Depletion of RAD17 sensitizes pancreatic cancer cells to gemcitabine

Pancreatic cancer is a devastating disease. With a mortality rate that is almost identical to incidence and a five-year survival rate of less than 6%, it accounts for the fourth most common form of cancer-related deaths in the developed world (Siegel et al., 2013). Reasons for its poor prognosis are late diagnosis, highly invasive and metastatic potential and resistance to chemotherapeutics (Hung et al., 2012). Pancreatic ductal adenocarcinoma (PDAC) is the most common form of pancreatic cancer. The only potentially curative therapy is resection of the primary tumour followed by chemotherapy to try to eliminate residual tumour cells and metastases. At the time of diagnosis only about 20% of patients undergo tumour resection (Kern et al., 2011). The majority is diagnosed at an advanced stage of disease, with local invasion and distant metastases being present, which precludes surgery. Adjuvant chemotherapy with gemcitabine (GEM), the standard treatment for the last 15 years (Wang et al., 2011), can extend five-year survival to 25% in some cases (Hackert et al., 2009). In the last decade, gemcitabine has also been the standard in first-line treatment of non-resectable tumours. However, the clinical effect of gemcitabine monotherapy remains limited, with a response rate below 20% (Glienke et al., 2012).

Many clinical studies have tried to improve gemcitabine treatment by combining it with classical chemotherapeutics. However, clinical benefit for advanced PDAC has been limited so far (Merl et al., 2010; Stathis and Moore, 2010). A meta-analysis of 15 clinical trials looking at overall survival revealed that only patients with a good performance status could benefit from combinations of gemcitabine with platinum analogues or fluoropyrimidines (Heinemann et al., 2008). Likewise, combinations with targeted drugs have yielded only marginal effects. These include small molecules and therapeutic antibodies targeted against the extracellular receptors EGFR, VEGF or HER2, which are overexpressed in pancreatic cancer. Although when combined with bevacizumab (targeting VEGF) or trastuzumab (targeting HER2) there was no clinical benefit over gemcitabine alone (Ko et al., 2012; Safran et al., 2004), a small but significant 2-week improvement was observed when combined with the tyrosine kinase inhibitor erlotinib in patients with advanced PDAC (Moore et al., 2007). Overall, the rationale for gemcitabine combination therapies has been motivated by empirical data from other cancer entities in combination with end-point expressional data (Philip, 2008).

Over the past few years, advances in functional RNAi screening technologies have caused a paradigm shift in the field of target discovery (Iorns et al., 2007; Kassner, 2008). Rather than looking at gene expression changes from an end-point perspective, the functional approach addresses the whole process of drug-induced cytotoxicity. Thus, it is much more suitable for discovery of targets available for potential combination therapy. Following this principle, we conducted a synthetic lethal RNAi screen to identify new targets that – when depleted by RNAi – sensitize pancreatic cancer cells to gemcitabine.

Methodologically, two modules of the barcoded Decipher small hairpin RNA (shRNA) library were screened. Each module comprised 27,500 shRNA expression constructs targeting over 5,000 genes, with at least five constructs per gene. Using stringent filter criteria during analysis and validation, we identified knockdown of the cell cycle checkpoint protein RAD17 homolog (from Schizosaccharomyces pombe; RAD17) to act in a synthetically lethal manner with gemcitabine in the cell lines BxPC-3, MiaPaca-2 and primary JoPaca-1 cells. RAD17 is involved in detecting gemcitabine-induced stalled replication forks and acts upstream of checkpoint kinase 1 (CHK1). Knockdown and inhibition of CHK1 has been shown to act in a synthetically lethal manner with gemcitabine in pancreatic cancer cells (Azorsa et al., 2009; Venkatesha et al., 2012), which was also confirmed in this study. Mechanistically, RAD17 in complex with replication factor C (RFC) subunits 2–5 is responsible for loading a DNA clamp composed of RAD9, RAD1 and HUS1 at stalled replication forks, a result of gemcitabine-induced masked chain termination. This early event in detection of stalled replication forks mediates activation of ataxia telangiectasia and Rad3 related (ATR) kinase, which is essential for downstream signalling, leading to cell cycle arrests. Our results show that the synergistic effect of RAD17 knockdown and gemcitabine is driven by forced mitotic entry of cells in which gemcitabine-induced stalled replication forks remain undetected. This leads to asymmetric DNA distribution during anaphase followed by DNA fragmentation and finally cell death by mitotic catastrophe.

