Detecting coevolution through allelic association between physically unlinked loci

Abstract

Coevolving interacting genes undergo complementary mutations to maintain their interaction. Distinct combinations of alleles in coevolving genes interact differently, conferring varying degrees of fitness. If this fitness differential is adequately large, the resulting selection for allele matching could maintain allelic association, even between physically unlinked loci. Allelic association is often observed in a population with the use of gametic linkage disequilibrium. However, because the coevolving genes are not necessarily in physical linkage, this is not an appropriate measure of coevolution-induced allelic association. Instead, we propose using both composite linkage disequilibrium (CLD) and a measure of association between genotypes, which we call genotype association (GA). Using a simple selective model, we simulated loci and calculated power for tests of CLD and GA, showing that the tests can detect the allelic association expected under realistic selective pressure. We apply CLD and GA tests to the polymorphic, physically unlinked, and putatively coevolving human gamete-recognition genes ZP3 and ZP3R. We observe unusual allelic association, not attributable to population structure, between ZP3 and ZP3R. This study shows that selection for allele matching can drive allelic association between unlinked loci in a contemporary human population, and that selection can be detected with the use of CLD and GA tests. The observation of this selection is surprising, but reasonable in the highly selected system of fertilization. If confirmed, this sort of selection provides an exception to the paradigm of chromosomal independent assortment.

Figure 1

CLD and GA Test Statistic Distributions The black curve shows the asymptotically expected null test statistic distribution, the gray bars are histograms of the empirical distribution of the test statistics between all SNP pairs on chromosomes 1 and 7, and the red bars are a histogram of test statistics between SNPs in ZP3 and ZP3R for (A) X₁² and (B) X₄².

Figure 2

Q-Q Plot Comparing ZP3-ZP3R with Random SNP Pairs The Q-Q plots compare the (A) X₁²-based and (B) X₄²-based permutation p values between ZP3-ZP3R and an equal number of random SNP pairs between chromosomes 1 and 7.

Figure 3

Q-Q Plot Comparing ZP3-ZP3R with Random Gene Pairs These Q-Q plots compare the (A) X₁²-based and (B) X₄²-based permutation p values between ZP3-ZP3R and PIP5K1A-DPY19L1. The dotted lines indicate significance thresholds with α = 0.05.

Figure 4

Comparative Gene Pair CDFs These plots show the cumulative distribution functions (CDFs) of permutation – log(p) computed between the gene pair of interest in red, 20 comparison gene pairs in gray, and the average CDF of all comparison gene pairs in black. In the top row, – log(p) between ZP3R and ZP3 (red) is compared to – log(p) between random gene pairs on chromosomes 1 and 7 (gray) and the average comparative gene pair – log(p) distribution (black) for (A) X₁²-based and (B) X₄²-based permutation p values. In the second row, – log(p) between PIP5K1A and DPY19L1 (red) is compared to – log(p) between random gene pairs on chromosomes 1 and 7 (gray) and the average comparative gene pair – log(p) distribution (black) for (C) X₁²-based and (D) X₄²-based permutation p values. In the third row, – log(p) in ZP3R-ZP3 (red) are compared to – log(p) between ZP3R and 20 genes on chromosome 7 (gray) and the average – log(p) distribution between ZP3R and chromosome 7 genes (black) for (E) X₁²-based and (F) X₄²-based permutation p values. In the bottom row, – log(p) in ZP3R-ZP3 (red) is compared to – log(p) between ZP3 and 20 genes on chromosome 1 (gray) and the average – log(p) distribution between ZP3R and chromosome 7 genes (black) for (G) X₁²-based and (H) X₄²-based permutation p values.

Figure 5

Power Curves Assuming the selective model described in the text, the exact power of the exact test and the asymptotically derived power of the asymptotic tests were computed for both (A) X₁² and (B) X₄² for 50 values of s ranging from 0 to 1. The dashed curves show exact power, and the solid curves show asymptotically estimated power. Violet, red, blue, black, and green curves are calculated with the use of n = 50, 200, 1000, 1480, and 3000, respectively.