Ubiquitin-mediated DNA damage response is synthetic lethal with G-quadruplex stabilizer CX-5461

CX-5461 is a G-quadruplex (G4) ligand currently in trials with initial indications of clinical activity in cancers with defects in homologous recombination repair. To identify more genetic defects that could sensitize tumors to CX-5461, we tested synthetic lethality for 480 DNA repair and genome maintenance genes to CX-5461, pyridostatin (PDS), a structurally unrelated G4-specific stabilizer, and BMH-21, which binds GC-rich DNA but not G4 structures. We identified multiple members of HRD, Fanconi Anemia pathways, and POLQ, a polymerase with a helicase domain important for G4 structure resolution. Significant synthetic lethality was observed with UBE2N and RNF168, key members of the DNA damage response associated ubiquitin signaling pathway. Loss-of-function of RNF168 and UBE2N resulted in significantly lower cell survival in the presence of CX-5461 and PDS but not BMH-21. RNF168 recruitment and histone ubiquitination increased with CX-5461 treatment, and nuclear ubiquitination response frequently co-localized with G4 structures. Pharmacological inhibition of UBE2N acted synergistically with CX-5461. In conclusion, we have uncovered novel genetic vulnerabilities to CX-5461 with potential significance for patient selection in future clinical trials.


Results
Loss of function screen of core DNA repair factors with G4 ligands. In order to identify new DNA repair-associated genetic interactions with CX-5461 with high sensitivity, a pooled targeted single guide RNA (sgRNA) library was generated (Fig. 1a). This targeted library was comprised of 480 genes corresponding to DNA damage sensing, DNA repair, DNA replication, cell cycle checkpoints, chromosome segregation, and chromosome structure and remodeling as well as a set of positive control or essential genes (Table 1 and Supplementary Data) [40][41][42][43] . Since CX-5461 also inhibits RNA polymerase I complex-mediated transcription of rRNA, we included POLR1A, POLR1B, UBTF, RRN3, and TAF1A genes to represent gene expression (transcription) pathways. The library contained a total of 2930 sgRNAs with 6 sgRNAs targeting each gene and 50 non-targeting sgRNAs (NT-sgRNAs) (Supplementary Data). The quality of the pooled library was determined by next-generation sequencing which revealed that more than 99% of sgRNAs were present in the library with less than an eightfold difference in the representation of 80% of them (Supplementary information Fig. 1a) and by functional assessment of known essential genes (methods).
To characterize the factors for which loss confers sensitivity to G4 drugs in cancer cells, we performed dropout screens in HCT116 cells with various drugs (Fig. 1b). To control for chemical scaffold-specific off-target effects, cells were treated with two G4 drugs of completely different structural classes, pyridostatin (PDS) and CX-5461, at two pre-determined inhibitory concentrations (IC) for each drug (IC 50 , the half-maximal inhibitory concentration and IC 30 ). Since CX-5461 (but not PDS) also inhibits RNA polymerase I at high concentrations 27 , cells were also treated in parallel with BMH-21, an RNA polymerase I inhibitor, which binds GC rich DNA but is not a DNA G4 stabilizer at IC 30 , to identify and exclude gene-drug fitness effects resulting from RNA polymerase I inhibition. We have previously shown that BMH-21 does not induce DNA damage in epithelial cells and despite inhibiting RNA polymerase I it does not exhibit synthetic lethality with HR deficiency. The drug treatments were started in triplicates on day 5 after transduction and continued for a total of 14 days. Cells were also harvested intermittently at days 7 and 10 of the treatment, and sgRNA representation analysis was done as described for the dropout screens (Supplementary information Fig. 1).
By consolidating the most depleted genes from different treatment groups of CX-5461 and PDS, and excluding those that were shared with the top hits from the BMH21 screen, we identified the genetic vulnerabilities specific to the G4 interacting drugs. First, for each treatment condition, the effects were relatively ranked using negative depletion scores and P-values calculated using MAGeCK at day 19 relative to vehicle (Supplementary Data). The MAGeCK depletion scores represented the RRA (robust rank aggregation) values of the individual genes in negative selection, taking into account the performance of the 6 individual sgRNAs per gene. Top depleted genes for each condition, ranked by the depletion scores, were then consolidated together to form a final list comprising of 81 genes ( Table 2). The depletion scores for each treatment condition were ordered according to the scores of CX-5461 treatment group at IC 50 (Fig. 1c). Apart from PDS-IC 50 , many of the top depleted genes were shared between CX-5461 and PDS, but not BMH-21 (Fig. 1d.) Candidates of interest included genes that were exclusively shared between different CX-5461 and PDS treatment groups (Fig. 1d). Hits in common with the BMH-21 treatment group, including RAD9A and RAD51, were excluded from future analysis as they likely represent genes not involved purely in G4 stabilization associated repair. We thus identified 58 genes including those known to be involved in G4 resolution, such as BRCA2 and REV1. As expected, multiple members of the HR pathway such as RAD54L, RAD52, PALB2, RAD51B exhibited negative fitness with G4 ligands. We also observed genes whose homologs were previously identified in screens performed in C. elegans (ATM, MUS81, POLQ) 27 . Gene hits common to all CX-5461 and PDS treatment conditions included RAD54L, H2AFX, POLQ, ATM, LIG4, and RNF168.  www.nature.com/scientificreports/ Next, we searched for cancer-related alterations in our top candidate genes using the cBio-Portal for Cancer Genomics database 44,45 . Most of these genes are mutated in many types of cancers (Supplementary information Fig. 3). Although all cancer types with loss of function or decreased gene expression of these genes could potentially benefit from CX-5461 therapy, we decided to focus on genes that could also serve as therapeutic targets. We searched for the druggability of the top hits from the screen using the Drug Gene Interaction Database (DGIdb) and found four genes that were predicted to interact with drugs: ATM, ATR, UBE2N, and PSMB4. The ATM/ ATR pathway has been shown to be activated by CX-5461 in acute lymphoblastic leukemia cells 28 . ATR inhibition enhances CX-5461-associated apoptosis in these cells. UBE2N, along with RNF168, is involved in DNA damage response-related ubiquitin signaling, and this pathway is frequently altered in cancers (Supplementary information Fig. 3). These alterations are accompanied by an associated increase (in case of amplification) or decrease (in case of deletion) in mRNA expression.  www.nature.com/scientificreports/ Taken together our screen identified known genes and pathways involved in G4 stabilization as anticipated from previous mechanistic studies, confirming the biological sensitivity of the screen, as well as several additional pathway members, including notably POLQ and two key members of the ubiquitin DNA damage sensing pathway, which we further followed up.
