Soft selective sweeps: Addressing new definitions, evaluating competing models, and interpreting empirical outliers

Abstract

The ability to accurately identify and quantify genetic signatures associated with soft selective sweeps based on patterns of nucleotide variation has remained controversial. We here provide counter viewpoints to recent publications in PLOS Genetics that have argued not only for the statistical identifiability of soft selective sweeps, but also for their pervasive evolutionary role in both Drosophila and HIV populations. We present evidence that these claims owe to a lack of consideration of competing evolutionary models, unjustified interpretations of empirical outliers, as well as to new definitions of the processes themselves. Our results highlight the dangers of fitting evolutionary models based on hypothesized and episodic processes without properly first considering common processes and, more generally, of the tendency in certain research areas to view pervasive positive selection as a foregone conclusion.

Fig 1. Frequencies of beneficial alleles C and T in the 30th and 60th generation post-onset of selection.

The y-axis gives the selection coefficient of the beneficial mutation C, and the x-axis the selection coefficient of the beneficial mutation T. Thus, along the diagonals s(C) = s(T), whereas off the diagonals, there is a selective differential between the beneficial mutations. As shown in panel (a), a beneficial mutation is likely to reach 50% frequency (their definition of a selective sweep) within 30 generations when selection coefficients are very strong. Panel (b) shows the proportions of replicates in which both beneficial nucleotides are at greater than 5% frequency given that the panel (a) condition has been met (their definition of a soft sweep). As shown, this is most likely when the selective effects are equal and falls off sharply with any selective differential. Finally, panel (c) follows the population to the 60th generation, demonstrating that an appreciable fraction of the scenarios meeting the ‘soft sweep’ definition of Feder, Pennings, & Petrov [3] in panel (b), in fact result in only a single nucleotide being brought to high frequency.

Fig 2

The distributions of the H12 statistic (from 100-kb regions, calculated in 401 SNP sliding windows following [2]) in the simulated Raleigh and Ithaca populations under the slightly modified best-fit demographic models inferred by (a) Duchen et al. 2013 [40] and (b) Arguello et al. 2019 [33]. For each demographic model, 2 scenarios are shown: strict neutrality with a constant recombination rate of 0.5 cM/Mb, as assumed by Garud, Messer, & Petrov [2] (100 replicates) and purifying selection in functional regions using the DFE of deleterious mutations inferred by Johri et al. 2020 [38], with variable recombination rates sampled from the D. melanogaster genome (100 replicates). The x-axes are truncated at 0.05 to show a clearer visualization of the means of the distributions. DFE, distribution of fitness effects.