WWOX sensitises ovarian cancer cells to paclitaxel via modulation of the ER stress response

There are clear gaps in our understanding of genes and pathways through which cancer cells facilitate survival strategies as they become chemoresistant. Paclitaxel is used in the treatment of many cancers, but development of drug resistance is common. Along with being an antimitotic agent paclitaxel also activates endoplasmic reticulum (ER) stress. Here, we examine the role of WWOX (WW domain containing oxidoreductase), a gene frequently lost in several cancers, in mediating paclitaxel response. We examine the ER stress-mediated apoptotic response to paclitaxel in WWOX-transfected epithelial ovarian cancer (EOC) cells and following siRNA knockdown of WWOX. We show that WWOX-induced apoptosis following exposure of EOC cells to paclitaxel is related to ER stress and independent of the antimitotic action of taxanes. The apoptotic response to ER stress induced by WWOX re-expression could be reversed by WWOX siRNA in EOC cells. We report that paclitaxel treatment activates both the IRE-1 and PERK kinases and that the increase in paclitaxel-mediated cell death through WWOX is dependent on active ER stress pathway. Log-rank analysis of overall survival (OS) and progression-free survival (PFS) in two prominent EOC microarray data sets (Tothill and The Cancer Genome Atlas), encompassing ~800 patients in total, confirmed clinical relevance to our findings. High WWOX mRNA expression predicted longer OS and PFS in patients treated with paclitaxel, but not in patients who were treated with only cisplatin. The association of WWOX and survival was dependent on the expression level of glucose-related protein 78 (GRP78), a key ER stress marker in paclitaxel-treated patients. We conclude that WWOX sensitises EOC to paclitaxel via ER stress-induced apoptosis, and predicts clinical outcome in patients. Thus, ER stress response mechanisms could be targeted to overcome chemoresistance in cancer.

Epithelial ovarian cancer is the most lethal gynaecological malignancy. Use of platinum and taxane-based chemotherapy result in high response rates, but 70% of patients relapse and develop drug-resistant disease. 1 Paclitaxel stabilises microtubule assembly, resulting in a mitotic block of cell cycle leading to apoptosis. 2 The cytotoxic effects of taxanes are not just because of its antimitotic function 3 but are, in part, mediated by endoplasmic reticulum (ER) stress/unfolded protein response (UPR). [4][5][6][7][8] UPR is a programme initiated by the accumulation of unfolded proteins in ER to re-establish homeostasis by activation of chaperones and translation inhibition. 9 The WWOX (WW domain containing oxidoreductase) gene on chromosome 16q23-24 is located at the same locus as the common fragile site FRA16D. 10 WWOX loss increases tumour susceptibility in several mouse models. [11][12][13] WWOX expression is lost or downregulated in most cancers because of genomic disruption or epigenetic silencing, and recently The Cancer Genome Atlas data sets have highlighted 44 novel somatic mutations in WWOX in various cancer types, several of which lead to changes in the protein function. 14-18 WWOX is highly expressed in secretory epithelia, in reproductive, exocrine and endocrine organs and also in neuronal bodies throughout the central nervous system. Mutations in WWOX have been reported in several neurological pathologies. 18,19 WWOX is lost in 30% of ovarian carcinomas and this is associated with disease progression, and poor prognosis. 20 We previously demonstrated that WWOX transfection of PEO1 ovarian cancer cells abolished their in vivo tumorigenicity because of altered interaction of tumour cells with surrounding ECM. 21 This did not correlate with decreased in vitro growth or survival, but was as a result of reduced integrin α3 levels. 21 There was also no effect of WWOX on apoptosis following cisplatin exposure. 21 As standard chemotherapy treatment in ovarian cancer is a combination of platinum and paclitaxel, we examined the impact of WWOX transfection on paclitaxel response. We report that WWOX transfection into the non-expressing PEO1 ovarian cancer cell line 21,22 causes sensitisation to paclitaxel, as demonstrated by increased apoptosis following exposure with paclitaxel. This is independent of the antimitotic function of taxanes, but is related to apoptosis because of ER stress induced by paclitaxel. To mimic the natural course of events involving WWOX loss in cancer, we used siRNA knockdown of endogenous WWOX to study its role in paclitaxel-induced cell death in SKOV-3 and OVCAR-4 cell lines. We conclude that (1) both IRE-1 and PERK arms of UPR get activated on exposure of ovarian cancer cells to paclitaxel and active ER stress is a requirement for WWOX-mediated cell death caused by paclitaxel and (2) WWOX expression may predict patient outcome to taxane-based chemotherapy.

