CD95L derived si- and shRNAs and the CD95L mRNA kill cancer cells through an RNAi mechanism by targeting survival genes

>80% of a large number of siRNAs and shRNAs targeting CD95 or CD95 ligand (CD95L) induce a form of cell death that is characterized by the simultaneous activation of multiple death pathways and preferentially affects transformed and cancer stem cells. We now show that these si/shRNAs kill cancer cells through canonical RNAi by targeting the 3’UTR of critical survival genes in a unique form of off-target effect. By testing 4666 shRNAs derived from the CD95 and CD95L mRNAs and an unrelated control gene, Venus, we have located the most toxic sequences in the open reading frame of CD95L. Consistently, CD95L mRNA is highly toxic to cancer cells after complete deletion of CD95. Our data provide the first evidence for mRNAs to affect cell fate through RNAi in mammalian cells. In addition, they suggest that cancer cells can be targeted with specific toxic RNAi active sequences present in the genome.

Introduction 4 toxic si/shRNAs derived from CD95 and CD95L seemed to exclude an OTE and pointed at a survival activity of CD95 and CD95L.
We wondered whether exogenously added recombinant CD95L protein could protect cells from the toxicity of CD95L derived shRNAs. When NB7 cells were incubated with different concentrations of a soluble form of CD95L (S2), toxicity exerted by shL1 was not affected ( Figure S2A). NB7 neuroblastoma cells were chosen for these experiments because they lack expression of caspase-8 (Teitz et al., 2000) and hence are completely resistant to the apoptosis inducing effects of CD95L.
To test whether human CD95L or CD95 proteins could protect human cancer cells from death, we introduced silent mutations into the targeted sites of three very toxic shRNAs, shL1 and shL3 (both targeting CD95L), and shR6 (targeting CD95). We first introduced eight silent mutations into the sites targeted by either shL1 or shL3 ( Figure S2B) and expressed these proteins in NB7 ( Figure S2C). Both mutant constructs were highly resistant to knockdown by their cognate shRNA but still sensitive to the knockdown by the other shRNA ( Figure S2C). Over-expression of these shRNA-resistant versions of the CD95L ORF did not protect the cells from shL1 or shL3, respectively ( Figure S2D). We then mutated the CD95 mRNA in the targeted site of shR6 ( Figure S2E). Expression of neither wt nor mutated CD95 in MCF-7 cells ( Figure S2F) reduced the toxicity when cells were infected with the pLKO-shR6 or another toxic virus, pLKO-shR7 ( Figure S2G). These data suggested that neither exogenously expressed or added CD95L or exogenously expressed CD95 protein can protect cells from toxic shRNA derived from these genes.
To determine whether the targeted sites in CD95 and CD95L were actually required for the shRNAs to kill we used CRISPR/Cas9 gene editing to excise sites targeted by different shRNAs and siRNAs in both alleles of the CD95 and CD95L genes. We first deleted a 41 nt piece of the CD95L gene in 293T cells, that contained the target site for shL3 ( Figure 1A, 1C). While internal primers could not detect CD95L mRNA in three tested clones, primers outside of the deleted area did detect CD95L mRNA ( Figure 1D, and data not shown). Three clones with this shL3 Δ41 deletion were pooled and tested for toxicity by shL3 expressed from a Tet-inducible plasmid (pTIP-shL3). Compared to a pool of control cells transfected only with the Cas9 plasmid, the 293T shL3 Δ41 cells were equally sensitive to the toxic shRNA ( Figure 1G). This was also observed when the clones were tested individually (data not shown).
To exclude the possibility that shL3 was inducing cell death due to an activity unique to shL3 and/or 293T cells, we deleted the same 41 nt in CD95L in the ovarian cancer cell line HeyA8; We also removed 64 nt containing the target site for the siRNA siL3 in the CD95L coding sequence, and a 227 nt region containing the target site for shR6 in CD95 in HeyA8 cells ( Figure 1A, 1B and Figure S3). In all cases homozygous deletions were generated ( Figure 1E). To confirm the deletion of the shR6 target site, we infected HeyA8 cells treated with the Cas9 plasmid only and HeyA8 with the deleted shR6 site with shR6, shR2 (both targeting the CD95 ORF) and shR6' (targeting the CD95 3'UTR). Five days after infection, CD95 mRNA was quantified by real time PCR (Figure 1F). Using a primer located outside the 227bp deletion, the mutated CD95 mRNA was still detectable in the cells. While shR2 and shR6' caused knockdown of CD95 mRNA in both the Cas9 expressing control and the R6 Δ227 k.o. cells, shR6 could only reduce mRNA expression in the parental cells. These data document that HeyA8 CD95 shR6 Δ227 cells no longer harbor the sequence targeted by shR6. Now having HeyA8 cells lacking one of three RNAi targeted sites in either CD95 or CD95L, we could test the role of the CD95 and CD95L genes in protecting HeyA8 cells from the death induced by either shRNAs (shL3 and shR6, two different vectors: pLKO or the Tet inducible pTIP) or the siRNA siL3. In all cases, the shRNA or siRNA that targeted the deleted region was 6 cells to toxic si/shRNAs and of Dicer -/cells to toxic siRNAs is consistent with a model in which toxic shRNAs require RISC to kill cells and Drosha -/and Dicer -/cells which are almost completely devoid of miRNAs  allow much more efficient uptake of the toxic si/shRNAs into the RISC. To test what the contribution of the seed sequence in the siRNAs was to their toxicity we generated a set of chimeric siRNAs in which we replaced nucleotides of the toxic siL3 siRNA with nucleotides of a nontoxic scrambled RNA. We did this starting either from the seed end or from the opposite end ( Figure 2H). HeyA8 cells expressing both the Venus-CD95L sensor (to monitor level of knockdown) and a Nuc-Red plasmid to fluorescently label nuclei (to monitor the effects on cell growth) were transfected with 5 nM of the chimeric siRNAs; total green fluorescence and the number red fluorescent nuclei were quantified over time. The siL3 control transfected cells showed an almost complete suppression of the green fluorescence and high toxicity. In the top panel of Figure 2H the data are summarized in which siL3 nucleotides were stepwise replaced with siScr nucleotides from the seed match end. Both RNAi and toxicity were profoundly reduced when three of the terminal siL3 nucleotides were replaced with the siScr nucleotides in those positions suggesting that the seed region (6mer highlighted in blue) is critical for both activities (top panel in Figure 2H). Consistently, when siL3 nucleotides were replaced with siScr nucleotides from the non-seed end, neither RNAi nor the toxicity was diminished until replacements affected residues in the seed region (bottom panel in Figure 2H). These data suggest that the 6mer seed sequence of siL3 was critical for its CD95L specific RNAi activity and also its toxicity. These data provide strong evidence that the toxicity observed was a sequence specific event, and not the result of general toxicity resulting from a toxic sequence motif in the siRNA, or from poisoning the RISC. Rather, this toxicity followed the rules of canonical RNAi.

Toxic CD95L and CD95 derived sh-and siRNAs cause downregulation of survival genes.
A general OTE by RNAi has been reported Jackson et al., 2006;Lin et al., 2005). However, this has not been found to cause toxicity in most cases and the targeted mRNAs were difficult to predict . The fact that 22 of the tested CD95 and CD95L targeting sh-and si/DsiRNAs were toxic to many cancer cells evoking similar morphological and biological responses ) generated a conundrum: Could an OTE trigger a specific biology? In order to test this we expressed two toxic shRNAs -one targeting CD95L (shL3) and one targeting CD95 (shR6) -in cells lacking their respective target sequences and subjected the RNA isolated from these cells to an RNA-Seq analysis. In order to detect effects that were independent of cell type, delivery of the shRNA, or targeted gene, we expressed shL3 in 293T (ΔshL3) cells using the Tet inducible vector pTIP and shR6 in HeyA8 (ΔshR6) cells using the pLKO vector. In each case, changes in RNA abundance were compared to cells in which we expressed a non-targeting shRNA in matching vectors. Total RNA was harvested in all cases at either the 50 hour time point (before the onset of cell death) or at the 100 hour time point (during cell death) ( Figure 3A). In order to achieve high stringency, the data were then analyzed in two ways: first, using a conventional alignment-based analysis to identify genes for which the mRNA changed more than 1.5 fold (and an adjusted p-value of less than 0.05) and second, by a read-based method, in which we first identified all reads that changed >1.5 fold and then subjected each read to a BLAST search to identify the gene it was derived from. Only RNAs that were detected by both methods were considered ( Table S1). The combination of the analyses resulted in one RNA that was upregulated and 11 mRNAs that were downregulated ( Figure 3B). Using an arrayed qPCR approach, most of these detected mRNA changes were validated for both cell lines ( Figure S4A). Interestingly, for nine of the eleven genes, published data suggested that they are either highly upregulated in cancer and/or critical for the survival of cancer cells as their inhibition or knockdown resulted in either growth reduction or induction of various forms of cell death (see legend of Figure S4 for details). Significantly, six of these eleven downregulated genes were recently identified in two genomewide RNAi independent lethality screens to be critical for the survival of cancer cell lines (Blomen et al., 2015;Wang et al., 2015) (Figure 3B and Figure S4B) ( Table S2). Considering that these two screens only identified 6.6% of human genes to be critical for cell survival, we found a significant enrichment (54.5%, p<0.0001) of these survival genes among the genes downregulated during the cell death induced by either shL3 or shR6. All six survival genes are either highly amplified or mutated in human cancers ( Figure S5A). In addition to these six genes, GNB1 and HIST1H1C were reported to be required fitness genes in recent high-resolution CRISPR-based screen (Hart et al., 2015). A kinetic analysis showed that most of the deregulated mRNAs were downregulated early with a strong effect at around 26 hours, two days before the onset of cell death ( Figure S4C). This suggested that the cells were dying because of the silencing of multiple critical survival genes providing an explanation for why multiple cell death pathways were activated. We are therefore calling this type of cell death DISE (for death induced by survival gene elimination).
To confirm that some of the downregulated genes were survival genes for HeyA8 cells, we transfected the cells with siRNA SmartPools targeting each of the eleven genes. Individual knockdown of seven of the targeted genes resulted in reduced cell growth when compared to cells transfected with a pool of scrambled siRNAs ( Figure 3C). To mimic the effect of the CD95 and CD95L-derived shRNAs, we treated HeyA8 cells with a combination of siRNA pools targeting these seven genes. Remarkably, 1 nM of this siRNA mixture (35.7 pM of each individual siRNA) was sufficient to effectively reduce cell growth of the cells ( Figure S5B) and to cause substantial cell death (Figure S5C), suggesting it is possible to kill cancer cells with very small amounts of siRNAs targeting a network of these survival genes.
To test the generality of this phenomenon we inducibly expressed another CD95L derived shRNA, shL1 in 293T cells using the pTIP vector, transfected HeyA8 cells with 25 nM siL3 and subjected the cells to RNA-Seq analysis 100 hours and 48 hours after addition of Dox or after transfection, respectively. In order to determine whether survival genes were downregulated in all cases of sh/siRNAs induced cell death, we used a list of 1883 survival genes and 423 genes not required for survival (nonsurvival genes) recently identified in a CRIPSR lethality screen (Table S2). We subjected the four ranked RNA Seq data sets to a gene set enrichment analysis using the two gene sets ( Figure 3D). In all cases survival genes were significantly enriched towards the top of the ranked lists. In contrast, nonsurvival genes were not enriched. One interesting feature of DISE that emerged was the substantial loss of histones. Of the 16 genes that were significantly downregulated in cells treated with four sh/siRNAs, 12 were histones ( Figure 3E). While it might be expected that dying cells would downregulate highly expressed genes such as histones, we believe that losing histones is a specific aspect of DISE because a detailed analyses revealed that the downregulated histones were not the most highly expressed genes in cancer cells ( Figure S6). In addition, almost as many genes with similar high expression were found to be upregulated in cells after DISE induction.

