Fig 1.
Seed based off-target effects in pooled siRNA screens.
(A) An on-target siRNA match is generally understood as a perfect match of nucleotides 1–19 of an 21 nucleotide long siRNA guide strand within the coding sequence of an intended transcript [8]. We define an off-target heptamer seed match as a perfect match of nucleotides 2–8 of the guide strand within the 3‘UTR of an unintended transcript. (B) While an on-target siRNA effect is limited to one or few different transcripts, mostly for one gene, a match for a seed can occur in thousands of different transcripts and several times within one 3‘UTR. (C) For pooled screens the elucidation of seed-based off-target effects is much more complex than for single screened siRNAs. The seeds of the three pool siRNAs may match thousands of transcripts and may translate into unintentional transcript silencing. For an on-target pool situation (left) it is always known from which transcript knock-down the phenotype results (yellow flash symbols near the transcript) while for the off-target situation it is unknown from which on- or off-target knock-down of transcripts the phenotype for a pool results (yellow and grey flash symbols near the pool).
Fig 2.
High content screening for targets regulating E-cadherin expression.
(A) ZEB1 is a direct repressor of E-cadherin. Therefore, ZEB1 was chosen as a positive screening control while a non-targeting siRNA (scrambled control) and the transfection reagent were used as negative screening controls. (B) PANC-1 cells imaged after knock-down of ZEB1 (top left) and after application of transfection media (top right). E-cadherin (green staining) expression intensity is significantly increased in the membrane of cells after knock-down of ZEB1 while it is low using non-targeting controls. (Blue color shows Hoechst stained nuclei). Signals were quantified by MetaXpress image analysis (bottom) and normalized for the total number of cells. The vertical dashed line indicates the hit threshold (0.04 normalized expression units). Error bars indicate the standard deviation of mean relative protein expression values. (C) Quantification of CDH1 mRNA expression after knock-down of ZEB1 as fold of control. The dotted vertical line indicates the expression of CDH1 mRNA after treatment of cells with control reagents.
Fig 3.
Primary screening results and expression of screened targets.
(A) Overlaid box and violin plot showing the primary screen phenotype distribution. Colored circles show effects of the ZEB1 positive control pool (red), the CDH1 pool (gold) and the CDK5R1 pool (blue), respectively. The dashed grey line indicates the hit threshold. (B) Histogram of log values of primary screening results combined with the expression status for a subset of 8,977 genes. The red dashed line indicates the hit threshold.
Table 1.
Contingency table.
Fig 4.
SENSORS principle—Prediction of off-targets and off-target effects from large scale RNAi data by two example genes.
(A) Assuming that off-target effects are mediated by heptamer seed matches of siRNAs in 3‘UTRs of unintended transcripts, the first step of our approach was the mapping of all heptamer seed matches of the pooled siRNA library to all 3‘UTRs of the transcriptome. This also includes transcripts for which no intentional pool or siRNA existed in the library. Two transcripts of ZEB1 and CDH1 are shown exemplified. The different height of the sketched seeds (small black lines above the 3‘UTRs; sketch for visualization–not based on measured data) visualizes a different average phenotype of the pools which the seed is part of. (B) In the second step the non-parametric z-score of the Wilcoxon rank sum statistic U was used to score each transcript’s tendency to be the cause for off-target effects or in short–to be an off-target. U is the robust, non-parametric sum of phenotypic ranks of seeds that occur within a 3’UTR. The z-score statistic is thus a metric indicating a U statistic that is unexpectedly deviating from the mean of all rank sums. (C) Combining primary screening results with off-target z-scores can predict additional targets and false positive results.
Fig 5.
Seed enrichment visualizations for 4 high scoring off-targets.
Density curves show the tendency of high scoring positive (red) and negative (blue) off-targets. The x-axes show the rank of the indicated numbers of seeds while the density of the respective ranks is shown on the y-axes. The difference in trends of high and low scoring off-targets is clearly visible by left- and right-skewed densities, respectively. ZEB1 (top left) and CDH1 (bottom right) were the most significant off-targets observed.
Fig 6.
CDK5R1 false positive prediction and validation.
(A) ZEB1 and KRAS were the most significant off-targets in our screen causing an E-cadherin up regulation while CDH1 and MYBL1 are strong negative off-targets causing a loss of E-cadherin expression. The red dashed line is the hit threshold for primary screening data (shown on the y-axis). Pools that fell within the orange zone (i.e. pools showing a primary score above the primary screen threshold but that have no significant off-target z-score) and that have at least one seed matching into the strong positive off-targets are considered likely false positives (red circles). These pools were deconvoluted and validated experimentally. (B) Common seed analysis for the CDK5R1 pool. While no other siRNAs with the seed sequence GTACCTC exhibited a significant phenotypic score, some of the siRNAs with the seed sequence AACAATG (match in ZEB1 3’UTR) showed a similar phenotype to the CDK5R1 pool (red points). One seed sequence is only present in the CDK5R1 pool. (C) Deconvolution of CDK5R1 siRNAs. The siRNA containing the seed AACAATG (si16899) was the only one showing a significant up regulation of E-cadherin expression, while all other siRNAs targeted against CDK5R1 showed no phenotype. (D) C911 control. The C911 control for si16899 kept the phenotype of the unaltered siRNA, indicating that the observed phenotype is due to a seed sequence-mediated off-target effect. The ZEB1 C911 siRNA showed no phenotype indicating that the ZEB1 phenotype is a true positive (on-target) result.
Fig 7.
MYBL1 is a transcriptional activator of E-cadherin.
For five of the seven siRNA pairs used to validate MYBL1 we observed a significant (marked by asterisks, p < 0.05) down-regulation of the CDH1 gene using on-target siRNAs in comparison with the modified C911 siRNAs. MYBL1 si1 caused a very low expression of CDH1 for both the unaltered siRNA and the C911 control siRNA. Thus, we expected the loss of CDH1 after knock-down of MYBL1 with si1 to be strong, but we observed a non-working C911 control for unknown reasons. The difference between unmodified and C911 siRNA specific control siRNA can be regarded as the observed true on-target effect, while the C911 control is a specific control on a single reagent level.