Approximate Bayesian Inference Reveals Evidence for a Recent, Severe Bottleneck in a Netherlands Population of Drosophila melanogaster

Abstract

Genome-wide nucleotide variation in non-African populations of Drosophila melanogaster is a subset of variation found in East sub-Saharan African populations, suggesting a bottleneck in the history of the former. We implement an approximate Bayesian approach to infer the timing, duration, and severity of this putative bottleneck and ask whether this inferred model is sufficient to account for patterns of variability observed at 115 loci scattered across the X chromosome. We estimate a recent bottleneck 0.006N e generations ago, somewhat further in the past than suggested by biogeographical evidence. Using various proposed statistical tests, we find that this bottleneck model is able to predict the majority of observed features of diversity and linkage disequilibrium in the data. Thus, while precise estimates of bottleneck parameters (like inferences of selection) are sensitive to model assumptions, our results imply that it may be unnecessary to invoke frequent selective sweeps associated with the dispersal of D. melanogaster from Africa to explain patterns of variability in non-African populations.T HERE is considerable interest in using population genetic approaches to identify regions of the genome that underlie population-or species-specific adaptations. More generally, such scans of the genome can also answer basic evolutionary questions such as the relative importance of selection and nonadaptive factors (such as changes in the size and structure of populations over time) in shaping patterns of genome variability. The rapid increase in our ability to survey population-level nucleotide variability for larger and larger portions of the genome yields increasing statistical power to distinguish among alternative population genetic models.The most common approach to hypothesis testing has been model based, comparing statistics summarizing features of the observed data, such as the distribution of polymorphism frequencies, to the expectations of a null model (e.g., Tajima 1989;Fu and Li 1993;Fay and Wu 2000). The most commonly used null model assumes a large, panmictic, population with neutral mutations occurring according to the infinite-sites model, hereafter referred to as the standard neutral model (SNM) (see Hudson 1983). Significant departures from the SNM are usually interpreted as evidence for the recent action of natural selection (e.g., Harr et al. 2002;Glinka et al. 2003). However, natural selection is not a unique alternative hypothesis to the SNM. Deviations from the demographic assumptions of the SNM, such as population bottlenecks, can result in the observation of significant departures in the same direction as that expected under selective sweep models (e.g. In Drosophila melanogaster, the observation that variation is substantially reduced outside of Africa (Begun and Aquadro 1993) has focused much attention on understanding the cause of this difference between populations. The species is believed to have an African origin and to have colonized other continents only recently (David and ...

Effects of Linked Selective Sweeps on Demographic Inference and Model Selection

“…It is therefore reasonable to assume that positive selection could have a substantial confounding effect on a variety of population genomic methods for demographic inference in practice, beyond those considered here. In the empirical literature, numerous recent studies of demographic history have found support for contractions and recent expansions of natural populations (Thornton and Andolfatto 2006;Fagundes et al 2007;Gravel et al 2011;Tennessen et al 2012;Duchen et al 2013). While such population size changes are probably common, and our results do not call the major findings of these studies into question, they do suggest that natural selection can exaggerate the inferred intensity of these changes.…”

“…Recent studies of large population samples capable of observing very rare alleles found evidence that this growth has accelerated dramatically within the last several thousand years (Coventry et al 2010;Tennessen et al 2012;Gao and Keinan 2016). Similarly, recent studies in Drosophila melanogaster show evidence of a severe out-of-Africa bottleneck (Begun and Aquadro 1993), but occurring within the last 20,000 years (Li and Stephan 2006;Thornton and Andolfatto 2006).While demographic inference from population genomic data in its various forms has proven to be a successful technique, a unifying assumption of these various inference methods (ABC, SFS-based, and SMC-based approaches) is that the genetic data in question are strictly neutral and free from the effects of linked selection in the genome. While this is an important simplifying assumption, it may be the case that in many populations a sizeable fraction of the genome is influenced by natural selection (Hahn 2008;Sella et al 2009;Corbett-Detig et al 2015).…”

“…Recent studies of large population samples capable of observing very rare alleles found evidence that this growth has accelerated dramatically within the last several thousand years (Coventry et al 2010;Tennessen et al 2012;Gao and Keinan 2016). Similarly, recent studies in Drosophila melanogaster show evidence of a severe out-of-Africa bottleneck (Begun and Aquadro 1993), but occurring within the last 20,000 years (Li and Stephan 2006;Thornton and Andolfatto 2006).…”

Controlling the False-Positive Rate in Multilocus Genome Scans for Selection

“…This model has five parameters: the population mutation rate (u ¼ 4N 0 m, where N 0 is the effective size of the ancestral population), the population recombination rate (r ¼ 4N 0 r), the time at which the derived population recovered from the bottleneck (t r ), the duration of the bottleneck (d), and the severity of the bottleneck ( f, 0 , f # 1). In this study, we use u ¼ 0.01/site, r ¼ 0.1/site, t r ¼ 0.004, d ¼ 0.015, and f ¼ 0.03, as these bottleneck parameters are compatible with data from European samples of D. melanogaster (Thornton and Andolfatto 2006). To perform these simulations, a program was written using the coalescent simulation functions in libsequence (Thornton 2003).…”

“…The beneficial mutation fixed in the recent past at t ¼ 0.0041 or 0.015, and we examined two strengths of selection-a ¼ 2fN 0 s ¼ 100 or a ¼ 1500. We assume N 0 ¼ 2.4 million (Thornton and Andolfatto 2006), and therefore our values of a correspond to s % 7 3 10 À4 and % 0.01, respectively. We are therefore studying the effect of a recent and relatively strong (2fN 0 s?1) sweep occurring at all loci in the history of the derived population.…”