A negative selection RNAi screen was performed in BxPC-3 cells, targeting over 10,000 genes. The individual shRNA expression constructs were transduced into the cells by means of lentiviral infection. Negative selection pressure was induced by growing the cells with 6 nM gemcitabine (IC20) for 72 hours. Cytotoxic effects throughout the screen were monitored by quantification of the shRNA expression constructs through the associated barcode sequences. First, baseline cell viability for each shRNA expression construct was determined by detecting the barcode abundance prior to treatment (timepoint t₀). Second, synthetic lethality was calculated by dividing normalized barcode abundance of the treatment end point (t_end-GEM) by the control end point (t_end-CTRL) yielding a synthetic lethality (SL) ratio. Average SL ratios were calculated for a minimum of three independent shRNA expression constructs against the same gene. These synthetic lethal interactions were ranked by strength and significance determined by t-test comparison to internal negative control shRNAs against luciferase (Fig. 1). In total, there were 70 potential candidate genes with an average SL ratio significantly below a log₂ of −0.35. Of all potential candidates, knockdown of RAD17 showed the most significant synthetic lethal effect (Fig. 1). The complete lists are shown in supplementary material Tables S1 and S2.

To assess the quality of the screening procedure, we analysed the screening data for potential candidates by GO-annotation cluster enrichment analysis (Huang et al., 2009). We found that the most significantly enriched GO-annotation clusters contained genes involved in DNA damage and DNA repair signalling (Table 1). Candidate genes of these clusters were selected for validation. In response to gemcitabine-induced replication fork stalling, checkpoint kinase 1 (CHK1) plays a key role in downstream signalling in the ATR–CHK1 pathway. Among the selected candidates, four other proteins are directly involved in this signalling pathway, namely RAD17, HUS1, WEE1 and RCF3 (see Fig. 2B for an overview).

Potential candidate genes were analysed individually for targeted gene knockdown using three independent shRNA expression constructs. As inhibition and knockdown of CHK1 is known to potentiate gemcitabine toxicity, shRNAs targeting CHK1 were used as a positive control. Genes for which all three shRNA expression constructs showed synthetic lethality were RAD17, WEE1 and HUS1, with a relative cell viability below 75%. The constructs affecting RFC3 did not result in a significant reduction of viability (Fig. 2A). The strongest effect was observed for the knockdown of RAD17, which reduced relative viability below 75% for all three and below 50% for two constructs. Therefore, we focused on RAD17 for further validation and characterization experiments. Depletion of RAD17 and CHK1 protein following shRNA expression was confirmed by western blot analysis (Fig. 2C). Differential expression analysis showed no off-target gene regulation for any of the three RAD17-targeting shRNA constructs.

RAD17 was validated in the pancreatic cancer cell lines BxPC-3 and MiaPaca-2 as well as the primary cell line JoPaca-1 (Fredebohm et al., 2012) using increasing concentrations of gemcitabine and CHK1 knockdown as a positive control. In all three cell lines, knockdown of RAD17 enhanced gemcitabine cytotoxicity significantly for the three independent shRNA expression constructs. Increased cytotoxicity was most pronounced for 6, 8 and 10 nM gemcitabine. The strongest synthetic lethal effect upon depletion of RAD17 was shown in MiaPaca-2 cells, in which cell viability was reduced from 80% to 20% when treated with 10 nM gemcitabine (Fig. 3). E6/E7-immortalized pancreatic ductal epithelial cells (HPDE) were also susceptible to synthetic lethality.

RAD17 plays a key role in the ATR–CHK1 pathway that is involved in response to stalled replication forks. Knockdown of RAD17 and CHK1, respectively, leads to a clear reduction of cells arrested in early S phase by 50 nM gemcitabine for 24 hours (Fig. 4). To follow up on this population and to determine the cause of increased cytotoxicity, we quantified mitotic cells after knockdown of RAD17 and CHK1 with and without gemcitabine using phosphorylated histone 3 (pH3) as a mitotic marker (Hans and Dimitrov, 2001). Knockdown of RAD17 in combination with 50 nM gemcitabine for 24 hours led to a shift of the mitotic fraction from a tetraploidic (4N) to a less than tetraploidic DNA content (<4N), which was also observed in the CHK1-positive control (Fig. 5). To a lesser extent, this shift was observed for cells in which CHK1 was depleted but that were not treated with gemcitabine. Quantification of this effect revealed that 20–30% of mitotic cells had been forced into mitosis without a fully duplicated genome. Additionally, the sub-G1 fraction of dead cells and debris was increased for cells treated with gemcitabine and depleted of CHK1 or RAD17. Depletion of CHK1 without treatment also resulted in an increase in sub-G1 fraction, more so than depletion of RAD17 (Fig. 5).

