Biased assessment of ongoing admixture using STRUCTURE in the absence of reference samples

Abstract

Detection of hybridization and introgression is important in ecological research as in conservation and evolutionary biology. STRUCTURE is one of the most popular software to study introgression and allows estimating what proportion of the genome of each individual belongs to each ancestral population, even in cases where no reference sample from the ancestral nonadmixed populations is previously identified. In spite of its frequent use, some studies have indicated that ancestry estimates may not always be reliable. We simulated population data under different conditions with regard to the genetic differentiation between ancestral populations, number of loci considered, number of alleles per marker and hybridization rate, and analysed data with STRUCTURE. When reference samples were not included, the comparison of the known degree of admixture for each simulated individual and the value estimated with STRUCTURE revealed a strong underestimation of the level of introgression, classifying many admixed individuals as nonadmixed. This derives from an inaccurate estimation of the ancestral allele frequencies. When samples from the nonadmixed ancestral population were included as reference in the analyses, the bias in the estimations was reduced. The most accurate estimates were obtained when potentially admixed samples were few in relation to reference samples. Thus, whenever possible, a very large proportion of nonadmixed reference samples should be included in admixture assessments and different approaches should be combined. The misestimate of the amount of introgression can impair our understanding of the evolutionary history of species and misguide conservation efforts.

Does hybridisation with an invasive species threaten Europe’s most endangered reptile? Genomic assessment of Aeolian lizards on Vulcano island

“…Assessments of hybrid zones based on a few (<100) loci can yield inaccurate estimates of introgression rate (Boecklen & Howard 1997), stressing the need for methods which extensively cover the genome. Despite the low information content in each individual SNP, the large number of markers provided by approaches such as RAD-seq offers a cost-effective method to achieve accurate hybrid classification (Twyford & Ennos, 2012) while genotyping a large number of individuals (Ravagni et al, 2021). Other studies observed discrepancies in the hybridisation rate measured using different marker typologies (e.g.…”

“…However, these early studies only used a small panel of allozyme markers (only four "diagnostic" loci; Capula, 1993). The assessment of hybridisation patterns requires very large (>100) panels of markers, as a smaller number of markers can yield both severe underestimation and overestimation of hybridisation rates (Ravagni et al, 2021). Unfortunately, such assessments are so far lacking.…”

Population genomic monitoring provides insight into conservation status but no correlation with demographic estimates of extinction risk in a threatened trout

“…Importantly, we included reference samples (Ravagni et al, 2021) of both rainbow trout (O. mykiss) and Yellowstone cutthroat trout (YCT; O. bouvieri), trout not native to this region but which have been introduced to the LCT range and present a primary threat to the genetic integrity of LCT via introgression. Effective population sizes were interpreted cautiously due to reliance on small and multicohort samples, as we discuss, but provide a standing gage of genetic health and evolutionary potential for LCT.…”

“…We first applied modern sequencing techniques to recently collected (~2015–2017) tissue samples to estimate common genetic monitoring metrics across most remaining conservation populations ( N = 60) of the Lahontan cutthroat trout (LCT, Oncorhynchus clarkii henshawi ). Importantly, we included reference samples (Ravagni et al, 2021) of both rainbow trout ( O. mykiss ) and Yellowstone cutthroat trout (YCT; O. bouvieri ), trout not native to this region but which have been introduced to the LCT range and present a primary threat to the genetic integrity of LCT via introgression. Effective population sizes were interpreted cautiously due to reliance on small and multi‐cohort samples, as we discuss, but provide a standing gage of genetic health and evolutionary potential for LCT.…”

Novel Populations from Simulated Admixed Populations

“…The two main approaches of simulation [26] are forward time simulation, where the evolution of initial population is tracked under various genetic models over multiple generations; and backward time (also known as coalescent simulation), where the simulation begins with the current time and work backwards into history. Hence, it is plausible to study the emergence of novel populations from simulated admixed populations but studies in this area have been sparse as recent simulation studies mainly focus on the admixture models [27][28][29][30][31] or determining population history [29,[31][32][33][34] and adaptations to novel environments [35].…”