Werner syndrome helicase is a selective vulnerability of microsatellite instability-high tumor cells

Targeted cancer therapy is based on exploiting selective dependencies of tumor cells. By leveraging recent functional screening data of cancer cell lines we identify Werner syndrome helicase (WRN) as a novel specific vulnerability of microsatellite instability-high (MSI-H) cancer cells. MSI, caused by defective mismatch repair (MMR), occurs frequently in colorectal, endometrial and gastric cancers. We demonstrate that WRN inactivation selectively impairs the viability of MSI-H but not microsatellite stable (MSS) colorectal and endometrial cancer cell lines. In MSI-H cells, WRN loss results in severe genome integrity defects. ATP-binding deficient variants of WRN fail to rescue the viability phenotype of WRN-depleted MSI-H cancer cells. Reconstitution and depletion studies indicate that WRN dependence is not attributable to acute loss of MMR gene function but might arise during sustained MMR-deficiency. Our study suggests that pharmacological inhibition of WRN helicase function represents an opportunity to develop a novel targeted therapy for MSI-H cancers.


Introduction
Defects in components of the DNA repair machinery, such as BRCA1/2 mutations or impaired DNA mismatch repair (MMR), are a common characteristic of tumor cells, accelerating the accumulation of DNA mutations or chromosomal aberrations that are required for neoplastic growth and transformation (Kinzler and Vogelstein, 1997). Plasticity of genome stability pathways permits tumor cells to tolerate the loss of individual DNA repair genes and leads to synthetic lethality (SL) upon targeting the compensating repair mechanism (Nickoloff et al., 2017). The first clinically approved drugs exploiting such a SL interaction are Poly(ADP-Ribose) Polymerase (PARP) inhibitors for therapy of BRCA1/BRCA2-deficient tumors (Kaufman et al., 2015;Lord and Ashworth, 2017).
MMR deficiency is caused by inactivation of genes of the DNA repair machinery involved in the resolution of nucleotide base-base mismatches during DNA replication (Jiricny, 2006;Kunkel and Erie, 2015). MMR defects lead to characteristic variations in the length of tandem nucleotide repeats across the genome, known as microsatellite instability (MSI) (Ellegren, 2004). Germline mutations in MMR genes, most commonly MLH1, MSH2, MSH6 and PMS2, are causative for Lynch syndrome, a cancer predisposition condition associated with increased lifetime risk to develop colorectal cancer (CRC) or other tumor types including endometrial and gastric carcinoma (Hampel et al., 2005;Lynch and Krush, 1971;Lynch et al., 2015). In sporadic, nonhereditary CRC, MSI is frequently observed due to epigenetic silencing of MLH1 (Cunningham et al., 1998;Herman et al., 1998;Kane et al., 1997;Kuismanen et al., 2000). MSI-high (MSI-H) tumors display a hypermutator phenotype (Cancer Genome Atlas Network, 2012), which entails increased immunogenicity, amendable to therapy with immune checkpoint inhibitors (Le et al., 2015). However, targeted therapies directly exploiting the MMR-deficient status of tumor cells do not exist.
Werner syndrome helicase (WRN) is a member of the RecQ DNA helicase subfamily (Croteau et al., 2014;Yu et al., 1996). RecQ helicases are involved in multiple DNA processing steps including DNA replication, double-strand break repair, transcription and telomere maintenance and are therefore considered to serve as 'genome caretakers' (Chu and Hickson, 2009;Croteau et al., 2014). The critical function of this protein family in genome maintenance is underscored by the fact that defects in three of the five family members -WRN, Bloom Syndrome RecQ Like Helicase (BLM) and RecQ Like Helicase 4 (RECQL4) -give rise to human disease syndromes associated with developmental defects and cancer predisposition (Brosh, 2013;Oshima et al., 2017). Specifically, patients with Werner syndrome display a premature ageing phenotype including arteriosclerosis, type II diabetes and osteoporosis and are prone to develop tumors of mesenchymal origin, such as soft tissue sarcoma or osteosarcoma (Goto et al., 2013;Hickson, 2003;Lauper et al., 2013). WRN is unique among RecQ family helicases in possessing 3'À5' exonuclease activity (Huang et al., 1998;Kamath-Loeb et al., 1998;Shen et al., 1998).
In contrast to the previously described tumor-suppressive role of WRN, we demonstrate in this study that WRN possesses a context-dependent critical pro-survival function for cancer cells. By leveraging a recently defined map of cancer cell specific vulnerabilities (McDonald et al., 2017) and a comprehensive molecular characterization of cancer cell models (Barretina et al., 2012;Streit et al., 2019) we identify WRN helicase as a selective dependency in MSI-H cancer cell lines.

