Influence of RNA circularity on Target RNA-Directed MicroRNA Degradation

Abstract

A subset of circular RNAs (circRNAs) and linear RNAs have been proposed to 'sponge' or block microRNA activity. Additionally, certain RNAs induce microRNA destruction through the process of Target RNA-Directed MicroRNA Degradation (TDMD), but whether both linear and circular transcripts are equivalent in driving TDMD is unknown. Here, we studied whether circular/linear topology of endogenous and artificial RNA targets affects TDMD. Consistent with previous knowledge that Cdr1as (ciRS-7) circular RNA protects miR-7 from Cyrano-mediated TDMD, we demonstrate that depletion of Cdr1as reduces miR-7 abundance. In contrast, overexpression of an artificial linear version of Cdr1as drives miR-7 degradation. Using plasmids that express a circRNA with minimal co-expressed cognate linear RNA, we show differential effects on TDMD that cannot be attributed to the nucleotide sequence, as the TDMD properties of a sequence often differ when in a circular versus linear form. By analysing RNA sequencing data of a neuron differentiation system, we further detect potential effects of circRNAs on microRNA stability. Our results support the view that RNA circularity influences TDMD, either enhancing or inhibiting it on specific microRNAs.

Graphical Abstract

Figure 1.

The circular topology of Cdr1as determines the outcome of its effect on miR-7′s stability. (A) Cdr1as total output levels (linear plus circular) measured by RT-qPCR upon transduction of either a scrambled shRNA or shCdr1as alone, or shCdr1as rescued with a linear version of Cdr1as (linCdr1as) lacking the shCdr1as site, in cortical primary neurons. n = 6 culture wells (from three independent primary cultures) for control and cirCDR1as KD; n = 4 culture wells (from two independent primary cultures) for linCdr1as rescue. (B) MiR-7 abundance measured by Taqman RT-qPCR in the same samples as in (A). n = 6 culture wells (from three independent primary cultures) for control and cirCDR1as KD; n = 4 culture wells (from two independent primary cultures) for linCdr1as rescue. (C) Cdr1as total output levels measured by RT-qPCR upon over-expression of linear Cdr1as (linCdr1as) in primary hippocampal neurons. Control corresponds to a non-related (GFP-expressing) linear transcript. n = 6 culture wells (from two independent primary cultures) for each condition. (D) MiR-7 abundance measured by Taqman RT-qPCR in the same samples as in (C). n = 6 culture wells (from two independent primary cultures) for each condition. Data are presented as mean ± SEM. Statistical significance was determined by unpaired Student's t tests (ns: P > 0.05, * P ⇐ 0.05, ** P ⇐ 0.01, *** P ⇐ 0.001, **** P ⇐ 0.0001).

Figure 2.

System to artificially express circRNAs reducing their overlapping, cognate linear RNA expression. (A) Top: illustration of the linear RNA expressing construct used as a positive control for TDMD (TDMD inducer). Bottom: illustration of the circRNA-expressing construct; depicted with coloured arrows are the sets of primers used to measure the different transcript variants (circular [1], Total Output-TO [2] and linear [3]). (B) RT-qPCR measuring Total output (TO, primer pair #2), linear (primer pair #3) and circular (divergent primer pair #1) RNA levels upon expression of the circRNA-expressing constructs from the tetracycline-inducible promoter (TREp) bearing perfectly matched or seed-mutant miR-124 sites for selective linear RNA degradation (see Supplementary Figure S2A). Primers are depicted in Figure 2A. n = 5 culture wells (from three independent primary cultures) of cortical neurons for each condition; and n = 7 culture wells (from four independent primary cultures) of hippocampal neurons for each condition. (C) Total reporter output (left) and circRNA levels (right) upon expression of the circRNA-expressing construct or the linear RNA construct (TDMD inducer). The constructs were expressed from the tetracycline-inducible promoter (TREp) and the synapsin (Syn) promoter respectively in order to achieve similar total output levels for both constructs n = 9 culture wells (from three independent primary cultures) of cortical neurons for each condition; n = 9 culture wells (from four independent primary cultures) of hippocampal neurons for each condition. Missing points are failed culture wells/RT-qPCR reactions. In (B, C), data are presented as mean ± SEM. Statistical significance was determined by unpaired Student's t tests (ns: P > 0.05, *P ⇐ 0.05, **P ⇐ 0.01, ***P ⇐ 0.001, ****P ⇐ 0.0001).

Figure 3.

