Recombinational landscape and population genomics of Caenorhabditis elegans

Abstract

Recombination rate and linkage disequilibrium, the latter a function of population genomic processes, are the critical parameters for mapping by linkage and association, and their patterns in Caenorhabditis elegans are poorly understood. We performed high-density SNP genotyping on a large panel of recombinant inbred advanced intercross lines (RIAILs) of C. elegans to characterize the landscape of recombination and, on a panel of wild strains, to characterize population genomic patterns. We confirmed that C. elegans autosomes exhibit discrete domains of nearly constant recombination rate, and we show, for the first time, that the pattern holds for the X chromosome as well. The terminal domains of each chromosome, spanning about 7% of the genome, exhibit effectively no recombination. The RIAILs exhibit a 5.3-fold expansion of the genetic map. With median marker spacing of 61 kb, they are a powerful resource for mapping quantitative trait loci in C. elegans. Among 125 wild isolates, we identified only 41 distinct haplotypes. The patterns of genotypic similarity suggest that some presumed wild strains are laboratory contaminants. The Hawaiian strain, CB4856, exhibits genetic isolation from the remainder of the global population, whose members exhibit ample evidence of intercrossing and recombining. The population effective recombination rate, estimated from the pattern of linkage disequilibrium, is correlated with the estimated meiotic recombination rate, but its magnitude implies that the effective rate of outcrossing is extremely low, corroborating reports of selection against recombinant genotypes. Despite the low population, effective recombination rate and extensive linkage disequilibrium among chromosomes, which are techniques that account for background levels of genomic similarity, permit association mapping in wild C. elegans strains.

Figure 1. Recombination rate domains.

Marey maps for each chromosome show genetic position of each marker (black points) as a function of physical position. Genetic position is measured in centiMorgans as defined on the recombinant inbred advanced intercross line population; these are not meiotic distances. Gray lines show the fits of segmented linear regressions, which estimate the boundaries of the recombination domains and their relative recombination rates. The shaded boxes above each plot show the genetically defined positions of the pairing centers .

Figure 2. Simulated chromosomes.

(A) The Marey maps for actual chromosome III data (black) and 10 chromosome III datasets simulated with discrete, constant-rate recombination domains (colors) show that variation within domains and indistinct boundaries between domains are expected. (B) The observed genetic length of chromosome III is smaller than expected. The histogram shows the lengths of 1000 chromosome III datasets simulated assuming one crossover per meiosis.

Figure 3. Allele frequencies in the recombinant inbred advanced intercross lines.

(A) The frequency of the N2 allele at each marker along the chromosomes. The expected frequency, 0.5, is represented by the gray line. (B) A close view of the allele frequencies on the left side of chromosome II shows a significant skew toward N2 alleles. The histogram on the right represents the maximum allele frequency skew for 1000 simulated chromosome II datasets. The red line represents the p = 0.01 significance threshold from the simulations.

Figure 4. Breakpoint counts exhibit linkage to genomic intervals.

Lod scores from nonparametric interval mapping are plotted as a function of genetic position for the four breakpoint count traits that exhibit significant linkage. Horizontal lines represent trait-specific genome-wide significance thresholds (p = 0.05) estimated by structured permutation.

Figure 5. The Hawaiian isolate CB4856 has a large excess of rare alleles.

Each of 1460 SNPs is plotted according to the frequency of the minor allele (black) or the frequency of the CB4856 allele (blue). Under panmixis, our SNP ascertainment should cause both sets of points to fall on straight lines, connecting allele frequencies 1/41 and 20/41 for minor allele frequency and 1/41 and 40/41 for Hawaii allele frequency. The plot shows that there is a large excess of rare alleles and that these rare alleles are CB4856 alleles.

Figure 6. Wild isolate genomes.

(A) The effects of SNP ascertainment on haplotypes. All SNPs were ascertained by comparing N2 and CB4856, and must therefore have arisen by mutation on the genealogical branches connecting those two strains. When a third strain is considered, there are three possible genealogies, but all SNP-generating mutations must reside on the ascertained branches, shown in red. The allelic states of the ascertained strains are shown as blue (CB4856) and orange (N2), and the wild isolate allele will be shared with either strain with probabilities that depend on the genealogy. (B) Expected wild isolate haplotypes from each of the genealogies under ascertainment. Typical haplotypes are represented as strings of SNP alleles colored by whether they are identical to N2 or to CB4856. In genealogies 1 and 3, most mutations will fall on the long outgroup branch, and the wild isolate will resemble the strain with which it shares a recent ancestor. In genealogy 2, the two ascertained branches have equal length with respect to the wild isolate, yielding an equal probability of each allele at each position. (C) Haplotypes of wild isolates. Each of the 41 distinguishable haplotypes is represented as a row for each chromosome. N2 carries haplotype 1 (all orange alleles) and CB4856 carries haplotype 41 (all blue). Putative deletions are red. The bracket above the X chromosome labels the interval across which haplotypes 29 and 39 exhibit haplotypes consistent with genealogy 3.

Figure 7. Population Structure.

Distinguishable wild isolate haplotypes are represented as rows. At left, the allelic composition of each haplotype is represented by orange and blue bars. At right, population assignments from structure are shown for each haplotype, with the most N2-like ancestral population orange and the most CB4856-like blue. Likelihoods for alternative numbers of ancestral populations (K) are shown below each plot. At K = 1, lnL = −23076.

Figure 8. Population genetic and meiotic recombination rate estimates.

(A) The population effective recombination parameter estimate (per base pair) is plotted in black for sliding windows of 2 Mb centered on each SNP. Estimates are derived from the rate of decay of linkage disequilibrium with physical distance. Red bars indicate the estimates of for whole recombination rate domains (arms and centers), and green bars indicate , the estimated meiotic recombination rate per base pair, inferred for each domain from the recombination fraction observed in RIAILs (Figure 1; Table 1). (B) Domain-specific estimates of ρ and c are correlated, and is about 40% the magnitude of .

Figure 9. Linkage disequilibrium among unlinked sites.

Every pair of unlinked SNPs with a significant r² at the specified false discovery rate is plotted. The axes represent the physically ordered SNPs spaced equally and not by distance.

Figure 10. Association mapping in wild C. elegans.

Negative log p-values for each of 907 SNPs are shown for two traits and three tests of association. SNPs are ordered by physical position from chromosome I through X. Gray lines represent Bonferonni-corrected p = 0.05 significance thresholds. In the upper plot, the highly significant red points cover identically positioned green points, and these points are plotted at an arbitrary −ln(p) of 40 because the perfect genotype-phenotype association yields an infinite −ln(p) under the mixed-model LRTs. The key applies to both panels.