Results
WRN dependency is associated with MSI-H status of cancer cells WRN was identified as a potential selective dependency in a subset of 398 cancer cell models in a recent pooled shRNA viability screen covering approximately 8000 genes (Project DRIVE) (https:// oncologynibr.shinyapps.io/drive/; McDonald et al., 2017). A genomic or expression-based biomarker predictive for WRN dependency was unknown. Depletion of WRN exclusively affects viability of a subset of CRC, gastric and endometrial cancer cell models reflected by RSA (redundant siRNA activity) sensitivity scores À3, indicative of cell essentiality ( Figure 1A). Intriguingly, CRC, gastric and endometrial cancers are the three human malignancies with the highest frequency of MSI-H status (Cortes-Ciriano et al., 2017). This raised the possibility that WRN represents a selective dependency in MSI-H cell lines.
In order to explore this hypothesis we developed a Random Forest model using an MSI feature list defined by Boland and Goel (Boland and Goel, 2010). This model classifies WRN sensitive and insensitive cell lines with an accuracy of 0.89 and a recall rate for sensitive lines of 0.69. Importantly, no true insensitive cell lines are classified as sensitive (Figure 1-figure supplement 1A). An analysis of variable importance revealed MLH1 expression as the feature most highly associated with the classification outcome, in line with the frequent inactivation of the MLH1 gene in MSI-H CRC (Cunningham et al., 1998;Herman et al., 1998;Kane et al., 1997;Kuismanen et al., 2000) (Figure 1-figure supplement 1A). Consistently, WRN dependency anti-correlates with MLH1 mRNA expression levels among the cell models used in Project DRIVE (Figure 1-figure supplement 1B; p=1.02*10 À4 , Fisher's exact test, stratification of MLH1-low and -high expressing cell models according to median MLH1 expression [TPM 37.44]).
Next, we wanted to experimentally validate the MSI status in a select set of cell lines. To this end, we used a fluorescent PCR-based analysis of five mononucleotide microsatellite markers to determine the MSS/MSI-H status of a subset of CRC, gastric and endometrial cancer cell models (Supplementary file 1). In addition, we utilized a comprehensive MSS/MSI-H status annotation of CRC cell models reported by Medico and colleagues (Medico et al., 2015). Analysis of gene dependency and MSS/MSI-H status data revealed that WRN dependency was strongly associated with MSI-H status across CRC, gastric and endometrial models (p=4.91*10 À8 , Fisher's exact test). Of the 18 cell lines classified as MSI-H, 15 cell lines (83%) were sensitive to WRN depletion using an RSA value of À3 to define WRN dependency ( Figure 1B). In contrast, WRN is dispensable for viability in all MSS cell models ( Figure 1B). Our analysis suggests that MSI-H status is a strong predictor for WRN sensitivity of cancer cells.

