A nonparametric test reveals selection for rapid flowering in the Arabidopsis genome

Abstract

The detection of footprints of natural selection in genetic polymorphism data is fundamental to understanding the genetic basis of adaptation, and has important implications for human health. The standard approach has been to reject neutrality in favor of selection if the pattern of variation at a candidate locus was significantly different from the predictions of the standard neutral model. The problem is that the standard neutral model assumes more than just neutrality, and it is almost always possible to explain the data using an alternative neutral model with more complex demography. Today's wealth of genomic polymorphism data, however, makes it possible to dispense with models altogether by simply comparing the pattern observed at a candidate locus to the genomic pattern, and rejecting neutrality if the pattern is extreme. Here, we utilize this approach on a truly genomic scale, comparing a candidate locus to thousands of alleles throughout the Arabidopsis thaliana genome. We demonstrate that selection has acted to increase the frequency of early-flowering alleles at the vernalization requirement locus FRIGIDA. Selection seems to have occurred during the last several thousand years, possibly in response to the spread of agriculture. We introduce a novel test statistic based on haplotype sharing that embraces the problem of population structure, and so should be widely applicable.

Figure 1. The Effect of Geography and FRI Alleles on Flowering Time

The map displays the position of 168 accessions from Europe, northern Africa, and western Asia. The histogram inset shows the distribution of the number of leaves at flowering without vernalization for these accessions. In both map and histogram, accessions carrying the fri _Col or fri _Ler alleles are indicated in red or blue, respectively, while the shade of gray in the remaining accessions corresponds to the number of leaves (i.e., 0–20). Bin sizes are halved (i.e., 0–10 leaves) in the histogram relative to the map.

Figure 2. PHS Plotted against Frequency for all Nonredundant Alleles in Our Genome-Wide Dataset (10,961 Alleles)

The percentile lines (95th, 97.5th, and 99th) are computed from the respective percentiles (and average frequency) of 100 overlapping sliding windows (100-point offset between adjacent windows) of 1,000 points along the data (sorted by frequency). The fri _Col allele is shown in red, while the fri _Ler allele and another allele that marks it are shown in blue. fri _Ler is the allele below the 95th percentile line.

Figure 3. Haplotype Sharing of Alleles Plotted by Frequency along Chromosome 4

Points represent the location of alleles, while corresponding lines indicate the average length of haplotype identity flanking the allele. Those alleles present in the upper 99th percentile are indicated in black. The fri _Col allele and associated haplotype are indicated in red. This haplotype is over 600 kb, or 8 cM, long. The allele marking fri _Ler and its associated haplotype are indicated in blue. Only those alleles with a positive score are plotted. Locations of the sequence fragments are indicated at the bottom of the figure.