Validation of POLQ in mediating G4 stabilizer sensitivity. To evaluate the spectrum of drug sensitivity mediated by POLQ-deficiency as well as to validate the screen results, new sgRNAs were designed. As an improvement from the GeCKO v2 design, the new designs aimed to both reduce off-target effects as well as increase sgRNA stability and activity [46][47][48] . Sensitivities to G4 stabilizers were determined by treating WT and POLQ-null HCT116 cells with varying concentrations of CX-5461, PDS, and BMH-21 for 7 days. The cell viabilities were then assessed using the WST-1 assay. Dose-response curves were plotted to determine IC 50 values. POLQ suppression by sgRNAs significantly decreased the IC 50 values of CX-5461 from 30.27 nM to an average of 11.97 nM across the 5 new sgRNAs (ANOVA followed by Dunnett's multiple comparisons tests; P-value ≤0.001; Supplementary information Fig. 4a). In a similar manner, IC 50 values of PDS were significantly decreased from 10.78 to 1.77 µM (ANOVA followed by Dunnett's multiple comparisons tests; p ≤0.001; Supplementary information Fig. 4b). In comparison, there was no difference in the sensitivity of POLQ-null cells to BMH-21 (ANOVA followed by Dunnett's multiple comparisons tests; p >0.05), further supporting the fact that POLQ was mediating drug sensitivities specific to G4 stabilizers (Supplementary information Fig. 4c). In a parallel experiment, POLQ-targeting siRNAs were transfected into HCT116 cells one day prior to drug treatments. Cells treated with siRNAs then underwent the same 7-day drug treatments to assess drug sensitivities. Similar to the sgRNA experiments, siRNA-mediated POLQ suppression decreased the IC 50 Fig. 2a). While we observed differences in drug sensitivity, we sought to confirm that the sgRNAs and siRNAs were efficiently suppressing POLQ. For each drug dose response experiment with either sgRNA or siRNA treatment described above, we determined the gene expression levels by extracting the RNAs and performing qRT-PCR on the cDNAs. The relative gene expression of POLQ was assessed with respect to non-targeting controls (non-targeting sgRNA or siRNA), using GAPDH expression as endogenous controls. In both sgRNA and siRNA treatments, POLQ expression was suppressed down to 20% (Supplementary information Fig. 5a, b). The fact that sgRNA and siRNA yielded similar POLQ expression levels may not be surprising. The advantage of CRISPR/Cas9-mediated suppression lies in the permanent modification to genomic DNA, whereas siRNA-mediated suppression is always transient. However, studies have shown that in experiments spanning a short time frame, siRNA-mediated suppression is not necessarily weaker 49,50 . This could be attributed to the fact that in shorter time frames, for sgRNA-mediated disruption in the DNA, pre-existing mRNAs are continuously translated until they are turned over or depleted 51 .
To assess whether POLQ is also important in mediating G4 stabilizer sensitivities in other cell backgrounds, we performed the same drug dose response experiments in U2OS cells, and isogenic mouse embryonic fibroblasts or MEFs for the following genotypes: wild-type (WT), KU70 −/− , POLQ −/− , and double knockout or DKO for KU70 and POLQ. For U2OS cells, suppression of POLQ by siRNA was confirmed by qRT-PCR (Supplementary information Fig. 5c). In U2OS cells, the IC 50 of CX-5461 was significantly decreased from 112.2 to 51.8 nM (ANOVA followed by Dunnett's multiple comparisons tests; p ≤0.001; Supplementary information Fig. 6a).
Similarly, IC 50 of PDS was significantly decreased from 7.51 to 4.23 µM (ANOVA followed by Dunnett's multiple comparisons tests; p ≤0.001; Supplementary information Fig. 6b). Again, sensitivities of POLQ-null cells to BMH-21 were unchanged (Supplementary information Fig. 6c). For MEFs, following the same drug dose response experiments, we determined that KU70 and POLQ single or double knockout cells were sensitized to both CX-5461 and PDS, but not BMH-21 (Fig. 2b). Compared to 18.43 nM in WT, the IC 50 of CX-5461 in Ku70-, POLQ-, and double-knockout cells were 5.00 nM, 6.85 nM, and 6.15 nM, respectively (ANOVA followed by Dunnett's multiple comparisons tests; p ≤0.05; Fig. 2b). Compared to 30.76 µM in WT, the IC 50 of PDS in Ku70-, POLQ-, and double-knockout cells were 8.90 µM, 7.54 µM, and 5.94 µM, respectively (ANOVA followed by Dunnett's multiple comparisons tests; n=3.; p ≤0.05; Fig. 2b). Our previous paper uncovered the increased sensitivity to CX-5461 in DNA-PK-knockout and LIG4-knockout cells 27 , which are two other members in the NHEJ pathway. The sensitivity of KU70-null cells to G4 ligands further supports the importance of NHEJ pathway in repairing CX-5461-induced DNA damage. As for POLQ, increased CX-5461 sensitivity after POLQ suppression was observed in four different cell backgrounds: C. elegans 27 , HCT116, U2OS, and MEF; these results suggest that POLQ-defective cells are indeed sensitive to G4 ligands.