WRN depletion by siRNA selectively impairs viability of MSI-H CRC and endometrial cancer cell lines
To experimentally corroborate the WRN dependency of MSI-H cancer cells, we applied short-interfering RNA (siRNA)-mediated knock-down of WRN in a panel of three MSS (SK-CO-1, CaCo-2, SW480) and three MSI-H (HCT 116, RKO, SNU-C4) CRC cell lines. In agreement with the results from Project DRIVE, WRN depletion using a mixture of four siRNA duplexes (Pool) or an individual siRNA (#1) targeting WRN profoundly affected viability in MSI-H, but not in MSS CRC models ( Figure 2A). In contrast, depletion of the known essential mitotic kinase PLK1, had a detrimental effect on viability of both MSS and MSI-H cell lines. Efficient depletion of WRN protein following siRNA transfection was confirmed by immunoblotting ( Figure 2A). The selective dependency on WRN was mirrored in colony formation assays with two MSS (LS1034, SK-CO-1) and two MSI-H cell lines (HCT 116,RKO)   Similar to MSS cancer models, non-transformed telomerase-immortalized human retinal pigment epithelial cells (hTERT RPE-1) did not display sensitivity to knock-down of WRN (Figure 2-figure supplement 1B). It is noteworthy, that the depletion of the related RecQ helicase BLM significantly impaired viability of hTERT RPE-1, but not HCT 116 cells. These RNAi experiments demonstrate that depletion of WRN abrogates viability in MSI-H but not MSS or non-transformed cells. To assess a potential mechanism to bypass WRN dependence in MSI-H cancer cells, we tested whether codepletion of p53 and WRN in the TP53-wild-type MSI-H CRC line HCT 116 would reverse the sensitivity to WRN knock-down. However, WRN/p53 co-depletion exacerbated the reduction in viability compared to WRN knock-down alone ( To determine whether reconstitution of MLH1 and MSH3 in the MSI-H CRC line HCT 116 would revert WRN dependence, we utilized variants of HCT 116 harboring copies of human chromosome 2 (+ch2, control line), chromosome 3 (+ch3, MLH1 reconstitution) or chromosome 3 and 5 (+ch3+ch5, MLH1 and MSH3 reconstitution) derived from normal fibroblasts (Haugen et al., 2008;Koi et al., 1994). In both HCT 116 +ch3 and HCT 116 +ch3+ch5 cell lines, we noticed only a modest increase of viability upon WRN knock-down compared to the parental cells or the HCT 116 +ch2 control line ( In a reciprocal approach, we monitored whether individual or combined knock-down of MLH1 and MSH3 would elicit differential effects on the viability of WRN-proficient versus WRN-knock-out cell models. To this end, using CRISPR-Cas9 we generated two WRN knock-out SW480 monoclonal lines (WRN KO #1 and #2) and a non-edited, WRN-proficient SW480 monoclonal control line (parental). Loss of WRN protein in the two knock-out lines was confirmed using immunoblot assays (

CRISPR-Cas9-mediated inactivation of WRN confirms the selective dependency of MSI-H CRC models on WRN
We carried out CRISPR-Cas9 depletion assays in MSS and MSI-H CRC models to independently confirm the selective WRN dependencies observed in shRNA/siRNA studies. Cell lines were stably transduced with Cas9 followed by transduction of lentiviral particles co-expressing GFP and single guide RNAs (sgRNAs) targeting WRN or the essential replication factor RPA3 ( Figure 3A). To investigate the relevance of the different protein domains in WRN by CRISPR scanning (Shi et al., 2015),   domain specific sgRNAs were used to target the exonuclease, helicase, RecQ helicase family DNAbinding (RQC) and Helicase and RNase D C-terminal (HRDC) domains and the C-terminal helix-turnhelix (HTH) motif. Negative selection of sgRNA expressing cells was monitored over 14 days via flow cytometry-based quantification of GFP expressing cells, and normalized to the effect of the RPA3 positive control sgRNA ( Figure 3B). We did not observe depletion of cells harboring WRN targeting sgRNAs in the MSS CRC cell line HT-29. In contrast, MSI-H HCT 116 cells expressing WRN sgRNAs depleted to a similar level as observed for the RPA3 positive control. sgRNAs directed against the exonuclease, helicase and RQC domains were most effective, implying a functional or structural requirement of both these domains in the context of WRN dependency. Interestingly, strong depletion effects were also observed for sgRNAs targeting the C-terminal HTH motif ( Figure 3B). A similar WRN sgRNA depletion pattern was observed in the MSI-H CRC cell line RKO, while we found far less pronounced depletion effects in the MSS CRC model SK-CO-1 (Figure 3-figure supplement 1). In agreement with the RNAi studies, our CRISPR-Cas9 experiments suggest that WRN provides an essential gene function in two MSI-H CRC cell lines but not in two MSS CRC lines.

