When we visualize evolution, we typically imagine a single, cleanly branching tree of life. However, if you sequence different genes from the same set of species, you will often find that individual gene histories tell wildly contradictory stories.
For instance, genomic analyses of hornbeams in Plastomes of nine hornbeams and phylogenetic implications by Ying Li et al. and cotton in Comparative Chloroplast Genomics of Gossypium Species: Insights Into Repeat Sequence Variations and Phylogeny by Ying Wu et al. frequently reveal stark topological conflicts between trees built from chloroplast DNA and those built from nuclear DNA. This phenomenon, known as gene tree discordance, is not merely experimental noise; it is a fundamental consequence of how genes inherit across generations.
The Mechanism of Incomplete Lineage Sorting
The primary biological driver of this discordance is Incomplete Lineage Sorting (ILS), or deep coalescence. To understand why it happens, we must look at populations rather than species.
Within any ancestral population, genes exist as a pool of diverse variants (alleles). When a population splits into new lineages during a rapid speciation event—as explored by Julie E. Horvath et al. in Development and application of a phylogenomic toolkit: Resolving the evolutionary history of Madagascar’s lemurs—these ancestral variants are randomly sorted into the new descendant groups.
If the time between consecutive speciation events is long, alleles have ample time to "coalesce" (trace back to a single common ancestor) within each branch, ensuring that the gene tree matches the species tree. However, if the speciation events happen in rapid succession, ancient polymorphisms can pass through the bottleneck of a speciation event without sorting. When they finally coalesce further back in time, the resulting gene tree can easily group species in a way that violates the true historical order of species splits.
Why Concatenation Fails
Historically, the standard workaround for gene tree discordance was concatenation: gluing all genetic sequences together into one massive "super-matrix" and running a standard tree-building algorithm. The underlying intuition was that the true "species signal" would naturally drown out the conflicting noise of individual gene histories.
However, in Likelihood-based tree reconstruction on a concatenation of aligned sequence data sets can be statistically inconsistent, Sebastien Roch et al. mathematically proved that this intuition is fundamentally flawed. Under widespread ILS, maximum likelihood estimation on concatenated data is statistically inconsistent. This means that as you add more genomic data, the algorithm does not converge on the truth; instead, it becomes more confident in the wrong tree topology.
Furthermore, in Phylogenetic mixtures on a single tree can mimic a tree of another topology, Mike Steel demonstrated that mixing different evolutionary processes on a single tree can perfectly mimic an entirely different topology, rendering naive statistical pooling dangerous. As shown by Noah M. Reid in Phylogenetic estimation error can decrease the accuracy of species delimitation: a Bayesian implementation of the general mixed Yule-coalescent model, relying on a single locus or ignoring these underlying gene tree conflicts severely biases downstream tasks like species delimitation.
The Confounding Effects of Duplication and Loss
The problem deepens when ILS co-occurs with structural genomic changes like gene duplication and loss. If an algorithm assumes that gene tree discordance is driven only by duplication and loss, it will misinterpret ILS.
In Unified modeling of gene duplication, loss, and coalescence using a locus tree, Matthew D. Rasmussen explained that when ancestral polymorphism persists across short branches, a naive duplication-loss model views this retained variation as an ancient duplication followed by multiple independent losses in different lineages. This causes a massive overestimation of gene family expansion and contraction rates.
To solve this, the author introduced the concept of a locus tree. A locus tree tracks the physical positions in the genome independently of the genes themselves, creating a multi-layered framework that models duplication, loss, and coalescence simultaneously.
Species Tree
└── Locus Tree (Tracks genomic positions; models Duplication/Loss)
└── Gene Tree (Tracks individual sequences; models Coalescence/ILS)
Reconciliation via Mathematical Frameworks
To accurately reconstruct the history of life, modern phylogenomics relies on reconciliation algorithms that explicitly nest gene trees inside species trees using the multispecies coalescent model.
Tools like DLRS, introduced by J. Sjöstrand et al. in DLRS: gene tree evolution in light of a species tree, use Bayesian frameworks to calculate the probability of a gene tree given a species tree, accounting for rate variations and duplications.
For larger datasets where probabilistic models become computationally prohibitive, parsimony-based approaches offer an elegant alternative. In Most parsimonious reconciliation in the presence of gene duplication, loss, and deep coalescence using labeled coalescent trees, Yi-Chieh Wu et al. developed DLCpar, which uses a "Labeled Coalescent Tree" structure to find the most parsimonious history of duplications, losses, and deep coalescence events. By explicitly penalizing non-coalescence alongside structural mutations, these frameworks prevent the false inflation of genomic changes and correctly isolate the true species tree from the messy histories of individual genes.