Re-validation of UBE2N, RNF168, MMS22L and MUS81 in growth competition assays. In order to further re-validate the results of the drug dropout screens, we selected four genes: UBE2N and RNF168 representing the DNA damage response-associated ubiquitin signaling pathway, MUS81 which is a DNA structure-specific endonuclease, and MMS22L which localizes to replication forks, as MMS22L-TONSL heterodimer, especially during replication stress 52 , to be examined individually using competitive growth assays (CGA) (Supplementary information Fig. 7). Growth competition assays provide a quantitative and robust method for determining fitness effects of drug-gene interactions. We also included an NT-sgRNA, NT5. Individual sgRNAs included those used in the library as well as new sgRNAs targeting the genes of interest (design described in methods). HCT116 cells were transduced with three individual sgRNAs representing each gene. Transduced Interestingly, in further experiments, we did not observe any drug-related sensitized phenotype in a MUS81deficient HCT116 cell line (Supplementary information Fig. 11), indicating that MUS81 might be a false positive or not have a strong drug interaction. Its also possible that cells adapt rapidly to Mus81 loss and stable knockouts already have escape mechanisms. Altogether, our findings were in agreement with our screen results with the strongest re-validated gene-drug effect being associated with UBE2N and RNF168 gene targeting.
Selective sensitivity to G4-stabilizing drugs in multiple cancer cell lines upon UBE2N depletion. Next, we characterized the dose kinetics of CX-5461, PDS and BMH-21 in human cancer cell lines upon UBE2N depletion using WST-1 cell proliferation assay. Utilizing a previously characterized UBE2N knockout (UBE2N −/− ) HCT116 cell line 53 , we observed that UBE2N −/− cells were more sensitive to CX-5461 as compared to the UBE2N +/+ cells (IC 50 Table 3).
Additionally, CRISPR sgRNA targeting of UBE2N in breast cancer cell line MDA-MB-231, using three different sgRNAs, revealed similar sensitivities to CX-5461 and PDS ( Fig. 2d and Table 3). UBE2N-targeted cells exhibited selective sensitivity, as compared to cells containing a non-targeting sgRNA (sgNT40), to CX-5461 (IC 50   Drug-dose response of HCT116 cells transfected with individual POLQ-targeting siRNAs, followed by 7 day treatment with CX-5461, PDS, or BMH-21. Drug-dose response curves are shown on the left. Barplots on the right show the calculated IC50 values of each cell genotype (mean + 95% CI), compared to non-targeting siRNAs. ***p ≤ 0.001, analyzed by GraphPad Prism with One-way ANOVA followed by Dunnett's multiple comparisons test, with a single pooled variance; n = 3. (b) Validation of PolQ sensitivity in MEF knockout cell lines. Drug-dose response of isogenic knockout MEF cells undergoing 7 day drug treatments with CX-5461, PDS, or BMH-21. Drug-dose response curves are shown on the left. Barplots on the right show the calculated IC50 values of each cell genotype (mean + 95% CI), compared to other genotypes. *p ≤ 0.05, analyzed by GraphPad Prism with One-way ANOVA followed by Dunnett's multiple comparisons test, with a single pooled variance; n = 3. (c) Validation of UBE2N sensitivity in HCT116 knockout cell lines. HCT116 UBE2N-wildtype (+ / +) and -null cells (− / −) were treated with serial dilutions of CX-5461, PDS, and BMH-21 for a week followed by WST-1 assay. (d) MDA-MB-231 cells were transduced with lentivirus targeting the UBE2N gene or a non-targeting control (NT40) and treated with serial dilutions of CX-5461, PDS and BMH-21 for one week followed by WST-1 assay. The results are representative of at least two independent experiments, and three technical replicates for each experiment. (e) HCT116 cells were transduced with individual sgRNAs including a non-targeting control (NT5), three targeting UBE2N, two for RNF168, three for MUS81, one for MMS22L. Transduced and non-transduced mixed cell populations were subjected to different drug treatments for one week at indicated doses and flow analysis was performed to determine the fraction of cells expressing red fluorescence. The mean fold change relative to vehicle treatment is shown for each mixed population (bar) along with the standard error of the mean (SEM). The results are representative of at least two independent experiments, and three technical replicates for each experiment. Bootstrap analysis was performed (details in Supplementary information) and bootstrapped confidence intervals for drug effect after gene knockout are shown in Supplementary information    Fig. 14). All these results indicate that RNF168 is likely to be involved directly in the repair of CX-5461-induced DNA damage. Next, we investigated nuclear ubiquitination since UBE2N and RNF168 promote mono-and poly-ubiquitination at sites of DNA DSBs. To determine if nuclear ubiquitin signaling is triggered in response to CX-5461-induced DNA damage, we examined ubiquitin conjugates at nuclear DNA DSBs using an antibody specific for mono-and poly-conjugated ubiquitin (FK2) and on chromatin, using immunofluorescence and Western blotting, respectively. Nuclear FK2 foci were observed in approximately 39.93 % ± 4.29 % and 48.47 % ± 3.94 % of U2OS cell nuclei after 0.1 µM CX-5461 treatment for 1 and 4 h, respectively, as compared to 6.29 % ± 3.29 % of vehicle-treated cells (Fig. 3b). A significant increase in the percentage of cells with nuclear FK2 foci, albeit at a lower proportion than after 0.1 µM CX-5461, was also observed with 1 µM CX-5461. Approximately, 44.77 % ± 12.95 % and 25.42 % ± 5.41 % of drug-treated cells demonstrated an increase in nuclear FK2 foci after 1 and 4 h, respectively, as compared to the vehicle treatment. An increase in the percentage of cells with FK2 foci was also observed with PDS treatment (Supplementary information Fig. 13a), suggesting that the DNA damage ubiquitination response is induced by multiple G4-stabilizers of different structural classes.
We further analyzed RNF168/UBE2N-dependent ubiquitination at different time points on acid-extracted histones from CX-5461-treated HCT116 cells by Western blots using FK2 antibody (Fig. 4a, Supplementary  Fig. 15a, c). In contrast with the focus forming assay, the high variation in Western blot signals between contrasts precluded observations of statistically significant differences when comparing the whole lane intensity of FK2 after CX-5461 treatment in UBE2N +/+ cells, versus the vehicle treatment ( Fig. 4a and Supplementary Fig. 15d), in spite of an apparent trend (1.5-1.9 fold) towards an increase in FK2 intensity (see bootstrap power analysis, methods). Histone ubiquitination was suppressed in the absence of UBE2N after 3 and 5 h of treatment at 1 µM and 0.1 µM, indicating that UBE2N is required for ubiquitination events resulting from CX-5461-induced DNA damage (Fig. 4a, Supplementary information Fig. 15c). FK2 signal intensity in Western blots decreased dramatically after treatment with USP2cc-the catalytic domain of a deubiquitinase, indicating that the FK2 signal in Western Blot is ubiquitination specific (Fig. 4b).
