Fig 1.
Graphical representation of siRNA molecule.
(A) siRNA duplex with presumed binding sites for RISC proteins, Ago and TRBP [6]. (B) Interaction pattern between siRNA guide strand and on-target transcript and (C) interaction between siRNA guide strand and its seed-dependent off-target transcript. The mechanistic overview shows the guide strand of siRNA (A) as the strand which enters RISC, and acts in downstream silencing events (B and C). Nucleotide position numbers mirror those used in the text and consequent figures.
Fig 2.
Correlation between Tm values of sequence subsections within siRNA duplexes and the corresponding off-target silencing efficiency.
The start of the subsection within a duplex is plotted on the Y axis (‘Start Position’) whereas the end of the subsection is plotted on the X axis (‘End Position’). The position numbering mirrors that used in Fig 1. The analysis was performed separately for each siRNA concentration–(A) 0.05, (B) 0.5, (C) 5 and (D) 50 nM. The siRNA sequences used in the analysis, together with corresponding knockdown percentages, are listed in S1 Table.
Fig 3.
Seed and non-seed region-dependent off-target effect analyses for siVIM-270 and siVIM-805.
(A, B) Expression profiles of off-target genes with 3’UTR sequences that perfectly match the corresponding siRNA seed region (i.e. off-targets) were compared to genes without such sequence. (C, D) The correlation between GC content in all the subsections within mRNA targets of siRNA non-seed region (8–21) and fold change of off-target effect was calculated in a similar manner to that shown in Fig 2. The correlation coefficient was grouped using quantiles as boundary values and target position corresponds to the numbering shown in Fig 1. GC contents in the non-seed regions at positions 8–15 showed the highest correlation with off-target effect (Fig 3C; siVIM-270, r = 0.22, p-value = 1.13E-12, Fig 3D; siVIM-805, r = 0.15, p-value = 1.57E-05). (E, F) Off-target transcripts for siVIM-270 and siVIM-805 were divided into four groups defined by the number of GC nucleotides in their non-seed regions (positions 8–15). Quantiles were used as the boundary values for classification; ‘Low’ (GC content < 3), ‘Medium’ (GC content = 3), ‘High’ (GC content = 4) and ‘Very High’ (GC content ≥ 5)(out of a total of 8). The number of off-target genes was 1065 for siVIM-270 (E) and 823 for siVIM-805 (F).
Fig 4.
Analysis of off-target effects based on sequence similarity between an siRNA non-seed region and its corresponding target sequences.
(A, B) Cumulative distribution of off-target transcripts grouped by their non-seed base-pairing (Fig 4A; siVIM-270 off-target effects, 2AU (70), 1AU (52), None (27), 1GC (133), 2GC (151), Fig 4B; siVIM-805 off-target effects, 2AU (55), 1AU (52), None (20), 1GC (52), 2GC (46)). Off-target transcripts with more than 2AU or 2GC match were omitted due to their low number. (C, D) The average GC contents for non-seed region (positions 8–15) were calculated for each group of off-target transcripts. (E, F) The cumulative distribution of off-target transcripts of siVIM-270 with 1GC match (133 transcripts)(E) and 2GC matches (151 transcripts)(F), were sub-divided based on their GC contents at positions 8–15. ‘Low’ subgroups have GC content lower than the average while ‘High’ subgroups have GC content higher than the average.
Table 1.
Contribution of miRNA non-seed region to gene silencing.
The list of putative targets perfectly matching the miRNA seed region (positions 2–8) was intersected with a list of experimentally validated miRNA targets [23]. Genes present on both lists were placed in the ‘Validated Targets’ group, while genes predicted to interact with miRNA but which have not been experimentally confirmed, were placed in ‘Remaining Genes’ group. The GC content in the positions 8–15 was calculated for both groups and compared. The difference was calculated by subtracting values of the ‘remaining’ group from the values in the ‘validated’ group.
Fig 5.
Summary of ‘highly effective’ and ‘off-target effect-reduced’ design rules for siRNA molecules.
Based on our previous analysis and the results shown in Figs 2–4, we recommend designing molecules with: (1) A or U at position 1, (2) G or C at position 19, (3) low predicted Tm/AU richness at positions 2–7, (4) absence of GC stretches longer than 9 nt, (5) high predicted Tm in siRNA duplex/high GC content in guide strand at positions 8–15, and (6) high average GC content for all off-target sequences corresponding to positions 8–15 in the siRNA non-seed region.