Toxic CD95L and CD95 derived sh-and siRNAs target survival genes in their 3'UTR.
To test whether the introduced CD95 and CD95L-derived shRNAs directly targeted genes through canonical RNAi, we subjected the two gene lists obtained from the RNA-Seq analysis (the cell lines treated with either shL3 or shR6 at the 50 hour time point) to a Sylamer analysis (van Dongen et al., 2008) designed to find an enrichment of miRNA/shRNA targeted sites in the 3'UTR of a ranked list of genes ( Figure 4A). This analysis identified a strong enrichment of the cognate seed match for shL3 and shR6 in cells treated with one of these two shRNAs. The analyses with cells treated with shRNAs for 100 hours looked similar but less significant, suggesting early targeting by the shRNAs followed by secondary events (data not shown). Enrichment in 6mers and 8mers were both detected (only 8mers shown) in the 3'UTRs but not the ORF of the ranked genes (data not shown).
Interestingly, the detected seed matches were shifted by one nucleotide from the expected seed match based on the 21mer coded by the lentivirus. RNA-Seq analysis performed for the small RNA fraction confirmed that in all cases (shScr and shL3 in pTIP, and shScr and shR6 in the pLKO vector) the shRNAs in the cells were being cleaved in a way that resulted in the predominant formation of an siRNA that was shifted one nucleotide away from the shRNA loop region (black arrow heads in Figure S7A). This allowed us to design toxic siRNAs based on the sequences of shL3 and shL6. These shRNA-to siRNA converts were toxic to HeyA8 cells ( Figure S7B) confirming that the observed toxicity was not limited to the TRC shRNA platform but based on a sequence specific activity of the sh/siRNAs. This Sylamer results for shL3 and shR6 were confirmed with cells treated with either shL1 or siL3. In both cases when the ranked RNA Seq data were subjected to a Sylamer analysis again the seed matches of the sh/siRNA introduced was significantly enriched in the 3'UTR of downregulated RNAs ( Figure S7C). In none of the Sylamer analyses of the four data sets did we see enrichment of seed matched in the 3'UTR of downregulated RNAs that matched the passenger strand. In all cases the only significantly enriched sequences matched the seed sequences in the guide RNA of the sh-and siRNAs. This suggests that the observed toxicity was not the result of passenger strand loading, one of the mechanisms that has been implicated in OTE of RNAi (Gu et al., 2012).
Our data suggested that DISE inducing sh-and siRNAs caused a early loss of survival genes and at the same time downregulated RNAs through canonical RNAi targeting their 3'UTR. However, it was not clear whether the most highly downregulated survival genes were targeted in their 3'UTR by RNAi active sequences. We had found that as little as 6 nucleotides determined whether an siRNA killed cancer cells (see Figure 2H). We found in 10 of the 11 targeted genes identified in the screen described in Figure 3A multiple 6mer seed matches for either shL3 and/or shR6 ( Figure 4B). It is therefore likely that the two shRNAs, shL3 and shR6, killed cells by targeting a network of genes enriched in critical survival genes through RNAi. The only gene without an shL3 or shR6 seed match was HIST1HC1. Interestingly, only four of the histones downregulated in cells after treatment with any of the four tested sh/siRNAs had a 3'UTR (underlined in Figure 3E) suggesting that most histones were not directly targeted by the sh/siRNAs.
To determine whether survival genes beyond these 11 genes were downregulated due to a direct targeting by the toxic sh/siRNAs we determined for each treatment whether there was a connection between the presence or absence of a predicted seed match of the sh/siRNA introduced and the significance of downregulation (>1.5 fold downregulated, p<0.05) among survival genes using the Fisher's Exact test ( Figure 4C). In almost all cases this analysis revealed that survival genes containing a predicted seed match in their 3'UTR were statistically more likely to be selectively downregulated than survival genes without such a motif. The analysis with shL1 treated cells did not reach statistical significance, likely due to the fact that this shRNA was found to be very toxic and the 100 hour time point may have been too late to observe evidence of significant targeting. This interpretation is supported by the observation that the significance for both shL3 and shR6 to target survival genes was higher at 50 hours when compared to the 100 hour times points ( Figure 4C) and that the Sylamer analysis of the shL1 treated cells was less significant after 100 hours of treatment than any of the other Sylamer analyses ( Figure S7C). Overall the analyses suggested that cells die by DISE due to an early silencing of survival genes through targeting seed matches in their 3'UTR followed by the downregulation of histones.

Identification of toxic shRNAs in the CD95L and CD95 mRNAs.
The majority of commercially available si-Dsi-, and shRNAs derived from either CD95 or CD95L were highly toxic to cancer cells. We therefore asked whether these two genes contained other sequences with a similar activity. In order to test all shRNAs derived from either CD95L or CD95, we synthesized all possible shRNAs, 21 nucleotides long, present in the ORF or the 3'UTR of either CD95L or CD95 starting with the first 21 nucleotides after the start codon, and then shifting the sequence by one nucleotide along the entire ORF and 3'UTR ( Figure 5A). We also included shRNAs from a gene not expressed in mammalian cells and not expected to contain toxic sequences, Venus. All 4666 oligonucleotides (700 Venus, 825 CD95L ORF, 837 CD95L 3'UTR, 987 CD95 ORF, and 1317 CD95 3'UTR shRNAs) were cloned into the Tet inducible pTIP vector ( Figure 5B) as five individual pools. We first tested the activity of each pool to be toxic and to target the Venus sensor protein (fused to either the ORF of CD95 or CD95L). NB7 neuroblastoma cells were again used because of their resistance to CD95L induced apoptosis. NB7-Venus-CD95L cells infected with the Venus-targeting shRNA pool showed some reduction in fluorescence when Dox was added, however, the shRNA pool derived from the CD95L ORF was much more active in knocking down Venus ( Figure S8A). No significant green fluorescence reduction was detected in cells after infection with the shRNA pool derived from the CD95L 3'UTR since the targeted sequences were not part of the sensor. Similar results were obtained when NB7-Venus-CD95 cells were infected with the Venus, CD95 ORF, and CD95 3'UTR targeting shRNA pools. To determine the ability to reduce cell growth (as a surrogate marker for toxicity), we infected NB7 parental cells with each of the five pools (parental cells were used to avoid a possible sponge effect by expressing either CD95L or CD95 sequences that were part of the Venus sensors). Interestingly, the pool of 700 shRNAs derived from Venus did not cause any toxicity ( Figure S8B). In contrast, the pool of the shRNAs derived from CD95L significantly slowed down growth, while no toxicity was observed when cells were infected with the pool of shRNAs derived from the CD95L 3'UTR. In the case of CD95, both the shRNAs derived from the ORF and the 3'UTR showed some toxicity. However, the shRNAs derived from the 3'UTR caused greater toxicity compared to those derived from the ORF. The data suggests that overall the shRNAs derived from the CD95L ORF and the CD95 3'UTR contain the most toxic shRNAs.
To determine the toxicity of each of the shRNAs in the pools, NB7 cells were infected with the libraries of shRNA viruses (MOI<1), and after puromycin selection cells were pooled 1:1:1 (Venus ORF/CD95L ORF/CD95L 3'UTR pools or Venus ORF/CD95 ORF/CD95 3'UTR pools) to allow for competition between shRNAs when Dox was added ( Figure 5B). Cells were cultured for 9 days with and without Dox to allow for cell death to occur. Next generation sequencing was performed on a PCR fragment containing the shRNA barcode (isolated from genomic DNA) to determine the relative abundance of each shRNA in three pools: 1) the cloned plasmid libraries, 2) cells after infection and culture for 9 days without Dox, and 3) cells infected and cultured with Dox for 9 days. A total of 71,168,032 reads were detected that contained a complete sequence of one of the cloned shRNAs. Almost all shRNAs were represented in the cloned plasmids (Table S3). The shRNAs in the CD95L pool (comprised of the Venus, CD95L ORF, and CD95L 3'UTR subpools) and the CD95 pool (comprised of the Venus, CD95 ORF, and CD95 3'UTR subpools) were ranked from highest (most toxic) to lowest underrepresentation. During this and subsequent analyses we noticed that in many cases Dox addition did cause a reduction of shRNAs, indicating an increase in toxicity; However, in other cases infection alone without the addition of Dox was toxic. This effect was likely due to the well-described leakiness of the Tet-on system (Pham et al., 2008) which we confirmed for the shR6 shRNA ( Figure S9A). To capture all toxic shRNAs, we therefore decided to split the analyses into two halves: 1) the changes in abundance after infection (infection -Dox) and 2) the changes in abundance after Dox addition (infection +Dox). The results for all shRNAs are shown in Figure S9B. Grey dots represent all shRNAs and red dots represent only the ones that were significantly underrepresented at least 5 fold. Interestingly, the highest abundance of downregulated shRNAs was found in the CD95L ORF and the CD95 3'UTR pools of shRNAs, which is consistent with the increased toxicity that was observed when NB7 cells were infected with either of these two pools (see Figure S8B). The shRNAs of these two toxic pools were highly enriched in the underrepresented shRNAs in the two pooled experiments (CD95L and CD95). Their toxicity was also evident when all shRNAs in each pool (2362 shRNAs in the CD95L and 3004 shRNAs in the CD95 pool) were ranked according to the highest fold downregulation ( Figure 5C). The three subpools in each experiment are shown separately; Again this analysis identified the subpools with the highest enrichment of underrepresented shRNAs as the ORF of CD95L and the 3'UTR of CD95 ( Figure 5C). In both analyses, the one with shRNAs underrepresented after infection is boxed in blue and the ones after Dox addition is boxed in orange.
This analysis allowed us to describe the toxicity landscape of CD95L and CD95 ORFs and their 3'UTRs ( Figure 5D). All shRNAs that were significantly underrepresented at least five fold (red dots in Figure S9B) are shown along the CD95L pool ( Figure 5D, left) and the CD95 pool ( Figure 5D, right) sequences. Blue marks show underrepresented shRNAs after infection, and orange marks show underrepresented shRNAs after culture with Dox. For both CD95L and CD95, toxic shRNAs were located in distinct clusters. The highest density of toxic sequences was found in the stretch of RNA that codes for the intracellular domain of CD95L (underlined in green in Figure 5D).

