Inferring Balancing Selection From Genome-Scale Data

Abstract

The identification of genomic regions and genes that have evolved under natural selection is a fundamental objective in the field of evolutionary genetics. While various approaches have been established for the detection of targets of positive selection, methods for identifying targets of balancing selection, a form of natural selection that preserves genetic and phenotypic diversity within populations, have yet to be fully developed. Despite this, balancing selection is increasingly acknowledged as a significant driver of diversity within populations, and the identification of its signatures in genomes is essential for understanding its role in evolution. In recent years, a plethora of sophisticated methods has been developed for the detection of patterns of linked variation produced by balancing selection, such as high levels of polymorphism, altered allele-frequency distributions, and polymorphism sharing across divergent populations. In this review, we provide a comprehensive overview of classical and contemporary methods, offer guidance on the choice of appropriate methods, and discuss the importance of avoiding artifacts and of considering alternative evolutionary processes. The increasing availability of genome-scale datasets holds the potential to assist in the identification of new targets and the quantification of the prevalence of balancing selection, thus enhancing our understanding of its role in natural populations.

Keywords: composite likelihood ratio tests; genetic variation; heterozygote advantage; natural selection; population genomics; summary statistics.

Fig. 1.

BLS timeline. References in the figure: Fisher (1922); Levene (1953); Dempster (1955); Dobzhansky (1955); Livingstone (1964); Hubby and Lewontin (1966); Kimura (1968); Gillespie and Langley (1974); Lande (1976); Hughes and Nei (1988, ; Asthana et al. (2005); Bubb et al. (2006); Andrés et al. (2009); Leffler et al. (2013); Bergland et al. (2014); DeGiorgio et al. (2014); Teixeira et al. (2015); Chakraborty and Fry (2016); Siewert and Voight (2017, ; Bitarello et al. (2018); Cheng and DeGiorgio (2019, , ; Isildak et al. (2021).

Fig. 2.

Timescales and genomic signatures of BLS and methods that can detect them. References in the figure: [1] Leffler et al. (2013); [2] Cheng and DeGiorgio (2019); [3] Tajima (1989); [4] Siewert and Voight (2017); [5] Bitarello et al. (2018); [6] Cheng and DeGiorgio (2019); [7] Hudson et al. (1987); [8] DeGiorgio et al. (2014); [9] Isildak et al. (2021); [10] (Soni et al. 2022). In humans, the ranges are roughly defined as follows (see table 2): recent ($<{10}^6$ years); long-term ($>{10}^6$ years); ultra long-term ($>7\times {10}^6$ years).