Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 9 (5), 1966-1980

Genome-wide Analysis of Drosophila Circular RNAs Reveals Their Structural and Sequence Properties and Age-Dependent Neural Accumulation


Genome-wide Analysis of Drosophila Circular RNAs Reveals Their Structural and Sequence Properties and Age-Dependent Neural Accumulation

Jakub O Westholm et al. Cell Rep.


Circularization was recently recognized to broadly expand transcriptome complexity. Here, we exploit massive Drosophila total RNA-sequencing data, >5 billion paired-end reads from >100 libraries covering diverse developmental stages, tissues, and cultured cells, to rigorously annotate >2,500 fruit fly circular RNAs. These mostly derive from back-splicing of protein-coding genes and lack poly(A) tails, and the circularization of hundreds of genes is conserved across multiple Drosophila species. We elucidate structural and sequence properties of Drosophila circular RNAs, which exhibit commonalities and distinctions from mammalian circles. Notably, Drosophila circular RNAs harbor >1,000 well-conserved canonical miRNA seed matches, especially within coding regions, and coding conserved miRNA sites reside preferentially within circularized exons. Finally, we analyze the developmental and tissue specificity of circular RNAs and note their preferred derivation from neural genes and enhanced accumulation in neural tissues. Interestingly, circular isoforms increase substantially relative to linear isoforms during CNS aging and constitute an aging biomarker.