Predicting shRNA toxicity -The toxicity index (TI).
Our data suggest that toxic shRNAs derived from either CD95L or CD95 kill cancer cells by targeting a network of genes critical for survival through canonical RNAi. Therefore, we wondered how many 8mer seed sequences derived from these toxic shRNAs would have corresponding seed matches in the 3'UTR of critical survival genes in the human genome. shRNAs containing these sequences could potentially be toxic to cells. To calculate such a hypothetical toxicity index, we used the ranked CRISPR data set (Wang et al., 2015) with 1883 SGs and 423 nonSGs. Based on our RNA Seq analyses, we hypothesized that the survival genes contained more putative seed matches of the shRNA in their 3'UTRs than the nonsurvival genes. We found 3'UTR information of 1846 of the survival genes and 416 of the nonsurvival genes.
In order to establish a Toxicity Index (TI) for each shRNA, we first generated a list containing a normalized ratio of occurrences of any possible 8mer seed match in the 3'UTRs of the survival and non-survival gene groups. This resulted in a ratio for each of the 65,536 possible 8mer combinations (Table S4), the TI. We then assigned to each of the 4666 shRNAs in our screen its TI and ranked each pool within the two experiments of our screen according to the highest TI and again separated the two groups of shRNAs, that were toxic just after infection and after addition of Dox ( Figure 5E, Table S5). In each ranked list, we could now assess whether the experimentally determined toxicity of shRNAs correlated with the in silico predicted TI. Remarkably, the highest enrichment of toxic shRNAs was found amongst those with higher TI for the subpool of shRNAs targeting the CD95L ORF followed by shRNAs in the subpool targeting the CD95 3'UTR. To confirm the significance of this finding, we repeated the analysis 10,000 times by randomly assigning 8mers and their associated TIs to the two shRNA pools and again sorted the data from highest to lowest TI. The reported p-values were calculated based on these permutated datasets using Mann-Whitney U tests.
We noticed that survival genes tended to be more highly expressed than nonsurvival genes (data not shown). To address the question whether toxic si/shRNAs only targeted survival genes or all genes that are highly expressed, we recalculated the TI based on a set of 850 highly expressed and expression matched survival and nonsurvival genes ( Figure S10A). This alternative TI tracked slightly less well with the toxic shRNAs we identified but the enrichment of toxic shRNAs towards the top of the list ranked according to the new TI was still statistically significant ( Figure S10B). This analysis demonstrates that survival genes contain more seed matches for toxic shRNAs in their 3'UTR than nonsurvival genes regardless of the expression level. Our data suggests that to a certain extent it is possible to predict the experimental toxicity of shRNAs based on the in silico calculated TI.

CD95L mRNA is toxic to cancer cells.
Expression of CD95L in most cancer cells is known to kill them through induction of apoptosis. Consequently, expressing CD95L in HeyA8 cells, which are highly sensitive to CD95 mediated apoptosis, resulted in induction of apoptosis within a few hours of infecting the cells with lentiviral CD95L (Figure 6A, left panel). Severe growth reduction was seen without any signs of apoptosis (not shown) when a CD95L mutant was expressed carrying a Y218R point mutation (expressed at similar levels as wt CD95L, Figure 6B) which prevents the CD95L protein to bind to CD95 (CD95L MUT in Figure 6A) (Schneider et al., 1997). To prevent full length CD95L protein to be produced by the CD95L mRNA we also introduced a premature stop codon right after the start codon in the CD95L MUT vector (CD95L MUT NP). This construct (confirmed to produce mRNA with no detectable full length CD95L protein, Figure 6B) was equally active in reducing the growth of HeyA8 cells when compared to the CD95L MUT vector ( Figure 6A, left panel). This result suggested that the CD95L mRNA could be toxic to HeyA8 cells without the CD95L protein inducing apoptosis. This was confirmed by expressing the three CD95L constructs in the presence of the caspase inhibitor zVAD-fmk ( Figure 6A, center panel). Even with suppressed apoptosis all three constructs were equally toxic to HeyA8 cells. Finally, we tested three HeyA8 CD95 protein k.o. clones ( Figure S3). In these cells without the addition of zVAD-fmk, wt CD95L and CD95L MUT NP were equally active in severely reducing the growth of the cells (Figure 6A, right panel and data not shown). These data suggested that it is the CD95L mRNA that killed the cells. Massive cell death could be seen in the CD95 -/cells treated with CD95L mRNA. This cell death resembled DISE induced by shRNAs  and cells showed the same inability to pass through mitosis without dying (Movies S1-S4). Cell death was confirmed by quantifying nuclear fragmentation ( Figure 6C) and we also detected a significant increase in ROS in CD95L expressing cells (Figure 6D), one of the main features of ongoing DISE . To exclude the possibility that truncated CD95 protein or any part of the CD95 mRNA would play a role in this toxicity, we deleted the entire CD95 gene in MCF-7 cells ( Figure S11). Wt CD95L killed these complete CD95 knock-out cells as well as CD95 shR6 deletion cells ( Figure S11F). To determine the cause of cell death induced by CD95L mRNA in HeyA8 CD95 -/cells, we performed an RNA Seq analysis. Similar to cells treated with CD95L derived si/shRNAs, we observed that downregulated genes were enriched in survival genes versus the nonsurvival control genes ( Figure 6E). In addition, cell death induced by CD95L mRNA resulted in a substantial loss of 11 of the 12 histones already detected to be downregulated in cells treated with sh/siRNAs ( Figure 6F). A Metascape analysis confirmed that nucleosome assembly, regulation of mitosis and genes that are consistent with the involvement of histones were among the most significantly downregulated RNAs across all cells in which DISE was induced by any of the four sh/siRNAs or by CD95L ( Figure 6G). These GO clusters are consistent with DISE being a form a mitotic catastrophe with cells unable to survive cell division .

CD95L mRNA kills cancer cells through RNAi
Very similar to the death induced by CD95L derived shRNAs, Drosha -/cells were hypersensitive to the expression of CD95L MUT NP ( Figure 7A) and now virtually all cells died ( Figure 7B). To test the hypothesis that Drosha -/cells were more sensitive to DISE induction because their RISC was not occupied by large amounts of miRNAs and to determine whether CD95L derived toxic siRNAs were incorporated into the RISC, we pulled down AGO1-4associated RNAs and analyzed their composition in wt and Drosha -/cells after expressing the DISE inducing mutant CD95L mRNA. For the pull down we used a peptide recently described to bind to all four AGO proteins (Hauptmann et al., 2015). As expected in wt HCT116 cells large amounts of small RNAs (19-23nt in length) were detected bound to the AGO proteins ( Figure  7C). Both AGO1 and AGO2 were efficiently pulled down. In contrast, in the Drosha -/cells which cannot process canonical miRNAs, only very small amounts of small RNAs were detected confirming the absence of miRNAs in the RISC. Surprisingly, the amount of pulled down AGO proteins was severely reduced despite that fact that the Drosha -/cells express normal levels of AGO2 (see Figure 2E). This suggests that the peptide did not have access to the AGO proteins in Drosha -/cells, either because it only binds to AGO proteins that are complexed with RNA or alternatively because AGO proteins without bound RNAs are no longer in the detergent extract used for peptide pull down.
The analysis of all AGO-bound RNAs showed that in the wt cells >98.4% of bound RNAs were miRNAs. In contrast in Drosha -/cells only 34% of bound RNAs were miRNA derived ( Figure 7D, data not shown). These include miRNAs that are processed independently of Drosha such as miR-320a  which in the Drosha -/cells was found to be a major RNA species bound to AGO proteins ( Figure 7D). In both wt and Drosha -/cells a significant increase in CD95L derived short RNAs was detected that were bound to the AGO proteins. In wt cells they corresponded to 0.0006% of the bound RNAs and in the Drosha -/cells to 0.043% of all AGO bound RNAs, a relative increase of >70 fold. Toxicity of CD95L was therefore not due to overloading the RISC. In the absence of most miRNAs the total amount of RNAs bound to AGO proteins in the Drosha -/cells was roughly 10% of the amount bound to AGOs in wt cells ( Figure  7D). The reduction of AGO-bound miRNAs in Drosha -/cells was paralleled by a substantial increase in binding of other small RNAs to the AGO proteins ( Figure 7E). Interestingly, the amount of AGO-bound CD95L derived small RNAs was >100 times higher in the Drosha -/cells when compared to the wt cells (red columns in Figure 7E).
These data support our hypothesis that Drosha -/cells are more sensitive to DISE due to their ability to take up more toxic small RNAs into the RISC in the absence of most miRNAs. Interestingly, not only did AGO proteins in Drosha -/cells bound more CD95L derived small RNAs but the length of the most abundant bound RNA species increased from 20 to 23 nt in the Drosha -/cells ( Figure 7F). To determine the sites within the CD95L mRNA that gave rise to small AGO bound RNAs we blasted all small AGO bound RNAs detected in all conditions against the CD95L ORF. This allowed us to align all CD95L derived reads to the CD95L sequence ( Figure S12). Many regions in CD95L mRNA gave rise to small RNAs. As expected many of the same small RNAs detected in wt cells were also detected in Drosha -/but at much higher counts (green peaks in Figure 7G, top panel). By comparing the AGO bound CD95L derived RNAs that had the same start site as the shRNAs in our toxicity screen we noticed that a large number of the AGO bound RNAs were derived from regions in the CD95L ORF that gave rise to toxic shRNAs (red peaks in Figure 7G, center panel). We also found a number of reads that were located at positions in CD95L that did not contain toxic shRNAs (green peaks in Figure 7G, center panel). This could be due to the fact that the toxicity screen was performed in NB7 cell whereas the AGO pull down experiments was performed in HCT116 cells. Our data suggest that the CD95L mRNA when overexpressed is toxic to cancer cells due to formation of small RNAs that are bound to AGO proteins presumably incorporated into the RISC and killing cells through RNAi. This process is independent of Drosha. To determine whether Dicer is required for either processing of CD95L or loading the small RNAs into the RISC we expressed the mutant CD95L mRNA in Dicer -/-HCT116 clones. Dicer -/cells were susceptible to CD95L mRNA induced DISE suggesting that the processing of CD95L into small RNAs is independent of both Drosha and Dicer. In summary, our data suggest that CD95 and CD95L derived siand/or shRNAs and the entire CD95L ORF are toxic to cancer cells by targeting critical survival genes through an RNAi mechanism that involves AGO proteins but is independent of both Drosha and Dicer. Furthermore, the data suggest that a high miRNA content by "filling up" the RISC might render cells less sensitive to DISE.