To determine if conjugated ubiquitin was indeed accumulating at the sites of DNA damage following CX-5461 exposure, we examined colocalization of FK2 foci with well-characterized markers of DNA DSBs, the phosphorylated histone H2AX (γH2AX) and 53BP1. γH2AX is also important in the chromatin remodeling events during DNA repair. Additionally, RNF168-mediated ubiquitination of histone H2A at K15 is crucial for 53BP1 recruitment to DNA damage sites 54 . To this end, we quantified cells with co-localized FK2 and γH2AX or 53BP1 foci in U2OS cells treated with CX-5461 (0.1 µM) for 1 h (Supplementary information Fig. 13b). Significantly higher number of γH2AX and 53BP1 foci were detected in cells exposed to CX-5461 (data not shown) than the background DNA damage foci in vehicle-treated cells, as previously descibed 27 . Over 80% of FK2 foci also co-stained with γH2AX after CX-5461 treatment, as compared to 25% after vehicle treatment (p <0.01) (Supplementary information Fig. 13b). The co-localizatin of FK2 foci and 53BP1 foci is fairly high even in vehicletreated cells, implying that 53BP1 marked endogenous DNA damage loci are also modified by ubiquitination. Because of the high co-localization level in vehicle treated condition, we did not observe significantly different FK2 and 53BP1 co-localization events between CX-5461 and vehicle exposures (Supplementary information  The graph on the right shows the overlap of BG4 and FK2 foci as either percent of total FK2 foci that are also BG4 + (BG4 + FK2 foci) or percent of total BG4 foci that are also FK2 + (FK2 + BG4 foci  Fig. 13b). However, the median increase of 63.59% in FK2 foci that co-stain with 53BP1 in CX-5461-treated cells, as compared to 43% in vehicle-treated cells, may still be biologically relevant. These findings demonstrate that CX-5461 mediated ubiquitination occurs at chromosome loci marked with DNA damage foci. We previously demonstrated that CX-5461-induced DNA damage is replication dependent 27 . To quantify the number of S-phase cells with FK2 foci, we employed 5-ethynyl-2′-deoxyuridine (EdU) incorporation into DNA during active DNA synthesis in U2OS cells. We observed that after 1 h of treatment with 0.1 µM CX-5461, almost all cells (99%) with FK2 foci were EdU-positive (Fig. 3c), suggesting that these cells were mostly in S-phase.
We next assessed whether conjugated ubiquitin foci formation in response to CX-5461 exposure occurs at chromosomal G4 loci (Fig. 3d). By using an engineered, G4 structure-specific antibody BG4 55 in U2OS cells, we observed colocalization of FK2 and BG4 foci in 31.8% of CX-5461-treated cells, as compared to 13.65% of vehicle-treated cells 27 . These findings suggest that conjugated ubiquitin accumulates at not only DNA damage sites but also at G4 structures upon exposure to G4 ligands.

UBE2N-dependent K63-linked chromatin ubiquitination in response to CX-5461.
Having established that exposure to CX-5461 triggers DDR-associated UBE2N/RNF168-mediated chromatin ubiquitination, we next examined the chromatin modifications underlying the resolution of CX-5461-induced DNA damage. UBE2N and RNF8 catalyze the formation of K63-linked ubiquitin chains on linker histone H1 flanking DNA DSBS which then recruits RNF168 to the site of damage 53 . We analyzed the induction of endogenous K63-linked ubiquitination (K63Ub) after CX-5461 treatment in chromatin fractions isolated from HCT116 cells using Western blots (Fig. 4a, middle panel, Supplementary information Fig. 15b, c, and Supplementary Fig. 12). An increase in K63-linked ubiquitination was observed 2 h after drug treatment. Loss of UBE2N reduced the accumulation of K63-linked ubiquitination (Fig. 4a, Supplementary information Fig. 15c, d). Treatment with USP2cc also greatly reduced the signal intensity of K63 Ub in Western blots, indicating that the K63 Ub signal is ubiquitination specific (Fig. 4b) To extend these observations, we additionally used U2OS cells with a stably integrated doxycyclin-inducible, green fluorescent protein (GFP)-tagged tandem ubiquitin-binding entity (K63-Super UIM) which has been shown to bind specifically to K63 linkages only 53 (Fig. 4c, d). Indeed, we observed increased GFP-K63-Super-UIM foci in 47.71 % ± 19.8 % nuclei after CX-5461 treatment as compared to 13.68 % ± 4.47 % nuclei in the vehicle treatment. However, because of data variation, the difference was not statistically significant (p = 0.056, two-tailed t-test). Next, utilizing an anti-ubiquitin antibody specifically targeting K63 ubiquitination in immunofluorescence assay, we also discovered that the number of k63Ub foci per cell greatly increased after CX-5461 treatment (p <0.001, Tukey multiple comparisons of means with 95% family-wise confidence level, Fig. 4e, f). A high percentage of overlap between K63Ub foci and DNA damage marker 53BP1 foci was observed 52.29 % ± 4.0 % (Fig. 4g), suggesting that the CX-5461 treatment-induced DNA damage is associated with UBE2N-RNF8-mediated K63-linked ubiquitination.

Increased sensitivity to CX-5461 in response to a small-molecule inhibitor of UBE2N.
