WRN Helicase is a Synthetic Lethal Target in Microsatellite Unstable Cancers

Synthetic lethality, an interaction whereby the co-occurrence of two or more genetic events lead to cell death but one event alone does not, can be exploited to develop novel cancer therapeutics1. DNA repair processes represent attractive synthetic lethal targets since many cancers exhibit an impaired DNA repair pathway, which can lead these cells to become dependent on specific repair proteins2. The success of poly (ADP-ribose) polymerase 1 (PARP-1) inhibitors in homologous recombination-deficient cancers highlights the potential of this approach in clinical oncology3,4. Hypothesizing that other DNA repair defects would give rise to alternative synthetic lethal relationships, we asked if there are specific dependencies in cancers with microsatellite instability (MSI), which results from impaired DNA mismatch repair (MMR). Here we analyzed data from large-scale CRISPR/Cas9 knockout and RNA interference (RNAi) silencing screens and found that the RecQ DNA helicase WRN was selectively essential in MSI cell lines, yet dispensable in microsatellite stable (MSS) cell lines. WRN depletion induced double-strand DNA breaks and promoted apoptosis and cell cycle arrest selectively in MSI models. MSI cancer models specifically required the helicase activity, but not the exonuclease activity of WRN. These findings expose WRN as a synthetic lethal vulnerability and promising drug target in MSI cancers.


Main Text:
Defects of DNA mismatch repair (MMR) promote a hypermutable state characterized by frequent insertion/deletion mutations, which occur preferentially at nucleotide repeat regions known as microsatellites, and single-nucleotide variant (SNV) mutations 5 . This class of hypermutation, termed microsatellite instability (MSI), contributes to the development of several cancers, including 15% of colon cancers 7 , 22% of gastric cancers 8 , 20-30% of endometrial cancers 9,10 , and 12% of ovarian cancers 11 . MSI cancers can arise in the setting of germline mutations in the MMR genes MSH2, MSH6, PMS2, or MLH1, a condition known as Lynch Syndrome 5 . More commonly, MSI cancers develop following somatic inactivation of an MMR gene, typically MLH1 loss via promoter hypermethylation 5 . Recently, MSI cancers have been associated with significant responses to immune checkpoint inhibition. However, 45-60% of patients with MSI cancers do not respond to immune checkpoint blockade and the use of these agents can also be limited by their toxicity 12,13 . Hence, there remains a significant need to develop further therapies targeting MSI tumors.
We hypothesized that MMR deficiency may create unique vulnerabilities in MSI cancers. To evaluate this hypothesis and identify candidate therapeutic targets for MSI cancers, we analyzed two independent large-scale cancer dependency datasets, Project Achilles and Project DRIVE. The Project Achilles dataset consisted of 391 cancer cell lines screened with a near genome-wide CRISPR/Cas9 library and Project DRIVE interrogated 398 cancer cell lines with an RNAi library to define genes essential for proliferation and survival of individual cancer cell lines 14,15 . By comparing essential genes in cell lines with or without specific features such as MSI, these functional genomic datasets provide an opportunity to identify synthetic lethal interactions (Fig. 1a). Moreover, the large number of cell lines included in these screens helps to more robustly identify dependencies associated with distinct features of cancer and not merely idiosyncratic dependencies isolated to individual cell lines.
Before we could test this approach, we needed to differentiate cell lines with MSI from microsatellite stable (MSS) models. The clinical MSI assay is performed by a single multiplex polymerase chain reaction (PCR) and analyzed by capillary electrophoresis 16 . However, these data were not available for all screened cell lines. Instead, we utilized MSI classifications from Phase II of the Cancer Cell Line Encyclopedia (CCLE) project 17 , which analyzed nextgeneration sequencing (NGS) data to determine the total number of genetic deletions and fraction of deletions located within microsatellite regions. These analyses, which were reproduced to facilitate comparison with dependency data, identified two distinct groups that were characterized as MSI and MSS (Fig. 1b). Moreover, these MSI classifications were highly concordant with available PCR-based MSI phenotyping 18,19 and with the predicted presence of MMR loss as defined by loss of mRNA expression, deletion, and/or damaging mutations to MLH1, MSH2, MSH6, and/or PMS2. In total, 45 unique MSI and 480 unique MSS cell lines were represented by one or both datasets.
