "hard polytomy").
• Branches (Edges): Represent ancestral lineages evolving through time. In a cladogram, branch lengths are arbitrary and convey only topological branching order; in a phylogram, branch lengths are proportional to the expected number of nucleotide or amino-acid substitutions per site; and in a chronogram (ultrametric tree), branch lengths are scaled to geological or calendar time.
• Sister Taxa and Rooting: Two clades or tips that share an immediate common ancestor not shared with any other taxon are sister groups. For n terminal taxa, there are (2*n* − 5)!! = 1 × 3 × 5 × ⋯ × (2*n* − 5) possible unrooted bifurcating trees and (2*n* − 3)!! rooted bifurcating trees (for n = 10 taxa, 34,459,425 rooted topologies exist). An unrooted tree specifies relationships without evolutionary directionality; to convert it into a rooted tree, systematists include an outgroup—a taxon known from independent evidence to have diverged prior to the MRCA of the ingroup—placing the root along the branch connecting the outgroup to the ingroup.

Hennigian Cladistics: Monophyly, Paraphyly, and Polyphyly

Willi Hennig demonstrated that valid natural classifications must reflect genealogy rather than overall phenetic similarity, because overall similarity conflates two distinct types of homologous character states: plesiomorphies (ancestral character states retained from a distant ancestor) and apomorphies (derived character states that originated as evolutionary novelties). Specifically, only synapomorphies—shared derived character states inherited from the immediate common ancestor of a group (such as mammary glands and three middle-ear ossicles in mammals, or feathers in theropod dinosaurs/birds)—provide valid evidence for grouping taxa into a clade. Shared ancestral states (symplesiomorphies, such as the retention of five digits in lizards and humans) and unique single-lineage traits (autapomorphies) are phylogenetically uninformative for resolving internal branching within a group, while independently evolved similarities (homoplasies arising from convergent evolution or evolutionary reversal) actively mislead tree inference if unrecognized.

Based on these principles, taxonomic groupings fall into three topological categories:

1. Monophyletic Group (Clade): Contains an ancestral species and all of its descendant species (e.g., Mammalia, Aves, or Archosauria including crocodilians and birds). Only monophyletic clades are recognized as valid formal taxa in cladistic systematics.
2. Paraphyletic Group (Grade): Contains an ancestral species and some, but not all, of its descendants—typically excluding a descendant clade that has undergone rapid morphological divergence. Classic examples include traditional "Reptilia" (when Aves are excluded, since crocodilians share a more recent common ancestor with birds than with lizards or turtles) and "Pisces" (bony and cartilaginous fishes when tetrapods are excluded).
3. Polyphyletic Assemblage: A non-natural group composed of taxa derived from two or more distinct ancestors whose common ancestor is excluded, united mistakenly on the basis of convergent homoplasies (such as grouping endothermic "warm-blooded" birds and mammals together).

Distance-Based and Maximum Parsimony Methods

Following multiple sequence alignment (via MAFFT or MUSCLE) to establish positional homology across columns, phylogenetic reconstruction proceeds either by distance-matrix clustering or by discrete character optimization. Distance methods first compress the alignment into an n × n matrix of pairwise evolutionary distances (d_ij), applying mathematical corrections for unobserved multiple substitutions ("multiple hits") at the same site—such as the Jukes–Cantor (JC69) formula d = −(3/4) ln(1 − 4*p*/3), where p is the observed proportion of differing sites, or the Kimura 2-Parameter (K80) model, which assigns distinct rates to transitions (purine↔purine or pyrimidine↔pyrimidine) and transversions (purine↔pyrimidine). While UPGMA (Unweighted Pair Group Method with Arithmetic Mean) assumes an unrealistic strict clock, Neighbor-Joining (NJ) (Saitou and Nei, 1987) iteratively joins pairs of taxa that minimize total tree length without assuming rate constancy.

Maximum Parsimony (MP) operates directly on discrete alignment columns, evaluating candidate tree topologies using Fitch's (1971) dynamic-programming algorithm to identify the tree requiring the minimum total number of nucleotide or morphological state changes. Although effective when substitution rates are low, Maximum Parsimony is statistically inconsistent in the Felsenstein zone—a parameter regime where two unrelated lineages evolve rapidly along long branches while separated by a short internal branch. Because there are only four nucleotide states (A, C, G, T), random parallel substitutions accumulate along the two long branches more rapidly than true synapomorphies accumulate along the short internal branch, causing parsimony to group the two long branches together erroneously—an artifact termed long-branch attraction (LBA).