Results
WWOX sensitises ovarian cancer cells to paclitaxel. WWOX reconstitution in PEO1 ovarian cancer cells (WWOX-7, WWOX-8) significantly enhanced paclitaxelmediated cell death (Figures 1a, Po0.01) compared with vector-transfected control cells (Vector-9, Vector-p2). We examined caspase activation, Annexin V positivity and an antibody array to measure the relative levels of apoptosisrelated proteins in these cell lines. WWOX transfection significantly increased caspase-3/7 activation following exposure with paclitaxel 24 h at either 8 or 16 nM, as compared with the vector-transfected control lines (Figures 1b, Po0.05). Increased levels of cleaved caspase-3 was observed in WWOX-transfected lines on treatment with paclitaxel ( Figure 1c). Following paclitaxel exposure, WWOX-transfected cells also displayed significantly higher levels of early apoptotic cells (Annexin V-positive, PI-negative) (Figures 1d, Po0.05). The antibody array demonstrated no consistent differences in the expression of apoptosis-related proteins in basal conditions, but upon paclitaxel exposure, a lower B-cell lymphoma-extra large (Bcl-xL)/Bcl-2-associated x (Bax) ratio in WWOX-transfected cells was observed (Figure 1e).
Two independent siRNA oligos were used to knockdown WWOX in WWOX-8 cell line. This decreased the apoptotic response to these cells to 8 and 16 nM paclitaxel (not shown) as compared with control transfections (Figures 2a, *Po0.05 and ***Po0.005) and cleaved caspase-3 expression (Figure 2b), confirming that paclitaxel chemoresponse is influenced by WWOX expression. In parallel, WWOX status has no impact on cisplatin response ( Figure 2c) to complement the previous data generated with stable transfectants. 21 To mimic the loss/reduction of WWOX as it happens in the course of cancer, we knocked down endogenous WWOX in ovarian cancer cells to monitor the impact on cell survival. We screened several ovarian cancer cells to detect WWOX expression ( Figure 2d WWOX-driven differential response to paclitaxel is independent of the antimitotic action of taxanes, integrin α3 and transforming growth factor-β1 levels. We next examined mitotic arrest in paclitaxel-treated cells to determine whether reduced survival following WWOX overexpression was due to an increased proportion of cells arrested in mitosis ( Figure 3a). WWOX status had no impact on the response of PEO1 cells to the kinesin Eg5 inhibitor monastrol, a selective antimitotic agent 23 (Figure 3b). Thus, the effect of WWOX on paclitaxel response is distinct from the antimitotic function of taxanes. We hypothesised that a differential response to paclitaxel treatment following WWOX transfection might be due to ITGA3 regulation. 21 However, we observed no significant impact of ITGA3 knockdown in PEO1 cell line on responses to paclitaxel (Figures 3c and d). We examined the the levels of transforming growth factor-β1 (TGFB1), which has previously been shown to be a critical mediator of paclitaxel response in ovarian cancer. 24 We could not detect measurable levels of secreted TFGB1 in culture medium in PEO1 clones, but quantitative RT-PCR (QPCR) revealed variable TGFBI levels and no correlation with WWOX expression (Figure 3e).
Paclitaxel induces ER stress response and WWOX determines cell fate in response to prolonged ER stress. Treatment of PEO1 cells with 8 nM paclitaxel led to induction of ER stress, as indicated by upregulation of GRP78 chaperone and phosphorylation of eukaryotic translation initiation factor 2A (eIF2A), as well as c-Jun N-terminal kinase (JNK) and p38 mitogen-associated protein kinase (MAPK) activation 25 (Figure 4a and Supplementary Figure 1). Similar changes were observed in a parallel time-course experiment following exposure to tunicamycin, a classic inducer of ER stress 25 (Figure 4b). WWOX-transfected cells displayed significantly reduced survival rates when exposed to tunicamycin (200 ng/μl) (Figure 4c To investigate ER stress induced by paclitaxel and the role of WWOX in this context, we carried out analysis of (1) PEO1 cells (Figures 5a) and (2) by knocking down endogenous WWOX expression in SKOV-3 and OVCAR-4 cell lines ( Figure 5b). We examined the expression of the two key ER transmembrane kinases IRE-1 and PERK, which activate ER stress responses. Both IRE-1 and PERK showed induction or increase of phosphorylation under the influence of paclitaxel in all cell lines compared with untreated cells. When probed with a phosphorylated IRE-1 antibody, a band at 110 kDa appeared in paclitaxel-treated cells. The identity of this band was confirmed using another antibody detecting total IRE-1 protein (Figures 5a and b). In a control treatment of the cells with thapsigargin (a strong inducer of ER stress), a slightly upward shifted pIRE-1 band was observed (Supplementary Figure 6), suggesting differences in the phosphorylation status of the protein under paclitaxel and thapsigargin treatments. It      WWOX expression predicts PFS in paclitaxel-treated ovarian cancer patients. We used data from two large cohorts of primary ovarian tumours to investigate the associations between WWOX expression and clinical outcome to chemotherapy. The data set referred to as 'TCGA', featuring over 500 primary tumour samples, was obtained from The Cancer Genome Atlas (https://tcga-data. nci.nih.gov/tcga/) 30 The data set referred to as 'Tothill', featuring nearly 300 primary tumour samples, was obtained from Gene Expression Omnibus (series accession GSE9891). 