Discussion
Most of the current use of RNAi is aimed at achieving a highly specific silencing with little OTE. In fact, OTE represents one of the largest impediments to the use of RNAi in phenotypic screening applications. We now demonstrate that DISE is a unique form of OTE that results in the simultaneous activation of multiple cell death pathways in cancer cells. The discovery that DISE involves loss of multiple survival genes now provides an explanation for the unique properties we described for this form of cell death, especially the observation that cancer cells have a hard time developing resistance to this cell death mechanism Murmann et al.).
There are a number of rules that have been elucidated for designing sh/siRNAs (Bramsen et al., 2009) to avoid undesired effects such as OTE (Petri and Meister, 2013), general toxicity due to the presence of toxic sequence motifs Petri and Meister, 2013), poisoning/saturating of the RISC (Grimm et al., 2006), or evocation of an IFN response (Marques and Williams, 2005). The following arguments and evidence support our conclusion that DISE is a manifestation of a novel, functionally important, conserved mechanism of genome regulation, and not the result of one of the above mentioned effects: 1) The sheer number of toxic shRNAs embedded in CD95L or CD95. A number of genome-wide shRNA and siRNA lethality screens have revealed that 2-5% of shRNAs targeting human genes are toxic to cells. We recently reported the identification of 651 genes out of about 18,000 targeted genes that are critical for the survival of eight different cancer cell lines . Hence, targeting 97% of the genes using the shRNAs of the TRC library was not toxic. Many of the genes targeted by these shRNAs were actually established survival genes (as discussed in ). That means that the number of shRNAs that are toxic due to a possible OTE or general toxicity would be expected to be very small. In contrast, we found that >80% of the shRNAs and siRNAs that were designed to target either CD95 or CD95L exhibited toxicity in multiple cell lines. 2) High concentrations of siRNAs can saturate the RISC, preventing access of crucial endogenous miRNAs (Khan et al., 2009). We have demonstrated that in general 5 nM of CD95L derived siRNAs are sufficient to kill cancer cells, and we have seen very efficient cell death with as little as 1 nM of siRNA (see Figure 1I). It is therefore unlikely that we are poisoning the RISC. It has been reported that in siRNA overexpression experiments, changes in mRNA expression can be caused by blocked access of endogenous miRNAs to the RISC, such as the highly expressed miRNA family, let-7 (Khan et al., 2009). However, we can exclude such an effect in our analysis, as there was no significant enrichment (or depletion) of the let-7 seed match motif (or that of any other miRNA) in our analyses (black lines in Figure  4A). 3) No IFN response was observed. We have performed multiple RNA-Seq and gene array analyses of cells in which DISE was induced by multiple sh-and siRNAs targeting CD95 or CD95L. In none of these analyses did we detect an increase in any of the known IFN response genes (Schoggins et al., 2011) (data not shown). In addition, we demonstrated that the latest generation of Dicer optimized 27mer DsiRNAs that do not elicit an IFN response (Kim et al., 2005) and the shRNAs expressed from within the cells shown to have low IFN triggering activity (Robbins et al., 2006) have the same toxic activities as the standard 21mer siRNAs (see Figure S1B). 4) Mutation of just one position destroys activity. A major argument against DISE toxicity being caused by overloading the RISC, an IFN response or the presence of known toxic sequences, lies in the analysis of the chimeras we generated between siL3 and a non-toxic scrambled oligonucleotide (see Figure 2H). This analysis demonstrated that the seed match positions of siL3 are critical for its toxicity. In fact, just replacing one nucleotide in a critical position in the center of the seed match almost completely abolished toxicity of the siRNA. Part of our data was generated using a widely used first generation stem loop shRNA platform, the TRC library. The TRC shRNAs have recently been found to be prone to cause OTE. Gu et al. showed that the loop design of this system results in imprecise Dicer cleavage and, consequently, the production of different mature small-RNA species that increase passenger loading, one major source of OTE (Gu et al., 2012). More recently it was reported that most guide RNAs derived from the TRC hairpin were shifted by 4 nt 3' of the expected 5' start site (Watanabe et al., 2016). While we did see a shift in processing of these stem loop shRNAs we did not see such a high level of imprecision of cleavage of our toxic shRNAs. In fact, 99.4% of the shR6 guide RNA started at the same nucleotide position ( Figure S7A). We did observe that the majority of processing of our pTIP and pLKO based shRNAs was shifted by one nucleotide (Figure S7A). This shift was consistent with the defined seed matches that were found by the Sylamer analyses. In general, one major seed match was detected with one other minor species (this was less obvious only for shL1, Figure S7C). Furthermore, all four Sylamer analyses only detected enrichments in the 3'UTR of downregulated mRNAs that were consistent with the guide strand targeting the mRNA and not the passenger strand. In all cases including in cells transfected with the siRNA siL3, the mainly enriched sequence motifs were either 7, or 8mers present in the 3'UTR of the targeted mRNA.
Interestingly however, while the majority of tested siRNAs, DsiRNAs and shRNAs were toxic to many cancer cells a more recent RNAi screen did not report shRNAs against CD95 or CD95L as being toxic (Morgens et al., 2016). The authors of this study used a second generation shRNA platform based on a miR-30 backbone. Indeed we recently generated miR-30 based shRNA versions of some of our most toxic shRNAs (shL1, shL3, shL4, shR5, shR5, and shR7) and found none of them to be toxic to HeyA8 cells (data not shown). However, they also had very low activity to silence the Venus based CD95L and CD95 sensors (data not shown) suggesting that the amount of shRNAs produced by cells may not be sufficient to induce DISE.
Our data on DISE are consistent with a number of properties of RNAi OTE previously reported. Similar to DISE, OTE mediated silencing requires a 6/7nt seed sequence of complementarity Jackson et al., 2006;Lin et al., 2005) and it targets mRNAs in the 3'UTR . Our data on shRNAs, siRNAs, and DsiRNAs suggest that DISE is not limited to one platform and requires sequence specific targeting. This conclusion is also consistent with a previous report that suggested that sequence-dependent offtarget transcript regulation is independent of the delivery method (Jackson et al., 2006). The authors found the same enrichment of 6mers and 7mers in 3'UTRs of targeted mRNAs for siRNAs and shRNAs (Jackson et al., 2006).
The most compelling argument against an effect limited to si/shRNAs, however, is the discovery that the mRNA of CD95L, which we showed to be enriched in RNAi active toxic sequences, itself is toxic to cancer cells. Overexpression of CD95L ORF in most cells will induce massive apoptosis and when it was discovered previously that CD95L delivered using viruses very efficiently killed multiple cancer cells that were completely resistant to CD95 mediated apoptosis, this was interpreted as a form of CD95L induced apoptosis that occurred from within the cells (ElOjeimy et al., 2006;Hyer et al., 2000;Sudarshan et al., 2005;Sun et al., 2012). We now conclude that the mRNA of CD95L kills cells in the same way as CD95Lderived si/shRNAs based on the following observations: 1) The morphology of the cells undergoing DISE induced by shRNAs and by overexpressing CD95L is virtually identical (see supplementary movies). 2) Both treatments result in generation of ROS. 3) The kinetics of these two forms of cell death are similar (days rather than hours). 4) We detect large numbers of cells that die just after the first attempt to divide (data not shown). 5) Drosha -/cells are hypersensitive to si-and or shRNAs and CD95L RNA mediated toxicity. 6) Both forms of cell death result in a preferential loss of critical survival genes followed by loss of histones. 7) We have identified multiple CD95L derived small RNAs (mostly 19-23 nt) that are bound to AGO proteins in cells undergoing DISE. Many of them colocalize with CD95L sequences we have identified as toxic to cancer cells. Most importantly, we found a substantially increased loading of such small CD95L derived RNAs into the RISC in the absence of Drosha consistent with the increase susceptibility of these cells to DISE.
We previously reported that Dicer Exo5-/-HCT116 cells (with deleted Exon 5) were at least as sensitive to induction of DISE (by either shL3 or shR6) than wt cells suggesting that Dicer deficient cells could be killed by DISE . It is has been reported that these Dicer deficient cells are hypomorphs (Ting et al., 2008) and indeed we had detected a small residual Dicer expression by Western blotting . We have now revisited this issue with HCT116 cells rendered completely deficient for Dicer using CRISPR/Cas9 gene editing . The fact that these Dicer -/cells were now completely resistant to the toxic effects of shL3 or shR6 demonstrates complete absence of Dicer protein. The comparison of the results with the two Dicer deficient cells now allows us to conclude that very small amounts of Dicer which resulted in a severe reduction of cellular miRNA levels  are sufficient to allow DISE to occur. Similar to the Drosha k.o. cells in the absence of mature miRNAs which seem to inhibit DISE, Dicer -/cells are hypersensitive to DISE induced by si/shRNAs.
Examining the actual length and location of all CD95L-derived sequences (see Figure S12) it appears that there are multiple defined areas that can give rise to small RNAs that are bound to AGO proteins. At present the mechanism of this specific degradation and/or RISC uptake is unknown. We can exclude the involvement of the components of the canonical miRNA biogenesis pathway as both Drosha and Dicer k.o. cells are fully sensitive to the toxic effects of CD95L mRNA.
We are proposing an entirely new concept of killing cancer cells that is based on the toxicity of CD95 and CD95L derived small RNAs. Naturally, there are many open questions such as: 1) Is DISE part of an anti cancer mechanism? We are proposing that DISE kills cancer cells in a way they usually cannot escape from. We have not found a way to block cancer cells from dying by DISE. We provide strong evidence to suggest that this is due to the simultaneous targeting of multiple survival genes that result in the activation of multiple cell death pathways. It will be difficult to prove that the cells are dying due to the preferential targeting of survival genes. It may never be possible to express multiple survival genes at the same time at physiological levels to render cancer cells resistant to the action of countless small RNAs. This prediction alone makes DISE a promising new strategy to kill cancer cells.
2) Does CD95L induce DISE in vivo? We and others have noticed upregulation of CD95L in multiple stress related conditions such as treatment with chemotherapy ( (Friesen et al., 1999) and data not shown). While the amount of CD95L RNA and the level of upregulation alone may not be enough to induce DISE, it could be the combination of multiple RNAs that are generated that kills cells by DISE. We view CD95L as just one of many RNAs that have this activity.
3) Are there other genes in the human genome with activities such as CD95L? We recently identified other genes in the genome that contain DISE inducing shRNAs (Patel and Peter). It is therefore possible that when cells are subjected to genotoxic or oncogenic stress that they generate numerous small RNAs that can be taken up by the RISC and in combination execute DISE.
At a minimum our data provide the first evidence of an overexpressed cDNA to be toxic to all tested cancer cells through an RNAi dependent mechanism. It was first shown in plants that overexpressed transgenes can be converted into RNAi active short RNA sequences (Hamilton and Baulcombe, 1999). Our data on the effects of overexpressed CD95L RNA maybe the first example of transgene determining cell fate through the RNAi mechanism in mammalian cells.