WRN dependency in MSI-H CRC is linked to its helicase function
To further dissect the relevance of WRN exonuclease and helicase function in WRN-dependent cell models, we generated FLAG-tagged, siRNA-resistant WRN (WRNr) expression constructs harboring loss-of-function mutations within the exonuclease-(E84A, Nuclease-dead) and helicase-domain (K577M, ATP-binding deficient), or both domains (E84A/K577M, Double-mutant) (Gray et al., 1997;Huang et al., 1998) ( Figure 4A). Wild-type and mutant forms of WRNr were transduced in HCT 116 cells and monoclonal lines with matched stable WRNr expression were generated. Expression and nuclear accumulation of transgenic WRNr variants among the cell lines was confirmed using immunofluorescence analysis ( Figure 4B). Immunoblotting revealed that transgenic WRNr wild-type as well as the mutant WRN proteins were expressed at levels higher than the endogenous WRN counterpart ( Figure 4B). Two wild-type WRNr expressing clones were selected based on the respective high and low expression of the transgene in order to cover the range of mutant WRNr variant expression observed in the selected panel of clones. As expected, viability of empty vector control-transduced cells was strongly reduced upon depletion of WRN ( Figure 4C). Importantly, both the high and low expression level of wild-type WRNr was sufficient to render HCT 116 cells inert to knock-down of endogenous WRN ( Figure 4C). This demonstrates the on-target effect of the WRN siRNA duplex and indicates that the transgene-mediated rescue of WRN function is not protein level sensitive. Exogenous expression of the nuclease-dead form of WRNr almost completely rescued the effect of endogenous WRN depletion. In stark contrast, although expressed at similar or higher levels than WRNr wild-type, both the ATP-binding deficient and double-mutant form of WRNr were unable to restore viability following depletion of endogenous WRN ( Figure 4C). In RKO cells expressing WRNr variants, we observed a slightly stronger dependency on WRN helicase ATP-binding function compared to exonuclease activity upon knock-down of endogenous WRN (

Loss of WRN causes mitotic defects and nuclear abnormalities in MSI-H cells
In order to investigate the cellular basis for the viability reduction of MSI-H cancer cells upon WRN depletion, we monitored the consequences of WRN loss-of-function using immunofluorescence analysis of the nuclear membrane protein LAP2b and Hoechst DNA staining in MSS and MSI-H CRC cell lines. SW480 did not display any phenotypic differences upon transfection with NTC and WRN-targeting siRNAs ( Figure 5A). Strikingly, in HCT 116 and RKO cells we observed formation of chromatin bridges and micronuclei upon WRN knock-down, both potential consequences of failed sister genome partitioning during mitosis ( Figure 5A). Enlarged nuclei, indicative of failed mitosis, were additionally observed upon WRN knock-down in HCT 116 cells. Quantification of the frequency of chromatin bridges and micronuclei revealed that WRN depletion did not affect baseline levels of these aberrant nuclear morphologies in the MSS CRC cancer models SK-CO-1 and SW480, while the   Figure 4 continued on next page frequency of both chromatin bridges and micronuclei was strongly increased upon WRN knockdown in the MSI-H CRC cell lines HCT 116 and RKO ( Figure 5B and C). In non-transformed hTERT RPE-1 cells, WRN depletion led to a slight increase of chromatin bridge and micronucleus formation, although far less pronounced compared to the MSI-H CRC models ( Figure 5B and C). Supportive of the immunofluorescence studies, live cell imaging revealed a strong increase of the incidence of lagging chromosomes and chromosome bridges during mitosis in WRN-depleted MSI-H CRC cell lines HCT 116 and RKO, but not in MSS SW480 cells ( Figure 6). We conclude that WRN depletion in MSI-H cells results in nuclear morphology and integrity aberrations that are manifested during cell division. The correlation of these defects with the observed cell viability reduction in MSI-H models suggests that nuclear abnormalities are causally linked to the anti-proliferative effect of WRN loss.