Gemcitabine-resistant JoPaca-1 cells were depleted of RAD17 or CHK1, treated with 0.5 and 1 µM gemcitabine for 24 hours and stained for pH3. Cells actively undergoing mitosis were selected and representative images recorded. Although RAD17 knockdown had little to no effect on normal mitosis, in cells forced into mitosis by a combination of RAD17 knockdown and gemcitabine anaphase was typically asymmetric (Fig. 6). An increase in gemcitabine concentration led to DNA fragmentation and dispersion following mitotic catastrophe. This phenotype was abundantly observed in gemcitabine-treated MiaPaca-2 cells depleted of RAD17 (Table 2). Analogous to increased sub-G1 fractions we observed an increase in apoptotic cells (Fig. 7). Interestingly, CHK1 knockdown by itself led to an abnormal amount of JoPaca-1 cells with condensed DNA but no meta- or anaphase phenotypes.

In this study, we identified knockdown of RAD17 as a potentiating factor of gemcitabine-induced cytotoxicity in pancreatic cancer cells. By means of pooled shRNA screening, over 10,000 genes were tested for synthetic lethality with gemcitabine, the standard chemotherapeutic drug used for treatment of advanced pancreatic cancer (Wang et al., 2011). The screen revealed knockdown of 70 genes with significant synthetic lethal effects in combination with gemcitabine. GO annotation cluster enrichment analysis showed a significant enrichment in DNA damage response and repair pathways. Considering the known DNA damaging effect of gemcitabine, this confirmed our methodological approach and the screening conditions applied. More specifically, the ATR–CHK1 pathway known to be directly involved in response to gemcitabine-induced stalled replication forks was clearly overrepresented. An individual validation confirmed strong synthetic lethality of gemcitabine and knockdown of RAD17 and CHK1. In particular, the successful validation of RAD17 in the primary cell line JoPaca-1 underlines the relevance of this finding, as the cells show a strong resemblance to their primary tumour origin (Fredebohm et al., 2012). Validation of RAD17 knockdown using three independent shRNA expression constructs minimizes the chance of any off-target effects being responsible for the observed phenotypes (Echeverri et al., 2006). A genome-wide differential expression analysis of mRNA extracts from RAD17-depleted cells and non-silencing controls showed no aberrant gene regulation.

Further characterization of the synergistic effect pointed to forced mitotic entry as the cause of increased cytotoxicity. This has been shown for inhibition of CHK1 and WEE1 (Aarts et al., 2012), and was shown here for RAD17. Immunofluorescence analysis of cells that are forced into mitosis shows a characteristic phenotype of mitotic catastrophe. Although RAD17 depletion without gemcitabine treatment did not result in abnormal mitosis, depletion of CHK1 resulted in premature DNA condensation. Similarly, DNA segregation has been observed in Chk1^−/− knockout mice as well, where it results in embryonic death (Takai et al., 2000). The amount of dead cells in the sub-G1 fraction confirms these results and points to increased lethality for CHK1-knockdown compared to RAD17-knockdown. Cell death following mitotic catastrophe has been suggested to be connected to apoptosis mediated by activation of caspase 2 (Castedo et al., 2004). Although apoptotic cells were higher in mitotically enriched cells undergoing mitotic catastrophe, we could not detect increased caspase-2 activity (data not shown).