Although several small-molecule covalent inhibitors of UBE2N have been developed [56][57][58] , NSC697923 has been shown to specifically inhibit UBE2N activity via the covalent modification of the conserved active site cysteine residue critical for the catalytic activity 59 . This small-molecule inhibitor exhibits a cytotoxic effect on diffuse large B-cell lymphoma (DLBCL), neuroblastoma and malig nant melanoma cells [60][61][62] . The effect of simultaneous combinations of CX-5461 and NSC697923 was examined in HCT116 cells using a three-dimensional dose-response surface method 63,64 . To test a large range of drug doses, a diagonal constant ratio combination design using equipotency ratios (e.g., CX-5461 IC 50 at 7.4 nM and NSC697923 IC 50 at 166.1 nM) was employed so that the effect of each drug would be equal in the combination 65 . The combination of the two drugs showed synergistic response across a range of combinations in HCT116 cells treated for 4 days, with a peak at 1.9 nM of CX-5461 and 166.1 nM of NSC697923 (Fig. 5a, Supplementary Data). In addition, a strong synergy volume of 124.8 nM 2 % at 95% CI) was observed (Fig. 5b, right panel). On the other hand, the overall antagonism volume was insignificant at − 1.62 nM 2 %.

Discussion
We have previously shown that CX5461 and other G4 binders cause selective cytotoxicity in HR-deficient cancers in part by induction of DNA breaks which cannot be repaired. This molecule is currently in early phase clinical trials to determine the therapeutic effect in patients with HR repair defects, with preliminary signs of clinical activity 39 . However the full spectrum of DNA repair deficiencies which could sensitize tumours to CX5461 and other G4 binders is not yet known. To fully saturate for genetic synthetic lethality we conducted a targeted CRISPR-Cas9 dropout screen contrasting G4 drugs, CX-5461 and PDS, with loss of function in 480 core genes involved in genome maintenance, DNA repair and replication. The inclusion of counter-screen compound BMH-21, an RNApol1 inhibitor but non-G4 binder, permitted prioritization for follow-up of hits that do not involve inhibition of RNA polymerase I, which is a known off-target effect of CX-5461 at high concentrations 27 . The sensitivity of the targeted screen was confirmed by re-identification of lethal G4 drug interaction with BRCA2, REV1, POLQ, ATM and ATR [25][26][27]32,66 . Among the G4 binder sensitizers we noticed convergent hits in UBE2N and RNF168, co-functional members of the histone ubiquitination DNA damage response pathway, which have not been previously reported in connection with G4 binders. The mechanism underlying selective elimination of tumor cells with compromised UBE2N-and RNF168-mediated ubiquitination, upon exposure to G4 ligands, was further characterized. The importance of UBE2N-, RNF8-and RNF168-mediated ubiquitin signaling and post-translational histone ubiquitination in DNA DSB repair is known. UBE2N, an E2 ubiquitin-conjugating enzyme, and RNF8 and RNF168, E3 ubiquitin ligases, are recruited to DNA DSBs [67][68][69][70] . They trigger a non-proteolytic ubiquitination cascade which contributes to DNA repair and repair pathway choice by recruiting additional factors to the DSB sites. UBE2N acts together with one of its two non-catalytic variants, UBE2V1 or UBE2V2, previously known as UEV1 and MMS2, respectively 71,72 . However, UBE2V1 only participates in cytoplasmic ubiquitination. UBE2N-UBE2V2 heterodimer binds RNF8 which facilitates the formation of K63-linked ubiquitin chains on H1-type linker histones providing a binding platform for RNF168 via UDM1 domain 53,67,73-75 .  www.nature.com/scientificreports/ RNF168 then catalyzes the ubiquitination of H2A-type histones at K13/K15, leading to recruitment of downstream DNA repair factors, including 53BP1 which promotes NHEJ 54,76,77 . However, during S/G2 phase, RAP80 recruitment to the extended K63-linked ubiquitin chains and loss of G1-specific RIF1-dependent suppression of HR leads to recruitment of BRCA1 complex and ultimately HR 68,76,77 . We were unable to achieve sufficient knockout/knockdown of RNF8 despite multiple guide designs in different cell lines to confirm the expected role, given the involvement of RNF168 and UBE2N.
The contribution of regulatory chromatin ubiquitination to G4 structure resolution and to the repair of G4-associated DNA damage has not been investigated. To explore the underlying mechanism of lethal gene-drug interactions, we examined activation of nuclear ubiquitin signaling and subsequent chromatin ubiquitination upon exposure to G4 drugs. We showed that CX-5461 treatment resulted in significant induction of RNF168 foci in nuclei. Additionally, we observed the accumulation of conjugated ubiquitin (FK2) at DNA DSBs and of K63-linked ubiquitin chains within chromatin after CX-5461 treatment. Importantly, a significant number of CX-5461-induced FK2 foci overlapped with G4 sites marked by BG4 antibody, strongly suggesting that chromatin ubiquitination is induced in response to DNA DSBs that is associated with G4 structures stabilized by CX-5461.
RNF168 has been suggested to link H2A ubiquitination to PALB2-and RAD51-dependent HR by directly interacting with PALB2 and loading PALB2 onto damaged DNA which is redundant with BRCA1 function during HR [78][79][80][81] . Interestingly, PALB2 was one of the hits in our screens although we did not explore its role in detail. RNF168 deficiency predisposes BRCA1 heterozygous mice to cancers, and these cells are hypersensitive to PARP inhibitors 82 . Forced targeting of PALB2 to damaged chromatin bypasses RNF168 requirement for genome maintenance in BRCA1 haploinsufficiency. Previously, we showed that some BRCA1-mutated tumors that are resistant to PARP inhibitors and platinum salts are sensitive to CX-5461 27 . Inhibition of chromatin ubiquitin pathway may further increase the sensitivity of these tumors to CX-5461.
Other than regulating histone ubiquitination in DNA DSB repair, UBE2N is also involved in post-replication repair (PRR) dealing with blocks to replication fork progression 83 . Interestingly, we identified PCNA from our PDS screen, which is also a member of the PRR pathway. Likely, CX-5461-stabilized G4 structures impede DNA replication and activate PRR. PCNA mono-ubiquitination promotes TLS polymerase to bypass replication blocks with errors 84 , while UBE2N mediated poly-ubiquitination of PCNA directs an error-free lesion bypass. The detailed mechanism of how UBE2N is involved in bypassing stabilized G4 structures during replication needs further investigation.