Structural chromosome aberrations in MSI-H cells after WRN loss-offunction
The observed nuclear integrity and mitotic defects caused by WRN depletion in MSI-H cells could be the consequence of preceding genome maintenance aberrations. This hypothesis is reinforced by the important role of RECQ family helicases, including WRN, in genome integrity (Chu and Hickson, 2009 Figure 7A). Quantification of the number of chromatin breaks per karyotype demonstrated a strong increase in the fraction of cells harboring 1-5 or >5 chromosome breaks in the two MSI-H CRC models upon WRN depletion ( Figure 7B). In the MSS CRC model SW480 and hTERT RPE-1 non-transformed cells only a minor increase of karyotypes with chromosome breaks was detected ( Figure 7B).
To test whether signs of DNA damage can already be detected in WRN-depleted interphase cells, we analyzed the intensity of nuclear g-H2AX signals. We observed a strong induction of g-H2AX upon siRNA-mediated depletion of WRN in MSI-H, but not MSS, CRC cancer cell lines (Figure 7-figure supplement 1). Strikingly, in both HCT 116 and RKO MSI-H CRC lines, WRN depletion led to an increase in nuclear g-H2AX signal similar to or higher than upon treatment of cells with the DNA double-strand break-inducing agent etoposide (Figure 7-figure supplement 1A and B). Importantly, nuclear g-H2AX levels in non-transformed and MSS hTERT RPE-1 cells were strongly increased by treatment with etoposide, while loss of WRN had no effect in this model (Figure 7figure supplement 1B).
Our analyses suggest that WRN helicase is essential for maintaining genome integrity in MSI-H cells by preventing chromosome breaks and erroneous chromosome fusions. The observed mitotic chromosome aberrations in WRN-depleted MSI-H cells can also explain the aforementioned nuclear morphology and mitotic defects, including micronuclei, lagging chromosomes and chromatin bridges. The detection of DNA damage markers in interphase nuclei of WRN-depleted MSI-H cells suggests that the aforementioned mitotic and post-mitotic aberrations originate from genome integrity defects in interphase. The correlation of chromosome aberrations and nuclear abnormalities with MSI-H status following loss of WRN function indicate that genome integrity defects are responsible for the profound reduction in viability of MSI-H cancer cells.