Replication fork stalling is a direct result of masked chain termination caused by the incorporation of gemcitabine triphosphate into the nascent DNA strand. This arrest in DNA synthesis needs to be resolved for the cell to replicate its genome and faithfully continue into G2 phase for preparation of mitosis. However, proofreading exonucleases cannot remove the incorporated gemcitabine monophosphate because of its penultimate position (Huang et al., 1991). Moreover, this damage cannot be repaired by any known DNA repair mechanisms: base excision repair, nucleotide excision repair, homologous recombination or non-homologous end joining work (Crul et al., 2003). RAD17 is involved in early events following masked chain termination and replication fork stalling that ultimately lead to intra-S phase arrest through the ATR–CHK1 pathway. The first step in recognition of the stalled fork is the recruitment of ataxia telangiectasia and Rad3-related protein (ATR) by binding of its close partner ATR-interacting protein (ATRIP) to replication protein A (RPA) bound to exposed single-strand DNA (Cortez et al., 2001; Zou and Elledge, 2003). The second step of activating the ATR–CHK1 signalling cascade starts with the recruitment of polymerase α (Pol α) to the leading strand by topoisomerase-2 binding protein 1 (TopBP1), which also binds to RPA (Yan and Michael, 2009). Pol α then creates primer-template junctions on the leading strand and mediates recruitment of the heterotrimeric DNA clamp complex of RAD9–HUS1–RAD1 (9-1-1) (Yan and Michael, 2009). In order to form a ring around the DNA, the 9-1-1 clamp needs to be loaded. This is mediated by a complex consisting of RAD17 and the small subunits of replication factor C (RFC2-5) called the clamp loader complex (Lee and Dunphy, 2010; Shiomi et al., 2002). The clamp loader complex binds to RPA-coated single-strand DNA at the 5′-end of a primer template junction where it loads the 9-1-1 clamp onto the DNA (Majka et al., 2006). In order to activate ATR and downstream effector kinases such as CHK1, cooperation between RAD17/RFC2-5, 9-1-1 and TopBP1 is important. Furthermore, RAD17 function is a prerequisite for this connection, as it is required for the interaction between TopBP1 and RAD9 and for increased recruitment of TopBP1 to the stalled replication fork (Lee and Dunphy, 2010).

This pathway not only regulates arrest during S phase upon DNA damage, but also affects the G2–M transition through its key regulator CHK1. Triggering activation of CHK1 engages two powerful mechanisms influencing cell-cycle regulation. Firstly, active CHK1 causes intra-S phase arrest by degradation of CDC25A phosphatase, which is responsible for activation of CDK2-cyclin A/E complexes (Boutros et al., 2007). Secondly, it inhibits G2–M transition by activation of WEE1, which is responsible for inhibition of CDK1/cyclin B complex (Boutros et al., 2007; O'Connell et al., 1997). Thus, early inhibition of this pathway results in cessation of cell cycle arrest and procession into premature mitosis.

Our results underline the importance of RAD17 in this mechanism. Also, they suggest RAD17 as a new target for gemcitabine combination therapy. Targeting the ATR–CHK1 pathway has become a promising approach for pancreatic cancer and other cancer entities (Carrassa and Damia, 2011; Garrett and Collins, 2011). Inhibitors of CHK1 are currently under clinical investigation [comprehensively reviewed by Garrett and Collins (Garrett and Collins, 2011)]. Similarly, a potent inhibitor of WEE1 kinase (MK-1775) shows promising results in preclinical experiments and is now included in phase I clinical studies with gemcitabine and platinum-based drugs. Combination of CHK1 and WEE1 inhibition has shown synergistic cytotoxicity (Aarts et al., 2012; Davies et al., 2011) illustrating the promising potential of targeting this pathway. Along this line, development of RAD17 inhibitors should provide a new option for combination with gemcitabine as well as CHK1 and WEE1 inhibitors.

The synthetic lethal effect of CHK1 and WEE1 inhibition in combination with DNA-damaging agents has been shown to be dependent on p53 deficiency (Aarts et al., 2012; Chen et al., 2006; Landau et al., 2012; Ma et al., 2012). The tumour suppressor protein p53 is a key regulator of cell-cycle arrest in response to DNA damage. Upon activation, it acts as a transcription factor of CDK inhibitors such as p21 (CDKN1A) and p16 (CDKN2A) (Alberts, 2002). Through this route, functional p53 signalling may override unscheduled mitotic entry triggered by inhibition of the ATR–CHK1 signalling pathway. Concordantly, E6/E7-immortalized HPDE cells with defective p53 signalling (Ouyang et al., 2000) were susceptible to synthetic lethality of CHK1 or RAD17 knockdown in combination with gemcitabine. BxPC-3 and MiaPaca-2 cell lines both carry deleterious mutations in the DNA-binding domain of p53 (Petitjean et al., 2007) necessary for transcriptional regulation. JoPaca-1 cells, however, have a mutation solely at codon 72 of the proline-rich domain that is important for induction of apoptosis but not cell-cycle arrest (Baptiste et al., 2002; Fredebohm et al., 2012; Walker and Levine, 1996; Zhu et al., 1999). However, the functional DNA-binding domain of p53 in JoPaca-1 cells does not prevent sensitivity to the synthetic lethality of CHK1 and RAD17 inhibition and gemcitabine.