We then compared genetic dependencies between MSI and MSS cell lines in each screening dataset. The Project Achilles and Project DRIVE datasets both independently identified WRN, which encodes a RecQ DNA helicase, to be the top preferential dependency in MSI cell lines (qvalue 5.3x10 -21 and 2.6x10 -63 , respectively, Fig. 1c). These findings remained true when we evaluated the data using the PCR-based MSI classifications. In contrast, none of the four other RecQ DNA helicases were identified as a preferential dependency in MSI cell lines.
MSI is most commonly observed in colorectal, endometrial, gastric, and ovarian cancers, which were well-represented by cell lines in the screening datasets. Indeed, when these data were stratified by lineage and MSI status, MSI cell lines from these four lineages (n = 34) showed greater dependence upon WRN than their MSS counterparts (n = 130). We also identified 11 MSI cell lines from lineages where MSI is less common (4 leukemia, 2 prostate, and single models of other lineages). However, these MSI cell lines were phenotypically distinct from the MSI cell lines of the four MSI-prone lineages. These 'atypical' MSI models were both less dependent on WRN (Fig. 1d) and possessed a substantially lower density of deletion mutations in microsatellite regions, despite also possessing events predictive of MMR inactivation. These observations raised the possibility that WRN dependency is not simply a result of MMR deficiency, but it may require specific cancer lineages and/or a more profound mutator phenotype. Indeed, even within the MSI cell lines from the four MSI-predominant lineages, WRN dependency correlated with the density of microsatellite deletions (Spearman's rho = -0.54, p = 0.0012).
These data suggested that WRN is a selective dependency in MSI cancers, raising the hypothesis that WRN could be a novel drug target for this class of cancers. To assess WRN as a genetic dependency, we validated three sgRNAs targeting WRN by demonstrating depletion of the WRN protein by immunoblot (IB) (Fig. 2a). We then evaluated the effects of CRISPR/Cas9-mediated WRN knockout in 5 MSS and 5 MSI cell lines, all from the four MSIpredominant lineages, with an 8-day viability assay. The effects of WRN knockout were comparable to the positive 'pan-essential' controls in the MSI lines. By contrast, WRN silencing in MSS models approximated the effects of negative controls, targeting intergenic regions (Fig.   2b). Similarly, in a 10-day competitive growth assay, CRISPR/Cas9-mediated WRN depletion substantially impaired the viability of MSI cell lines despite negligible effects in MSS cell lines ( Fig. 2c). Complementing the CRISPR/Cas9 data, RNAi-mediated WRN silencing with short hairpin RNA (shRNA) impaired MSI, but not MSS, cell viability (Figs. 2d, 2e), consistent with our hypothesis that WRN loss is synthetic lethal with MSI.
To test whether the phenotype elicited by the sgRNAs was specifically due to inactivation of WRN, we developed an sgRNA that binds to an exon-intron junction of WRN (sgWRN-EIJ) and hence would target endogenous WRN but not exogenous WRN cDNA. Cas9-expressing KM12 cells, a MSI colorectal cancer model, were then transduced with GFP or myc-tagged WRN cDNA followed by distinct sgRNAs. Indeed, sgWRN-EIJ silenced endogenous WRN but not exogenous WRN cDNA (Fig. 2f). Correspondingly, WRN cDNA but not GFP expression rescued KM12 cell viability from sgWRN-EIJ (Fig. 2g). In contrast, sgWRN2, targeting both endogenous and exogenous WRN, equally impaired both GFP-expressing and WRN cDNA-expressing cell viability. These observations argue that the selective lethal phenotype induced by sgRNAs targeting WRN is attributable to its effects upon WRN and not an 'off-target' effect.
