An improved method for circular RNA purification using RNase R that efficiently removes linear RNAs containing G-quadruplexes or structured 3' ends

Abstract

Thousands of eukaryotic protein-coding genes generate circular RNAs that have covalently linked ends and are resistant to degradation by exonucleases. To prove their circularity as well as biochemically enrich these transcripts, it has become standard in the field to use the 3'-5' exonuclease RNase R. Here, we demonstrate that standard protocols involving RNase R can fail to digest >20% of all highly expressed linear RNAs, but these shortcomings can largely be overcome. RNAs with highly structured 3' ends, including snRNAs and histone mRNAs, are naturally resistant to RNase R, but can be efficiently degraded once a poly(A) tail has been added to their ends. In addition, RNase R stalls in the body of many polyadenylated mRNAs, especially at G-rich sequences that have been previously annotated as G-quadruplex (G4) structures. Upon replacing K+ (which stabilizes G4s) with Li+ in the reaction buffer, we find that RNase R is now able to proceed through these sequences and fully degrade the mRNAs in their entirety. In total, our results provide important improvements to the current methods used to isolate circular RNAs as well as a way to reveal RNA structures that may naturally inhibit degradation by cellular exonucleases.

Figure 1.

Hundreds of linear RNAs are resistant to RNase R digestion. (A) HeLa total RNA was incubated in KCl-containing buffer -/+ RNase R followed by preparation of RNA-seq libraries. (B) RNA-seq data generated from control (red) or RNase R treated samples (blue) were used to produce a normalized coverage value over individual nucleotides (Reads per Kilobase per Million [RPKM]). The DIS3, HIPK3, QRICH1, PPP1R8 and ACTR2 loci are shown. Gray arrows below gene models indicate the direction of transcription. Regions that generate an exonic circular RNA (circRNA) or a circular intronic RNA (ciRNA) are denoted in purple and brown, respectively. RNase R stalling in the terminal exons of PPP1R8 and ACTR2 is highlighted in blue. (C) For the top 25% of highly expressed genes in each RNA-seq replicate, the RPKM ratio (RNase R/Control) was calculated. A density plot showing the distribution of RPKM ratios is shown along with a Venn diagram denoting the number of genes with a RPKM ratio ≥1 in each replicate. (D) Comparison of RPKM ratios for the 1,015 genes that had a RPKM ratio ≥1 in both RNA-seq replicates. Genes encoding non-histone mRNAs (blue), snRNAs (red), histone mRNAs (green), and other noncoding RNAs (gray) are denoted. (E) The set of 860 non-histone genes with a RPKM ratio ≥1 in our data was compared to previously published RNA-seq experiments that analyzed RNA expression levels –/+ RNase R treatment. Non-histone mRNA genes with a RPKM ratio (RNase R/Control) ≥1 in each data set are shown.

Figure 2.

Many mRNAs remain resistant to RNase R digestion after A-Tailing or longer incubation times. (A) To determine if addition of single-stranded poly(A) tails enables more efficient digestion by RNase R, purified HeLa total RNA was incubated with E-PAP prior to RNase R digestion (white bars). As controls, RNA was incubated only in the reaction buffers (black bars) or subjected to RNase R digestion alone (gray bars). 300 ng of the remaining RNA was used for reverse transcription followed by qPCR to measure the relative abundances of the indicated transcripts. RNase R sensitive mRNAs (orange), circRNAs (purple), ciRNAs (brown), snRNAs (red), ncRNAs (gray), and RNase R resistant mRNAs (blue) are noted. (B) HeLa total RNA was treated at 37°C with buffer (containing KCl) only or RNase R for the indicated amounts of time. 300 ng of the remaining RNA was then used for reverse transcription followed by qPCR to measure the relative abundances of the indicated transcripts. All data were normalized to the control samples and are shown as mean ± SD, n = 3.

Figure 3.

RNase R stalls within the body of many mRNAs. (Top) UCSC genome browser tracks depicting the last exons of four genes that fail to be completely digested by RNase R. Gray arrows below gene models indicate the direction of transcription. RNA-seq data generated from control (red) or RNase R treated samples (blue) are shown along with RNase R stalling sites predicted by DaPars (orange), G-quadruplexes annotated by Guo and Bartel (purple) or Kwok et al. (brown), and the northern blot probes (green). (Bottom) HeLa total RNA was treated for 15 min at 37°C with either buffer (containing KCl) only or RNase R and then subjected to Northern blotting. Red and blue arrows indicate full length and partially digested transcripts, respectively. Linear and circular CDYL transcripts were used to confirm RNase R activity.

Figure 4.

