CRISPR-Cas9-based mutagenesis frequently provokes on-target mRNA misregulation

The introduction of insertion-deletions (INDELs) by non-homologous end-joining (NHEJ) pathway underlies the mechanistic basis of CRISPR-Cas9-directed genome editing. Selective gene ablation using CRISPR-Cas9 is achieved by installation of a premature termination codon (PTC) from a frameshift-inducing INDEL that elicits nonsense-mediated decay (NMD) of the mutant mRNA. Here, by examining the mRNA and protein products of CRISPR targeted genes in a cell line panel with presumed gene knockouts, we detect the production of foreign mRNAs or proteins in ~50% of the cell lines. We demonstrate that these aberrant protein products stem from the introduction of INDELs that promote internal ribosomal entry, convert pseudo-mRNAs (alternatively spliced mRNAs with a PTC) into protein encoding molecules, or induce exon skipping by disruption of exon splicing enhancers (ESEs). Our results reveal challenges to manipulating gene expression outcomes using INDEL-based mutagenesis and strategies useful in mitigating their impact on intended genome-editing outcomes.


INTRODUCTION
Technologies enabling the directed introduction of double-stranded DNA breaks such as CRISPR/Cas9 have transformed our ability to systematically identify DNA sequences important in biology 1,2 . The repair of these double-stranded breaks by non-homologous end-joining (NHEJ) results in insertion-deletions (INDELs) of unpredictable length that upon introduction into exonic sequences could alter the coding frame and install a pre-mature termination codon (PTC). Ribosomes that encounter a PTC in nascent mRNAs, recognized by the assembly of a complex that includes proteins from the ribosome and a 3' exon-splice junction complex, induces the destruction of the mutant mRNA 3,4 . On the other hand, INDELs that preserve the reading frame may yield proteins with altered sequences and thus shed light on determinants important for its function 5 .
Exonic sequences are laden with regulatory features that control many facets of the mRNA lifecycle including splicing and folding, two mRNA attributes that influence protein sequence composition and sites of initiation/termination, respectively [6][7][8] . Yet, the frequency with which these elements once impacted by INDELs influence gene expression outcomes remains mostly unknown. Another potential obstacle to precision gene-editing using INDEL-type mutagenesis is the presence of pseudo-mRNAs, mRNAs harboring a PTC that can nevertheless incorporate introduced INDELs thus altering their potential to produce proteins 9 .
To determine the extent to which these molecular events confound our ability to predict gene expression outcomes from CRISPR/Cas9 editing, we took inventory of the posttranscriptional and -translational effects of frameshift-inducing INDELs in a panel of CRISPRedited cells lines. We observed changes in the array of transcripts or proteins expressed from CRISPR-targeted genes in ~50% of the cell lines studied. A mechanistic account of these phenomena is presented here.

