Reading Phylogenetic Trees

A trout shares a common ancestor with you that it does not share with a shark. So a trout is more closely related to you than to a shark, even though trout and shark both look like fish.

That sentence sounds wrong until you can read a phylogenetic tree. This lesson teaches you to read one: where to look for relatedness, which features of a drawing mean nothing, how to place a trait on a tree, and how much to trust the tree itself.

The idea: a tree is a hypothesis about ancestry

A phylogenetic tree is a branching diagram of hypothesized ancestry among a set of taxa (species or larger groups). The living or sampled taxa sit at the tips. Each internal node, where a branch splits, stands for an inferred common ancestor of every lineage above it. The root is the ancestor of everything on the tree. A clade is an ancestor and all of its descendants, and two clades that are each other’s closest relatives are sister groups. The most recent common ancestor (MRCA) of a set of tips is the latest node from which all of them descend.

The nearest analogy is a family tree drawn for cousins only. Everyone shown is alive today, the grandparents are inferred rather than photographed, and you judge who is closer kin by how recently two people’s lines meet.

The word “hypothesis” matters. Nobody watched these lineages split. A tree is estimated from characters, usually DNA or protein sequences (see sequencing and genomes), and it can be tested and overturned like any other historical explanation, as testing evolutionary explanations describes.

Most trees are rooted with an outgroup: a taxon assumed to lie outside the group being studied. The root gives the tree a direction in time, from ancestral to derived.

Reading relatedness from the MRCA, and four misreadings

Worked problem, part 1. Here is a rooted tree, written in parenthetical form as (lamprey, (shark, (trout, (frog, human)))). Each pair of parentheses is a clade. Lamprey is the outgroup. Drawn out, with the internal nodes labelled R (root), J, B and T:

       ┌─────────────── lamprey
  R ───┤
       │   ┌─────────── shark
       └ J ┤
           │   ┌─────── trout
           └ B ┤
               │   ┌─── frog
               └ T ┤
                   └─── human

Step 1. Is trout closer to shark or to human? Trace each pair back to where their lines meet. Trout and human meet at node B. Trout and shark meet at node J, which is older, because J is ancestral to B. The trout-human MRCA is more recent, so trout is more closely related to human (and equally to frog). That trout and shark sit next to each other on the page and look alike is irrelevant.

Step 2. Swap frog and human. Rotate node T so that human is drawn above frog. Node T still has the same two descendants, and every clade has the same members. Nothing about the relationships has changed.

Node R (root) J B T Rotate this node

Tip 1 lamprey shark trout frog human Tip 2 lamprey shark trout frog human

Character Jaws (present in shark, trout, frog, human) X, mapping (a): gain, then loss in frog X, mapping (b): two separate gains X, mapping (c): gain at the base, two losses

In this sketch: parsimony only, with the outgroup (lamprey) state taken as ancestral; tip states for X are made up; branch lengths mean nothing; bootstrap values are the lesson’s two example numbers.

Find the most recent common ancestor; do not read relatedness from how the tips are drawn. Rotating a node, tip order and node counts change nothing, and a bootstrap value is a percentage of replicate trees, not a probability that the clade is true. This is a sketch of the logic.

That step points to the first of four common misreadings.

1. Rotation. Any node can be spun like a mobile without changing the tree. A tree can be drawn with human at the top or the bottom, and the history is the same.
2. Adjacency. Tips next to each other on the page are not, for that reason, closest relatives. Shark sits beside trout in the drawing above, but trout’s sister group is the frog-human clade.
3. Counting nodes. The number of nodes between two tips depends on which other species were sampled. Add five more fish to the tree and the node count between trout and human rises, though their MRCA has not moved. Find the MRCA; do not count steps.
4. “More evolved.” Every living tip has had the same amount of time since the root. The human and rainbow-trout lineages have each been evolving since their MRCA, at node B. Lampreys are not primitive animals. Some of their characters, such as the absence of jaws, are ancestral; others are their own derived features. Characters, not whole organisms, are ancestral or derived.

Branch lengths need one more warning. On a cladogram, only the branching order means anything, and branch lengths are drawn for convenience. On a phylogram, branch lengths are scaled to estimated change, usually in substitutions per site, not years. Converting change into time needs a molecular clock and a calibration, the subject of molecular evolution.

Characters on branches: homology, convergence and parsimony

A character is a feature that varies among taxa, such as the presence of jaws, and its forms are states. The state found in the outgroup is taken as ancestral; a state that arose within the group is derived. Changes in state happen along branches, so you map them there.

Homology is similarity due to a single origin in a common ancestor. Convergence is similarity that arose independently in separate lineages. Parsimony prefers the mapping that needs the fewest changes. It is a rule for choosing among explanations, not a claim that evolution takes the shortest path.

Worked problem, part 2.