Nevertheless, the lack of mutations in the DNA-binding domain of p53 in JoPaca-1cells may account for the slightly reduced synthetic lethal effect compared with that in BxPC-3 and MiaPaca-2 cells. Future studies will have to determine which mutations are most sensitive in the functional dependence of CHK1–ATR pathway inhibition on p53-regulated cell-cycle arrest. Once this is clarified, clinical application of CHK1/WEE1, and potentially RAD17 inhibitors, should be accompanied by molecular diagnostics of patient p53-mutational status.

As shown here and elsewhere (Azorsa et al., 2009), knockdown of CHK1 has a proliferation inhibiting effect on its own, characterized by abnormal chromosomal segregation. Therefore, inhibition of CHK1 is likely to cause undesired cytotoxicity in normal cells causing increased side effects. RAD17 knockdown by itself did not inhibit cell proliferation nor did it show abnormal DNA segregation. Given the central role of CHK1 in many DNA-damage response pathways (Carrassa and Damia, 2011), targeting the replication fork stalling pathway at the position of RAD17, upstream of CHK1 and WEE1, might provide a more specific approach.

The pancreatic cancer cell lines MiaPaCa-2 and BxPC-3 were obtained from ATCC (Rockville, MD, USA). The primary cell line JoPaca-1 was isolated as described previously (Fredebohm et al., 2012) and used at low passage (<P10) to retain primary characteristics. HPDE cells were kindly provided by Professor Francisco Real (Spanish National Cancer Research Centre, CNIO). All cell lines were free of mycoplasms, viruses and cell line contamination as tested by PCR (Schmitt and Pawlita, 2009). Cells were cultured in standard medium and supplements, which were obtained from Life Technologies, Darmstadt, Germany: BxPC-3 in Roswell Park Memorial Institute (RPMI) 1640 medium (cat. no. 21875), MiaPaCa-2 in Dulbecco's modified Eagle's medium (DMEM; cat. no. 41965) supplemented with 1 mM sodium pyruvate (cat. no. 11360), and JoPaca-1 in Isocove's modified Dulbecco's medium (IMDM; cat. no. 21056); HPDE in keratinocyte serum-free medium (KSFM), 0.15 ng/ml epidermal growth factor receptor and 25 µg/ml bovine pituitary extract (cat. no. 17005075). All media were supplemented with 10% foetal calf serum (cat. no. 10500) and 1% penicillin–streptomycin (cat. no. 15140).

We targeted 10,000 genes for shRNA-mediated knockdown using modules 1 and 2 of the Decipher library (Cellecta, Mountain View, CA, USA) in two screens under identical conditions. Each gene was targeted by five shRNA expression constructs, which are identifiable by barcode sequences. The target cell line BxPC-3 was infected with lentiviral particles containing either module 1 or 2 at a multiplicity of infection of 0.3. Positively infected cells were selected by addition of 0.5 µg/ml puromycin 24 hours after infection. Selection took place for 48 hours, after which puromycin was removed and cells were allowed to recover for another 48 hours. After recovery, cells were divided into treatment and control groups. At this point, a sample of 20 million cells was taken for the t₀ reference time point, shock-frozen and kept at −80°C. Cells were kept under selective pressure of 0.5 µg/ml puromycin, which was added after cells had attached to the substrate. A 72-hour treatment with 6 nM gemcitabine, or DMSO as a control, commenced 48 hours after passaging. After an additional 72 hours without treatment, control and treated cells were harvested, shock-frozen and kept at −80°C.