Prevalence of unanticipated gene expression outcomes following on-target CRISPR/Cas9-mediated gene editing in commercially available cell lines.
To service several ongoing research programs, we had assembled a panel of commercially available HAP1 cell lines harboring frameshift-inducing INDELs that presumably eliminate effective protein production from the targeted gene by promoting non-sense mediated decay (NMD) of the encoded mRNA ( Fig. 1A; Supp. Table 1). HAP1 cells harbor a single copy of each chromosome thus reducing the challenges frequently associated with achieving homozygosity in diploid cells for genetic studies 10 . To confirm the effects of the INDEL on target gene expression, we used two antibodies each recognizing a different epitope within the targeted protein ( Fig. 1B; Supp. Table 2). We observed in some cell lines the anticipated loss of protein presumably due to the introduced INDEL but in other instances the appearance of novel proteins detectable by Western blot analysis using a single or both antibodies (4/13 cell lines or ~30%; Fig. 1B). For example, in the case of the TOP1, SIRT1, CTNNB1, and LRP6 knockout cell lines, we observed the substitution of the canonical protein for a faster migrating novel protein detected by Western blot analysis.
Given our inability to account for the emergence of these novel proteins based on the annotated genetic alteration introduced by CRISPR/Cas9, we next examined the effects of the INDEL on mRNA splicing given that exonic sequences harbor splicing regulatory elements 8,11,12 ( Fig. 1C; Supp. Table 3). In the case of the TOP1 knockout cell line where we had observed the appearance of a novel TOP1 protein, we also witnessed the emergence of a novel mRNA species. Sequencing a cDNA-derived amplicon from the novel splice variant revealed the absence of the INDEL-containing exon suggesting the mutant protein was generated by an INDEL-induced exon exclusion event (Supp. Table 4). The truncated TOP1 protein exhibited disproportionate loss of protein in the nucleus compared with in the cytoplasm, but nevertheless retained catalytic activity ( Fig. 2A-C). The retention of catalytic activity by the truncated TOP1 protein is consistent with the designation of TOP1 as an essential gene in HAP1 cells from a gene trap mutagenesis screen that would preclude its elimination in viable cells 10 13 . In the case of the VPS35 and TLE3 cell lines, we observed changes in the splice variants harboring the CRISPR-targeted exons although no detectable novel proteins emerged (see Fig. 1C).
In contrast to the TOP1 clones, the CTNNB1 and LRP6 cell lines exhibited no detectable change in mRNA splicing associated with the targeted exons suggesting the novel proteins are a consequence of alternative translation initiation (ATI) events presumably induced by the introduced INDELs (see Fig. 1C). Consistent with this hypothesis, the mutant LRP6 protein is not glycosylated perhaps as a consequence of default expression in the cytoplasm in the absence of its N-terminal signal sequence (Supp. Fig. 1A, 1C). Similarly, the novel β-catenin protein co-migrates on SDS PAGE with an engineered β-catenin protein initiating from Met88 (Supp. Fig. 1B). In summary, in ~50% of CRISPR-edited cells acquired from a commercial source we observed unexpected changes in protein expression or mRNA splicing that challenge the notion that these reagents could be used to report the cellular effects of complete genetic ablation (Fig. 2D). Although not investigated here, conceivably the mutant proteins could also contribute to neomorphic cellular phenotypes.
ATI and pseudo-mRNAs compromise a CRISPR/Cas9 based strategy to achieve a gene knockout.
We had complemented our efforts to generate cells genetically null for various genes-ofinterest with de novo CRISPR/Cas9-based gene targeting projects. As part of our focus on the tumor suppressor kinase LKB1, we observed the emergence of unexpected protein productsboth smaller and larger proteins than the canonical protein -that were not readily explained by the presence of CRISPR-introduced INDELs (Fig. 3A-C). Given the INDELs created in LKB1 are localized to the first protein coding exon (Fig. 3D) and the antibody recognizing the C-but not the N-terminus epitope reported the shortened LKB1 protein on SDS PAGE (see Fig. 3B-C), we concluded that an ATI event induced by CRISPR/Cas9-introduced INDELs likely resulted in an LKB1 protein lacking a portion of its N terminal sequence (ATI LKB1 protein).
We also noted in MIA, but not HAP1 cells, a slower migrating protein recognized by LKB1 antibodies emerged in CRISPR/Cas9-edited clones with frameshift-inducing INDELs (see Fig.  3C; Super LKB1 protein). The appearance of Super LKB1 protein coincided with the appearance of a new mRNA splice variant that contained a 131 bp exon not included in the transcript that encodes the canonical LKB1 protein (Fig. 3E). Consistent with this exon belonging to an LKB1 pseudo-mRNA not previously annotated in MIA cells, the addition of cycloheximide (CHX) to disrupt NMD in parental MIA cells resulted in the emergence of an LKB1 splice variant that includes this exon (Supp. Fig. 2A). Thus, the same INDELs that induced a frameshift in the canonical transcript now removed a PTC from an LKB1 pseudo-mRNA and capacitated it for protein production (see Fig. 3E). We noted that HAP1 cells did not transcribe an mRNA containing this exon thus our introduction of INDELs into exon 1 did not result in the production of the Super LKB1 protein (Supp. Fig. 2B). An understanding of both the transcriptome and the pseudo-transcriptome in cells is thus critical to anticipating the net effect of frameshift-inducing INDELs introduced by CRISPR/Cas9 9 .
To understand how CRISPR/Cas9-introduced INDELs may have produced the ATI LKB1 protein, we generated cDNAs harboring each of the two INDELs that were found in the edited cells expressing these proteins (MIA cells, clone M2) in order to remove any potential contribution of altered mRNA splicing to the production of the mutant proteins (Fig. 3F). When either INDEL was introduced into the canonical LKB1 cDNA sequence, we observed the expression of a protein that co-migrated with the ATI LKB1 protein. This unexpected protein product also co-migrated with an engineered protein that initiates at methionine 51 (see Fig.  3F). We noted that a cDNA harboring the 1 bp insertion that provoked the ATI LKB1 protein was not associated with a redistribution of the initiation site suggesting leaky scanning is not likely responsible for the ATI event (Supp. Fig. 3). Given that the PTC is located 3' to the predicted ATI site, we also assume this is not a re-initiation phenomenon in which translational termination is followed by the ribosome re-launching translation at a secondary AUG codon 14 . At the same time, we also evaluated the effects of these mutations on cDNAs that encode the predicted pseudo-mRNA sequence (with the 131 bp additional exon). As anticipated, we observed the emergence of proteins that co-migrated with Super LKB1 protein given that either mutation with the additional 131 bp sequence would eradicate a PTC present in the unaltered pseudo-mRNA sequence (see Fig. 3F).
We next considered whether alternative secondary structures of the mutant mRNAs might induce this ATI at methionine 51 (Met51; Fig. 3G). Hairpin structures in the 5' untranslated region (UTR) near the 5'm7G cap have been shown to inhibit translation initiation [15][16][17] although very stable hairpins also can inhibit translation initiation farther away from the 5'm7G cap 16 . Conversely, hairpin structures 3' to AUG start codons can facilitate translation initiation, even at non-canonical start sites 18-20 . We used TurboFold II 21 , an algorithm that predicts RNA secondary structures conserved in a set of homologous RNA sequences and aligns the sequences, to model the possible RNA structures of several mutant clones that generated the Met51 initiated protein compared to those that did not (see Fig. 3G, Supp. Fig. 4). Among the sequences that produced the smaller protein, a common 3' hairpin downstream from the methionine 51 ATI site was identified at a distance that has been previously described to facilitate translation initiation 18,20 . The 3' hairpin may facilitate ATI by extending the pause time of the 40s ribosomes and allowing them to engage the Met51 start site. In the clones that do not exhibit ATI, the highly stable 5' hairpin upstream of the ATI site may cause stalling of 40S ribosomes and prevent access to the Met51 start site whereas the 80S elongation ribosome that started translation at the canonical AUG start site can resolve the 5' hairpin to continue translation of the LKB1 protein as previously described 16 . Consistent with this model, an LKB1 cDNA that includes substitutions at the same position as the INDEL failed to produce the ATI LKB1 protein and a predicted 3' hairpin structure (Supp. Fig. 5). Taken together, our data suggests that ATI could be a consequence of altered mRNA structure induced by INDELs. Given targeting the gene sequences proximal to the initiating ATG is considered a robust approach for gene elimination campaigns using CRISPR/Cas9 22 , our observation may be applicable to other CRISPR-editing projects including those that yielded the CTNNB1 and LRP6 HAP1 clones (see Fig. 1).