Step 3. Jaws. Jaws are absent in lamprey and present in shark, trout, frog and human. Taking the outgroup’s state, absence, as ancestral, one gain on the branch from R to J explains all five tips. That is 1 change, and jaws in sharks and humans are homologous under this mapping.

Step 4. A made-up character X. Suppose X is in state 1 in trout and human and state 0 in lamprey, shark and frog. These states are illustrative. With 0 ancestral, try each mapping:

• (a) Gain on the branch from J to B, which gives state 1 to trout, frog and human; then a loss on the branch to frog. 2 changes. Here X in trout and human is homologous.
• (b) A separate gain on the branch to trout and another on the branch to human. 2 changes. Here X in trout and human is convergent.
• (c) Gain on the branch from R to J, then losses in shark and in frog. 3 changes.

No mapping does it in one, because no single branch leads to trout and human alone. So the minimum is 2, and mappings (a) and (b) tie. Parsimony alone cannot tell homology from convergence here. You would need more evidence: the developmental or genetic basis of X, fossils with intermediate states, or more taxa.

Real convergence is inferred the same way. In the Greater Antilles, Anolis lizards fall into habitat types called ecomorphs (sets of species with similar habitat use and body form), such as twig and trunk-ground specialists. Losos and colleagues, in 1998, compared these types with a molecular phylogeny of species from Cuba, Hispaniola, Jamaica and Puerto Rico, and inferred that ecomorphs arose independently on different islands. Lizards of the same type on different islands were usually not each other’s closest relatives, with two possible exceptions noted by the authors. Not every type occurs on every island: for example, Jamaica has no grass-bush species.

One caution. Different genes in the same genome can have different histories, so a tree from one gene can disagree with the species tree. That disagreement is a property of gene genealogies, not convergence. Gene-family histories of this kind are covered in how signaling evolved.

How trees are built and how confident to be

Building a tree starts with a multiple sequence alignment: sequences of one gene, lined up so that each column holds positions inferred to share an origin. The alignment is itself a hypothesis of homology at every site. The variants in it arise by mutation. Three families of methods then choose among trees:

• Maximum parsimony finds the tree needing the fewest character changes.
• Maximum likelihood uses an explicit model of sequence change and finds the tree under which the observed alignment is most probable.
• Bayesian inference estimates a posterior distribution of trees and model parameters given the data, the model and a prior.

Then comes the question of support. Joseph Felsenstein, in 1985, applied the bootstrap: resample the alignment’s columns with replacement to make many replicate data sets of the same size, infer a tree from each, and record the percentage of replicate trees that contain each clade.

Worked problem, part 3.

Step 5. Suppose the vertebrate tree carries bootstrap values of 98 on (frog, human) and 72 on (trout, (frog, human)). Those clades appeared in 98% and 72% of the replicate trees. They do not mean that the clades are 98% or 72% likely to be true. A bootstrap value tells you how consistently the characters in hand recover a clade under the method used. If the model is wrong in a systematic way, the same wrong clade can come back from every replicate.

That is not a hypothetical worry. Kapli and Telford (2020) simulated protein data sets on a known animal tree with sponges branching first, letting rates of change differ among sites. A model that treats all sites alike recovered the wrong tree in 98 of 100 data sets, and the wrong grouping usually had high bootstrap support. A richer model recovered the true tree in 88 of 100.

The root of the animal tree shows the uncertainty in practice. Schultz and colleagues (2023) mapped the linkage of genes on chromosomes, using unicellular relatives of animals as outgroups, and reported shared derived linkages grouping sponges with cnidarians, placozoans and bilaterians. That places comb jellies as the sister group of all other animals. Copley (2025) argued that their statistical test overstated the significance of some of those linkages. Read this as a live dispute.

Finally, a node with more than two branches, a polytomy, usually means the data were not strong enough to resolve the order of splits. It is rarely evidence that three lineages split at once.

Design problem

Similar twig-dwelling lizards, with short limbs and slender bodies, occur on four Caribbean islands. Design a test that distinguishes one origin of the twig form followed by dispersal between islands from repeated independent origins on each island.

One way to do it

Sample every lizard species you can on all four islands, plus mainland outgroups, and sequence several genes. Do not use genes or traits that were used to define the twig ecomorph, or the test becomes circular. Infer the tree by maximum likelihood or Bayesian methods and assess support by bootstrap or posterior probabilities.

The two hypotheses predict different trees. One origin plus dispersal predicts that the twig species form a single clade. Repeated origins predict that each island’s twig species is nested among other species from its own island, with the twig species of different islands in different clades. Map the twig form onto the tree with the outgroup’s state as ancestral and count the changes each hypothesis needs.

Decide in advance what support you require. A clade of twig species with bootstrap support of 50 does not settle the matter; strong, consistent support across genes and methods comes closer. Even then, convergence on a tree is an inference with uncertainty. This is the logic of Losos and colleagues (1998) for Greater Antillean Anolis.