Discussion
Treatment paradigms for MSI-H tumors have recently shifted with the approval of the immune checkpoint agents pembrolizumab, nivolumab and ipilimumab, targeting programmed cell death 1 (PD-1) and cytotoxic T-lymphocyte-associated Protein 4 (CTLA-4) in this patient segment (Le et al., 2017;Le et al., 2015;Overman et al., 2018). Pembrolizumab constitutes the first cancer therapy approval based on a patient selection biomarker irrespective of the tumor type, highlighting MSI-H status as a therapeutically trackable and clinically implemented feature of tumor cells (Goswami and    Knijnenburg et al., 2018). However, while MSI-H status is linked to a CpG island methylator (CIMP) phenotype and an increased mutational burden, a functional dependency enabling selective targeting of MSI-H cancer cells remains elusive. The results of this study uncover a novel vulnerability of MSI-H tumor cell models and indicate that pharmacological inhibition of WRN ATPase/helicase function might serve as an attractive targeted therapeutic strategy in MSI-H cancer.
Our data suggest that similar to the tumor agnostic activity of checkpoint blockade, MMR deficiency and MSI-H status represent a genetic determinant for WRN dependency regardless of tumor type. As indicated by the MLH1/MSH3 reconstitution and depletion studies, the genetic interaction of WRN and MMR genes cannot readily be recapitulated using isogenic cell models and thus seems to be distinct from acute and hard-wired synthetic lethal interactions, such as described in cancer cells for the alternate BAF complex ATPases SMARCA2/SMARCA4 or the cohesin subunits STAG1/ STAG2 (Oike et al., 2013;van der Lelij et al., 2017;Benedetti et al., 2017). In the Random Forest model (see Figure 1-figure supplement 1A), loss of MLH1 expression was the feature most strongly associated with WRN dependency. However, we cannot rule out that alterations frequently co-occurring with MMR-deficiency or MSI-H status (Boland and Goel, 2010;Knijnenburg et al.,  Upon WRN loss-of-function we observe a strong and rapid decrease in the viability of MSI-H cell models that is accompanied by nuclear abnormalities and cell division defects. In particular, we find that WRN-depleted MSI-H cancer cells display chromosome breaks, chromatin bridges and micronuclei indicative of genome instability that is highlighted during cell division. The strong induction of g-H2AX in WRN-depleted MSI-H interphase cells suggest that pervasive DNA damage, e. g. DNA double strand breaks, accumulate already in interphase prior to entry into mitosis. The occurrence of these defects and phenomena in MSI-H but not MSS cells upon WRN inactivation suggests that these aberrations might be causally linked to the selective reduction in viability in MSI-H cells. While rescue studies using WRN variants clearly indicate WRN helicase function as the critical enzymatic activity in MSI-H cell models, CRISPR domain scanning indicates a potential structural requirement of the exonuclease domain and the HTH loop of WRN. WRN is a member of the RecQ helicase family which fulfils pleiotropic functions in DNA repair (Chu and Hickson, 2009). MMR activity is required for activation of the G2/M checkpoint in response to DNA damage prior to entry into mitosis (O'Brien and Brown, 2006). WRN function in MSI-H cells might therefore be critical for the resolution of DNA damage events and to prevent premature entry into mitosis. Of note, cell lines derived from Werner syndrome patients display defective mitotic recombination and are susceptible to genome instability (Prince et al., 2001). However, in MSS cancer and non-transformed cells WRN depletion had no or very mild effects on viability, suggesting that pharmacological inhibition of WRN might allow for an MSI-H cancer-directed therapy that spares normal cells and tissues.
The chromosome breaks and radial chromosomes observed in MSI-H cells upon WRN depletion indicate the generation and/or persistence of DNA double strand breaks. Future research is required to dissect the molecular basis for this effect. It is conceivable that WRN is required to process and resolve DNA repair or replication intermediates that arise in MMR-deficient cells or that MMR is required to cope with intermediates emerging upon compromised WRN function. The identification of the molecular basis of the WRN-MSI-H relationship will also help to understand why some rare outlier MSI-H cell lines (see Figure 1B) do not respond to WRN inactivation.
Werner syndrome patients show an increased lifetime risk to develop tumors, pointing to a tumor-suppressive function of WRN (Goto et al., 2013). Interestingly, homozygous Wrn-null mice display no overt phenotype and do not recapitulate the premature ageing or cancer predisposition conditions of Werner syndrome, unless crossed into a Terc-null background (Chang et al., 2004;Lebel and Leder, 1998;Lombard et al., 2000). Mutations in Werner syndrome are almost exclusively truncating nonsense, splicing or frameshift mutations affecting WRN nuclear localization, suggesting that concomitant loss of WRN helicase and exonuclease function might be required for the onset of Werner syndrome (Fu et al., 2017;Huang et al., 2006;Matsumoto et al., 1997;Yokote et al., 2017). This indicates that inhibition of WRN helicase function might have a therapeutic index for the treatment of MSI-H cancer without inducing Werner syndrome related phenotypes.
Our study highlights the power of combining deep functional genomic screening data with tumor cell line profiling to identify new targets with an associated predictive biomarker in oncology. Given the possibility to develop potent and selective small molecule inhibitors of WRN helicase (Aggarwal et al., 2013;Aggarwal et al., 2011;Rosenthal et al., 2010), our findings outline a novel strategy for the treatment of a clinically defined subset of patients harboring MSI-H/MMR-deficient tumors. Since genome instability can elicit cytoplasmic nucleic acid sensor pathways and innate immune responses (Mackenzie et al., 2017), the induction of cancer cell selective nuclear aberrations by WRN inactivation could provide a synergistic combination option with the approved immunotherapy agents for the benefit of MSI-H cancer patients.