ATI suppresses NMD and facilitates expression of a C-terminally truncated protein from an mRNA harboring a PTC.
Despite the introduction of a frameshift-promoting INDEL in LKB1, we presumed that an ATI event, which restores codon usage to its native phase, would fail to elicit NMD during the pioneer round of translation. At the same time, having avoided destruction, the mutant mRNA is now able to support repeated rounds of translation including presumably short polypeptides initiating at the canonical start site and ending at the PTC. Given our initial Western blot analysis of the LKB1 CRISPR-edited clones did not capture low molecular proteins (see Fig. 3B,C), reexamination of LKB1 proteins in our CRISPR-edited clones indeed revealed the presence of a small LKB1 polypeptide. This protein (short LKB1) co-migrates with an engineered protein that initiates at the canonical start site but terminates at the presumed PTC introduced by the INDEL (Supp. Fig. 6A).
We compared the effects on mRNA stability of an INDEL associated with ATI with an INDEL that yielded no detectable LKB1 polypeptides (Supp. Fig. 6B-C) in order to determine if ATI suppressed NMD as a potential mechanism for promoting C-terminally truncated proteins. Comparing the levels of the two LKB1 mRNAs, we observed greater loss of the mRNA in the CRISPR-edited clone lacking any detectable ATI events (Supp. Fig. 6D). We observed little difference induced by CHX exposure in LKB1 mRNA abundance in the ATI-associated cells when compared to parental cells suggesting that NMD is not acting on the mRNA with an ATIprovoking mutation (see Supp. Fig. 6D). On the other hand, in the case of the CRISPR-edited cell line that expresses no LKB1 polypeptides, we observed a 10-fold change in LKB1 mRNA in the presence of CHX suggesting the mutant mRNA in this case is subject to robust NMD action (Supp. Fig. 6D). In total, we observed the production of three polypeptides in lieu of the canonical LKB1 protein following the introduction of a frameshift-inducing INDEL: Super LKB1, ATI LKB1, and Short LKB1 (Supp. Fig. 6E). More generally, our observations also suggest that introducing INDELs early in the transcript increases the potential for an ATI event that is able to clear off all of the splice junction complexes during the pioneer round thus enabling the synthesis of polypeptides with truncations in the C-terminal sequence.