In addition to ubiquitin mediated DDR, we validated loss of function in POLQ as synthetic lethal with CX-5461 treatment. POLQ is not essential for normal cell survival, but is over-expressed in about 70% of breast cancers regardless of HR status 85,86 . Previous studies with non-medicinal structure G4 binders have hinted at a role for polQ/TMEJ in G4 mediated damage responses, here we show that the clinical stage molecule reflects this activity. The combination of POLQ inhibitors and G4 binders could thus be an attractive treatment strategy to complement PARP inhibitor treatment. POLQ mediated MMEJ pathway has been recognized as an important pathway for DSB repair. Recently, the role of POLQ in replication stress and fork reversal has been explored. Our results suggest that POLQ might also be involved in replication bypass through G4 regions in the genome.
Recently a genome-wide shRNA screen using PDS and PhenDC3, which are not drug like molecules but are both G4 ligands, identified several genes that when depleted enhanced cell death with these G4 ligands 87 . In spite of the differences between shRNA and sgRNA/CRISPR loss of function screens, we also identified some of the genes described in the shRNA screens including BRCA2, PALB2 and POLQ. Other overlapping genes included WEE1, PCNA and PSMC4, which although significant hits within a single treatment condition with PDS or CX-5461 in our screens, were nonetheless excluded from our final priority list because of the stringent requirement to be common to at least two treatment conditions. The non-overlapping findings from these two independent studies highlights the importance of using multiple approaches to identify a comprehensive list of gene interactions.
Another recent paper analyzed CX-5461 response in C. elegans 88 . A chemical genetic screen was performed in C. elegans and homologs of human genes that are sensitive to CX-5461 were identified, but different drug sensitivity profiles between C. elegans and human were discovered. For example, NHEJ deficiency is sensitive to CX-5461 in human cells but not in C. elegans, indicating distinct drug response to G4 stabilizers in different species.
Collectively, through focused screening and re-validation, we have identified additional genetic vulnerabilities to G4 stabilizers in human cells, including POLQ as well as genes involved in DNA replication, DNA damage repair, checkpoint, and DDR-associated ubiquitination. Our results suggest that G4 stabilizer drugs could be effective for cancers with a broad range of genome instability and these results potentially expand biomarker driven patient inclusion and combination therapy with G4 stabilizer drugs.

Methods
Plasmids. The lentiCRISPRv2 was a gift from Feng Zhang (Addgene plasmid #52961) 89 . A fluorophorecontaining variant of lentiCRISPRv2 was generated by replacing the puromycin resistance marker with the DsRed-Express2 (DRX2) fluorescence marker (LCV2-DRX2). Gel extraction with the QIAEXII Gel Extraction Kit (Qiagen) was performed on the linearized 13 kb lentiCRISPRv2 vector following a BamHI and PmeI restriction digest to remove the puromycin marker, along with downstream WPRE and UTR elements. The DRX2 insert was prepared by PCR-amplifying a 1.6 kb oligo fragment (gBlock, IDT) which included DRX2 sequence as well as the lost WPRE and UTR sequences. Joining of the DRX2 insert and the cut vector was performed using NEBuilder HiFi DNA Assembly (NEB), as per manufacturer's instructions. After transforming the cloning reaction into Stellar chemically competent cells (Clontech) and plating on Amp resistant LB Agar plates, the final Drugs. CX-5461 (Selleckchem, cat# S2684) and BMH-21 (eMolecules) were prepared in 50 mM NaH 2 PO 4 , as described previously 27 . Pyridostatin trifluoroacetate (Sigma-Aldrich, cat# SML0678) was prepared in water, NSC697923 (Sigma-Aldrich, cat# SML0618) in DMSO, and puromycin (ThermoFisher, cat# A1113803) in water.
Sub-genomic library sgRNA oligonucleotide design and synthesis. 20mer oligonucleotide sequences representing sgRNAs were selected from Human GeCKO v2 pooled library 89 . sgRNA designs missing from the GeCKO library for DNA2, TERF2, DBF4, and TRRAP genes were added 92 . DNA oligonucleotides were synthesized as a pool using a B3 synthesizer (CustomArrayInc) platform, PCR amplified using ArrayF and ArrayR primers and Phusion HS Flex (NEB) and size selected using 1% agarose gel as described previously 89,93 . Pooled oligos were cloned into lentiCRISPR v2-puromycin plasmid, digested with BsmbI, at 42:1 insert-to-vector ratio, using Gibson assembly 93  Sequencing library preparation and next generation sequencing. 1  Library performance during screening. To assess library performance we conducted gene dropout screens without any G4 drugs selection to characterize the following: (1) optimal duration of puromycin selection, (2) changes in the abundance of NT-sgRNAs, (3) essential gene depletion over time, and (4) optimal length for drug screens. To this end, we performed two independent and longitudinal screens in HCT116 cells (Supplementary information Fig. 1b). Cells were transduced at a low MOI of 0.3 and subjected to puromycin selection for 3 or 6 days. A fraction of cells was harvested every three days until day 19 for sgRNA representation analysis by next-generation sequencing. sgRNA abundance and statistical analyses were performed using CaRpools and MAGeCK 94,96 .
When compared with sgRNA abundance on day 1 (pre-puromycin selection), we did not observe any remarkable difference in gene depletion on day 19 between populations treated with puromycin for 3 days or 6-days (Supplementary information Fig. 1c; Pearson R = 0.87), thus prompting us to apply only 3 days of puromycin selection for the drug screens. In addition, the abundance of NT-sgRNAs did not change as compared to the initial pool irrespective of the duration of puromycin selection (Supplementary information Fig. 1c), and in independent dropout screens (Supplementary information Fig. 1d). In comparison, gene depletion (P-value <0.05) was observed after 3 days of puromycin selection (Supplementary information Fig. 1e). By day 19, over 10% of the genes for cell populations undergoing 3 days of puromycin selection showed significant dropout with reference to either day 1 (pre-selection) (Supplementary information Fig. 1e) or day 4 (post-selection) (data not shown). The top depleted genes included POLR2I, GAPDH and ANAPC5. Of the top 10% of depleted genes in our screens, almost 92% of the genes were fitness genes, with 38 out of 48 genes previously described as core fitness genes and 6 as HCT116 cell line-dependent fitness genes (Supplementary Table 1) 97 .