Together, these findings suggest that the WRN-dependent state in MSI/MMR-deficient cancers could be therapeutically exploited by targeted inhibition of WRN. The WRN protein functions as both a 3' to 5' exonuclease and 3' to 5' helicase in multiple processes including DNA repair, DNA replication, and telomere maintenance 20 . Thus, we sought to determine which enzymatic function of WRN is essential for MSI cancer cell lines. Using our ability to reintroduce WRN cDNA to rescue the effects of targeting endogenous WRN, we expressed exonuclease-dead (E84A), helicase-dead (K577M), or dually exonuclease/helicase-dead (E84A/K577M) versions of WRN cDNA 20 in KM12. We found that inactivation of the exonuclease domain did not attenuate rescue, suggesting this enzymatic function of WRN is dispensable in MSI cancers. By contrast, inactivation of the WRN helicase prevented rescue (Fig. 2g). By replicating the dependency of MSI cancers on the WRN protein with genetic inhibition of the helicase function of WRN, these results nominate the helicase domain as a candidate drug target.
We next sought to understand the cellular consequences of WRN loss, using both CRISPR/Cas9 and shRNA-mediated silencing of WRN in MSI and lineage-matched MSS models. Cell cycle analysis with EdU/DAPI staining revealed that WRN silencing reduced the proportion of MSI cells in S phase. Concomitantly, we observed an increase in cells in G1 or G2/M phases, suggesting cell cycle arrest at either the G1 or G2/M phases (Fig. 3a).  (Fig. 3c). Further investigation with immunofluorescence (IF) in the MSI models after WRN silencing demonstrated increased p53 phosphorylation at S15, a key signal for p53 activation 21 .
In contrast, we observed no substantial change in the phospho-p53 signal intensity in the MSS models after WRN depletion (Figs. 3d, 3e). After WRN depletion in the TP53 wild-type MSI cells, we also observed increased staining of cyclin-dependent kinase inhibitor p21, providing another indication of p53 activation. Conversely, we observed no substantial change in p21 signal intensity after WRN silencing in our MSS and TP53-mutant MSI cells. We then reevaluated our dependency data, stratifying by MSI and p53 status. While p53-intact MSI cell lines were more sensitive to WRN loss than their p53-impaired counterparts, this analysis demonstrated that both wild-type and mutant TP53 MSI cell lines were dependent on WRN (Fig.   3f). These data suggest that while WRN loss leads to p53 activation, p53 activity is not solely responsible for WRN dependence.
The finding of increased p53 phosphorylation at S15, a phosphorylation target of the DNAdamage response kinases ATR and ATM 21 , and subsequent p53 activation led us to hypothesize that WRN loss in MSI cancer cells may lead to DNA damage. This hypothesis is consistent with the well-reported roles of p53 and WRN in responding to DNA damage and preserving DNA integrity [22][23][24] . Indeed, biallelic germline inactivation of WRN leads to Werner Syndrome, a disease characterized by premature aging and increased cancer incidence due to impaired DNA damage repair and telomeric shortening leading to chromosomal aberrations 25,26 .
We therefore asked if DNA damage may be responsible for the impaired viability observed in WRN-depleted MSI cells. Consistent with this hypothesis, WRN silencing in MSI cells, but not MSS cells, substantially increased ɣH2AX and 53BP1 foci, markers of double-strand DNA breaks (DSB) (Figs. 4a-c). These findings were further corroborated by increased phospho-ATM (S1981) foci formation and Chk2 phosphorylation at T68, indicating a cellular DSB response known to activate p53 and anti-proliferative signaling 27 (Figs. 4a, 4d). We also observed increased ɣH2AX in MSI cells transduced with shRNA against WRN, arguing that the DSBs are not just a consequence of CRISPR/Cas9 endonuclease activity (Fig. 4e). These observations explain why p53-impaired MSI cells are still sensitive to WRN depletion since DSBs are toxic to cells, independent of p53 status 28 .