To determine the abundance of cells before and after treatment and in the DMSO control, barcode sequences were amplified from genomic DNA and subjected to high-throughput sequencing. Amplification of barcode sequences was achieved by two rounds of PCR according to the following protocol. Cell pellets from samples obtained at the t₀, t_end-GEM and t_end-CTRL time points were thawed and subjected to ultrasonication at 50% intensity for 10 seconds on ice to shear genomic DNA into fragments of 1–10 kb. Genomic DNA was isolated from cell pellets using DNeasy Blood & Tissue Kit (69581, Qiagen, Hilden, Germany). For the first round of PCR, 50 µg of fragmented genomic DNA were used in eight 100 µl reactions containing: 1 µl Titantium Taq polymerase and corresponding PCR buffer (639210, Clontech of Takara Bio, Saint-Germain-en-Laye, France), 250 µM dNTP mix (M3015.4100, Genaxxon, Ulm, Germany), 3 µM HTS forward primer: 5′-TTCTCTGGCAAGCAAAAGACGGCATA-3′ and 3 µM reverse primer: 5′-TGCCATTTGTCTCGAGGTCGAGAA-3′ in nuclease-free water (AM9938, Life Technologies, Darmstadt, Germany). Desalted purity primers were obtained from Sigma Aldrich (Munich, Germany). Reaction mixtures were prepared on ice and subjected to a first round PCR starting with activation of the polymerase at 94°C for 3 minutes, followed by 16 cycles of denaturation at 94°C for 30 seconds, annealing at 65°C for 10 seconds, and elongation at 72°C for 20 seconds. Final extension was done at 68°C for 2 minutes. Second round PCR was done using 2 µl of the first-round reaction as template in three 100 µl reactions per condition. This time, primers GEX forward 5′-CAAGCAGAAGACGGCATACGAGA-3′ and GEX reverse 5′-AATGATACGGCGACCACCGAGA-3′ were used, while all other reagents remained the same. Thermal cycler settings were also unchanged, with the exception of a reduced elongation time of 10 seconds and a reduction of the cycle number to ten. Amplified barcode sequences were purified using PCR purification and gel extraction kits (28104 and 28704, Qiagen). Barcodes of module 1 and 2 were combined for each time point and quantified by next-generation sequencing (Genome Analyzer IIx, Illumina, San Diego, CA, USA). Each module contained a two-base-pair identifier attached to the barcode sequence for later discrimination.

Sequences were analysed to determine the abundance of each barcode using the software ‘Barcode Deconvoluter’ provided by Cellecta. The software counts the number of 18-nucleotide barcode sequences with a tolerance of two mismatches and assigns each barcode to the corresponding shRNA construct. Furthermore, each module was identified by two module-specific base pairs attached to the barcode sequence. Further analysis was done in Microsoft Excel. In order to compare reads of different modules and time points, each group of modules and time points was normalized to the average number of reads. Barcodes below 100 reads for t₀ and t_end-CTRL were filtered out for reasons of quality assurance. Finally, SL ratios were calculated by dividing normalised reads of t_end-GEM by normalised reads of t_end-CTRL and taking the log₂ of this ratio. Genes passing these filter criteria with at least three shRNA expression constructs were averaged and a P-value calculated in comparison to internal luciferase control constructs using unpaired, two-sided, unequal variance t-test statistics. Candidates for further analysis were selected on the basis of the strength and significance of the SL ratio. Following this principle, genes were included when significance was below a P-value of 0.05 and the log₂ SL ratio was below −0.35. Selection of candidates for validation was assisted by functional annotation enrichment analysis (Huang et al., 2009).

Lentiviral particles were produced using a second generation system described and engineered by Didier Trono's group (Zufferey et al., 1997). HEK-293T cells were seeded in DMEM without antibiotics at a density of 65,000 cells per cm² in a six-well format for single constructs or 175 cm² flasks for production of viral pools. Helper plasmids psPAX2 and pMD2.G were co-transfected with the construct carrier pRSI9 the following day using Lipofectamine (18324010, Life Technologies) and Plus Reagent (11514015, Life Technologies) according to the manufacturer's instructions. The next day, the medium was exchanged for high-serum DMEM (30% FCS). Supernatant containing lentiviral particles was harvested the following day, filtered through a 0.45 µm PES membrane (295–4545, Thermo Scientific) and stored in aliquots at −80°C.

For each candidate gene, three shRNA expression constructs were cloned into pRSI9-U6-(sh)-UbiC-TagRFP-2A-Puro vector according to the supplier's protocols and validated by Sanger sequencing. All shRNA sequences used for candidate validation are listed in supplementary material Table S3. Lentivirus was produced and titrated for each cell line to yield an optimal dilution factor for multiplicity of infection between 1 and 2 for 24 hours of infection. In a first round of validation, BxPC-3 was treated 72 hours after infection with 6 nM gemcitabine for 72 hours. Cell viability, determined by the resazurine assay (Czekanska, 2011), was normalised to the non-silencing control (shCTRL) for treatment and control conditions, respectively, and expressed in percentages. shRNAs against RAD17 and CHK1 were further validated in MiaPaca-2 cells and the primary cell line JoPaca-1 using concentrations of gemcitabine between 0 and 18 nM.