We interpret the hypersensitivity of both Drosha -/and Dicer -/cells to DISE in the following way: Most of the small RNAs in the cells that are loaded into the RISC are miRNAs. Using Ago pull-down experiments we determined 98.4% of AGO associated RNAs in HCT116 cells to be miRNAs (99.3% in HeyA8 cells, data not shown). It was recently reported that Drosha -/cells showed a reduction of miRNA content from roughly 70-80% to 5-6% and Dicer -/cells had a reduction down to 14-21% . Since neither Drosha -/nor Dicer -/cells express reduced AGO2 protein levels (see Figure 2E), it is reasonable to assume that their RISC can take up many more of the toxic DISE inducing RNAs than the RISC in wt cells explaining the super toxicity of both DISE inducing si/shRNAs and CD95L mRNAs in these cells.
We previously showed that expression of either shL3 and shR6 induced DISE in immortalized normal ovarian fibroblasts much more efficiently than in matching nonimmortalized cells , suggesting that this form of cell death preferentially affects transformed cells. Our data now provide an interesting model to explain the higher sensitivity of cancer cells to DISE when compared to normal cells. It is well documented that cancer cells in general have global downregulation of miRNAs when compared to normal tissues (Lu et al., 2005). This might free up the RISC for DISE inducing RNAs and would imply that miRNAs may protect normal cells from DISE.
Overall our data allow us to predict that any small RNA with DISE inducing RNAi activity that does not require Dicer processing can kill cancer cells regardless of its Dicer or Drosha status. In fact, in an accompanying manuscript we demonstrate that DISE can be triggered in vivo to treat ovarian cancer in xenografted mice by delivering CD95L-derived siRNAs using nanoparticles (Murmann et al.). No toxicity was observed in the treated mice. These data suggest that it might be possible to develop a novel form of cancer therapy based on the DISE OTE mechanism.
Cell lines. The ovarian cancer cell line HeyA8, the neuroblastoma cell line NB7, and the breast cancer cell line MCF-7 were grown in RPMI 1640 medium (Cellgro #10-040-CM), 10% heatinactivated FBS (Sigma-Aldrich), 1% L-glutamine (Mediatech Inc), and 1% penicillin/streptomycin (Mediatech Inc). The human embryonic kidney cell line 293T was cultured in DMEM (Cellgro #10-013-CM), 10% heat-inactivated FBS, 1% L-Glutamine, and 1% penicillin/streptomycin. HCT116 parental, Drosha -/-, and Dicer -/cells were generated by Narry Kim . HCT116 parental (cat#HC19023), a Drosha -/clone (clone #40, cat#HC19020) and two Dicer -/clones (clone #43, cat#HC19023 and clone #45, cat#HC19024) were purchased from Korean Collection for Type Cultures (KCTC). All HCT116 cells were cultured in McCoy's medium (ATCC, cat#30-2007), 10% heat-inactivated FBS, 1% L-Glutamine, and 1% penicillin/ streptomycin. NB7 cells overexpressing wild type CD95L and mutant CD95L cDNAs were generated by seeding cells at 50,000 to 100,000 cells per well in a 6-well plate and infecting cells with lentivirus generated in 293T cells from the pLenti-CD95L, pLenti-CD95L1MUT, pLenti-CD95L3MUT, pLenti-CD95L MUT , and pLenti-CD95L MUT NP vectors in the presence of 8 μg/ml polybrene. Selection was done with 3 μg/ml puromycin 48 hours after infection. HCT116 and HCT116 Drosha -/cells overexpressing CD95L MUT NP were generated by seeding cells at 100,000 cells per well in a 24-well plate and infecting cells with lentivirus generated in 293T cells from the pLenti-CD95L MUT NP vector in the presence of 8 μg/ml polybrene. Selection was done with 3 μg/ml puromycin 48 hours after infection. Both HeyA8 cells and HCT116 cells overexpressing CD95L were plated in 96-well plate for incuCyte analysis one day after selection in the presence of puromycin. MCF-7 cells overexpressing CD95 cDNAs were generated by seeding cells at 50,000 per well in a 6-well plate followed by infection with retrovirus generated in phoenix ampho 293 cells from the pLNCX2-CD95 or pLNCX2-CD95R6MUT vectors in the presence of 8 μg/ml polybrene. Selection was done with 200 μg/ml G418 48 hours after infection for 2 weeks. The HeyA8 cells used in Figure 2D carried a lentiviral Venus-siL3 sensor vector (Murmann et al.) and were infected with NucLight Red lentivirus (Essen Bioscience #4476) with 8 μg/ml polybrene and selected with 3 μg/ml puromycin and sorted for high Venus expression 48 hours later. HeyA8 ΔshR6 clone #2 sensor cells used in Figure 2A to C were infected with lentiviruses generated from the Venus-CD95L (full length) sensor vector to over-express the Venus-CD95L chimeric transcript. Cells were sorted for Venus expression 48 hours later. Plasmids and constructs. The venus-CD95L ORF and venus-CD95 ORF (full length) sensor vectors were created by sub-cloning the Venus-CD95L or the Venus-CD95 inserts (synthesized by IDT with flanking XbaI RE site on the 5' end and EcoRI RE site at the 3' end in the pIDTblue vector), which are composed of the Venus ORF followed by either the CD95L ORF (accession number NM_000639.2) or the CD95 ORF (accession number BC012479.1) as artificial 3'UTR (both lacking the A in the start codon), respectively, into the modified CD510B vector using XbaI and EcoRI. Ligation was done with T4 DNA ligase.
The pLenti-CD95L MUT vector was created by subcloning a CD95L cDNA insert with 2 nucleotide substitutions in codon 218 (TAT -> CGT) resulting in replacement of tyrosine for arginine, which has been described to inhibit binding to CD95 (Schneider et al., 1997). The pLenti-CD95L MUT NP vector insert contained both the Y218R mutation and a single nucleotide substitution at the second codon (CAG -> TAG), resulting in a premature stop codon right after the start codon.
Dox-inducible vectors expressing shRNAs were constructed by subcloning an annealed double-stranded DNA insert containing the sequence encoding the shRNA hairpin (sense strand: 5'-TGGCTTTATATATCTCCCTATCAGTGATAGAGATCGNNNNNNNNNNNNNNNNNNNNNCTC GAG nnnnnnnnnnnnnnnnnnnnnTTTTTGTACCGAGCTCGGATCCACTAGTCCAGTGTGGGCATGCT GCGTTGACATTGATT-3') into the pTIP-shR6 vector . BsaBI (NEB #R0537) and SphI-HF (NEB #R3182) were used to digest both the pTIP-shR6 vector (to excise the shR6 insert) and the insert followed by ligation with T4 DNA ligase. The template oligos were purchased from IDT. The poly-N represents the two 21bp sequences that transcribe for the sense (N) and antisense (n) shRNA.
The target sites of siL3, shL3, and shR6 were homozygously deleted from target cells by cotransfecting Cas9 plasmid with each corresponding pair of pSC-B-gRNA plasmids. Briefly, 400,000 cells were seeded per well on a 6-well plate the day prior to transfection. Each well was transfected with 940 ng of Cas9-GFP plasmid (pMJ920) (Jinek et al., 2013) and 450 ng of each pSC-B-gRNA plasmid using Lipofectamine 2000 (ThermoFisher Scientific #11668027). Media was replaced next day. One to two days later, cells were sorted for the top 50% population with the highest green fluorescence. Those cells were cultured for an additional week to let them recover. The cells were then sorted by FACS (BD FACSAria SORP system) directly into 96well plates containing a 1:1 ratio of fresh media:conditioned media for single cell cloning. Approximately two to three weeks later, single cell clones were expanded and subjected to genotyping. PCR using both a primer pair that flanked the region to be deleted and another pair containing one flanking primer and one internal primer was used to screen clones for homozygous deletion. For detection of the Δ41 deletion of the shL3 site, the flanking external primers were 5'-TCTGGAATGGGAAGACACCT-3' (Fr primer) and 5'-CCTCCATCATCACCAGATCC-3' (Rev primer), and the internal Rev primer was 5'-ATATACAAAGTACAGCCCAGT-3'. For detection of the Δ227 deletion of the shR6 site, the flanking external primers were 5'-GGTGTCATGCTGTGACTGTTG-3' (Fr primer) and 5'-TTTAGCTTAAGTGGCCAGCAA-3' (Rev primer), and the internal Rev primer was 5'-AAGTTGGTTTACATCTGCAC-3'. For detection of the Δ64 deletion of the siL3 site, the flanking external primers were 5'-CTTGAGCAGTCAGCAACAGG-3' (Fr primer) and 5'-CAGAGGTTGGACAGGGAAGA-3' (Rev primer), and the internal Rev primer was 5'-ATATGGGTAATTGAAGGGCTG-3'. For detection of the CD95 gene deletion, the flanking external primers were 5'-TGTTTAATATAGCTGGGGCTATGC-3' (Fr primer) and 5'-TGGGACTCATGGGTTAAATAGAAT-3' (Rev primer), and the internal reverse primer was 5'-GACCAGTCTTCTCATTTCAGAGGT-3'. After screening the clones, Sanger sequencing was performed to confirm that the proper deletion had occurred. Three clones were pooled for each si/shRNA target site deletion with the exception of HeyA8 ΔshR6 for which only clone #11 showed homozygous deletion of the shR6 site; clones #1 and 2 were not complete shR6 deletion mutants, but frame-shift mutations did occur in each allele (as in clone #11) making them CD95 knockout clones as depicted in Figure S3B. Knockdown with pTIP-shRNA viruses. Cells were plated at 50,000 to 100,000 cells per well in a 6-well plate. Cells were infected with lentivirus generated in 293T cells from the desired pTIP-shRNA vector in the presence of 8 μg/ml polybrene. Media was replaced 24 hours later. Selection was done 48 hours after infection with 3 μg/ml puromycin. Induction of shRNA expression was achieved by adding 100 ng/ml doxycycline (Dox) to the media.
Soluble CD95L protein rescue experiments. NB7 cells were seeded at 500 cells per well in a 96-well plate. Next day, cells were infected with the scrambled pLKO lentiviruses or pLKO-shL1 lentiviruses at an M.O.I. of 50 (to achieve 100% transduction efficiency under conditions omitting the puromycin selection step) in the presence of 8 µg/ml polybrene and 100 ng/ml of S2 CD95L for 16 hrs. Media was replaced the next day with media containing varying concentrations of recombinant S2 CD95L (50 ng/ml and 100 ng/ml). Cell confluence was measured over time in the IncuCyte Zoom.
To perform arrayed real-time PCR ( Figure S4), total RNA was extracted and used to make cDNA as described for standard real-time PCR. For Taqman Low Density Array (TLDA) profiling, custom-designed 384-well TLDA cards (Applied Biosystems #43422489) were used and processed according to the manufacturer's instructions. Briefly, 50 µl cDNA from each sample (200 ng total input RNA) was combined with 50 µl TaqMan Universal PCR Master Mix (Applied Biosystems) and hence a total volume of 100 µl of each sample was loaded into each of the 8 sample loading ports on the TLDA cards that were preloaded with assays from Thermofisher Scientific for human GAPDH control (Hs99999905_m1) and for detection of ATP13A3 (Hs00225950_m1), CAPZA1 (Hs00855355_g1), CCT3 (Hs00195623_m1), FSTL1 (Hs00907496_m1), FUPB1 (Hs00900762_m1), GNB1 (Hs00929799_m1), HISTH1C (Hs00271185_s1), NAA50 (Hs00363889_m1), NUCKS1 (Hs01068059_g1), PRELID3B (Hs00429845_m1), SNRPE (Hs01635040_s1), and TFRC (Hs00951083_m1) after the cards reached room temperature. The PCR reactions were performed using Quantstudio 7 (ThermoFisher Scientific). Since each of the port loads each sample in duplicates on the TLDA card and because two biological replicates of each sample were loaded onto two separate ports, quadruplicate Ct values were obtained for each sample. Statistical analysis was performed using Student's t test. To prepare RNAs from CD95L overexpressing cells HeyA8 ΔshR6 clone #11 cells were plated at 75,000 cells per well in six-well plates and were infected with lentivirus generated in 293T cells from the pLenti-CD95L vector in the presence of 8 μg/ml polybrene. Selection was done with 3 μg/ml puromycin 48 hours after infection. Cells were plated at 600,000 per 15 mm dish (Greiner CELLSTAR, cat#P7237, Sigma) after one day of puromycin selection. Total RNA was harvested at 50 hours after plating for RNAseq analysis Western blot analysis. Protein extracts were collected by lysing cells with RIPA lysis buffer (1% SDS, 1% Triton X-100, 1% deoxycholic acid). Protein concentration was quantified using the DC Protein Assay kit (Bio-Rad). 30 μg of protein were resolved on 12% SDS-PAGE gels and transferred to nitrocellulose membranes (Protran, Whatman) overnight at 25 mA. Membranes were incubated with blocking buffer (5% non-fat milk in 0.1% TBS/Tween-20) for 1 hour at room temperature. Membranes were then incubated with the primary antibody diluted in blocking buffer over night at 4°C. Membranes were washed 3 times with 0.1% TBS/Tween-20. Secondary antibodies were diluted in blocking buffer and applied to membranes for 1 hour at room temperature. After 3 more additional washes, detection was performed using the ECL reagent (Amersham Pharmacia Biotech) and visualized with the chemiluminescence imager G:BOX Chemi XT4 (Synoptics).