We next sought to determine which feature of MSI cells, the hypermutable state or MMR deficiency, leads to DNA damage and viability impairment upon WRN loss. We first considered the possibility that the hypermutation in MSI cells leads to recurrent mutation and inactivation of another helicase, creating a dependency upon the WRN helicase. Analogously, the second most significant dependency in MSI models from both datasets was RPL22L1, a previously defined dependency often found in MSI cancers due to frequent mutation of its paralog, RPL22 15,29 (Fig. 1c). However, we found no other helicase whose loss could account for the preferential dependency upon WRN in MSI cells. We also asked if cell lines with other hypermutable states similarly require WRN. Specifically, we evaluated cell lines with polymerase epsilon (POLE) exonuclease domain mutations that have been described to impair the POLE proofreading function. These mutations induce a markedly increased frequency of SNV mutations rather than the mixed pattern of insertion/deletion and SNV mutations associated with MMR deficiency 30,31 . Using published reports of POLE mutations implicated to cause a hypermutator phenotype, we analyzed the functional genomic datasets and identified 5 cell lines with reported proofread-impairing POLE mutations. All 5 cell lines were MSS from endometrial or colorectal lineages and insensitive to WRN loss, suggesting that hypermutability alone cannot account for WRN dependency.
We then explored whether MMR deficiency contributes the MSI cell's dependence upon WRN.
To assess this hypothesis, we compared the effects of WRN silencing upon a previously characterized model of MMR restoration. In this model, the MMR activity of the MLH1-and MSH3-mutated MSI cell line, HCT116, was functionally restored by introducing chromosomes 3 and 5 carrying a wild-type copy of MLH1 and MSH3, respectively 32,33 . WRN knockdown led to ɣH2AX accumulation and impaired the viability of both parental HCT116 and a control HCT116 cell line with an additional chromosome 2. By contrast, transfer of chromosomes 3 and 5 from normal fibroblasts suppressed ɣH2AX accumulation and partially rescued HCT116 viability from WRN depletion (Figs. 4f, 4g). These data suggest that ongoing MMR impairment contributes to the increased DSB and impaired viability observed in response to WRN silencing. One potential reason for this contextual requirement is that MMR deficiency contributes to the formation or accumulation of genomic structures that require the WRN helicase to unwind and thus prevent DNA damage. Such structures could include insertion-deletion loops 24 and/or displacement loops (D-loops) between homeologous DNA sequences 34,35 . Indeed, MMR impairment in yeast creates a dependency on Sgs1, the yeast homolog of WRN and BLM, to resolve homeologous D-loops normally rejected by an intact MMR system 36 . Beyond the potential role of WRN in preventing DNA damage, loss of WRN's capacity to participate in non-homologous end joining and/or homologous recombination (HR) could further contribute to DSB accumulation 37 .
Given the differential requirement for WRN between MSI and MSS cells, we hypothesized that WRN may have a greater role in maintaining genomic integrity following development of MSI.
WRN has been reported to respond to stress upon DNA damage by disseminating from the nucleolus towards the DNA in the nucleoplasm 38 (Fig. 3f, Fig. S1d,e) we computed aggregate WRN dependency scores for each cell line by averaging together RNAi and CRISPR dependency scores (which are both normalized so that the median score of panessential genes is set at -1) 14,43 .
Genomics Data: All genomic data used in the analysis has been published and can be found in the Cancer Cell Line Encyclopedia (CCLE) portal (https://portals.broadinstitute.org/ccle/data).
Gene expression data (taken from the file: CCLE_DepMap_18Q1_RNAseq_RPKM_20180214.gct) were transformed according to log(RPKM + 0.001). Gene-level relative copy number data were derived from a combination of whole-exome sequencing (WES) and SNP data to achieve maximal coverage across cell lines.
When both data types were available for a given cell line, we prioritized WES over SNP data.
Relative copy number data were also log-transformed for analysis. For mutation data, we utilized the merged mutation calls (file: CCLE_DepMap_18Q1_maf_20180207.txt), which combines information from multiple data sources and types. 1B, S1A), we averaged these values across the data sources available for each cell line after normalizing for systematic differences between data sources. In particular, we used linear regression models to estimate and remove scale and offset differences between data sources so that the normalized number of deletions (and number of deletions in MS regions) measured in each data source was equal on average. mRNA-sequencing: Cas9-expressing cells (KM12, SW48, and OVK18) were lentivirally transduced with the following sgRNAs: sgCh2-2, sgWRN2, and sgWRN3 (sequences provided below). Cells were selected with puromycin to ensure delivery of sgRNAs and RNA was purified 72 hours after transduction. This was performed in duplicate prior to cDNA library preparation and subsequent RNA-sequencing via the Illumina NextSeq 500.