Figure 1
Figure 1. Annotation and examples of Drosophila circular RNAs
(A) Example of a circularizing exon from scro. A paired-end read maps with an out-of-order junction spanning a splice junction (left gene model), consistent with derivation from a back-splicing event that generates a circular RNA (right gene model). (B) Analysis of all loci to which reads were mapped to putative out-of-order genomic junctions. These exhibit progressive enrichment of annotated CDS-CDS and 5′UTR-CDS splice sites amongst higher expressed loci. (C) Distribution of circular RNAs amongst genes. (D) Fraction of back-spliced relative to all spliced reads amongst well-accumulated circles (≥10 back-spliced reads). For reference, log10(−1) is 10% of reads, and log10(−2) is 1%. (E) Examples that illustrate the diversity in configurations of circularizing exons. The red bars above and orange bars below the gene models denote the number of back-spliced and forward-spliced reads at each junction, respectively, and are drawn to scale. Not shown are other alternative forward-spliced junctions, including skipped exons, or lower-expressed back-spliced junctions. (F) Gene-based assessment of the conservation for 1:1:1 D. melanogaster/D. yakuba/D. virilis orthologs to generate circular RNAs, based on analysis of head total RNA-seq data. All two-way species overlaps were significant by chi-squared test to p<2.2E-16, as was the three-way overlap. See also Figures S1–S2.
Figure 2
Figure 2. Experimental validation of circular RNAs
(A) Northern blots of total and polyadenylated (pA+) RNAs from ovary and head. Linear mRNAs are enriched in pA+ samples while circular species are depleted. (B) Northern analysis comparing total and exoribonuclease RNase R-treated RNAs from heads. Linear mRNA forms are depleted by RNase R while circular species are resistant. (C) Quantitative RT-PCR analysis of RNase R-treated samples. Signals were normalized to those obtained using controls samples that were not treated with RNase R. Most of the circular RNA amplicons were maintained or even increased following RNase R treatment. (D) End-point RT-PCR verifies back-splicing of single-exon and multi-exon circles. Genomic DNA (G) was used a negative control template for these reactions. These tests validate circularization of pan and ank2, which appeared to be sensitive in the RNase R tests.
Figure 3
Figure 3. Circular RNAs exhibit characteristic exonic positions and are internally spliced
(A, B) Analysis of exon positions. To identify functional correlations, we binned circular RNAs according to levels of back-spliced reads. Properties of background circles generated from random exons are shown in black. (A) Start exon of circular RNAs. Background circles exhibit bias to initiate with second exons (e.g., for genes with 3 exons, the only possibility is to circularize exon 2). However, circular RNAs exhibit enhanced involvement of second exons. (B) End exon of circular RNAs. Circular RNAs are most enriched for exons involving positions 2–4 of gene models, and the bias towards the 5′ end increases progressively with higher levels of junction-spanning reads. (C) Genomic annotations of circular RNAs. The dominant classes involve 5′-UTR-CDS exons and CDS-only exons. (D) Assessment of internal splicing of multi-exon circles. Shown is an example from ca-alpha1d, for which some mate-pairs of back-spliced reads are spliced and some derive from the intron. Total RNA-seq data below shows that intronic reads accumulate broadly across neighboring introns not involved in the circle. (E) Genomewide analysis shows that the vast majority of RNA circles from individual genomic loci are mostly spliced. (F) Barplot summary showing that aggregate circular RNAs are predominantly internally spliced.
Figure 4
Figure 4. Long flanking introns are a major determinant for RNA circularization
(A) Examples of well-expressed circular RNAs that are flanked by long upstream and downstream introns. (B) Comparison of upstream intron length. Drosophila introns exhibit a bimodal distribution of short lengths and a broad tail of longer lengths. Circular RNAs are biased to be flanked by longer upstream introns than the typical “long” intron class. (C) Circular RNAs are also biased to be flanked by long downstream introns. (D, E) The correlation of flanking intron length was examined across five different different cutoffs of junction-spanning reads. For both upstream (D) and downstream (E) flanking introns, circular RNAs supported by progressively higher numbers of junction-spanning reads were biased for progressively longer flanking intron lengths. Wilcoxons rank-sum tests showed that the increase in overall flanking intron length was significantly greater for each successive bin of increasing circular RNA expression examined. See also Figures S3–S5.
Figure 5
Figure 5. miRNA binding sites within circular RNAs
(A) A highly expressed circle from muscleblind (mbl), which involves its 5′ UTR and first coding exon and is flanked by long introns. Far greater numbers of back-spliced reads were recovered than for forward-spliced reads. The mbl 5′ UTR includes well-conserved binding sites for several neural miRNAs, including for the miR-279 seed family. (B) Summary of the top pan-Drosophilid-conserved miRNA binding sites located with coding regions of circular RNAs. In general, these are dominantly found on the sense strands of these genes. (C) Example of a coding circular RNA from sickie, which contains a remarkable density of deeply-conserved miR-190 binding sites. It is evident that each miR-190 site has been selectively constrained for primary sequence, relative to the neighboring coding regions. (D) Comparison of density of well-conserved 2–8 seed matches for well-conserved miRNAs in circularizing versus linear coding regions. Circularizing coding regions are subject to much higher miRNA targeting than bulk linear coding transcriptome. This holds true when restricting the comparison directly to linear coding regions of transcripts that undergo circularization. Comparison of different bins of circular RNA accumulation shows that the lowest expression bin has a lower site density compared to the higher bins, which exhibit similar density to 3′ UTRs.
Figure 6
Figure 6. Neural-biased expression and relevance of circular RNAs
(A) Heatmap of circle abundance across 103 Drosophila libraries. Each column represent a library, and each row represents a circular RNA ordered by highest abundance in an individual library. The main trends visible from this summary include a mild increase in circle accumulation across embryonic development, enhanced accumulation of circular RNAs in larval/pupal CNS relative to other dissected tissues of these stages, and predominant accumulation of circles in adult heads relative to all samples. In addition, the ordering of circular RNAs by rows highlights that few circles are well-expressed in a manner that is exclusive of adult heads. (B, C) Comparison of circular RNA expression in heads versus many individual developmental stages, tissues, or cell lines. Plots of numbers of loci (B) and frequency of loci (C) emphasize that few circular RNAs are not expressed in heads, and heads express many circles that are not detected elsewhere. (D) Gene Ontology (GO) term enrichments of circular RNAs reveal many gene sets that relate to neural development and/or neural function. See also Figure S6.
Figure 7
Figure 7. Age-dependent increase in circular RNAs in the nervous system
(A) Developmental timecourse emphasizes that dissected larval/pupal CNS accumulates much higher levels of circular RNAs than does any embryonic stage. Adult heads accumulate even higher levels circular RNAs, and these increase with adult aging, as measured in independent datasets of female and male head libraries. Note the two 20 day male libraries gave nearly identical numbers. (B, C) Analysis of individual circular RNA loci demonstrates their globally increased accumulation at 4 days (B) and 20 days (C) compared to 1 day. To ensure these shifts did not represent selective increases in transcription of genes that generate RNA circles, we plotted changes in RNA circles against the changes in host gene expression. Between 1 and 4 days, the circular RNAs showed significant increase in abundance (Wilcoxon p=1.1e-4) while the host mRNA transcripts were unchanged (p=0.88). Between 1 day and 20 days the increase in circle expression was even more significant (p<2e-16), while host mRNAs showed a slight decrease (p=0.035). (D) Age-dependent changes in selected RNA circles in total RNA-seq data. (E) qPCR assays validate age-dependent increases in accumulation of circular RNAs. See also Figure S7.

Similar articles

See all similar articles

Cited by 221 PubMed Central articles

See all "Cited by" articles

Publication types

LinkOut - more resources