Impact of artificial circRNA constructs on TDMD in primary neurons. (A) RT-qPCR Taqman assay showing miR-132 abundance upon transduction of the linear control (left) or the circRNA-expressing construct (right) carrying bulged (TDMD-compatible) or seed-mutant miR-132 sites. n = 9 culture wells (from three independent primary cultures) of cortical neurons for each condition; n = 9 culture wells (from four independent primary cultures) of hippocampal neurons for each condition. Missing points are failed culture wells/RT-qPCR reactions. (B) AGO2-Flag immunoprecipitation (RIP) followed by RT-qPCR in HEK293T cells. MiR-27a was used to normalize expression. Relative circRNA abundance was measured using circRNA-backspliced-junction specific divergent primers and normalized to miR-27a levels as an unrelated RISC-loaded miRNA not expected to be affected by the circRNA. Levels of non-specific U6 background binding to Ago2 are shown. As an IP quality control, FLAG/HA-AGO2 input levels were shown to be similar across transfected conditions and efficiently pulled-down using anti-FLAG beads (Supplementary Figure S3B). Accordingly, miR-27, but not U6 RNA, was efficiently co-immunoprecipitated, showing an even recovery of Ago-bound RNA across transfected conditions (Supplementary Figure S3C). n = 4 culture wells (from two experiments) of HEK293T cells for each condition. (C) Subcellular fractionation showing total versus cytoplasmic (left) and total versus nuclear (right) fractions, followed by RT-qPCR of the circRNA and linear RNA isoforms, normalized by GAPDH (for cytoplasm) or U6 (for nucleus). Fractionation efficiency was assessed via Western Blot and RT-qPCR (Supplementary Figure S3D, E). n = 4 culture wells (from two experiments) of HEK293T cells for each condition. In (A–C), data are presented as mean ± SEM. Statistical significance was determined by unpaired Student's ttests (ns: P > 0.05, * P ⇐ 0.05, ** P ⇐ 0.01, *** P ⇐ 0.001, **** P ⇐ 0.0001).

Figure 4.

Effect of artificial circRNA constructs on TDMD in HEK293T cells. (A) Illustration of the constructs expressing linear or circular RNA of different sizes from a backbone containing Drosophila laccase2 introns (27). (B) Northern blot analysis of samples from HEK293T cells expressing miR-218 constructs of the indicated length. n = 6 culture wells (from two independent experiments) of HEK293T cells for each condition. NT are non-transfected cells. (C) MiR-218-5p abundance upon transduction of the linear or circRNA-expressing constructs of different lengths carrying bulged (TDMD-compatible) or seed-mutant miR-218-5p sites. n = 6 culture wells (from two independent experiments) of HEK293T cells for each condition. (D) Northern blot analysis of samples from HEK293T cells expressing long miR-7 constructs. n = 3 culture wells (from one experiment) of HEK293T cells for each condition. (E) MiR-7-5p abundance upon transduction of the linear or circRNA-expressing constructs of different lengths carrying bulged (TDMD-compatible) or seed-mutant miR-7-5p sites. n = 6 culture wells (from two independent experiments) of HEK293T cells for each condition. Data are presented as mean ± SEM. Statistical significance was determined by unpaired Student's ttests (ns: P > 0.05, * P ⇐ 0.05, ** P ⇐ 0.01, *** P ⇐ 0.001, **** P ⇐ 0.0001).

Figure 5.

CircRNAs potentially stabilize dozens of microRNAs across neuron-like differentiation. (A) Scatter plot of miRNA expression fold changes (log₂) across differentiation of hESC H9 cells into forebrain (FB) neuron progenitor cells plotted against the number of effective sites, coloured by quartiles of increasing number of ‘effective’ sites within circRNAs. (B) Boxplot showing circRNA expression fold changes (log2) across differentiation separated by quartiles of increasing ‘circRNA-specific miRNA load index’. The analysis includes 236 miRNAs. For panel B, GLM with emmeans statistics are shown between the least paired and the remaining groups (ns: P > 0.05).

Figure 6.

Predicted TDMD-like site architectures within circRNAs are more frequent for the most highly paired miRNAs. (A) Examples of predicted TDMD-like sites within circRNAs using scanMiR. All five miRNAs belong to the ‘+++ paired’ quartile. (B) Pie charts showing the proportion of predicted linear RNA- and circRNA–miRNA interactions involving at least one predicted TDMD-like site against miRNAs within quartiles of increasing miRNA-specific Pairing coefficient. 5′UTRs, CDSs and 3′UTRs within linear RNAs of protein-coding genes were considered for this analysis. Shown are the p-values for the Fisher's Exact test between each quartile and the least paired quartile (‘- paired’).