Together, these results show that the subgenomic dropout screens were able to detect essential gene dropout, with high sensitivity and reproducibility, without a significant perturbation in the NT-sgRNA pools, thus providing a platform to perform focused functional genetic screens with drugs.
During comparative screening with different drugs we assessed the performance of individual sgRNAs in the top depleted genes of interest (Supplementary information Fig. 2). It was evident that not all sgRNAs targeting the same gene were depleted in a similar manner. As an example, for LIG4, we observed that two of the sgRNAs led to the enrichment of the respective cell populations across all treatment groups, including BMH-21. MAGeCK assessed the relative enrichment or depletion of individual sgRNAs. For LIG4, MAGeCK determined that 2 of the sgRNAs were enriched, with a positive fold-change, while the other 4 were depleted. The inefficient cutting of sgRNAs has been cited by other groups [98][99][100] , illustrating the importance of methods such as MAGeCK to first measure the performance of individual sgRNAs before deriving a gene-based ranking.
Subcloning, lentivirus production and titration for individual sgRNAs. Individual sgRNAs were designed using the Deskgen Cloud CRISPR Design Software (Desktop Genetics). sgRNA cloning into LCV2-DRX2 vector was performed as previously described 89,93 . Transformation was performed by adding 2 µl ligated product into 50 µl Stellar Chemically Competent Cells (Takara) as per manufacturer's instructions. Up to 100 µl of transformed cells in S.O.C. Medium (ThermoFisher) were plated onto LB Agar plates containing 100 mg/ ml Ampicillin and grown overnight at 37 °C. The following afternoon, multiple colonies from each plate of cloned sgRNA plasmids were inoculated and grown overnight at 37 °C in 1 ml LB broth with 100 mg/ml Ampicillin. Plasmid DNA was extracted from 800 µl bacteria culture using Qiagen Plasmid Miniprep Kit (Qiagen) or Monarch Plasmid Miniprep Kit (NEB), as per manufacturer's instructions. Plasmid samples were sent for Sanger sequencing validation, and the correct clones were expanded, by growing the remaining 200 µl bacteria culture in 200 ml LB Broth overnight, at 37 °C with shaking. Plasmid DNAs were then extracted using the PureLink HiPure Plasmid Maxiprep Kit (ThermoFisher). Lentivirus was generated as described above. To titrate lentiviruses, cells of interest were plated at 50,000-200,000 cells per well into a 24-well plate. Starting from 10 µl, fivefold dilutions of the virus were added into 250 µl of media (up to 8 dilutions). After 24 h, media were replaced with fresh media without virus. Seventy-two hours after viral transduction, flow cytometry analyses were performed using FlowJo to determine the percentage of cells expressing DRX2. Virus titration curves were generated by plotting the fluorophore-expressing fraction against the dilution factor. From the data points on the exponential part of the curve, viral titer was calculated by using (% positive cells × # cells per well × 1000/ volume of virus per well).
Competitive growth assays (CGA). LCV2-DRX2 vectors with individual sgRNAs were produced as described above. HCT116 cells were transduced in 24-well plates at a density of 150,000 cells per well with the virus at an MOI of 0.1-0.3. Three days after transduction, cells expressing DRX2 were flow sorted on an Aria Fusion or Aria III and mixed with non-transduced cells. First, the ratio of transduced to non-transduced cells was determined to allow the untreated mixtures to contain at least 40-50% transduced cells at the end of the www.nature.com/scientificreports/ assay by monitoring the proportion of DRX2+ cells 4 and 7 days after transduction using a Fortessa flow cytometer. The sgRNA MMS22L.2 was excluded from the final analysis because of the low proportion of DRX2+ cells in the mixed cell population. Growth-adjusted cell mixtures were split and plated in triplicate wells to be treated with drugs or vehicle as in the drug screens. Cells were harvested 7 days after drug treatment and analyzed by flow analysis to determine the fraction of cells expressing DRX2. The ratio of the mean percentage of DRX2+ cells in drug-treated and vehicle-treated cell populations was calculated to determine the relative fold-change in the mixed cell populations for each sgRNA of interest. Results were analyzed using R. Each experiment was repeated at least twice with three technical replicates per experiment.
Statistical analysis for CGA . Bootstrapping was performed in a blocked fashion, to reflect the conditions of the experiment. For an experiment involving N a assays, N a assay numbers were randomly selected with replacement (outer bootstrap block). Within the N a randomly selected assays, N ? replicate values were then randomly selected with replacement (inner bootstrap block). The resultant data set was of the same form as the original, with the same number of assays and replicate observations within each assay. 400,000 blocked bootstrap replicate data sets were constructed and the average proportion of cells surviving calculated in the logistic space. After transforming the averages back to the raw proportion space, differences between vehicle and drug conditions were calculated. From the distribution of the 400,000 bootstrapped differences, confidence intervals were determined.

Immunoblotting. For chromatin ubiquitination experiments, cells were resuspended in Triton Extraction
Buffer (TEB) at a concentration of 10 7 cells/mL and lysed on ice for 10 min on a horizontal shaker at 90 rpm. Cells were then washed in TEB and resuspended in 0.2N HCl at 4 × 10 7 cells/mL before extracting histones overnight at 4 °C. After centrifugation, the supernatant was saved and neutralized with 1/10th the volume of 2M NaOH. For each sample, 10 6 cells were boiled in 1X Laemmli buffer and 2-mercaptoethanol at a 1:10 ratio for 10 min at 90 °C before loading in each well of SDS-PAGE. For all other Western blots, cells were boiled in 1X Laemmli buffer for 10 min at 90 °C and protein was quantified using the Pierce 660 nm Protein Assay (ThermoFisher, cat#22660). 10 µg protein was loaded in each well of a 12% SDS-PAGE and transferred to nitrocellulose or PDMS membrane at 100 V for 1.5 h or at 30 V overnight. Primary antibodies were incubated at the concentrations listed in Table 3 as per manufacturer's instructions. After incubating with the appropriate secondary antibodies (Goat anti-mouse, Dako cat#P0447 and Abcam cat#ab97040; goat anti-rabbit, Abcam cat#ab6721; mouse anti-goat, Santa Cruz Biotechnology cat#sc-2354) at 1:10000 or 1:5000 in 5% non-fat dried milk-TBST for 1 h, membranes were visualized with Immobilon Western Chemiluminescent HRP Substrate (MilliporeSigma, cat#WBKL20500) and imaged with the ImageQuant LAS 4000 (GE Healthcare) using the ImageQuant TL software.
