Genome-wide analysis of Alu editability

Abstract

A-to-I RNA editing is apparently the most abundant post-transcriptional modification in primates. Virtually all editing sites reside within the repetitive Alu SINEs. Alu sequences are the dominant repeats in the human genome and thus are likely to pair with neighboring reversely oriented repeats and form double-stranded RNA structures that are bound by ADAR enzymes. Editing levels vary considerably between different adenosine sites within Alu repeats. Part of the variability has been explained by local sequence and structural motifs. Here, we focus on global characteristics that affect the editability at the Alu level. We use large RNA-seq data sets to analyze the editing levels in 203 798 Alu repeats residing within human genes. The most important factor affecting Alu editability is its distance to the closest reversely oriented neighbor-average editability decays exponentially with this distance, with a typical distance of ∼800 bp. This effect alone accounts for 28% of the total variance in editability. In addition, the number of Alu repeats of the same and reverse strand in the genomic vicinity, the expressed strand of the Alu, Alu's length and subfamily and the occurrence of reversely oriented neighbor in the same intron\exon all contribute, to a lesser extent, to the Alu editability.

Figure 1.

Distribution of Alu editibility. Editability is calculated as the ratio of inosines to (inosines+adenosines) in all reads coming from the Alu element of choice (see (9) for more details). Almost all Alu elements are edited to some extent, but editability is typically less than 1%.

Figure 2.

Editing level increases with decreasing distance to the nearest reversely oriented neighbor. (a) The distance dependence closely follows an exponential function, with a typical decay distance of ∼800 bp (see Equation (1)). The same trend is observed when looking only at strong editing sites, where the editing level is >25%. (b) A slight decrease in editability is observed for Alu elements which are very close to their reverse neighbor. (c–e) Too small a distance is detrimental for editing. A schematic illustration of two reversely oriented neighboring Alu elements that form a dsRNA structure. (c) A long dsRNA structure is formed when both Alu elements has an optimal distance between them. (d) When the two Alu elements are too close, with no spacing between them, the RNA flexural rigidity prevents full pairing. In this case, the RNA bases next to the neighboring ends of the Alu elements are likely to be less tightly bound, and thus less edited. (e) A case of nested Alu elements. A positive strand Alu resides in the middle of a negative strand Alu. Here too, although the distance between the elements is formally zero, pairing is negatively affected.

Figure 3.

Number of Alus in the neighborhood affects editability. (a) One finds a positive correlation of editability with the number of reversely oriented repeats in the genomic neighborhood (10 000 bp each side) and a negative correlation with the number of same strand elements. (b) The effect is even stronger when looking at the immediate neighborhood (2000 bp each side). Note that we plot the difference between the observed editing level and the average level for all Alu whose nearest neighbor is at the same distance (formula (1)). This difference could be positive or negative.

Figure 4.

Editibility versus Alu length. Alus much shorter than the typical length are less editable, as they form weaker dsRNA. Elements too long are also less edited, on average, since their neighbor can bind to only a part of the long element. The longer the neighboring repeat is, the stronger the editing.

Figure 5.

Pairing with a repeat in the same exon facilitates stronger editing. Editing is weaker when the closest neighbor resides in a different gene segment (exon\intron). Elements in an exon, with a nearest neighbor on the same exon, are very highly edited.

Figure 6.

Editing versus Alu strand. Poly-A strand is more editable. Moreover, the distance dependence is different between the two strands. Up to a distance of ∼700 bp poly-U Alu elements are significantly more editable, while the reverse is seen when the neighbor further apart.

Figure 7.

Editing versus similarity between the two neighbors. We used BLAST alignment as a proxy for sequence similarity. The more similar the Alu repeats are, the higher is the editing level. Pairs with extremely high identity score are less well edited. Possibly, this is due to the almost perfect dsRNA helix formed, lacking of A:C pairing in the secondary structure, which are preferred targeted by ADARs.

Figure 8.

Editing in the various Alu subfamilies. Similar picture is observed after correcting for distance to the nearest reversely oriented Alu.

Figure 9.

A typical genomic 10k bp neighborhood of an Alu. The UCSC track presents a part of the ATM gene, which shows four of its exons. This part of the gene includes 16 intronic Alus, 9 are in ‘+’ orientation and 7 are in the ‘-’ orientation. Below, appears a figurative representation of the putative paired-Alu dsRNA structures that might form. Alus are shown around the outer ring and are oriented in a clockwise direction with ‘+’ Alu indicated in red and ‘–’ in pale blue. Neighboring inverted non-diverse Alu closer than 3500 bp are connected by a line. Other tracks contain coverage (light-green bars) and editing levels (light-orange bars). Two of the Alus are not editable according to our criteria.