31 Raw gene expression data (Affymetrix CEL files) were obtained for each data set and normalised using robust multiarray average (RMA). 32 WWOX expression and OS: Fitting Cox proportional hazards regression models revealed significant association between higher WWOX expression and longer OS in patients treated with platinum in combination with taxane in both TCGA (n = 373, hazard ratio = 0.48, P = 0.036) and Tothill (n = 194, hazard ratio = 0.53, P = 0.028) cohorts. In both the cohorts higher expression of WWOX was associated with longer survival times even when the extent of disease after primary surgery was taken into account (TCGA hazard ratio = 0.48, P = 0.046; Tothill hazard ratio = 0.67, P = 0. 19 (Figures 7b  and d). We used the TCGA data set to investigate the impact on progression of WWOX expression being classed as detectable or absent relative to normal physiological levels.
Corresponding Kaplan-Meier curves in Figures 7e and f show that, in patients treated with platinum and taxane (Figure 7e) but not in patients treated with platinum alone (Figure 7f), this dichotomisation suggests longer PFS in patients with WWOX expression detectable at a level comparable to normal tissue.
WWOX expression influences ovarian cancer molecular subtypes: WWOX expression showed very significant variation in expression levels across the molecular subtypes reported in the Tothill study. 31 The poor outcome ovarian cancer subtype '1' had lower WWOX expression across three probesets (Figure 8a, P = 0.0035; left, Po1x10 − 4 ; middle and P = 0.00069; right), implying that the loss of WWOX expression has a role in development of aggressive characteristics of this subtype of ovarian cancer.

WWOX, associations with ER stress genes and influence on survival:
To investigate the clinical relevance of WWOX expression and ER stress genes on survival, we investigated the interaction between a panel of genes transcriptionally upregulated by ER stress (GRP78, XBP1, ATF6, eukaryotic translation initiation factor 2-alpha kinase 3 (EIF2AK3), ERN1 and DNA damage-inducible transcript 3) and association of WWOX expression. For each of the panel of genes, proportional hazards regression models were fitted to evaluate the additive effects of WWOX and the ER stress gene, and the interaction between the two genes. It was striking that the association between higher WWOX expression and better outcomes (longer OS and PFS) was dependent on the endogenous levels of GRP78 (Figures 8b-e) but none of the other factors examined (not shown). GRP78 is the best described marker of ER stress, regarded as a key regulator of multiple arms of the UPR, and transcriptionally regulated via conserved ER stress response elements in its promoter. 9 We found that the impact of WWOX expression on OS (Figures 8b and c) and PFS (Figures 8d  and e) was effectively seen in patients (Tothill data set) who had lower endogenous expression of GRP78 (Figures 8c  and e) and not in those who had higher GRP78 expression (Figures 8b and d). The patients were split according to GRP78 expression being above or below the median value. The median was 2.42-fold greater than the minimum value and 3.25-fold lower than the maximum value, making the total range of expression (lowest to highest) 7.89-fold. The samples with high GRP78 expression likely represent tumours with intrinsic ER stress leading to apoptosis, irrespective of chemotherapy, and it might be speculated that WWOX status has no impact on these tumours, which have activation of proapoptotic pathways before chemotherapy  commences. Also, tumours with high GRP78 may represent evolved tumours with intrinsic ER stress, which is more inclined towards a cytoprotective UPR function. Conversely, WWOX appears to strongly affect PFS in the 'low GRP78' tumours, and this may correspond to tumours with intrinsically low, but paclitaxel-inducible, ER stress. These data suggest an inter-relationship between WWOX and the UPR. We could not however detect any interaction between WWOX and GRP78 under the influence of paclitaxel in co-immunoprecipitation assays performed in WWOX-8 cells (Supplementary Figure 9).