CD95 surface staining.
A pellet of about 300,000 cells was resuspended in 2% BSA/PBS on ice. Cells were resuspended with either anti-APO-1 primary antibody or the IgG3 isotype control diluted to 10 μg/ml in 2% BSA/PBS. Cells were incubated on ice at 4°C for 30 minutes, washed with 2% BSA/PBS, and anti-IgG3-RPE was added diluted in 2% BSA/PBS. After an incubation for 30 minutes at 4°C, percent red cells were determined by flow cytometry (Beckman Coulter).

Cell death quantification (DNA fragmentation) and ROS production.
A cell pellet (500,000 cells) was resuspended in 0.1% sodium citrate, pH 7.4, 0.05% Triton X-100, and 50µg/ml propidium iodide. After resuspension, cells were incubated 2 to 4 hours in the dark at 4°C. The percent of subG1 nuclei (fragmented DNA) was determined by flow cytometry. For pLKO shRNA infection experiments, subG1 quantification was performed eight days after infection. For siRNA and DsiRNA transfection experiments, subG1 quantification was performed four days after transfection. ROS production was quantified as previously described . Cell growth and fluorescence over time. After treatment/infection, cells were seeded at 700 to 1,500 per well in a 96-well plate in triplicate. Images were captured at indicated time points using an IncuCyte ZOOM live cell imaging system (Essen BioScience) with a 10x objective lens. Percent confluence, red object count, and the green object integrated intensity were calculated using the IncuCyte ZOOM software (version 2015A).

RNA-Seq analysis. HeyA8
ΔshR6 clone #11 was infected with pLKO-shScr or pLKO-shR6. A pool of three 293T ΔshL3 clones was infected with either pTIP-shScr or pTIP-shL3. After selection with puromycin, the pTIP-infected 293T cells were plated with Dox in duplicate at 500,000 cells per T175 flask. The pLKO-infected HeyA8 cells were plated at 500,000 cells per flask. Total RNA was harvested 50 hours and 100 hours after plating. RNA libraries were generated and sequenced (Genomics Core facility at the University of Chicago). The quality and quantity of the RNA samples was checked using an Agilent bio-analyzer. Paired end RNA-SEQ libraries were generated using Illumina TruSEQ TotalRNA kits using the Illumina provided protocol (including a RiboZero rRNA removal step). Small RNA-SEQ libraries were generated using Illumina small RNA SEQ kits using the Illumina provided protocol. Two small RNA-SEQ sub-libraries were generated: one containing library fragments 140-150 bp in size and one containing library fragments 150-200 bp in size (both including the sequencing adaptor of about 130bp). All three libraries (one RNA-SEQ and two small RNA-SEQ) were sequenced on an Illumina HiSEQ4000 using Illumina provided reagents and protocols. The raw counts for RNAs were calculated by HTSeq. The sequenced reads were mapped to hg38 human genome using Tophat and bowtie2. Differential gene expression was analyzed by R Bioconductor DESeq2 package using shrinkage estimation for dispersions and fold changes to improve stability and interpretability of estimates. P values and adj P values were calculated using the DESeq2 package. Counts and sequences of small RNAs were extracted from raw fastq file by UNIX shell scripts and then grouped and summarized by SQL language in database tables. To extract small RNAs (shRNA sequences) adaptor sequences were remove using Trim Galore software.
In order to identify differentially abundant RNAs using a method unbiased by genome annotation, we also analyzed the raw 100 bp reads for differential abundance. First, the second end in each paired end read was reverse complemented, so that both reads were on the same strand. Reads were then sorted and counted using the core UNIX utilities sort and uniq. Reads with fewer than 128 counts across all 16 samples were discarded. A table with all of the remaining reads was then compiled, summing counts from each sequence file corresponding to the same sample. This table contained a little over 100,000 reads. The R package edgeR (http://bioinformatics.oxfordjournals.org/content/26/1/139) was used to identify differentially abundant reads, and then these reads were mapped to the human genome using blat (http://genome.cshlp.org/content/12/4/656.abstract) to determine chromosomal location whenever possible. Homer (http://homer.salk.edu/homer/) was used to annotate chromosomal locations with overlapping genomic elements (such as genes). Raw read counts in each sequence file were normalized by the total number of unique reads in the file.
To identify the most significant changes in expression of RNAs both methods of RNAs Seq analysis (alignment and read based) were used to reach high stringency. All samples were prepared in duplicate and for each RNA the average of the two duplicate was used for further analysis. In the alignment based analysis only RNAs were considered that had a base mean of >2000 reads and were significantly deregulated between the groups (adjusted p-value <0.05). Only RNAs were scored as deregulated when they were more than 1.5 fold changed in the shL3 expressing cells at both time points and in the shR6 expressing cells at either time points (each compared to shScr expressing cells) ( Table S1). This was done because we found that the pLKO driven expression of shR6 was a lot lower than the pTIP driven expression of shL3 (see the quantification of the two shRNAs in Figure S7A) and likely as a result of the cell responses in the shR6 expressing cells were reduced. In the read based analysis only normalized read numbers of >10 across the samples in each treatment were considered. Only RNA reads were further considered that had a variation between duplicates of less than 2 fold. Only RNAs reads were considered that showed >1.5 change between treatment groups in both cell lines and both time points (Table S1). Using the chromosomal localization of each read the gene of origin was determined. Finally, All RNAs were counted that showed deregulation in the same direction with both methods. This resulted in the identification of 11 RNAs that were down and 1 that was upregulated in cells exposed to the shRNAs shL3 and shR6. To determine the number of seed matches in the 3'UTR of downregulated genes the 3'UTRs of the 11 mRNAs were extracted from the Homo sapiens gene (GRCh38.p7) dataset of the Ensembl 86 database using the Ensembl Biomart data mining tool. For each gene only the longest deposited 3'UTR was considered. Seed matches were counted in all 3'UTRs using in-house Perl scripts.
GSEA used in Figure 3E and 5C was performed using the GSEA v2.2.4 software from the Broad Institute (www.http://software.broadinstitute.org/gsea); 1000 permutations were used The Sabatini gene lists ( Table 2) were set as custom gene sets to determine enrichment of survival genes versus the nonsurvival control genes in downregulated genes from the RNA seq data; GSEA was also performed on all the gene ontology groups (MSigDB c5.all.v6.0). Adjusted pvalues below 0.05 were considered significantly enriched. The GO enrichment analysis shown in Figure 5E was performed using the software available on www.Metascape.org using default running parameters.

AGO pull down
Between 10 and 25 x 10 6 HCT116 wt or Drosha -/cells stably expressing either pLenti or pLenti-CD95L MUT NP were lysed in NP40 lysis buffer (20 mM Tris, pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% (v/v) NP40, supplemented with phosphatase inhibitors) on ice for 15 minutes. The lysate was sonicated 3 times for 30 s at 60% amplitude (Sonics, VCX130) and cleared by centrifugation at 12,000g for 20 minutes. AGO1-4 were pulled down by using 500 mg of Flag-GST-T6B peptide (Hauptmann et al., 2015) and with 60 ml anti-Flag M2 magnetic beads (Sigma-Aldrich) for 2 hrs at 4°C. The pull-down was washed 3 times in NP40 lysis buffer. During the last wash, 10% of beads were removed and incubated at 95°C for 5 minutes in 2x SDS-PAGE sample buffer. Samples were run on a 4-12% SDS-PAGE and transferred to nitrocellulose membrane. The pull-down efficiency was determined by immunoblotting against AGO1 and AGO2 (Abcam 98056 and 32381). To the remaining beads 500 µl TRIzol reagent were added and the RNA extracted according to the manufacturer's instructions. The RNA pellet was diluted in 20 µl of water. The sample was split and half of the sample was dephosphorylated with 0.5 U/µl of CIP alkaline phosphatase at 37°C for 15 min and subsequently radiolabeled with 0.5 µCi γ-32 P-ATP and 1 U/µl of T4 PNK kinase for 20 min at 37°C. The AGO1-4 interacting RNAs were visualized on a 15% urea-PAGE. The remaining RNA was taken through a small RNA library preparation as previously descried (Hafner et al., 2012). Briefly, RNA was ligated with 3' adenylated adapters and separated on a 15% denaturing urea-PAGE. The RNA corresponding to insert size of 19-35 nt was eluted from the gel, ethanol precipitated followed by 5' adapter ligation. The samples were separated on a 12% Urea-PAGE and extracted from the gel. Reverse transcription was performed using Superscript III reverse transcriptase and the cDNA amplified by PCR. The cDNA was sequenced on Illumina HiSeq 3000. PCR primer sequences: CAAGCAGAAGACGGCATACGAGATCGTGATGTGACTGGAGTTCCTTGGCACCCGAG AATTCCA. To identify CD95L derived small RNAs all reads (4 conditions in duplicate) after removal of the adaptor sequences were BLASTED against the CD95L ORF (derived from NM_000639.1). Only reads were considered with an evalue of less than 0.05 and 100% identity across the entire length of the read.
Conversion of shL3 and shR6 to siRNAs. From the RNA seq analysis with HeyA8 ΔshR6 infected with pLKO-shR6 and 293T ΔshL3 clones infected pTIP-shL3, we analyzed the mature double-stranded RNAs derived from pLKO-shR6 and pTIP-shL3 and found that the most abundant RNA forms were both shifted by one nucleotide. Based on these most abundant species observed after cellular processing, we converted shL3 and shR6 sequences to siRNAs. The genomic target sequence in shL3 (21nt) is 5'-ACUGGGCUGUACUUUGUAUAU-3'. For the shL3=>siL3 sense strand, one G was added before the A on the 5' end while the last U on the 3' end was deleted, and second and third to the last ribonucleotides on the 3' end (UA) were replaced with deoxyribonucleotides for stabilization. For shL3=>siL3 antisense strand, the last three nucleotides on the 5' end (AUA) were deleted and one U and two dTs (UdTdT) were added after the last U on the 3'end. The shL3=>siL3 sense strand is 5'-GACUGGGCUGUACUUUGUAdTdA-3' and antisense strand is 5'-/5Phos/UACAAAGUACAGCCCAGUUdTdT-3'. The shR6=>siRNA was designed in similar fashion except that two Gs instead of one G were added to the 5' end of the sense strand while UUdTdT instead of UdTdT was added to the 3' end of the antisense strand. The genomic target sequence in shR6 (21nt) is 5'-GUGCAGAUGUAAACCAAACUU-3'. The shR6=>siR6 sense strand is 5'-GGGUGCAGAUGUAAACCAAACUU-3' and antisense strand is 5'-/5Phos/UUUGGUUUACAUCUGCACUUdTdT-3'. Both shL3=>siL3 and ShR6=>siR6 siRNA duplexes were purchased from Dharmacon.

