TL;DR
This paper reveals that intralocus recombination, a common biological process, can lead to an "inconsistency zone" where standard methods for inferring species trees from genetic data can fail, even with abundant data. This occurs when recombination rates differ significantly between closely related species, exacerbating the effects of incomplete lineage sorting.
What Problem Does This Paper Address?
Estimating the evolutionary history of species (phylogenetics) from genetic data is a complex task. One major challenge is "incomplete lineage sorting" (ILS), where gene lineages don't sort perfectly into species lineages, leading to conflicting signals in the data. Another factor, intralocus recombination (recombination within a single gene), has historically been assumed to be negligible or absent in many phylogenetic models. However, recent work suggests this assumption might be too strong, especially for certain types of genes and species. This paper investigates how intralocus recombination, particularly when it varies across lineages, impacts the accuracy of species tree inference methods, specifically focusing on those that rely on inferring relationships from "rooted triplets" (relationships among three species).
Methods
The authors employ a two-pronged approach:
- Analytical Proofs: They develop a mathematical framework based on the multispecies coalescent with recombination (MSCR) model. This model explicitly incorporates intralocus recombination. They then use this framework to analytically demonstrate the existence of an "inconsistency zone" for a common triplet-based method called R (Majority-Rule Rooted Triple). They prove that under certain conditions (specifically, differential recombination rates and short internal branches in the species tree), R can converge to the wrong species tree topology.
- Simulation Study: To validate their analytical findings and characterize the inconsistency zone, they conduct extensive simulations. They generate synthetic genetic data under various parameter settings (recombination rates, mutation rates, branch lengths) and evaluate how often the R method infers the correct species tree topology. This allows them to map out the conditions under which R fails.
Datasets
Not discussed in the paper. The study relies on simulated data generated according to the MSCR-JC(k) model.
Key Insights and Ideas
- Intralocus recombination is not always benign: While often ignored, intralocus recombination can be a significant source of phylogenetic error, especially when its rate varies across lineages.
- The "Inconsistency Zone": The paper formally defines and characterizes an "inconsistency zone" where standard triplet-based methods fail. This zone is not due to a lack of data but rather to specific biological conditions.
- Differential Recombination is Key: The inconsistency arises not just from recombination itself, but crucially from differential rates of recombination between closely related taxa. This asymmetry amplifies the effects of ILS.
- Short Internal Branches Matter: The problem is exacerbated when the time between speciation events (internal branch lengths) is short, as this is when ILS is already high.
- R* is Susceptible: The popular R* method, which relies on inferring relationships from rooted triplets, is shown to be particularly vulnerable to this inconsistency.
Key Results
- Analytical Proof of Inconsistency: The authors mathematically prove that for certain species phylogenies with differential intralocus recombination rates and short internal branches, R* is not statistically consistent, meaning it won't reliably infer the correct topology even with infinite data.
- Characterization of the Inconsistency Zone: Simulations confirm the analytical results, showing that the inconsistency zone exists for biologically plausible parameter values (e.g., eukaryotic recombination and mutation rates).
- Magnitude of Error: While the inconsistency can lead to incorrect inference, the simulations suggest that the magnitude of the error (i.e., how much the inferred topology deviates from the true one) is often relatively small.
- Robustness under Uniform Recombination: When recombination rates are similar across all lineages, the methods appear to be more robust, even with short internal branches.
Limitations and Future Work
- Idealized Model: The study focuses on a simplified, ultrametric three-taxa species phylogeny and uses the Jukes-Cantor model for mutations. Future work could explore the impact of variable population sizes and more complex mutation/recombination models.
- Distance-Based Inference: The theoretical results primarily apply to distance-based methods for inferring gene trees. Extending these findings to likelihood-based inference would be valuable.
- Unrooted Quartets: The paper briefly touches upon unrooted quartets, suggesting similar issues might arise, which warrants further investigation.
- Other Methods: While R* is analyzed, the implications for other summary coalescent-based methods (e.g., ASTRAL) are not fully explored.
Summary updated