Random Forest model
To explore the hypothesis that WRN sensitivity is associated with MSI, we employed an exploratory machine learning approach (Qi, 2012). We first divided the DRIVE WRN cell line sensitivity data (McDonald et al., 2017) into four distinct groups, using a k-means clustering algorithm. We chose four clusters to model insensitive (cluster 1), moderate insensitive (cluster 2), moderate sensitive (cluster 3) and sensitive (cluster 4) cell lines. We subsequently denoted clusters 1 and 2 as insensitive and clusters 3 and 4 as sensitive and chose an RSA score <À1.37 (maximum value of cluster 3) as a cutoff between sensitive and insensitive lines. Only a small fraction of cell lines, 30 out of 371 cell lines (8%), is sensitive in the entire data set, suggesting a pronounced class imbalance between sensitive and insensitive cell lines. To address this class imbalance, we focused our subsequent analysis on cell lines originating from CRC, as i) most sensitive cell lines are from this indication and ii) MSI has been extensively characterized in this indication (Boland and Goel, 2010;Medico et al., 2015). 36% (13 out of 36) of colon cancer cell lines are sensitive to WRN loss of function according to our k-means clustering based approach. We next assembled a MSI feature list. We used cell line gene expression and mutation data for the cell lines from a set of genes, i) involved in MMR (EXO1, MLH1, MLH3, MSH2, MSH3, MSH6) and ii) genetic target genes of MSI in CRC (Boland and Goel, 2010). We next trained a Random Forest model based on 50% of the data. On the full dataset, the model classifies WRN sensitive and insensitive cell lines with an accuracy of 0.89 and a recall rate for sensitive lines of 0.69.

MSI analysis
Genomic DNA was isolated using QIAampDNA mini kit (Qiagen, Hilden, Germany). Per reaction 2 ng of genomic DNA was used for fluorescent PCR-based analysis of the mononucleotide microsatellite marker length using the Promega MSI Analysis System, Version 1.2 kit. Microsatellite fragment length of a standard panel of markers  according to the 'Revised Bethesda Guidelines for Hereditary Nonpolyposis Colorectal Cancer (Lynch Syndrome) and Microsatellite Instability' (Umar et al., 2004) was analyzed using capillary electrophoreses (Applied Biosystems 3130xl Genetic Analyzer, 16-capillary electrophoresis instrument) and evaluated with GeneMapper Software 5 (Applied Biosystems).

siRNA transfection and cell viability
For knock-down experiments, cells were transfected with ON-TARGETplus SMARTpool siRNA duplexes (Dharmacon, Lafayette, CO, US) targeting WRN (L-010378-00), MLH1 (L-003906-00) or MSH3 (L-019665-00) using Lipofectamine RNAiMAX reagent according to the manufacturer's instructions (Invitrogen, Waltham, MA, US). For WRN knock-down, additionally an individual siRNA was used (J-010378-05). Chromosome spreads, immunoblotting, immunofluorescence and live cell imaging experiments were performed using a final siRNA concentration of 20 nM. Cell viability assays were performed using 10 nM siRNA in 96-well plates in a total volume of 100 ml per well. Viability was determined using CellTiter-Glo (Promega, Madison, WI, US). Seven days post-transfection, 100 ml CellTiter-Glo solution was added directly to the cell medium, mixed and incubated for 10 min prior to determination of the luminescence signal. Crystal violet staining was performed in 24-well format using a total volume of 1000 ml per well. Seven days post-transfection, cells were fixed with 4% formaldehyde in PBS and stained twice for 30 min with crystal violet (SIGMA, HT901). For codepletion of p53 and WRN 10 nM of the respective siRNA duplexes each were used for immunoblot and viability assay.