Construction of pTIP-shRNA libraries.
The pTIP-shRNA libraries were constructed by subcloning libraries of 143bp PCR inserts of the form 5'-XXXXXXXXXXXXXXXXXXXXXXXXXXATAGAGATCGNNNNNNNNN NNNNNNNNNNNNCTCGAGNNNNNNNNNNNNNNNNNNNNNTTTTTGTACCGAGCTCGGAT CCACTAGTCCAGTGTGGGCATGCTGCGTTGACATTGATT-3' into the pTIP-shR6 vector after excising the shR6 insert. The poly-N region represents the 21-mer sense and antisense shRNA hairpin. The intervening CTCGAG is the loop region of the shRNA. The 5 libraries targeting Venus, CD95L ORF, CD95L 3'UTR, CD95 ORF, or CD95 3'UTR were composed of every possible 21-mer shRNA (i.e. each nearest neighbor shRNA was shifted by 1 nucleotide). These libraries were synthesized together on a chip as 143 bp single-stranded DNA oligos (CustomArray Inc, Custom 12K oligo pool). Each shRNA pool had its own unique 5' end represented by the poly-X region. This allowed selective amplification of a particular pool using 1 of 5 unique Fr primers (CD95L ORF: 5'-TGGCTTTATATATCTCCCTATCAGTG-3', CD95L 3' UTR: 5'-GGTCGTCCT ATCTATTATTATTCACG-3', CD95 ORF: 5'-TCTTGTGTCCAGACCAATTTATTTCG-3', CD95 3'UTR:  and the common reverse primer 5'-AATCAATGTCAACGCAGCAT-3'. Phusion High Fidelity Polymerase was used to amplify each library pool; standard PCR conditions were used with an annealing temperature of 61°C and 15 cycles. PCR reactions were purified using PCR Cleanup kit (QIAGEN). The pTIP-shR6 vector and each of the amplified libraries were digested with SphI-HF and BsaBI. Digested PCR products were run on either a 2% Agarose gel or a 20% polyacrylamide (29:1) gel made with 0.5 x TBE buffer. PCR products were extracted using either Gel Extraction kit (QIAGEN) for extraction from Agarose gels or via electro-elution using D-Tube Dialyzer Mini columns (Novagen #71504). Purified PCR inserts were then ligated to the linearized pTIP vector with T4 DNA ligase for 24 hours at 16°C. The ligation mixtures were transformed via electroporation in MegaX DH10B T1 cells (Invitrogen #C6400) and plated on 24 cm ampicillin dishes. At least 10 colonies per pool were picked and sequenced to verify successful library construction. After verification, all colonies per library were pooled together and plasmid DNA extracted using the MaxiPrep kit (QIAGEN). The 5 pTIP-shRNA library DNA preps were used to produce virus in 293T cells.
Lethality screen with pTIP-shRNA libraries. NB7 cells were seeded at 1.5 x 10 6 per 145 cm 2 dish. Two dishes were infected with each of the 5 libraries with a transduction efficiency of about 10 to 20%. Media was replaced next day. Infected cells were selected with 1.5 μg/ml puromycin. Cells infected with the Venus, CD95L ORF, and CD95L 3'UTR-targeting libraries were pooled in a 1:1:1 ratio to make the CD95L cell pool. Likewise, cells infected with the Venus, CD95 ORF, and CD95 3'UTR-targeting libraries were pooled to make the CD95 receptor cell pool. The CD95 and the CD95L cell pools were plated separately each in 2 sets of duplicates seeded at 600,000 cells per 145cm 2 dish. One set received 100 ng/ml Dox, and the other one was left untreated (total of 4 dishes per combined pool; 2 received no treatment and 2 received Dox). Cells infected with the different libraries were also plated individually in triplicate with or without Dox on a 96-well plate to assess the overall toxicity of each pool. DNA was collected from each 145cm 2 dish 9 days after Dox addition.
The shRNA barcodes were amplified from the harvested DNA template using NEB Phusion Polymerase with 4 different pairs of primers (referred to as N, N+1, N+2, and N+3) in separate reactions per DNA sample. The N pair consisted of the primers used to amplify the CD95L ORF library (Fr: 5'-TGGCTTTATATATCTCCCTATCAGTG-3' and Rev: 5'-AATCAATGTCAACGCAGCAT-3'). The N+1 primers had a single nucleotide extension at each 5' end of the N primers corresponding to the pTIP vector sequence (Fr: 5'-TTGGCTTTATATATCTCCCTATCAGTG-3' and Rev: 5'-TAATCAATGTCAACGCAGCAT-3'). The N+2 primers had 2 nucleotide extensions (Fr: 5'-CTTGGCTTTATATATCTCCCTATCAGTG-3' and Rev: 5'-ATAATCAATGTCAACGCAGCAT-3'), and the N+3 primers had 3 nucleotide extensions (Fr: 5'-TCTTGGCTTTATATATCTCCCTATCAGTG-3' and Rev: 5'-AATAATCAATGTCAACGCAGCAT-3'). The barcodes from the pTIP-shRNA library plasmid preparations were also amplified using Phusion Polymerase with the N, N+1, N+2, and N+3 primer pairs. The shRNA barcode PCR products were purified from a 2% Agarose gel and submitted for 100 bp paired-end deep sequencing (Genomics Core facility at the University of Chicago). DNA was quantitated using the Qubit. The 4 separate PCR products amplified using N, N+1, N+2, and N+3 were combined in equimolar amounts for each sample. Libraries were generated using the Illumina TruSeq PCRfree kit using the Illumina provided protocol. The libraries were sequenced using the HiSEQ4000 with Illumina provided reagents and protocols. Raw sequence counts for DNAs were calculated 28 by HTSeq. shRNA sequences in the PCR pieces of genomic DNA were identified by searching all reads for the sense sequence of the mature shRNA plus the loop sequence CTCGAG. To avoid a division by zero problem during the subsequent analyses all counts of zero in the raw data were replaced with 1. A few sequences with a total read number <10across all plasmids reads were not further considered. In the CD95L pool this was only one shRNA (out of 2362 shRNAs) (L792') and in the CD95 20 shRNAs (out of 3004 shRNAs) were not represented (R88, R295, R493, R494, R496, R497, R498, R499, R213', R215', R216', R217', R220', R221', R222', R223', R225', R226', R258', R946', R1197', R423'). While most shRNAs in both pools had a unique sequence two sequences occurred 6 times (L605', L607', L609', L611', L613', L615', and L604', L606', L608', L610', L612', L614'). In these cases read counts were divided by 6. Two shRNAs could not be evaluated: 1) shR6 in the CD95 pool. It had a significant background due to the fact that the pTIP-shR6 was used as a starting point to clone all other shRNAs. 2) shL3 was found to be a minor but significant contaminant during the infection of some of the samples. For each condition two technical duplicates and two biological duplicates were available. To normalize reads in order to determine the change in relative representation of shRNAs between conditions, the counts of each shRNA in a subpool (all replicates and all conditions) was divided by the total number of shRNAs in each subpool (%). First, the mean of the technical replicates (R1 and R2) was taken. To analyze the biological replicates and to determine the changes between conditions, two analyses were performed: 1) The change in shRNA representation between the cloned plasmid library and cells infected with the library and then cultured for 9 days without Dox (infection -Dox). Fold downregulation was calculated for each subpool as [(plasmid %/-Dox1 %+plasmid %/-Dox2 %)/2]. 2) The difference in shRNA composition between the infected cells cultured with (infection +Dox) and without Dox. Fold downregulation was calculated for each subpool as [(-Dox1 %/+Dox1 %)+(-Dox1 %/+Dox2 %)+(-Dox2 %/+Dox1 %)+(-Dox2 %/+Dox2 %)/4]. Only shRNAs were considered that were at least 5 fold underrepresented in either of the two analyses (data in Table S3).
The toxicity index (TI). The TI is defined by the sum of the counts of a 6mer or 8mer seed match in the 3'UTRs of critical survival genes divided by the seed match counts in the 3'UTRs of nonsurvival genes. We used the 1882 survival genes recently described in a CRISPR/Cas9 lethality screen by Wang et al. (Wang et al., 2015). The survival genes were defined by having a CRIPSR score of <-0.1 and an adjusted p-value of <0.05. We chose as a control group to these top essential genes the bottom essential genes using inverse criteria (CRISPR score of >0.1 and adjusted p-value of <0.05) and are referring to them as the "nonsurvival genes". Both counts were normalized for the numbers of genes in each gene set. 3'UTRs were retrieved as described above. For the survival genes 1846 and for the nonsurvival genes 416 3'UTRs were found. For each gene only seed matches in the longest 3'UTR were counted. The TI was calculated for each of the 4096 possible 6mer combinations and each of the 65536 possible 8mer combinations (Table S4). These numbers were then assigned to the results of the shRNA screen (Table S5). An alternative TI was calculated based on the top 850 most highly expressed survival genes (all expressed >1000 average reads) and select 850 expression matched genes not described to be critical for cancer cell survival.

Sylamer analysis.
Sylamer is a tool to test for the presence of RNAi-type regulation effects from a list of differentially expressed genes, independently from small RNA measurements (van Dongen et al., 2008) (http://www.ebi.ac.uk/research/enright/software/sylamer). For short stretches of RNA (in this case length 6, 7, and 8 in length corresponding to the lengths of the determinants of seed region binding in RNAi-type binding events), Sylamer tests for all possible motifs of this length whether the motif occurrences are shifted in sequences associated with the list under consideration, typically 3'UTRs when analyzing RNAi-type binding events. A shift or enrichment of such a motif towards the down-regulated end of the gene list is consistent with upregulation of a small RNA that has the motif as the seed region. Sylamer tests in small increments along the list of genes, using a hypergeometric test on the counts of a given word, comparing the leading part of the gene list to the universe of all genes in the list. For full details refer to (van Dongen et al., 2008). Enriched motifs stand out from the back-ground of all motifs tested, as visible in the Sylamer plot. The plot consist of many different lines, each line representing the outcomes of a series of tests for a single word, performed along regularly spaced intervals (increments of 200 genes) of the gene list. Each test yields the log-transformed P-value arising from a hypergeometric test as indicated above. If the word is enriched in the leading interval the log-transformed value has its value plotted on the positive y-axis (sign changed), if the word is depleted the log-transformed value is plotted on the negative y-axis. 3' UTRs were used from Ensembl, version 76. As required by Sylamer, they were cleaned of low-complexity sequences and repetitive fragments using respectively Dust (Morgulis et al., 2006) with default parameters and the RSAT interface (Medina-Rivera et al., 2015) to the Vmatch program, also run with default parameters. Sylamer (version 12-342) was run with the markov correction parameter set to 4.