Exon symmetry influences CRISPR/Cas9 outcomes following aberrant exon skipping.
In the analysis of our assembled HAP1 cell line panel, we also observed ~30% of the clones exhibited exclusion of the targeted exon in the mRNA. Exons are replete with splicing regulatory motifs including exon splicing enhancers and suppressors (ESEs and ESSs, respectively). These degenerate hexameric sequences dictate the extent to which exons are included within a transcript 12,23 . We suspected that exon exclusion was at least in part due to the disruption of ESEs by an INDEL event. As part of our efforts focused on studying the SUFU tumor suppressor protein, we had generated a collection of cells that presumably were null for SUFU based on Western blot analysis (Fig. 4A, B). Yet, we noted that many of these clones exhibited exclusion of the targeted exon (Fig. 4C). The extent of exon exclusion notably differs suggesting other factors, perhaps RNA structure changes that contribute to exon splicing regulation, also may be compromised by the introduction of an INDEL at this position within the SUFU mRNA. We identified a cluster of potential ESEs in the targeted SUFU exon that was likely impacted by the INDEL in these clones (Fig. 4D). No ESSs were identified in this case. To determine how reliably we can induce exon exclusion by impacting a predicted ESE, we introduced INDELs at putative ESEs found in other SUFU exons and performed similar analysis of the protein and mRNA in RMS13 cell line ( Fig. 4E-L). In every instance, we observed exon exclusion by targeted disruption of a putative ESE.
When all the clones presented so far from both commercial and de novo engineered were considered with respect to predicted impact on an ESE and exon exclusion, we observed a strong correlation between these two events ( Fig. 4M; Supp. Fig. 7). A subset of the clones exhibiting alternative splicing also expressed novel polypeptides (see TOP1 and SIRT1; see Fig. 1B). We noted in both these cases that the exons were symmetric -meaning that exon exclusion would result in a transcript lacking that sequence but nevertheless retain the original reading frame. In the case of the SUFU clones, the majority of exons skipped were asymmetric thus likely resulting in the lack of protein expression. However, we noted one targeted and skipped exon (exon 2) was symmetric yet the resulting transcript failed to generate a detectable protein perhaps due to misfolding of the mutant protein (see Fig. 4E, F). Indeed, the skipped exon encodes part of an intrinsically disordered region of the protein that is essential for interaction with members of the pro-survival BCL2 family members 24 . From these SUFU clones, we expect that decreased SUFU mRNA seen in CRISPR-edited cells was due to NMD provoked by the introduction of a frameshift inducing INDEL, or exclusion of the targeted asymmetric exon and the introduction of a PTC in an NMD-enabling position within the gene.

