Advancedadvanced biology

Phylogenomics: Reconstructing the Tree of Life from Genomes

Learning objectives

  • Explain why gene trees and species trees can disagree, with incomplete lineage sorting as the key mechanism.
  • Compare concatenation and coalescent-based species-tree methods and their assumptions.
  • Describe how ortholog identification and alignment quality affect phylogenomic inference.
  • Evaluate controversies at deep nodes: the animal root, bird radiation, and human-chimp-gorilla.
  • Identify open problems: hybridization, horizontal transfer, and model adequacy.

From single genes to genomes

For decades, phylogenies were built from a handful of genes -- ribosomal RNA, a few proteins -- and deep relationships remained contentious. Phylogenomics changed the scale: thousands of loci from whole genomes, bringing statistical power that single genes could never provide. Early triumphs included resolving the animal root debates and the radiation of modern birds. But genomes brought a surprise: with enough data, different genes confidently support different trees. The human-chimp-gorilla trichotomy is the classic case -- about 15% of the genome groups humans with gorillas rather than chimpanzees. This is not error but biology: when speciation events occur close together in time, ancestral polymorphisms sort randomly into descendant lineages, a phenomenon called incomplete lineage sorting (ILS). Phylogenomics therefore had to become a statistical science of conflicting signals, not merely a bigger version of single-gene phylogenetics.

Gene trees versus species trees

A gene tree traces the ancestry of one locus; the species tree traces population splits. Under the multispecies coalescent, gene lineages within an ancestral population coalesce (find common ancestors) at random times, and if speciation intervals are short relative to effective population size, lineages may fail to coalesce before the next split -- producing gene trees that mismatch the species tree. The probability of discordance grows as internode times shrink: rapid radiations are inherently hard. Two methodological families handle this. Concatenation (supermatrix) stitches all genes into one alignment, assuming one tree -- statistically inconsistent under high ILS, converging on the wrong tree with more data. Coalescent summary methods (ASTRAL, MP-EST) instead estimate gene trees first, then find the species tree most consistent with their distribution -- consistent under the coalescent but sensitive to gene-tree estimation error. The field's consensus favors coalescent-aware methods for rapid radiations, with concatenation retained for its power when ILS is low.

Orthologs, alignment, and the garbage-in problem

Phylogenomics stands or falls on data quality. Orthologs -- genes related by speciation -- must be distinguished from paralogs created by duplication, because paralogs trace duplication history, not species history. Automated pipelines (OrthoFinder, BUSCO single-copy markers) approximate this, but hidden paralogy remains a major source of error, especially in plants with rampant whole-genome duplication. Alignment of thousands of loci introduces its own artifacts: misaligned regions create false signal, and aggressive trimming can discard genuine information. Model adequacy matters as much as data volume -- site-heterogeneous models (CAT, LG+C60) that allow different amino acid preferences across sites often overturn trees built under simpler models, famously repositioning the animal root between ctenophores and sponges depending on model choice. Long-branch attraction, where fast-evolving lineages artifactually group together, is amplified by genomic data. The lesson: phylogenomics is not stamp-collecting with more stamps; every step from orthology to model selection is an inference that can mislead.

Case studies: deep nodes under fire

Several flagship controversies illustrate the stakes. The root of the animal tree -- sponges or comb jellies (ctenophores) as sister to all other animals -- flips with dataset and model choice, with profound implications for whether neurons and muscles evolved once or multiple times. The neoavian radiation, when modern bird orders diversified after the K-Pg extinction, produced such extreme ILS that even genome-scale data required coalescent methods and still leaves some nodes uncertain. Within mammals, the placements of bats, tree shrews, and the root of placentals have all been revised by phylogenomics. Perhaps most instructive is what genomes confirmed: the monophyly of previously doubted groups, the Afrotherian radiation uniting elephants with aardvarks and tenrecs, and the fine structure of human evolution including Neanderthal and Denisovan admixture. Each resolved node strengthens confidence in the methods; each contested one maps where biology genuinely resists resolution.

Open frontiers: networks, not just trees

The tree metaphor itself is under revision. Hybridization is rampant -- oaks, Heliconius butterflies, and hominins all exchanged genes after splitting -- so species histories are better represented as networks with reticulation edges. Horizontal gene transfer, dominant in prokaryotes, forces microbiologists toward pangenome and gene-sharing graphs rather than single trees. Introgression deserts and adaptive introgression (e.g., Denisovan EPAS1 aiding Tibetan altitude adaptation) show gene flow can be evolutionarily creative. Methodologically, the frontier is model adequacy: do our substitution models capture real molecular evolution, or do we need machine-learned or mechanistic models? Total-evidence dating combining fossils, morphology, and genomes remains contentious. And a humbling limit persists: for the fastest ancient radiations, the true history may be a near-polytomy -- a genuine simultaneous burst -- that no amount of data can resolve into bifurcations. Phylogenomics is thus maturing from tree-building into the statistical study of how genomes record, distort, and occasionally erase history.

When data disagrees with itself

In phylogenomics, conflict among genes is data, not noise. The pattern of discordance -- which alternative trees appear and how often -- directly estimates ancestral population sizes and speciation tempos through the multispecies coalescent.

Key points

  • Phylogenomics uses genome-scale data, but different genes often support different trees.
  • Incomplete lineage sorting makes gene trees mismatch species trees during rapid radiations.
  • Coalescent summary methods are consistent under ILS; concatenation can converge on the wrong tree.
  • Orthology errors, alignment artifacts, and model misspecification are major error sources.
  • Deep nodes (animal root, bird radiation) remain contested; hybridization forces network thinking.
  • Some ancient rapid radiations may be genuine near-polytomies that no data can fully resolve.

Common misconceptions

✗ Myth: More genomic data always gives the true tree.

✓ Truth: Under incomplete lineage sorting, concatenation methods are statistically inconsistent -- more data converges on the wrong tree with high confidence. Correct methods matter more than data volume.

✗ Myth: A phylogenetic tree shows the complete history of species.

✓ Truth: Trees omit hybridization, introgression, and horizontal transfer, all common in nature. For many groups the true history is a network, and forcing it into a tree discards real biology.

✗ Myth: Mitochondrial DNA gives the species tree.

✓ Truth: mtDNA is a single non-recombining locus with its own history -- subject to ILS, selective sweeps, and introgression. It is one gene tree, useful but not authoritative about species relationships.

Frequently asked questions

What is incomplete lineage sorting?

When two speciation events occur close together, ancestral genetic polymorphisms may sort randomly into descendant species, so a gene's tree differs from the species tree. About 15% of the human genome groups us with gorillas rather than chimpanzees for exactly this reason.

Why do sponges versus comb jellies matter for the animal root?

Because the root determines how many times complex traits evolved. If ctenophores branched first, their neurons and muscles either evolved independently or were lost in sponges -- radically different scenarios for the origin of animal complexity.

How do researchers distinguish orthologs from paralogs?

Through reciprocal sequence similarity, synteny (conserved gene order), phylogenetic tests, and single-copy marker sets like BUSCO. No method is perfect, especially after whole-genome duplications, so hidden paralogy remains a leading source of phylogenomic error.

Can hybridization really overwrite species trees?

Yes. Extensive gene flow can make most of the genome reflect introgression rather than the original split, as seen in oaks and some butterflies. Network methods that model reticulation are replacing strictly bifurcating trees for such groups.