Statistical analyses.
Continuous data were summarized as means and standard deviations and dichotomous data as proportions. Continuous data were compared using t-tests for two independent groups and one-way ANOVA for 3 or more groups. For evaluation of continuous outcomes over time, two-way ANOVA was used with one factor for the treatment conditions of primary interest and a second factor for time treated as a categorical variable to allow for nonlinearity. Comparisons of single proportions to hypothesized null values were evaluated using binomial tests. Statistical tests of two independent proportions were used to compare dichotomous observations across groups. To test if higher TI is enriched in shRNAs that were highly downregulated, p-values were calculated based on permutated datasets using Mann-Whitney U tests. The ranking of TI was randomly shuffled 10,000 times and the W statistic from our dataset was compared to the distribution of the W statistic of the permuted datasets. Test of enrichment was based on the filtered data of at least 5-fold difference, which we define as a biologically meaningful. Fisher Exact Test was performed to assess enrichment of downregulated genes amongst genes with si/shRNA seed matches. All statistical analyses were conducted in Stata 14 or R 3.3.1.
Data availability. The accession numbers for the RNA-Seq and expression data reported in this paper are GSE87817 and GSE96620.   and total green object integrated intensity (right) over time of a HeyA8 CD95 knockout clone (ΔR6 cl#2) expressing the Venus-CD95L sensor after transfection with 25 nM of single-stranded sense, single-stranded antisense, or double-stranded (ds) siScr or siL3 siRNAs. The CD95L sensor is schematically shown and comprises the Venus ORF fused to the CD95L ORF lacking the A of the ATG start codon (X). (B) Percent cell confluence (left) and total green object integrated intensity (right) over time of the HeyA8 CD95L sensor cell used in Figure 2A after transfection with 5 nM siScr or siL3 double-stranded RNA (dsRNA) or doublestranded DNA (dsDNA). (C) Summary of experiments to test whether siL3 and siL2 siRNAs modified as indicated (left) were active (check mark) or not (X) in reducing green fluorescence or cell growth (both >70% reduction at end point) when transfected at 25 nM siRNAs (except for blunt end oligonucleotides which were used at 5 nM and compared to 5 nM of siL3) into HeyA8 CD95L sensor cells used in Figure 2A. Endpoints were 164 hours for blunt end siRNA transfection, 180 hrs for modified siL3 and 144 hrs for modified siL2 siRNA transfections. (D) Red object count over time of HeyA8 cells (expressing NucRed) after transfection with different ratios of siL3 and mutant siL3 (siL3MUT). (E) Percent cell confluence over time of HCT116 parental (left) or Dicer -/-(clone #43, another Dicer -/clone, #45, gave a similar result, data not shown), or Drosha -/-(right) cells after infection with either shScr, shL3 or shR6. Inserts show expression levels of Drosha/Dicer and AGO2 levels in the tested cells. (F) Western blot analysis of HCT116 wt, Dicer -/or Drosha -/cells 5 days after infection with either pLKO-shScr or pLKO-shR6. (G) Percent cell confluence over time of HCT116 wt, Dicer -/and Drosha -/cells after transfection with 25 nM siScr or siL3. (H) Percent reduction in Venus expression (green) and in cell number (red object count (red)) over time of HeyA8 cells expressing the Venus-CD95L sensor and red nuclei after transfection with 5 nM of different chimeric siRNAs generated by substituting nucleotides in the toxic siL3 with the scrambled siRNA sequence beginning at either the seed match end (top) or the opposite end (bottom) of siL3 after 188 hours. The schematic in the middle shows the sequence of siL3 and the Scr siRNA (both sense and antisense strands). The 6mer seed sequence region of siL3 (positions 2 to 7) is highlighted in light blue. Nucleotides shared by siScr and siL3 are shown in grey font.  Figure 3A. mRNAs shown in red were found to be essential cancer survival genes in two CRISPR lethality screens. The number of essential genes was enriched from 6.6% of the tested genes (Blomen et al., 2015;Wang et al., 2015) to 54.5% in our study (p=3 x 10 -6 according to binomial distribution). The number of shL3 and shR6 6mer seed sequences in the 3'UTRs is shown for each gene in the green and yellow ovals, respectively.  Sylamer plots for the list of genes in the shL3 experiment (top) and the shR6 experiment (bottom) ordered from down-regulated to up-regulated. The most highly enriched sequence is shown which in each case is the 8mer seed match of the introduced shRNA. The red line corresponds to a p-value threshold of 0.05 after Bonferroni correction for the number of words tested (65536). Bonferroni-adjusted p-values are shown. The unadjusted p-values are 1.58E-24 and 1.35E-26, respectively. The black line corresponds to the represents the sequences carrying the let-7 8mer seed match. (B) Location of the 6mer seed matches of either shL3 or shR6 in the 3'UTRs of the 11 genes (shown at scale) identified in the screen described in Figure 3A. Red font indicates a critical survival gene. (C) A series of six 2x2 contingency tables comparing whether or not a critical survival gene is downregulated after treatment with the indicated siRNA or shRNA to whether or not its 3'UTR contains at least 1 seed match for the introduced sh/siRNA. p-values were calculated using Fisher's Exact Test to determine any significant relationship between gene downregulation and presence of seed matches in 3'UTR. Figure 5. Identifying all toxic shRNAs derived from CD95L and CD95. (A) Schematic showing the cloned shRNAs covering the ORF of Venus and the ORFs and 3'UTRs of CD95L and CD95. The 3'UTR is displayed as a dashed line because it was not included in the full-length Venus-CD95L/CD95 sensors. (B) Work-flow of pTIP-shRNA library synthesis, shRNA screen and data analysis. (C) Ranked fold reduction of shRNAs spanning Venus and CD95L (ORF and 3'UTR) (left 3 panels) and Venus and CD95 (ORF and 3'UTR) (right 3 panels). The ranked lists were separated into the shRNAs derived from Venus (top), the ORFs (center) and the 3'UTRs (bottom). The p-value of enrichment for each ranked set of shRNAs is given. Only the parts of the ranked lists are shown with the downregulated shRNAs. For all 6 panels, the top section of each panel (boxed in blue) contains the data on shRNAs downregulated after infection of cells and cultured for 9 days without Dox when compared to the composition of the shRNA plasmid library and the bottom half (boxed in orange) contains the data on shRNAs downregulated after culture with Dox for 9 days when compared to the culture without Dox. P-values were calculated using Mann Whitney U tests with a one-sided alternative that the rank was lower. (D) The location of all shRNAs significantly downregulated at least 5 fold along the sequences of Venus, CD95L ORF, CD95L 3'UTR (left panel) and Venus, CD95 ORF, and CD95 3'UTR (right panel). The top half of each sub panel (blue ticks) shows the shRNAs downregulated after infection and the bottom half (orange ticks) contains the data on shRNAs downregulated after culture with Dox for 9 days. Significance of enrichment in the different subpanels is shown. p-values were calculated according to statistical tests of two proportions. Each data set was compared to the corresponding Venus distribution. Green line: sequence that corresponds to the intracellular domain of CD95L. (E) Fold downregulation versus ranked Toxicity Index for shRNAs of the Venus/CD95L pool (left three panels) and the Venus/CD95 pool (right three panels). Orange and blue tick marks indicate the same as in Figure 4D. To test if higher TI is enriched in shRNAs that were highly downregulated, p-values were calculated based on permutated datasets using Mann-Whitney U tests. The ranking of TI was randomly shuffled 10,000 times and the W statistic from our dataset was compared to the distribution of the W statistic of the permuted datasets.  Figure 3D). Histones underlined contain a 3'UTR. (G) Metascape analysis of 5 RNA Seq data sets analyzed. The boxed GO term clusters were highly enriched in all 5 data sets.  Figure 5D between the shRNAs downregulated in cells after infection and after Dox treatment.               Figure 3A with the critical survival genes found in the Sabatini and Brummelkamp studies (Blomen et al., 2015;Wang et al., 2015), all listed in Table S2. The Venn diagram was generated using http://bioinformatics.psb.ugent.be/webtools/Venn. (C) Kinetic quantitative PCR of the downregulated genes in the 293T ΔshL3 pTIP-shL3 cells. RNA was collected at 14 hrs, 26 hrs, and 50 hrs after treatment with Dox. NS, not significant, * p<0.05, ** p<0.001, *** p<0.0001.

3'-A C C A A A U G U A C A A C A A A C U -5'
The following describes the 11 genes that were significantly downregulated after introducing the toxic shRNAs shL3 or shR6 into cancer cells (see Figure 3B) and some of their cancer relevant activities: 1) CAPZA1 (capping actin protein of muscle Z-line alpha subunit 1) is an actin capping protein. CAPZA1 knockdown has been reported to cause disassembly of autophagosomes (Mi et al., 2015) It is overexpressed in malignant melanoma (Sun et al., 2011). 2) CCT3 (chaperonin containing TCP1 subunit 3) is part of a chaperon complex that folds various proteins including actin and tubulin. CCT3 is required for proper mitotic progression (Zhang et al., 2016). 3) FSTL1 (follistatin-like 1) is a putative activin-binding protein. Knockdown of FSTL1 in lung cancer cells resulted in mitotic arrest followed by apoptosis promoted by the activation of caspase-3 and -9 (Bae et al., 2016). FSTL1 is downregulated during cellular senescence of human mesenchymal stem cells (Yoo, Choi, & Kim, 2013). 4) FUBP1 (far upstream element binding protein 1). A lack of FUBP1 causes a cell-autonomous defect in the maintenance of fetal and adult hematopoietic stem cells (HSCs). FUBP1-deficient adult HSCs exhibit significant transcriptional changes, including upregulation of the cell-cycle inhibitor p21 and the pro- Values were calculated from samples done in quadruplicates shown as the mean ± S.E. (C) PI staining used to quantify percent subG1 for cells 4 days after transfection with 1 nM and 5 nM of combined siRNA pools targeting the 7 different survival genes as in Figure S5B. *** p<0.0001. Figure S6. Quantification of the mature shRNA forms. (A) Graphical representation of the percentage of the different Dicer cut sites to produce the mature passenger (top) and guide (bottom) strands of 3 shRNAs expressed from two vectors. All analyses were performed with cells 50 hrs after either Dox addition (in pTIP expressing cells) or infection with the pLKO virus. Letters in green: vector sequences; black: passenger and guide strands of shRNAs; Arrow heads label the most highly cleaved residues; the darker the arrow head the more highly cleaved. Numbers in yellow box represent total number of reads detected for passenger and guide strands. (B) Percent cell confluence in HeyA8 cells after transfection with shL3=>siL3 (shL3 converted to an siRNA) or shR6=>siR6 (shR6 converted to an siRNA). Conversion was based on the most common mature double-stranded RNA form produced as indicated by the results in Figure S6A. Insert: percent DNA fragmentation in the same samples. *** p<0.0001. (C) Sylamer plots for the list of genes in the shL1 experiment (293T cells 100 hrs after infection with pLKO-shL1) (left) and the siL3 experiment (48 hrs after transfection of HeyA8 cells with siL3) (right) ordered from down-regulated to up-regulated. The most highly enriched sequences are shown which in each case is the 7mer seed match of the introduced shRNA. The red line corresponds to a p-value threshold of 0.05 after Bonferroni correction for the number of words tested. Bonferroni-adjusted p-values are shown.   . Cells were infected with the Tet-inducible pTIP-shR6 virus, selected for two days with puromycin and then subjected to an analysis in the IncuCyte Zoom. No Dox was added. Two other inducible constructs (pTIP-shL1 and pTIP-shL3) were tested in the same way and no evidence of leakiness was observed (data not shown), supporting the finding in the shRNA screen that certain shRNA constructs display leakiness while others do not. (B) Scatterplot showing the fold change of shRNAs after infection of cells and culture for 9 days without Dox when compared to the composition of the shRNA plasmid library (X axis) and the fold change of shRNAs after culture with Dox for 9 days when compared to the culture without Dox (Y axis). The red dots are the shRNAs that were significantly downregulated at least 5 fold. The number of shRNAs labeled in red in each quartile is given in blue. Two of the shRNAs tested before are labeled in green. Figure S10. Does DISE preferentially target survival genes or all highly expressed genes? (A) Correlation between 1185 survival genes (genes identified as critical survival genes in two genome-wide lethality screens (Blomen et al., 2015;Wang et al., 2015) and expressed at least at 100 reads in all of the 16 control RNA Seq samples in this study) and 1185 expression matched nonsurvival genes (genes not identified as critical survival genes in two genome-wide lethality screens (Blomen et al., 2015;Wang et al., 2015) and expressed at least at 100 reads in all of the 16 control RNA Seq samples in this study). (B) Reanalysis of the CD95L ORF data in Figure 5E using two alternative ways to calculate the toxicity index (TI). Left: the analysis shown in Figure 5E with the data ranked using the original TI (using all known 3'UTRs for each gene group). Center: analysis with the data ranked using the original TI but based on only the longest 3'UTR for each gene. Right: analysis with the data ranked using the new TI based on expression matched SGs and nonSGs identified in A and using the longest 3'UTR for each gene. To test if higher TI is enriched in shRNAs that were highly downregulated, pvalues were calculated based on permutated datasets using Mann-Whitney U tests. clone #1 infected with empty vector or with the RNA only version of CD95L (wt CD95L killed the cells within a few hours, not shown). The three right panels: MCF-7 CD95 -/-FA4 or two MCF-7 clones in which we deleted the shR6 site resulting in an out-of-frame shift (generated as described in Figure 1B and S3) after infection with either vector or wt CD95L. 14 Supplementary Tables and movies: Table S1: Results of the RNA-Seq analysis used to generate Figure 3B.