We first excluded genes which had less than 1 counts per million in more than half of the samples. The weighted trimmed mean of M-values 48 method was used to normalize the library size of each sample, using the calcNormFactors function from the R package: edgeR 49  IF for WRN and fibrillarin was performed as previously described 38 . 1. Genome-wide loss-of-function screening identifies genes synthetic lethal with MSI. a, Schematic of dependency dataset analysis. Cell lines were grouped by feature and dependency scores for each gene were analyzed to identify feature-specific genetic dependencies. b, Screened cell lines were plotted by the deletion density and fraction of deletions occurring in microsatellite (MS) regions. MSI classification by next generation sequencing (NGS) and multiplex polymerase chain reaction (PCR) are indicated. c, False discovery rate (FDR) q-values were plotted against the mean difference of dependency Z scores between MSI and MSS cell lines for both Project Achilles CRISPR/Cas9 and Project DRIVE RNAi datasets. d, WRN dependency scores were plotted by cancer lineage and further subclassified by MSI and MSS status. Gene dependency scores are normalized to the control sgRNAs for each cell line. A value 0 represents the median dependency score of negative control sgRNAs and -1 represents the median dependency score of sgRNAs targeting panessential control genes. The lower and upper hinges correspond to first and third quartiles (25 th and 75 th percentiles). The upper and lower whiskers extend to the largest value within 1.5*IQR (inter-quartile range) from the hinge.  a, Cell cycle evaluation of representative Cas9-expressing MSI and MSS cell lines after lentiviral transduction with the indicated sgRNAs. b, Annexin V/propidium iodide (PI) staining evaluating early apoptosis and late apoptosis/non-apoptotic cell death in representative Cas9-expressing MSI and MSS cell lines 7 days after delivery of indicated sgRNAs. c, GSEA enrichment and depletion scores in WRN-depleted OVK18 cells plotted against WRN-depleted SW48 cells. Signature enrichment plots for the indicated Hallmark gene sets are shown for OVK18 and SW48 cells after WRN depletion. d, phospho-p53 (S15) immunofluorescence (IF) of indicated Cas9-expressing cell lines treated with indicated sgRNAs. Scale bar = 50 µm. e, Quantification of nuclear phospho-p53 (S15) staining intensity per cell. f, WRN dependency was evaluated for cell lines classified as MSS, MSI from an infrequent MSI lineage, or MSI from an MSIpredominant lineage and further subclassified by p53 status. (*) p < 0.05; (**) p < 0.005; (ns) p ≥ 0.05 by two-tailed Student's t-test. Hinges and whiskers as per Fig. 1d. a, IB of ɣH2AX, phospho(T86)-and total-Chk2, WRN, and GAPDH in representative cell lines after CRISPR/Cas9-mediated WRN depletion. b, ɣH2AX IF of representative cell lines after delivery of indicated sgRNAs. Scale bar = 50 µm. c, Quantification of nuclear ɣH2AX staining intensity per cell. d, phospho-ATM (S1981) IF of representative cell lines after delivery of indicated sgRNAs. Scale bar = 50 µm. e, IB of ɣH2AX, phospho(T86)-and total-Chk2, and GAPDH after RNAi-mediated WRN knockdown. f, IB of ɣH2AX, WRN, MLH1, MSH3, GAPDH in HCT116 with or without MMR restoration after lentiviral transduction of indicated shRNAs. g, Relative viability of HCT116 cells with and without MMR restoration 7 days after lentiviral delivery of the indicated shRNAs. Values are presented as means ± SE (n = 6). (*) p < 0.01; (**) p < 0.001; (***) p < 0.0002 by two-tailed Student's t-test. h, WRN IF of representative cell lines. Scale bar = 20 µm. i, Quantitative analysis of WRN colocalization to the nucleolar marker, fibrillarin, by Pearson's colocalization coefficients. (*) p < 0.001; (**) p < 0.02 by two-tailed Student's t-test.