Evolutionary transitions to self-fertilization influence the inference of introgression history
The availability of polymorphism data and statistical inference methods allows documenting the widespread occurrence of introgression and hybridization across the tree of life. However, these methods are primarily optimized for outcrossing species without generation overlap or seed banking, thereby ignoring the consequences of life-history traits (and their evolution) on genome-wide polymorphism patterns. We investigate how a transition from outcrossing to selfing, a common feature of plant species, may affect the inference of introgression history. We simulate six demographic models with different histories of gene flow under two mating-system scenarios: a constant high selfing rate and a transition-to-selfing scenario. Using an Approximate Bayesian Computation framework with random forests, we compare model choice based on genotypic summary statistics alone and in combination with coalescent statistics derived from coalescent tree sequences. Including coalescent information substantially improves model classification, especially for distinguishing secondary contact and continuous gene flow. Cross-classification of pseudo-observed datasets shows that ignoring a transition to selfing can lead to false demographic inferences, with transition-to-selfing data often misclassified as ancient gene flow or secondary contact when analyzed under a constant selfing model. We then apply this inference framework to genomic data from Arabis nemorensis and Arabis sagittata, two predominantly selfing species with evidence for post-split hybridization. Our analyses reveal a likely transition to selfing roughly 470,000-890,000 years ago, and a likely continuous level of gene flow after the species split. The latter results lead us to revisit our previous scenario of gene flow due to secondary contact between species inferred under constant selfing. Changes in mating systems and, by extension, life-history traits can therefore bias inference about introgression if they are not explicitly modeled. Tree-sequence-based coalescent statistics provide useful information for inferring complex demographic histories that involve both gene flow and transitions to selfing.