Discussion
The combination of platinum and taxanes is the gold standard for chemotherapy in ovarian cancer; however, cure rates remain low. Our data suggest that WWOX regulates taxane response in ovarian cancer by modulating the drug-related ER stress response. One-third of ovarian cancers exhibit loss of WWOX protein 20 and these would therefore be expected to respond less to paclitaxel. Turning to the clinical relevance of our findings, WWOX expression levels predicted longer OS and PFS in a population of paclitaxel-treated ovarian cancer patients.
Recently, WWOX has been shown to be responsible for cancer cells' response to chemotherapy on two accounts. WWOX sensitises squamous cell carcinoma cells to apoptosis induced by the antifolate chemotherapeutic agent methotrexate by regulating autophagy responses in cancer cells and absence of WWOX leads to chemotherapeutic drug resistance. 33 A second study demonstrated that WWOX through its WW1 domain interacts with the p53 homologue protein ΔNP63 known to induce chemoresistance in cancer cells. WWOX sequesters ΔNP63 in the cytoplasm, thus changing its cellular location and transactivation of its target genes. WWOX could reverse ΔNP63-induced chemoresistance to cisplatin by sensitising cancer cells to cisplatininduced cell death. 34 ER stress response is also activated in hypoxic or nutrientdeprived tumours, 9,25,[35][36][37][38] and thus WWOX loss might be the key to multistep carcinogenesis that enables tumours reaching a macroscopic size to survive under conditions of cellular stress due to stimuli such as hypoxia. WWOX, via its WW domain, physically interacts with hypoxia-inducible factor 1α subunit and modulates its levels and transactivation functions. 39 Multiple myeloma cells are stressed by unfolded protein overload because of their production of monoclonal immunoglobulins. Interestingly, WWOX is frequently lost or undergoes recurrent genomic alterations in multiple myeloma and this is associated with poor outcomes. [40][41][42][43][44][45][46] Tumours resembling human multiple myeloma were also noted in WWOX-deficient mice. 12 Here, through modulating WWOX expression in ovarian cancer cells we show that WWOX influences paclitaxelinduced cell death through ER stress pathway and an intact IRE-1 arm of UPR is crucial for WWOX to function. The bioinformatics data presented in Figure 8 captured the expression levels of GRP78 to be a determinant and an indicator of WWOX-mediated influences on patient survival. GRP78 activates all arms of ER stress including IRE-1 and PERK and its expression along with WWOX could be used as a predictive biomarker of patient responses to paclitaxel treatment. WWOX interacts with SCOTIN/SHISA5, 47 which was recently demonstrated to be a proapoptotic protein involved in ER stress response. 48,49 With two interaction-rich WW domains and an enzymatic tail, WWOX could be involved in protein transport, act as a chaperone or it may regulate redox status, particularly since WWOX knockout mice display severe metabolic abnormalities. 50,51 WWOX involvement in ER stress response is consistent with the stress response network hypothesis for common fragile site genes based on the conservation of their loci including the intronic or intragenic fragile site structure that could serve as stress sensors and involvement in stress response already ascribed to WWOX, fragile histidine triad and RAR-related orphan receptor. 52 In conclusion, our study provides novel insights into the function of the WWOX and suggests that defective proapoptotic signalling during ER stress response associated with loss of WWOX may confer paclitaxel resistance in a proportion of ovarian cancer patients. In general, adaptive UPR mechanisms could be targeted to increase chemotherapy efficacy where that cytotoxic mechanism is mediated by ER stress.

Materials and Methods
Cell lines. The PEO1 ovarian cancer cell line (derived within our previous laboratory in University of Edinburgh) is homozygously deleted for WWOX exons 4-8 and lacks WWOX protein expression. We developed independent stable WWOX-and vector-transfected lines as described previously. 21,22 Cells were maintained in RPMI containing 10% FCS, 100 U/ml penicillin, 100 μg/ml streptomycin (and 3 μg/ml blasticidin for stably transfected cells) and cultured at 37°C, 5% CO 2 . PEO1 line derivatives were recently authenticated by confirming the presence of a specific homozygous deletion within WWOX by RT-PCR; the presence of plasmid vector by blasticidin selection; and that WWOX was expressed in WWOX-transfected lines, but null in vector-transfected lines by qRT-PCR and immunoblotting. 21,22 Growth curves. Log-phase cells (6 × 10 3 per well) were seeded into 96-well plates. Following 24 h incubation, media were replaced with media containing paclitaxel, cisplatin, monastrol, tunicamycin, SP600125, KIRA6 or GSK2656157. Monastrol and SP600125 were obtained from Tocris Bioscience (Bristol, England, UK) and tunicamycin from Sigma-Aldrich (St. Louis, MO, USA), KIRA6 from Calbiochem (San Diego, CA, USA) and GSK2656157 from Selleckchem (Houston, TX, USA). Cells were quantified by sulforhodamine B (SRB) assay.
