Towards reliable detection of introgression in the presence of among-species rate variation

Abstract

The role of interspecific hybridization in the context of diversification dynamics has recently seen increasing attention. Genomic research has now made it abundantly clear that both hybridization and introgression - the exchange of genetic material through hybridization - are far more common than previously thought. Moreover, even highly divergent species were found to hybridize and backcross. These findings raise the question whether commonly used methods for the detection of introgression are applicable to such divergent hybridizing species, given that most of these methods were originally developed for analyses at the level of populations and recently diverged species. In particular, the assumption of constant evolutionary rates, which is implicit in many commonly used approaches, is more likely to be violated as evolutionary divergence increases. To test the limitations of introgression detection methods when being applied to divergent species, we simulated thousands of genomic datasets under a wide range of settings, with varying degrees of among-species rate variation and introgression. Using these simulated datasets, we were able to show that commonly applied statistical methods, including the D-statistic and tests based on sets of phylogenetic trees, produce false-positive signals of introgression between highly divergent taxa when these have different rates of evolution. These misleading signals are caused by the presence of homoplasies that occur at different rates when rate variation is present. To distinguish between the patterns caused by rate variation and genuine introgression, we developed a new test that is based on the expected clustering of introgressed sites and implemented this test in the program Dsuite.

Incomplete lineage sorting and hybridization as drivers of tree discordance in Petunia and related genera (Petunieae, Solanaceae)

“…Branch length methods are also susceptible to the influence of rate variation. For instance, QuIBL has demonstrated a propensity for producing false positives when using shorter alignments in comparison to alignments longer than 1,000 bp (Koppetsch et al 2023). Notably, our QuIBL dataset exhibited a median length of 494 bp, with10% of the alignments being longer than 900 bp.…”

Genealogical asymmetry under the IM model and a two-taxon test for gene flow

“…Despite the strengths mentioned above, the ABBA-BABA test and other asymmetry-based methods are fundamentally limited by the fact that they only consider a small fraction of the information contained within a sample of genomes (Lohse and Frantz 2014;Martin and Amos 2021). Consequently, these methods cannot distinguish between gene flow and certain scenarios of ancestral population structure (Durand et al 2011) or differences in mutation rate between populations (Xiong et al 2022;Frankel and Ané 2023 but see Koppetsch et al 2023).…”