Fig 2: Editing-dependent RNA structural changes near AGO2-miRNA target sites. a Profile of relative 7mer accessibility around editing sites located in 3′UTRs and within 100 nt from predicted miRNA target sites (blue), and around random A positions (red). Y-axis shows log2 fold change in the probability of accessibility of a 7mer (centered at the position indicated on the x-axis) between structures folded using the edited (G) vs. unedited (A) version of the sequences calculated using RNAplfold46 (see Methods). Shaded areas represent confidence intervals. Black bar indicates regions with a significant difference comparing the relative accessibilities around editing sites and random A’s (p value < 0.01, Wilcoxon Rank-Sum test). b Nucleotide frequencies opposite to the editing site in minimum free energy structures predicted with RNAfold68 using the unedited (blue bars) and edited sequences (red bars). ‘N’ is shown for structures where the editing site was unpaired and the opposite nucleotide was ambiguous (e.g., if the editing site was in a hairpin loop or bulge with multiple likely opposite nucleotides). c Correlation between relative 7mer accessibilities (calculated as in a) at predicted miRNA target sites (x-axis) and gene expression differences between edited (editing level ≥ 0.3) and non-edited individuals (y-axis, ratio calculated as edited/non-edited). Each circle represents average values of relative accessibility and expression ratio of a group of miRNA target sites (x-axis) and genes (y-axis) with similar distance (differing by ≤ 10 nt) between editing sites and miRNA target sites. Pearson correlation coefficient and significance of correlation are shown. d Scheme for the proposed structure-mediated regulation of target mRNA abundance depending on A-to-I editing

Fig 3: RNA editing-induced secondary structural changes affect RNA abundance. a Schematic diagram of the minigene system. Target sequences (3′ UTRs of GOLGA3 and GINS1) and their mutant versions were inserted into the firefly luciferase (Fluc) 3′ UTR. Renilla luciferase (Rluc) was used as a reference reporter. Fluc and Rluc were co-transfected into HEK 293 control cells, AGO2 overexpression (OE) and AGO2 knockdown (KD) cells. RT-qPCR was carried out 48 h after transfection (details described in Methods). Green region represents firefly luciferase, red region represents Renilla luciferase and blue represents inserted target sequences or mutant versions. b Predicted RNA secondary structure of GOLGA3 3′ UTR flanking the RNA editing site. The “A” nucleotide at the editing site is highlighted in red. Blue sequences correspond to predicted miRNA target site (of the miRNA seed region). Green dashed boxes illustrate the unedited (A), pre-edited (G) and mutant nucleotides and their counterpart bases in the predicted RNA secondary structure, respectively. c Similar as b, but for the GINS1 gene. d Western blot of AGO2 expression in control cells and cells with AGO2 OE or KD. e, f Relative RNA expression levels of minigenes with unedited, pre-edited or mutant versions of the sequences shown in b and c respectively. Fluc and Rluc vectors were co-transfected into HEK293 control (Ctrl) cells and cells with AGO2-OE (left panel) or AGO2-KD (right panel). Relative Fluc RNA abundance between pre-edited (or mutant) and unedited versions of the minigenes is shown (see x-axis labels). Error bars represent standard deviation based on three experimental replicates. P values were calculated using Wilcoxon Rank-Sum test. N.S. not significant (p ≥ 0.05)

Fig 4: Correlation between AGO2-miRNAs targeting and RNA editing levels. a Correlation of average editing levels (per individual) with AGO2 RPKM levels. Pearson correlation coefficients are shown. b Average editing level in dependence of the distance between an editing site and a predicted miRNA target site. The average was taken over all editing sites with distances ≤ the distance shown on the x-axis. Shaded areas represent standard error of the mean. P value is shown to compare editing level difference at editing sites close (<50 nt, black bar at top) and relatively far (500–1000 nt, black bar at top) from predicted miRNA target sites (Wilcoxon Rank-Sum test). c Editing level differences (AGO2 KD minus control, K562 cells) of two groups of editing sites identified in the Geuvadis RNA-Seq data. Left: editing sites located in 3′ UTRs within 500 nt from predicted miRNA target sites; right: editing sites in 3′ UTRs without predicted miRNA target sites. The number of genes in each group is shown on top. P value was calculated to compare the editing level differences between the two groups of editing sites (Wilcoxon Rank-Sum test). **: p < 1e-3. d Gene expression changes upon ADAR1 KD compared to control (K562 and HepG2 cells, respectively) in two groups of genes (with number of genes indicated on top). Group 1 consists of genes with editing sites (identified in the Geuvadis RNA-Seq data) within 500 nt from predicted miRNA target sites (left box in each panel); group 2 consists of genes without editing sites in 3′ UTRs (right box in each panel). Expression ratio was calculated as RPKM ratio of KD to control. P values were calculated to compare the log2 fold changes in RPKM values of the two groups of genes (Wilcoxon Rank-Sum test). **: p < 1e-7

Fig 5: CSDE1 is a target gene of miR-525-5p. (a) On miRDB, miRTarBase, and TargetScan, 32 overlapping target mRNA are identified, and CSDE1 is chosen as the gene potentially downstream of miR-525-5p. (b) Quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR) analyses validate that CSDE1 is increased and reduced after transfection with LINC01234 overexpression plasmid and small interfering RNA in breast cancer (BC) cells. (c) qRT-PCR analysis of the expression of CSDE1. (d) Immunohistochemical (IHC) analysis of the expression of CSDE1 in BC tissues and normal tissues. (e) Quantification of IHC results. (f) Correlation between CSDE1 expression and LINC01234 expression in BC tissues. (g) The potential binding sequences between miR-525-5p and 3′UTR of CSDE1. (h) miR-525-5p mimics cotransfected with CSDE1-WT notably reduce luciferase activity. (i) RNA pull-down results show that miR-525-5p can band with the 3′UTR of CSDE1. (j) RNA immunoprecipitation results reveal that CSDE1 and miR-525-5p are specifically enriched in the AGO2 antibody immunoprecipitated complexes. ∗∗p < 0.01; ∗∗∗p < 0.001.