RNase R often stalls at G-rich regions that form G-quadruplexes. (A) Average nucleotide frequencies in the regions flanking the 337 RNase R stalling sites that were predicted by DaPars. Location of stalling site is denoted by dashed vertical line. (B) Stalling site locations were shuffled 100 times in the last exons of highly expressed (top 25%) genes and the average nucleotide frequencies were calculated as in A. (C) The locations of the 337 stalling sites predicted by DaPars were compared to annotated G-quadruplexes defined by Guo and Bartel (red) or Kwok et al. (blue). The density plot shows the distribution of distances from the predicted stalling sites to the nearest G-quadruplex in each dataset. The Venn diagram denotes the number of stalling sites with an annotated G-quadruplex within ±200 nt. (D) Quadruplex forming G-rich sequences (QGRS) values were calculated for the 400 nt regions flanking the 255 stalling sites with an experimentally annotated G4 nearby (red), the 82 stalling sites without an experimentally annotated G4 nearby (blue), and 337 shuffled stalling sites (gray). Box plots show the 25th–75th percentiles and whiskers represent extreme data points no more than 1.5 times the interquartile range. P values were calculated by Mann–Whitney U-test.

Figure 5.

Replacing K⁺ with Na⁺ or Li⁺ in the reaction buffer enables RNase R to fully digest mRNAs containing G-quadruplexes. (A) HeLa total RNA was treated for 15 min at 37°C with buffer (containing KCl or LiCl) only or RNase R and then subjected to Northern blotting. Red and blue arrows indicate full length and partially digested transcripts, respectively. Linear and circular CDYL transcripts were used to confirm RNase R activity. (B) A 315 nt region containing the PPP1R8 G-quadruplex was inserted into the 3′ UTR downstream of eGFP. Plasmids were transfected into HeLa cells followed by isolation of total RNA, treatment for 15 min at 37°C with RNase R, and analysis by Northern blotting. (C) HeLa total RNA was treated at 37°C with buffer (containing NaCl or LiCl) only or RNase R for the indicated amounts of time. 300 ng of the remaining RNA was then used for reverse transcription followed by qPCR to measure the relative abundances of the indicated transcripts.

Figure 6.

A-Tailing followed by RNase R treatment in LiCl-containing buffer allows more efficient depletion of linear RNAs. (A) Purified HeLa total RNA was incubated with E-PAP prior to RNase R digestion in the presence of LiCl buffer (white bars). As controls, RNA was incubated only in the reaction buffers (black bars) or subjected to RNase R digestion alone (gray bars). 300 ng of the remaining RNA was used for reverse transcription followed by qPCR to measure the relative abundances of the indicated transcripts. RNase R sensitive mRNAs (orange), circRNAs (purple), ciRNAs (brown), snRNAs (red), ncRNAs (gray), and RNase R resistant mRNAs (blue) are noted. (B) For the 1015 genes that failed to be degraded by RNase R in the presence of KCl (RPKM ratio ≥ 1), the respective RPKM ratio (RNase R/Control) was determined after A-Tailing and RNase R treatment in the presence of LiCl (cyan). Data are grouped according to transcript class and then ranked by the RPKM ratio observed in the KCl dataset (pink). Each dot represents a specific gene, with PPP1R8 and ACTR2 marked by arrows. (C) RNA-seq tracks, highlighting the PPP1R8 and ACTR2 loci. Gray arrows below gene models indicate the direction of transcription. RNase R stalling sites predicted by DaPars (orange) and G-quadruplexes annotated by Guo and Bartel (purple) or Kwok et al. (brown) are shown. (D) Box plots depicting RPKM ratios (RNase R/Control) for non-histone mRNA, snRNA, histone mRNA, and other ncRNA gene loci when RNA was treated with RNase R in the presence of KCl (pink) or subjected to A-Tailing followed by RNase R treatment in the presence of LiCl (cyan). Box plots show the 25th–75th percentiles and whiskers represent extreme data points no more than 1.5 times the interquartile range. Mann–Whitney U-test was used to determine the statistical significance. Median fold decreases for each class were calculated and are denoted at the bottom. (E) Cumulative distribution functions of circRNA junction reads ratio (average number of circRNA junction reads predicted by CIRI2/total spliced reads) in RNA-seq datasets generated from the indicated treatment conditions. All genes that produced a circular RNA with at least 2 junction reads in one of the samples are included. Mann–Whitney U-test was used to determine the statistical significance. ^∗∗P < 2.2e–16.

Figure 7.

Coupling A-Tailing and RNase R digestion in LiCl-containing buffer enables more efficient enrichment of circular RNAs. (Left) When total RNA is treated with RNase R in KCl-containing buffer, many transcripts with G-quadruplex structures (denoted in pink) or structured 3′ ends (denoted in green) fail to be fully degraded. (Right) Upon inclusion of an A-Tailing step followed by RNase R digestion in LiCl-containing buffer, both of these classes of linear RNAs are more efficiently degraded. This results in isolation of a more pure population of circular RNAs.