CRISPR depletion assays
Stable Cas9 expressing cell lines, using either Blasticidin or Puromycin as selection markers, were transduced with vectors encoding GFP and sgRNAs targeting different domains of WRN (Supplementary file 3). On day three post-transfection the fraction of GFP positive cells, measured via flow cytometry analysis, was set to 100%. All values were normalized to the control RPA3_1.3 for relative depletion ratio. cDNA transgene vectors pLVX-WRN-3xFLAG-IRES-Puro, pLVX-WRN-3xFLAG-IRES-Puro K577M, pLVX-WRN-3xFLAG-IRES-Puro E84A and pLVX-WRN-3xFLAG-IRES-Puro E84A _K577M for siRNA-resistant transgene expression were generated by gene synthesis (GenScript, China) based on the WRN cDNA sequence NCBI NM_000553.5 followed by cloning into the parental pLVX vector (Clontech, Mountain View, CA, US). Codon optimization changes were introduced into WRN coding sequences to render the transgenes siRNA-resistant.

Immunofluorescence
For immunofluorescence, 72 hr post siRNA transfection cells were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.2% Triton X-100 in PBS for 10 min and blocked with 3% BSA in PBS containing 0.01% Triton X-100. Cells were incubated with primary LAP2ß or phospho-Histone H2A.X (Ser139) and secondary antibody (Alexa 488, Molecular Probes, Eugene, OR, US), DNA was counterstained with Hoechst 33342 (Molecular Probes, Eugene, OR, US; H3570). Coverslips and chambers were mounted with ProLong Gold (Molecular Probes, Eugene, OR, US). Images were taken with an Axio Plan2/AxioCam microscope and processed with MrC5/Axiovision software (Zeiss, Germany). For quantification of g-H2AX immunofluorescence, nuclei were identified based on Hoechst staining using segmentation in ImageJ 1.52a and corresponding g-H2AX mean intensities of nuclei were determined.

Chromosome spreads and live-cell imaging
For chromosome spread analysis, Nocodazole (1.5 mM final concentration) and 2 mM caffeine was added to the medium for 6 hr. Cells were harvested and hypotonically swollen in 40% medium/60% tap water for 5 min at room temperature. Cells were fixed with freshly made Carnoy's solution (75% methanol, 25% acetic acid), and the fixative was changed three times. For spreading, cells in Carnoy's solution were dropped onto glass slides and dried. Slides were stained with 5% Giemsa (Merck) for 4 min, washed briefly in tap water and air dried. For chromosome spread analysis two independent slides were scored blindly for each condition. Live-cell imaging was performed using Spinning Disk Confocal UltraView Vox Axio Observer equipped with Plan apochromat 20x/0.8 objective (Zeiss) and an electron-multiplying charge-coupled device 9100-13 camera (Hamamatsu Photonics). The microscope was controlled using Volocity software (Perkin-Elmer). DNA was counterstained with 100 nM SiR-Hoechst 3 hr before the start of imaging. At 24 hr post siRNA transfection cell nuclei were imaged in 2 z slice sections spaced 6 mm every 6 min for 48 hr. For the imaging, cells were seeded into glass bottom 24 well SensoPlate (Greiner) with imaging medium (phenol red free DMEM supplemented with 10% [vol/vol] FCS, L-glutamine 2 mM and 1% [vol/vol] penicillin-streptomycin). During live cell imaging, cells were maintained at 37˚C in a humidified 5% CO 2 atmosphere. Data availability All data generated or analysed during this study are included in the manuscript and supporting files.