Annexin V apoptosis assays. Log-phase cells (2.5 × 10 5 per well) were seeded into 6-well plates. Following 24 h incubation, media were replaced with media containing paclitaxel. After 24 h of drug exposure, floating and adherent cells were collected and Annexin V measured by flow cytometry using TACSTM Annexin V-FITC Apoptosis Detection Kit (R&D Systems, Minneapolis, MN, USA).
Measuring mitotic index. Cells were plated and treated as for Annexin V assays. Following drug exposure, the cells were fixed with 2% paraformaldehyde, blocked with 1% BSA and permeabilized with 1 × BDPerm/Wash solution (BD Biosciences, San Jose, CA, USA). Cells (2 × 10 5 ) were stained with 1:100 dilution of anti-MPM2 antibody (Millipore, Leicester, UK), followed by incubation with 1:500 dilution of secondary fluorescein isothiocyanate (FITC)-conjugated antibody (Sigma-Aldrich). Cells were analysed by fluorescence-activated cell sorting (FACS) and the percentage of MPM2-positive cells was designated as mitotic index.
Protein arrays. Cells were plated and treated as for Annexin V assays. Cell lysates from three experiments (100 μg protein each) were combined, incubated with human apoptosis protein or phosphoprotein arrays (ARY009 and ARY003; R&D Systems) and processed as per the manufacturer's protocol. The films were scanned with transmission-mode scanner and pixel densities analyzed with ImageQuant (GE Healthcare, Chicago, IL, USA).
Co-immunoprecipitation. The cells were cultured on 15-cm Petri dishes and collected when they reached 70-80% confluence (when using PEO1 stable transfectants the medium with blasticidin was replaced with media without the drug 24 h before plating for the experiment). The cells were washed with ice-cold PBS, and detached with a cell scraper into 5 ml volume of fresh PBS, centrifuged and lysed immediately or stored as a pellet in − 80°C. Cells collected from each dish were lysed in 400 μl of the following buffer for 30 min on ice (0.5-1% Triton X or 0.5-1% NP40, 10% glycerol, 150 mM NaCl, 50 mM Tris-HCl (pH = 7.2), 0.2 mM Na 3 VO 4 , 50 mM NaF, 2 mM EDTA, 1 mM PMSF, supplemented with 100 U/ml aprotinin, 10 μg/ ml leupeptin and 1 μg/ml pepstatin). The lysates were centrifuged at 13 000 r.p.m. at 4°C for 20 min and the supernatant was transferred to a new tube. The supernatant in each tube was precleared by incubation for 1 h with 10 μl of Protein A/G UltraLink Resin (Pierce, Waltham, MA, USA; no. 53132). Two micrograms of a relevant or control antibody as appropriate was added to each precleared sample and incubated at 4°C on a circular shaker for 3-6 h. Subsequently, 10 μl of the resin was added to each sample and incubated at 4°C on a circular shaker for 3 h to overnight. The beads pellets were washed four times with the lysis buffer and resuspended in 30 μl of 1xPAGE loading buffer (Fermentas, Waltham, MA, USA; no. R0611), 76 supplemented with DTT to the final concentration of 100 mM and boiled at 95°C for 5 min. The immunoprecipitates were analysed by immunoblotting as described above.
Bioinformatics and statistical analysis. The TCGA data set, featuring over 500 primary tumour samples, was obtained from The Cancer Genome Atlas (https://tcga-data.nci.nih.gov/tcga/). The 'Tothill' data set featuring nearly 300 primary tumour samples was obtained from Gene Expression Omnibus (series accession GSE9891). Raw gene expression data (Affymetrix CEL files) were obtained for each data set and normalised using RMA. 32 Statistics derived from Cox proportional hazards models were used to evaluate the quantitative association between a unit increase of some measure of term(s) of interest and either PFS time or OS time. Dichotomisation of patient cohorts into higher-than-median expression of a gene of interest and lower-than-median expression of the gene of interest was carried out for the purposes of illustrating effects in Kaplan-Meier plots. Statistical significance of differences in survival curves were evaluated using log-rank tests. Linear models were fitted with LIMMA to derive F-statistics to assess the variation of a given probe's measured gene expression levels across samples grouped according to the categorical clinical variables 'residual disease following primary surgery' and 'molecular subtype'.