Total RNA was isolated with the RNAeasy Isolation kit (Qiagen, Hilden, Germany), following the protocol suggested by the manufacturer. RNA integrity was evaluated using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, USA). The RNA was analysed on the HumanHT-12 v4 Whole Genome Expression BeadChip (Illumina, San Diego, CA). To synthesize first and second strand cDNA and for amplifying biotinylated cRNA, the Illumina Totalprep RNA Amplification Kit was used. Hybridisation to the BeadChip was performed according to the manufacturer's instructions without modification. In short, a maximum of 10 µl cRNA was mixed with a 20 µl GEX-HYB hybridisation solution. The preheated 30 µl assay sample was dispensed onto the large sample port of each array and incubated for 18 hours at 58°C. Subsequently, the samples were washed and scanned with a BeadArray Reader (Illumina). Raw data were exported from the Illumina Beadstudio software to R (Ritchie et al., 2011). The data were quantile normalised and log₂ transformed. The expression data are available in the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-1535.

Infected and selected cells were lysed in M-PER lysis buffer (78501, Pierce, Thermo Scientific) containing complete protease inhibitors (04719964001, Roche, Mannheim, Germany) 1 tablet per 10 ml, 1 mM NaF and 0.2 µM Na₃VO₄. SDS-PAGE and western blotting were done according to standard protocols (Sambrook and Russell, 2006). The following primary antibodies were used for validation of gene knockdown: mouse Chk1 (G-4; sc-8408, Santa Cruz, Heidelberg, Germany); rabbit Rad17 D3G6 (8561, Cell Signaling Technology, Frankfurt, Germany) and for loading control, antibodies against β-actin (rabbit polyclonal, ab8227, Abcam, Cambridge, UK and mouse monoclonal, A5316, Sigma Aldrich, Taufkirchen, Germany). Western blots were incubated with ECL plus (32132, Thermo Scientific) and exposed using a chemiluminescent image analyzer (LAS-4000 mini, Fuji Film, Düsseldorf, Germany). Images were improved in Photoshop-CS5 (Adobe Systems GmbH, Munich, Germany) using auto-level functions for tone and contrast enhancement. Phospho-histone-3-positive cells were quantified by fluorescence activated cell sorting (FACS; Canto II, BD Biosciences, Heidelberg, Germany) after fixation in 70% ethanol and permeabilisation with 0.25% Triton X-100 using phospho-histone H3 (Ser10) antibody conjugated to Alexa-Fluor-488 (3465, Cell Signaling Technology). DNA was counterstained with 7-amino-actinomycin D (7-AAD; cat. no. 559925, BD Biosciences). FACS data were acquired using FACSDiva (BD Biosciences) and analysed using FlowJo software (Tree Star, Ashland, OR, USA).

For immunofluorescence of pH3-positive cells, JoPaca-1 cells depleted of RAD17 or CHK1 were grown in culture slides (cat. no. 354559, BD Biosciences) and treated with 0.5 µM and 1 µM gemcitabine for 24 hours. Cells were fixed with 3.7% formaldehyde and permeabilised with 0.25% Triton X-100. Slides were incubated with the same Alexa-Fluor-488 pH3 antibody used in FACS experiments for 18 hours at 4°C. After removing the chambers coverslips were fixed on top of the slides by Vectashield® mounting medium containing 4′,6-diamidin-2-phenylindol (DAPI) to counter-stain nuclei (H-1500, Vector Laboratories, Burlingame, CA, USA). Images were taken with a confocal laser scanning microscope (LSM 700, Carl Zeiss Microscopy, Oberkochen, Germany).

Quantification of asymmetric mitosis was done using MiaPaca-2 cells depleted of RAD17 with or without 50 nM gemcitabine, compared with cells treated with shCTRL. Mitotic cells were enriched by mitotic shake-off at 250 r.p.m. for 30 minutes and collected with dead cells in the supernatant. Cells were fixed, permeabilised and stained for pH3 as described above. For quantification, five to ten images were taken at random locations on the slide.

The authors thank Philipp Kuhn, Sandra Widder and Viola Ladenburger for technical assistance. Lentiviral shRNA libraries were kindly provided and developed by Cellecta using National Institutes of Health funded research grant support.