CRISPinatoR: a web-based guide RNA design tool that exploits ESE disruption for achieving gene elimination.
Purposeful disruption of ESEs in asymmetric exons could improve gene knockout efficiency given that even INDELs that fail to alter the coding frame would have a second opportunity for introducing a PTC by skipping the exon altogether. To systematize this strategy, we developed the CRISPinatoR, a website that identifies asymmetric exons found in a given gene and CRISPR/Cas9 guide sequences that help to deliver double-stranded breaks within proximity of a putative ESE (Fig. 5A, Supp. Fig. 8). At the same time, the portal could be used to induce the skipping of an exon harboring a deleterious mutation in order to generate a novel protein that may retain function. We note that when analyzing genome-wide CRISPR libraries, that ratio of guides targeting symmetric and asymmetric exons was fairly consistent, suggesting that these algorithms do not factor in potential gene elimination efficiency based on exon symmetry (Supp. Fig. 9A-B). Similarly, the CRISPinatoR could be used to re-evaluate previously reported phenotypes using CRISPR/Cas9 based on the potential for the sgRNA for inducing exon skipping.
Targeting RNA regulatory elements for gene knockout agendas.
We pressure-tested the ability of CRISPinatoR to design guides that induce exon skipping for either degradation of mRNA or production of novel protein-encoding mRNAs by targeting asymmetric or symmetric exons, respectively. Using the WNT receptor LRP5 as a case study, we asked the CRISPinatoR to identify sgRNAs that presumably would be able to induce exon skipping in each exon class (Fig. 5B). We identified 6 clones that harbored INDELs at the anticipated LRP5 exonic sequence by targeted sequencing of isolated genomic DNA (Fig. 5C). Using RT-PCR analysis coupled with targeted sequencing, we observed exon skipping in clones associated with both guides (Fig. 5D; Supp. Fig. 10). We observed an absence of LRP5 protein in the clone exhibiting exclusion of an asymmetric exon (Fig. 5E). However, in the clone exhibiting exclusion of a symmetric exon, we observed the appearance of a faster migrating protein (see Fig. 5E). We confirmed that this new protein retains glycosyl moieties, suggesting that its signal sequence localized to the N-terminus is intact unlike in the case of the LRP6 edited HAP1 clone ( Fig. 5F; see Supp. Fig. 1A,C). The presence of a secreted protein and evidence for skipping of the CRISPR targeted exon suggest that the novel LRP5 protein formed would harbor a compromised β-propeller domain -one of two that contributes to WNT3A binding (Fig. 5G). Indeed, we observed response of a clone expressing the truncated LRP5 protein to exogenously supplied WNT conditioned medium using a WNT pathway reporter (Fig.  5H). The weakened response compared to WT HAP1 cells likely reflects reduced total LRP5 protein levels and/or reduced WNT binding affinity with deletion of exon 16 sequence. On the other hand, the cell expressing the LRP5 mRNA excluding the CRISPR-edited asymmetric exon showed a loss of WNT pathway response consistent with the absence of LRP5 protein production from an mRNA lacking an asymmetric exon (see Fig. 5H).

DISCUSSION
The microRNA-like behavior of short interference RNAs (siRNAs) has long posed a challenge to using RNA interference (RNAi) for selective gene product ablation in both early discovery and therapeutic settings 25,26 . Whereas this issue is not inherent in DNA-editing systems such as CRISPR/Cas9, we show here that this technological advantage is offset by the unanticipated effects stemming from the on-target changes that impact the regulation of the RNA product and the translation of the protein it encodes. Exon skipping events, for example, associated with CRISPR have been previously observed although the mechanistic basis for these phenomena was not well-understood 27 . Our incomplete understanding of RNA splicing regulatory mechanisms, inability to accurately predict RNA structural changes introduced by INDELs, and limited accounting of the pseudo-transcriptome challenge our ability to anticipate transcriptional and translation outcomes as a consequence of introducing INDELs in exonic sequences (Fig.  6). Indeed, we assume these trials extend to other INDEL-producing gene editing systems that have been applied in human cells such as the CRISPR endonuclease Cpf1 28 .
A number of considerations in guide design could be installed in our design workflow to increase the fidelity of DNA sequencing information for predicting protein translational outcomes. A map of RNA-regulatory motifs (such as ESEs) that might be impacted by a CRISPR/Cas9-delivered INDEL such as that generated by the CRISPinatoR for the human genome could help in improving gene elimination or protein engineering campaigns. We acknowledge that the impact of RNA structure and possibly other determinants that can influence the function of regulatory sequences involved in RNA splicing, for example, are not accounted for by our database. At the same time, an understanding of lineage-associated pseudo-transcripts that would be edited alongside the intended target transcripts, would also help to anticipate the emergence of novel protein products such as Super LKB1 from conversion of a pseudo-mRNA to a protein-encoding mRNA.
Perhaps the most daunting challenge that we encountered from our analysis of CRISPRedited cell lines is the emergence of IRESs likely due to INDEL-induced changes in RNA structure. We anticipate that the number of ATI events associated with INDELs will be higher than what is reported here given the shortage of antibodies useful for detecting native as well as potentially truncated proteins that emerge from ATI. In this regard, the use of translation inhibitors such as CHX combined with RT-PCR could be a simple method to flag mRNAs that harbor CRISPR/Cas9-introduced frame-shift inducing INDELs yet for reasons including ATI subversion are not substrates for NMD.
Our observations also have implications for the use of INDEL-based genome editing tools for gene rescue efforts where induced exon skipping can excise sequences that harbor a mutation thus producing a viable gene 29 . These outcomes are currently achieved by using two CRISPR guides that flank a mutated exon 30-32 , or target an exon-specific splice junction using a single guide 33 . However, the ability to use a single guide targeting an ESE to achieve a similar outcome should reduce the dangers of using two CRISPR guides and expand the number of single guide options with acceptable off-target risks. In this regard, guides identified by the CRISPinatoR targeting ESEs found in symmetric exons could be used systematically identify such opportunities in genes involved in disease.

