Purifying selection of long dsRNA is the first line of defense against false activation of innate immunity

Abstract

Background: Mobile elements comprise a large fraction of metazoan genomes. Accumulation of mobile elements is bound to produce multiple putative double-stranded RNA (dsRNA) structures within the transcriptome. These endogenous dsRNA structures resemble viral RNA and may trigger false activation of the innate immune response, leading to severe damage to the host cell. Adenosine to inosine (A-to-I) RNA editing is a common post-transcriptional modification, abundant within repetitive elements of all metazoans. It was recently shown that a key function of A-to-I RNA editing by ADAR1 is to suppress the immunogenic response by endogenous dsRNAs.

Results: Here, we analyze the transcriptomes of dozens of species across the Metazoa and identify a strong genomic selection against endogenous dsRNAs, resulting in their purification from the canonical transcriptome. This purifying selection is especially strong for long and nearly perfect dsRNAs. These are almost absent from mRNAs, but not pre-mRNAs, supporting the notion of selection due to cytoplasmic processes. The few long and nearly perfect structures found in human transcripts are weakly expressed and often heavily edited.

Conclusion: Purifying selection of long dsRNA is an important defense mechanism against false activation of innate immunity. This newly identified principle governs the integration of mobile elements into the genome, a major driving force of genome evolution. Furthermore, we find that most ADAR1 activity is not required to prevent an immune response to endogenous dsRNAs. The critical targets of ADAR1 editing are, likely, to be found mostly in non-canonical transcripts.

Fig. 1

Detection of putative long dsRNA structure across transcriptomes. a Transcriptomes of 49 organisms, from yeast to human, were analyzed. Using BLAST, we searched for long highly similar sequences within the same mRNA and pre-mRNA sequence. Reversely oriented sequences (red) are likely to pair and form a long intra-molecular dsRNA structure. As a natural control, we use same-strand tandem duplicated sequences (blue), which are not relevant to the secondary structure. b For example, looking at the pre-mRNA of the human otud7b gene, multiple same-strand (blue) and inverted (red) matches are found, most of these pair together repetitive elements (orange) within introns (thin black line). In contrast, the mRNA molecule shows very few hits, all of them pair tandem sequences within repetitive elements in the 3′ UTR. Green bars represent A-to-I editing events, all of which are located in regions that have an inverted sequence match in the pre-mRNA

Fig. 2

Putative dsRNAs are depleted in mRNA across organisms. Comparison of inverted duplicated sequences (potentially folding into dsRNA; red) to tandem duplicated sequences (control; blue) across a wide range of organisms. For each organism, we present the relative abundance of each of the two types of alignments. a In the mRNA molecules, the potential for dsRNA formation is strongly depleted in most organisms. b In contrast, pre-mRNA molecules exhibit, for most organisms, a number of potential dsRNA regions similar to that of the control

Fig. 3

Long and nearly perfect duplexes are extremely rare. a Number of genomic nucleotides within putative dsRNAs (IDS) and control (TDS) regions, summed over all organisms, as a function of the region length (right) and the similarity (left), for both mRNA (top) and pre-mRNA (bottom). Note the logarithmic scale. For regions longer than 300 bp and of very high identity, the depletion of putative dsRNAs becomes more pronounced, even in pre-mRNAs. b Comparison of long (> 300), nearly perfect (> 96%), inverted duplicated sequences (potentially folding into dsRNA; red) to tandem duplicated sequences (control; blue) across a wide range of organisms, for pre-mRNA molecules. In most organisms, these putative dsRNA structures are completely depleted. Gray indicates no data (zero IDS and zero TDS). c In mRNA, there are only 4 long and nearly perfect structures in all organisms combined (2419 bps), compared with 258 control TDSs (172,326 bps)

Fig. 4

Expression and editing of long and nearly perfect duplexes. Depletion of long and nearly perfect dsRNAs from the human transcriptome is more pronounced for regions that are expressed more strongly. In this figure, expression was calculated based on a pool of 30 GTEx samples, from 15 different tissues (Additional file 1: Table S6). a Distribution of dsRNA tightness (%identity) for several expression levels. Many of these regions are not expressed at all, and the ones that are (very weakly) expressed show a reduced fraction of nearly perfect (> 95%) structures. b Similarly, long structures are depleted in the expressed regions, compared with the ones that show no expression. c Twenty long and nearly perfect human structures were expressed at a level exceeding FPKM = 0.01 (roughly, 0.002–0.02 RNA molecules per cell [31]). Seven of these (including the two structures expressed at levels exceeding FPKM = 0.1) are indeed edited appreciably (black arrows), possibly bringing the edited structure below 96% identity. The single point marked by a dashed line corresponds to the data presented in d. d Actual editing pattern for one of the unwound structures. A putative dsRNA structure, located within an intron of klhdc1 (chr14:50213276-50215083), is expressed in our pool at a level of FPKM = 0.13. These transcripts are heavily edited, with an estimated number of 14 inosines per transcript, on average. Top: reads mapped to 1 of the 2 arms of the structure (see region coordinates noted in the panel). Data accumulates reads from the 30 pooled samples. Editing events (A-to-G mismatches) show up in brown. Bottom: pile-up of the reads coverage, 59 different editing sites are observed in this 291-bp-long region. They are marked by green and brown bars (standing for A and G fractions, respectively)