Maximum Likelihood, Bayesian Inference, and Statistical Support

Probabilistic methods overcome long-branch attraction by modeling sequence evolution as a continuous-time Markov process defined by an instantaneous rate matrix Q along branches of length t (where transition probabilities are given by the matrix exponential P(t) = exp(Qt)). The most general reversible nucleotide model is the General Time-Reversible (GTR) model, which incorporates four equilibrium base frequencies (π_A, π_C, π_G, π_T) and six symmetric exchangeability parameters, typically combined with a discrete Gamma distribution (+Γ) (parameterized by shape parameter α) to model site-to-site rate heterogeneity and a proportion of invariable sites (+I).

• Maximum Likelihood (ML): Introduced to phylogenetics by Joseph Felsenstein (1981) along with his pruning algorithm, ML searches for the tree topology T, branch lengths t, and model parameters θ that maximize the likelihood function L(T, t, θ) = P(Alignment | T, t, θ). Because evaluating all topologies is NP-hard, programs such as IQ-TREE and RAxML use nearest-neighbor interchange (NNI) and subtree pruning and regrafting (SPR) heuristics.
• Bayesian Inference (BI): Using Bayes' theorem, P(T, t, θ | Data) = [P(Data | T, t, θ) P(T, t, θ)] / P(Data), Bayesian algorithms (such as MrBayes and BEAST) avoid the intractable marginal likelihood denominator by employing Markov chain Monte Carlo (MCMC) sampling (typically Metropolis-coupled MCMC with heated chains) to sample trees in direct proportion to their posterior probability.
• Felsenstein's Non-Parametric Bootstrap: To quantify confidence in individual clades under ML, NJ, or MP, Felsenstein (1985) introduced the bootstrap: columns of the original L-site alignment are resampled randomly with replacement L times to generate 100 to 1,000 pseudoreplicate alignments. The percentage of bootstrap trees in which a given bipartition (node) appears represents its bootstrap support (values ≥70%–95% indicating robust clade recovery).

Even when individual gene trees are inferred accurately, a gene tree can disagree topologically with the underlying species tree due to three biological processes: incomplete lineage sorting (ILS) (deep coalescence of ancestral polymorphisms across rapid successive speciation nodes, modeled via the multispecies coalescent in ASTRAL), horizontal gene transfer (HGT) or hybridization/introgression, and unrecognized gene duplication and loss (confusing paralogs with orthologs).

Why is the observed proportion of nucleotide differences (p-distance) an underestimate of the true evolutionary distance between two divergent sequences?

As two lineages diverge over time, multiple substitutions occur at the same nucleotide position—including parallel changes, convergent changes, and back-mutations (reversions such as A → G → A) that restore the ancestral base. Consequently, the observed p-distance approaches an asymptotic saturation ceiling (75% difference for four equally frequent nucleotides) while the true number of historical substitutions continues to increase linearly with time, requiring logarithmic Markov corrections (such as JC69, K80, or GTR) to recover additive branch lengths.

How do Bayesian posterior probabilities differ from non-parametric bootstrap proportions?

A Bayesian posterior probability represents the true conditional probability that a clade is correct given the alignment, the substitution model, and the prior distributions, and typically reaches 0.95–1.00 on moderate datasets, though it can be inflated if the substitution model is misspecified. A Felsenstein bootstrap proportion measures repeatability under random column resampling of the alignment and is a more conservative estimator of clade accuracy.

Does rotating the branches around an internal node change the meaning of a phylogenetic tree?

No. Internal nodes of a phylogenetic tree act like hanging mobiles: rotating the two descendant branches around any internal pivot point changes the vertical or horizontal left-to-right ordering of the tip labels on the page, but leaves the set of common ancestors, sister-group relationships, and branch lengths identical. Evolutionary relatedness is determined strictly by the recency of shared common ancestors (internal nodes), never by the proximity of tip labels along the margin.

1. Hennig, W. (1965). "Phylogenetic systematics." Annual Review of Entomology, 10, 97–116. https://doi.org/10.1146/annurev.en.10.010165.000525
2. Felsenstein, J. (1981). "Evolutionary trees from DNA sequences: A maximum likelihood approach." Journal of Molecular Evolution, 17(6), 368–376. https://doi.org/10.1007/BF01734359
3. Felsenstein, J. (1985). "Confidence limits on phylogenies: An approach using the bootstrap." Evolution, 39(4), 783–791. https://doi.org/10.1111/j.1558-5646.1985.tb00420.x
4. Yang, Z., & Rannala, B. (2012). "Molecular phylogenetics: Principles and practice." Nature Reviews Genetics, 13(5), 303–314. https://doi.org/10.1038/nrg3186