Signature of balancing selection in Arabidopsis

Abstract

Natural selection and genetic linkage cause DNA segments to have genealogical histories resembling those of the selected sites. When a polymorphism maintained by selection is old, it will have an island of enhanced sequence variability surrounding it, which represents a detectable "signature of selection." We investigate the structure of single-nucleotide polymorphisms (SNPs) in a 20-kb interval containing the Arabidopsis thaliana disease resistance gene RPS5, a locus containing common alleles for the presence/absence of the entire locus. The alleles are considerably diverged at surrounding sites, indicative of an old polymorphism maintained by selection. The island of "enhanced" variability extends several kilobases to either side of the RPS5 deletion junction, and these SNPs are in nearly complete linkage disequilibrium with the RPS5 insertion/deletion. At a distance of 10 kb to either side of the locus, however, we find low levels of polymorphism and the absence of linkage disequilibrium between individual SNPs and RPS5 alleles. Our results show that the interval of enhanced variability surrounding this balanced polymorphism in Arabidopsis is large enough to be readily detected, but small enough to span the focal gene and few others. For this species it should be possible to identify the complete set of genes with long-lived polymorphisms, a potentially important subset of genes segregating for functional variants.

Fig 1.

The structure of the RPS5 gene regions. Open boxes depict location of known and predicted coding regions; filled boxes are regions included in this study. The numbers above solid boxes are nucleotide positions at the ends of the aligned sequences, with DJ for the RPS5 polymorphism set to 0.

Fig 2.

SNPs within and between RPS5 allelic classes, excluding singletons, and sequence at these positions in A. lyrata. Differences from the reference sequence Col-0 (GenBank accession no. ) are presented. The numbers of singleton polymorphisms are given in parentheses alongside accession names. (A) 5′-10kb and CR (upstream of DJ). (B) CR (downstream of DJ) and 3′-9kb. Some recombinational shuffling is evident within the CR, as indicated by long tracks of SNPs present in the 5′ region of the resistance alleles Kz-3 and Lip-0 that are characteristic of SNPs belonging to the susceptibility allele class, and tracks of SNPs present in the 3′ region of the susceptibility alleles Ab-27 and Rf-4 that carry SNPs characteristic of resistance alleles.

Fig 3.

Sliding window analysis of silent (synonymous and noncoding) sites in RPS5 flanking regions. Window size is 250 silent sites with a 25-site increment. Abscissa is base position relative to DJ in aligned sequences. (A) Observed nucleotide diversity between RPS5+ and RPS5− alleles and average divergence between A. thaliana and A. lyrata sequences, with Jukes-Cantor correction (39). (B) Results of a coalescent model with selection and recombination (3, 10). For each sliding window, plotted is the ratio of average pairwise difference between allelic classes (observed or predicted under balancing selection) to average pairwise difference expected under the standard neutral model (see Materials and Methods for details). Results for two scaled recombination rates are shown for balancing selection (BS) predictions: ρ = 6 × 10⁻⁴, our best estimate of this parameter, and ρ = 6 × 10⁻⁵, a 10-fold lower rate. Both expected curves assume an equilibrium frequency of the two allelic classes, P = 0.5; The scaled mutation rate (β) between RPS5+ and RPS5− is fitted for each curve to bring the predictions close to data at the site under selection (assumed to be the DJ).