A bioinformatics method identifies prominent off-targeted transcripts in RNAi screens

Abstract

Because off-target effects hamper interpretation and validation of RNAi screen data, we developed a bioinformatics method, genome-wide enrichment of seed sequence matches (GESS), to identify candidate off-targeted transcripts in primary screening data. GESS analysis revealed a prominent off-targeted transcript in several screens, including MAD2 (MAD2L1) in a screen for genes required for the spindle assembly checkpoint. GESS analysis results can enhance the validation rate in RNAi screens.

Figure 1. Summary of the Genome-wide Enrichment for Seed Sequence matches (GESS) method

The GESS algorithm begins by splitting the set of siRNAs into two sets: those with phenotype and those without phenotype. The user enters criteria for defining a matching transcript, including the siRNA strand(s), seed length (J) and seed match multiplicity (K). GESS calculates the percent of siRNAs in each set that shows seed matching with each sequence in the genome-wide database (SMF^a for active siRNAs and SMFⁱ for inactive siRNAs). Statistical significance of seed match enrichment (SME) among the set of active siRNAs compared to the set of inactive siRNAs is performed (see Methods).

Figure 2. GESS identifies major off-targeted transcripts in RNAi screen datasets

(a) The plot shows GESS analysis of 27,534 human mRNA 3′UTRs on the primary data from a screen that identified siRNAs inducing loss of SAC function. Each point represents one 3’ UTR, and indicates the percentage of active siRNAs containing a seed match to the 3’ UTR (SMF^a; percent of n = 49 total active siRNAs) plotted against the percentage of inactive siRNAs containing a seed match to the 3’ UTR (SMFⁱ; percent of n = 9,856 total inactive siRNAs). (b) The plot shows GESS analysis as above on data published from an siRNA screen for components required for mitotic arrest upon inhibition of the mitotic kinesin Eg5 in HeLa cells. P1c-seeds were used as the source of inactive siRNAs (n = 308, for both active and inactive siRNAs). (c) The plot shows GESS analysis as above, on data published from an siRNA screen for genes involved in the TGFβ pathway. Experimentally identified siRNAs that showed no phenotype (a cutoff of two standard deviations of activity was used to separate active from inactive siRNAs) were used for the set of inactive siRNAs (n = 391 active siRNAs, n = 18,869 inactive siRNAs). Significance threshold was determined independently for each data point, using the Benjamini and Hochberg (Simes’) method to correct the baseline value of α which is 0.05. Statistically significant outliers are depicted in red and their number is reported.

Figure 3. GESS-informed selection of siRNA pools enriches for genes that reproduce the primary phenotype upon targeting with additional siRNAs

The schematic shows that siRNA pools targeting 641 transcripts scored in a primary screen for genes required for homologous recombination. Upon deconvolution, pools targeting 99 genes showed the phenotype on at least two out of four siRNAs. Of these genes, 88 were further evaluated (11 genes were dropped because no additional siRNAs were commercially available, the original pool showed toxicity, or the retested genes were in the lower spectrum of primary screen scores). GESS analysis showed that the RAD51 3′UTR is sensitive to off-targeting. The schematic shows the rate at which the phenotype was reproduced with and without removal of siRNAs that contain a 7mer seed match to RAD51 3′UTR.