Cell Culture
HELA, MIA PaCa-2 and RMS13 cell lines were purchased from ATCC. WT and CRISPR-edited HAP1 knockout commercial cell lines were purchased from Horizon Discovery (Supp. Table 1). Puromycin was purchased from Fisher Scientific (ICN10055225). Cycloheximide was purchased from EMD Millipore (239765). NE-PER Nuclear and Cytoplasmic extraction reagent (78833) was purchased from ThermoFisher.

Western blot analysis
Cell lysates were generated with PBS/1% NP40 buffer supplemented with protease inhibitor cocktail (Sigma Cat. No. S8820). Protein sample loading buffer was added to cell lysates and proteins were separated on SDS-PAGE (BioRad Criterion TGX Precast Gel). Antibodies used for immunoblotting are listed in Supp. Table 2.
Transfection of sgRNAs 1x10 6 MIA PaCa-2 or HAP1 cells were seeded per 6 well plate and co-transfected with 0.5 µg pCas-Guide plasmid using Effectene transfection reagent (Qiagen). 24 hrs after transfection, cells were trypsinized and plated in 150 mm 2 culture dishes in various dilutions for clonal selection.

Clonal isolation of CRISPR-edited cells
Cells in 150 mm 2 culture plates were treated with 0.5 µg/ml of puromycin in order to enrich for cells expressing Cas9. Puromycin selection was maintained for 10 days after which single colonies were isolated and grown in a 96 well plate. Cells from single colonies were passaged multiple times until sufficient cells were available for analyzing genomic DNA, RNA and protein.

Genomic DNA extraction and genomic sequencing
Genomic DNA was extracted from the CRISPR-edited cells using Genomic DNA Minikit (Bioland Scientific) according to manufacturer's instructions and used as template for PCR amplification. PCR primers encompassing the CRISPR-targeted region were designed. PCR was performed with GoTaq Green Master Mix (Promega M7122) with following conditions: 98°C for 2 mins (initial denaturation), 25 cycles of 98°C for 30 secs, 56°C for 30 secs, 72°C for 30 secs (denaturation, annealing, extension) and final 70°C for 5 minutes (final extension). Gel electrophoresis in a 1.5% agarose gel was performed and the PCR products were purified from the gel using QIA Quick PCR Purification Kit (Qiagen) and cloned into pCR-TOPO plasmid using TOPO TA cloning kit for Subcloning (ThermoFisher Scientific). pCR-TOPO plasmids containing genomic DNA sequences were transformed into TOP10 competent cells and individual colonies were selected and sequenced.

RNA extraction and analysis
RNA extraction was performed using RNeasy Mini Kit (Qiagen) according to manufacturer's instructions. cDNA synthesis was performed on 1 µg of RNA using ProtoScript First Strand cDNA Synthesis Kit (Promega). Primers recognizing exons flanking the CRISPR-targeted exon (Supp. Table 3) were used to amplify the cDNA sequences isolated from the CRISPR-edited cells. PCR products were electrophoresed in a 1% agarose gel and the gel bands were isolated using QIA Quick PCR Purification Kit (Qiagen). Isolated DNA was cloned into pCR-TOPO plasmids using the TOPO TA cloning kit (ThermoFisher Scientific), and clones were sequenced at the UTSW Sequencing Core.        PTEN  TBK1  BAP1  PTEN  TBK1  BAP1   TLE3   PPM1A   TBK1   BAP1   PTEN   TLE3  PPM1A  BCL2L2  TLE3  PPM1A  BCL2L2  TLE3  PPM1A  BCL2L2   1000