Statistical analysis for FK2 and K63Ub immunoblots. Increasing levels of H3 were associated with decreasing levels of FK2 density measurements (Total FK2 (whole lane)), so H3 did not appear appropriate as a loading control. Total H2A measured via 2 bands at 15 kDa and 2 bands measured at 30 kDa (presumably dimers of H2A) provided a loading control positively associated with FK2 total. With two data points per experimental condition, independent linear fits within condition were not possible as all fits would have been degenerate, leaving no degrees of freedom for error estimation. Thus the simplifying assumption of equal slopes within all treatment conditions was made. Initial analysis on the raw western blot photo density data yielded estimates with confidence intervals (CI) extending to negative density values. Transforming the photo density values via base 2 logarithms removed the lower bound of zero, yielding an unconstrained data range from minus to plus infinity. Regression model fit diagnostics showed reasonable Gaussian behaviour for the log2 transformed data, with estimates and CIs well removed from data boundaries. Treatment condition differences in the log realm correspond to fold-change estimates back in the raw scale, facilitating interpretation of differences as compared to differences expressed in arbitrary densitometry values. With limited data available, basic linear model fits were obtained. Insufficient data was available to perform e.g. linear mixed effects models. All conditions were estimated as fixed effects in an ANCOVA model (FK2 total ∼ Total.H2A + Genotype * Treatment) with total H2A providing the ANCOVA covariate to allow for loading control adjustment. The model fit to the log2 transformed data yielded the following ANOVA table: Analysis of Variance Significance codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 .. 1 The positive association of log2(FK2 total) with log2(Total H2A) showed a P-value of 0.00052, indicating that the ANCOVA adjustment was important and necessary. The confidence intervals in Fig. 4b show that ubiquitination levels in the CRISPR knockout UBE2N −/− line were generally lower than those in the wild type line, and this is reflected by the Genotype P-value 0.017 in the ANOVA table. The differences associated with differing treatment conditions show wide CIs, all crossing the null hypothesis value for no treatment difference (Fold-change = 1.0 on the raw scale, or zero difference in intercept values on the log2 scale). Thus there is no statistical evidence of a difference in ubiquitination levels due to treatment conditions relative to the wild type vehicle control condition. The same statistical analysis was applied to quantify the K63-linked ubiquitination signal in Western blots.
Immunofluorescence. For RNF168 foci, U2OS cells were treated with CSK buffer (100 mM NaCl, 300 mM sucrose, 3 mM MgCl, 10 mM PIPES (pH 6.8), with proteinase inhibitors) for 4 min followed by fixation with 2% paraformaldehyde in TBS (50 mM Tris-HCl, pH7.5, 150 mM NaCl) for 20 min, and methanol for 1 min. After blocking in 3% BSA and 0.2% Tween-20 in TBS, RNF168 antibody (Table 3) was incubated overnight followed by incubation with secondary antibody for 1 h. For K63Ub foci staining, U2OS cells were treated with CSK buffer with supplements (DTT 1mM, proteinase inhibitors, phosphotase inhibitors and deubiquitination inhibitor NEM 10 mM) for 5 min on ice, then washed with TBS with supplements for 2 times, before fixation with 2% paraformaldehyde for 20 min. Cells were further treated with cold methanol for 1 min. After blocking in 3% BSA and 0.2% Tween-20 in TBS for 1 h, K63Ub antibody (Sigma 05-1308, clone Apu3) was incubated at 100 times dilution overnight followed by incubation with secondary antibody at 1000 dilution for 1 h. For colocalization, after K63 staining with primary and secondary antibodies, 53BP1 staining was performed as previously described 27 . Immunofluorescence of BG4, 53BP1 and γ-H2AX was described previously 27 . To visualize FK2 foci, U2OS cells grown on glass cover slides were fixed with cold methanol for 10 min and then with acetone for 30 s. Cells were permeabilized with 0.5% Triton X-100 for 25 min, then blocked for 1 h with 3% BSA and 0.2% Tween-20 in TBS. FK2 antibody (Table 3) was incubated at 1:10,000 dilution overnight. Mouse Alexa 488 secondary antibody was diluted at 1:5000 and incubated for 1 h. Because of background foci, damage foci positive cells were defined as γ-H2AX foci ≥ 5, 53BP1 foci ≥ 3, RNF168 foci ≥ 3, FK2 foci ≥ 10. At least 100 cells were counted for each condition, and at least two independent experiments were performed. P-value was determined by two-tailed unpaired t-test for the difference of means. For FK2 foci analysis in case of CX-5461 1 µM, Welch's two-tailed test was used to determine significance because of unequal variances. DNA damage foci per cell and foci overlap were counted by Spotcounting software developed by Steven Poon, which is publicly available at github: https:// github. com/ shahc ompbio/ SpotC ounti ngApp. For K63-linked ubiquitination, 30,000 U2OS cells with GFP-K63-Super-UIM were grown on coverslips and induced by doxycycline as described previously 53 . Drugs or vehicle were added 24 h after induction and cells were fixed in 4% paraformaldehyde for 15 min followed by permeabilization with 0.2% Triton X-100. Coverslips were mounted on glass slides with DAPI and images were obtained using an upright Colibri LED microscope. All images were taken using the same exposure time. Cell quantification was performed using ImageJ. For each condition, at least 100 cells were examined with a threshold of 10 or more foci per cell. Two independent experiments were performed. P-value was determined by two-tailed unpaired t-test for the difference of means.