Recombination-associated sequence homogenization of neighboring Alu elements: signature of nonallelic gene conversion

Abstract

Recently, researchers have begun to recognize that, in order to establish neutral models for disease association and evolutionary genomics studies, it is crucial to have a clear understanding of the genomic impact of nonallelic gene conversion. Drawing on previous successes in characterizing this phenomenon over protein-coding gene families, we undertook a computational analysis of neighboring Alu sequences in the genome scale. For this purpose, we developed adjusted comutation rate (aCMR), a novel statistical method measuring the excess number of identical point mutations shared by adjacent Alu sequences, vis-à-vis random pairs. Using aCMR, we uncovered a remarkable genome-wide sequence homogenization of neighboring Alus, with the strongest signal observed in the pseudoautosomal regions of the X and Y chromosomes. The magnitude of sequence homogenization between Alu pairs is greater with shorter interlocus distance, higher sequence identity, and parallel orientation. Moreover, shared substitutions show a strong directionality toward GC nucleotides, with multiple substitutions tending to cluster within the Alu sequence. Taken together, these observed recombination-associated sequence homogenization patterns are best explained by frequent ubiquitous gene conversion events between neighboring Alus. We believe that these observations help to illuminate the nature and impact of the enigmatic phenomenon of gene conversion.

FIG. 1.

Actual and bootstrapped aCMR for the 22 human autosomes and 2 sex chromosomes. The bars in black represent the observed average chromosomal aCMR, whereas the bars in gray represent the aCMR expected by chance alone (bootstrapped aCMR). Error bars represent one standard deviation. The bootstrapped values were calculated by averaging the aCMR of randomly selected Alu pairs from the given chromosomes (50 iterations, each with 20,000 samples). The insert illustrates the actual and expected aCMR for the PAR1 pseudoautosomal regions and the rest of the sex chromosomes. The aCMR for the PAR2 regions is not displayed due to the paucity of Alu sequences in this region.

FIG. 2.

Influence of interlocus distance between Alu pairs on aCMR for parallel Alu pairs and antiparallel Alu pairs. All 384,397 parallel Alu pairs are sorted by their interlocus distances and evenly distributed into 40 bins, and all Alu pairs in a bin are represented by one point, positioned by their average interlocus distance and aCMR. Similarly, all 329,896 antiparallel Alu pairs are also represented by 40 points. The most pronounced decrease in aCMR appears to occur in the 250–700 bp and >3,000 bp range, whereas aCMR is relatively stable in the 700–3,000 bp range.

FIG. 3.

Analysis of aCMR versus longest stretch of sequence identity shared between adjacent Alus. All 714,293 Alu pairs were binned according to each pair's maximal length of uninterrupted bp identity, and the averaged aCMR of all Alu pairs in each bin is plotted, with vertical bars representing one standard deviation. Bootstrapped values were calculated for randomly selected Alu pairs grouped by the same bins. Elevation of aCMR, above baseline, is apparent for Alu pairs sharing just 35–40 bp of uninterrupted identity and increases steadily after that. The elevation is more pronounced in parallel Alus, as opposed to their antiparallel counterparts.

FIG. 4.

Analysis of average distance between comutation sites along the Alu sequence. The genome-wide average distance between comutation sites in neighboring Alus is plotted versus aCMR alongside the average bootstrapped values. Only 77,450 Alu pairs with two or more comutations are plotted. Comutations become, on average, closer than expected, by chance alone, as the aCMR increases.

FIG. 5.

Weak-to-strong bias of comutations, stratified by sequence similarity of Alu pairs. Actual and bootstrapped proportions of weak to strong (AT to GC) and strong to weak (GC to AT) are shown. Alu pairs are binned by their longest stretch of sequence identity.

FIG. 6.

aCMR as a function of chromosomal position in and around the PAR1 regions of the sex chromosomes. Average aCMR along the entire Y chromosome and the equivalent 40 Mb region of X chromosome are shown in green. A sliding window of 15 Alu pairs was used to reduce pair to pair variations of aCMR. A pronounced elevation of aCMR is present in the PAR1 region, the only long region undergoing recombination on both the Y and X chromosomes. (Though the PAR2 region also undergoes recombination, the lack of sufficient Alu pairs in this region hampers the estimation of this regions aCMR.) Schematic drawing of segmental organization of chromosomes X and Y are adapted by permission from Macmillan Publishers Ltd: Nature (Ross et al. 2005), copyright 2005.

FIG. 7.

The percent sequence identities of neighboring Alu pairs, stratified by their interlocus distances. The sequence identities at all 283 positions are shown in the upper panel and that at 232 non-CpG nondiagnostic positions are shown in the lower panel. Each dot groups 1,921 Alu pairs.